Split (3)

train · 4.35k rows

article stringlengths 1.98k 169k	summary stringlengths 1.01k 4.15k	section_headings listlengths 2 38	keywords listlengths 0 12	year stringclasses 11 values	title stringlengths 30 189
Changes in temperature affect biochemical reaction rates and , consequently , neural processing . The nervous systems of poikilothermic animals must have evolved mechanisms enabling them to retain their functionality under varying temperatures . Auditory receptor neurons of grasshoppers respond to sound in a surprisingly temperature-compensated manner: firing rates depend moderately on temperature , with average Q10 values around 1 . 5 . Analysis of conductance-based neuron models reveals that temperature compensation of spike generation can be achieved solely relying on cell-intrinsic processes and despite a strong dependence of ion conductances on temperature . Remarkably , this type of temperature compensation need not come at an additional metabolic cost of spike generation . Firing rate-based information transfer is likely to increase with temperature and we derive predictions for an optimal temperature dependence of the tympanal transduction process fostering temperature compensation . The example of auditory receptor neurons demonstrates how neurons may exploit single-cell mechanisms to cope with multiple constraints in parallel . Changes in temperature considerably modulate physico-chemical processes and , consequently , also affect neural processing ( Schmidt–Nielsen , 1997; Robertson and Money , 2012 ) . The dependence of neural activity on temperature poses a particular challenge for animals without central heat regulation , like insects , who are permanently exposed to temperature fluctuations . These animals must have evolved intrinsic mechanisms at the behavioral , systems , or cellular level that help to circumvent temperature-induced behavioral modulations . Such compensatory mechanisms , however , may also come into play for homeothermic animals under pathological conditions , like fever or hypothermia in mammals . Nevertheless , our understanding of generic design principles that enhance robustness to temperature fluctuations remains limited . The goal of this study is to identify mechanisms and limitations of cellular temperature compensation at the level of firing rates . We start from a characterization of the temperature dependence of neural responses in an insect auditory system , which we find to be surprisingly robust to temperature changes . The absence of network inputs to these receptor neurons ( Vogel and Ronacher , 2007; Clemens et al . , 2011 ) suggests that a cellular mechanism underlies the observed temperature compensation and hence raises the more general question to what extent temperature compensation can be achieved at the level of individual cells . Based on generic conductance-based models of spike generation , we then show that the experimentally observed degree of temperature compensation can be explained by physiological properties intrinsic to single cells despite a substantial dependence of ion channels on temperature . Temperature dependence is usually quantified by the so-called Q10 value , which characterizes the relative change of a variable when temperature rises by 10°C . Several invertebrate species were found to have firing-rate Q10 values above 2 ( i . e . , to double their neurons' firing rate ) , which is in line with the fact that many underlying biochemical processes also exhibit Q10 values of two or more ( French and Kuster , 1982; Pfau et al . , 1989; Warzecha et al . , 1999; Hille , 2001; Spavieri et al . , 2010 ) . In contrast , we found that grasshopper auditory receptor neurons on average increased their firing rate by only ∼40–50% ( corresponding to a Q10 value of 1 . 4–1 . 5 ) . Receptor responses are shaped by a cascade of two major steps ( Gollisch and Herz , 2005 ) — ( 1 ) auditory transduction , which translates the vibrations of the tympanal membrane into receptor currents and ( 2 ) spike generation . Temperature compensation of the response must be achieved by compensatory mechanisms in these individual components or their combined output . Based on a computational analysis , we first investigate how the second component , that is cellular spike generation in terms of the translation from input current to firing rate , can be temperature compensated in generic model neurons and identify conductances whose temperature dependence favors robustness . As energy efficiency of signaling is an important constraint ( Attwell and Laughlin , 2001; Niven and Laughlin , 2008 ) , we also resolve whether the identified mechanisms for temperature compensation come at an additional metabolic cost and identify the key parameters of temperature dependence that increase energy efficiency of action-potential generation as well as of the maintenance of the resting potential . Moreover , we show that information transfer ( via spike rates ) is fostered by temperature increments . Second , we combine spike generation with a phenomenological model of mechanotransduction and predict properties of the temperature dependence of this nonlinear transformation that would allow for an efficient compensation in firing rates to the degree observed in our experimental data . As our model-based approach generalizes beyond the grasshopper system , our findings can be expected to reflect principles that could be implemented in many invertebrate and vertebrate species . Based on recordings of auditory receptor neurons in the metathoracic ganglion of the grasshopper Locusta migratoria , we quantified the dependence of the firing rate on temperature . Figure 1A shows voltage responses to stimulation at three different sound intensities and two different temperatures , as well as spike shapes . Interestingly , spike rates at a given sound intensity did not differ much between the low and high temperatures and mildly increased from low to high temperature , while spike width decreased . In general , firing rates of grasshopper receptor neurons are relatively high , saturating only at several hundreds of Hz . At a given temperature , the transfer function , that is the firing rate as a function of sound intensity , has a sigmoidal shape ( Figure 1B ) . Three parameters ( saturation rate , half-max sound level , and dynamic-range width ) are sufficient to capture the experimental transfer functions ( R2 >0 . 95 for all response curves ) . Comparing transfer functions at the two different temperatures revealed that their temperature dependence was surprisingly low ( Figure 1C ) : all corresponding median Q10 values were below 1 . 5; the sound intensity at half-maximal response ( half-max sound level ) remained almost unchanged , as did the median of the slope at half-max sound level . The width of action potentials , in contrast , was lower at the higher temperature . 10 . 7554/eLife . 02078 . 003Figure 1 . Cooling mildly affected electrophysiologically recorded firing rates generated by auditory receptor neurons in response to sound . ( A ) Voltage traces at 29 and 21°C for one neuron ( red and blue lines , respectively ) . Black horizontal lines mark time intervals of stimulus presentation; stimulus intensity as indicated . ( B ) Firing-rate as a function of sound intensity was well described by sigmoidal functions at both temperatures ( same neuron as in A ) . The three sigmoidal parameters ( saturation rate , sound level at half-maximum , and dynamic-range width ) were extracted from fits to the experimental data . ( C ) Statistics of the measured temperature dependence , Q10 ( x ) = ( x ( T+ΔT ) x ( T ) ) 10/ΔT , were computed for several quantities x . For a population of nine receptor neurons all three parameters of the sigmoidal function , as well as the spike rate and slope at the cold half-maximum were temperature compensated ( median Q10∈[1 , 1 . 5] , see also Figure 1—figure supplement 2 and Figure 1—figure supplement 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02078 . 00310 . 7554/eLife . 02078 . 004Figure 1—figure supplement 1 . Temperature calibration curve . Time course of temperature changes during the cooling-down procedure at the Peltier element ( blue ) and the tissue close to the tympanal membrane , where auditory receptor neurons attach ( red ) ; recordings from four animals . These curves were used as reference to estimate the temperature change during electrophysiological recordings . DOI: http://dx . doi . org/10 . 7554/eLife . 02078 . 00410 . 7554/eLife . 02078 . 005Figure 1—figure supplement 2 . Sound-intensity resolved p-values of statistical differences between firing rates at the two temperatures . Stimuli at each sound intensity were presented five times ( before and after the temperature change ) . It was tested whether the corresponding firing rates at the two temperatures belong to an identical distribution with equal medians ( ranksum test; p-values < 0 . 05 indicate a significant effect of temperature on firing rate ) . The figure illustrates the distribution of the test's p-values across all neurons at different levels of sound intensity ( the symbol + indicates an outlier ) . The effect of temperature on firing rate was most significant for high sound intensities . Presumably , a higher relative variability in firing rate obscured the effect of temperature at lower firing rates ( i . e . , at lower sound intensities ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02078 . 00510 . 7554/eLife . 02078 . 006Figure 1—figure supplement 3 . Statistical analysis of Q10 values . Statistical significance of observables shown in Figure 1C . Based on a Wilcoxon signed rank test the null hypotheses that the median of a distribution was 1 , 1 . 5 , or 2 were tested . For values of p<0 . 05 Q10 medians were statistically different ( from 1 , 1 . 5 , or 2 . 0 , respectively ) . Spike rates both at half-max and at saturation were significantly affected by temperature changes ( Q10>1 , p=0 . 001 ) , but not significantly different from 1 . 5 ( p≥0 . 4 ) , indicating temperature compensation . Q10 values of saturation spike rates , in particular , were also significantly lower than 2 . 0 . The median temperature dependencies of half-max sound level and slope at half-max were not significantly different from 1 , indicating temperature invariance . Dynamic-range width and action-potential width significantly increased with cooling ( medianQ10<1 , p<0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02078 . 006 Compared to the temperature dependencies previously observed in other species like moth auditory receptor neurons , locust stretch receptors , and fly H1 neurons and photoreceptors ( Q10 ( spike rate ) ∼2 , Pfau et al . , 1989; Coro and Perez , 1990; Warzecha et al . , 1999; Tatler et al . , 2000 ) , the grasshopper responses were temperature compensated; the dependence was similar to what has been referred to as warm-insensitive in hypothalamic neurons of the rat ( Q10 ( spike rate ) ∼1 . 3 , Curras and Boulant , 1989 ) . To understand this low dependence of receptor neuron firing rate on temperature , we next turned to mathematical modeling . The grasshopper auditory periphery consists of a relatively simple feed-forward network , in which the receptor neurons constitute the first layer . Receptor neurons are known to respond to vibrations of the tympanal membrane , but they do not receive input from the neuronal network . Mechanisms of temperature compensation must hence be cell-intrinsic . To resolve which intrinsic processes can be sufficient to account for the observed degree of temperature compensation , we first focussed on spike generation , leaving mechanotransduction aside . We analyzed in generic model neurons , how the temperature dependence of ionic conductances mediating spike generation can reduce the dependence of firing rate on temperature . We used the Connor–Stevens model ( Connor et al . , 1977; Dayan and Abbott , 2005 ) to simulate a type I spike generation process ( Izhikevich , 2007 ) as it is assumed for grasshopper receptor neurons ( Benda , 2002 ) . Besides a sodium and a leak conductance ( gNa and gL ) , this model comprises a delayed-rectifier and an A-type potassium conductance ( gK and gA ) , which are both known to be present in the grasshopper nervous system ( Ramirez et al . , 1999 ) . Temperature dependence was assumed to affect the opening and closing rates of all gates of the three ion channel types ( i . e . , the m and h gates for gNa , n gates for gK , a and b gates for gA ) , as well as their peak conductances ( g¯Na , g¯K , g¯A ) and that of the leak conductance , g¯L . For a systematic analysis ( refers to ( Prinz et al . , 2003 ) ) , we independently varied the temperature dependence of these parameters ( comprising a total of nine ) within physiologically realistic ranges: Q10 ( x ) ∈[2 . 0 , 4 . 0] for transition rates and Q10 ( g¯X ) ∈[1 . 2 , 2 . 0] for peak conductances ( Partridge and Connor , 1978; Hille , 2001 ) . For each combination of parameters , the transfer function ( input current to firing rate; i . e . , the f-I curve ) was computed at two temperatures: 18 and 28°C ( Figure 2A , B ) . 10 . 7554/eLife . 02078 . 007Figure 2 . Temperature dependence of spike generation in a conductance-based neuron model . ( A ) Voltage responses to step current stimuli of different amplitudes; blue: 18°C ( the reference temperature ) , red: 28°C . Top trace corresponds to a model with strongly temperature-dependent firing rate , middle trace to a temperature-compensated model . ( B ) f-I curves at both temperatures , corresponding to the examples shown in A . ( C ) Results of the sensitivity analysis for the RMSD . The largest impact is exerted by temperature dependencies of the potassium conductances ( Q10 ( n ) , Q10 ( g¯A ) , and Q10 ( g¯K ) ) . Signs +/− indicate the qualitative impact ( see main text for details ) . ( D ) Distribution of the RMSD , across all models . Note that a Q10 of 1 . 5 corresponds to an RMSD of ∼0 . 5 ( 50% relative change ) . ( E ) Parameter impacts on RMSD were robust against ±20% perturbation of the model's peak conductances at 18°C ( black symbols: perturbations of individual peak conductances; grey symbols: all combinations of ±20% changes to the four peak conductances ) . ( F ) Results of the sensitivity analysis for the temperature dependence of the slope of f-I curves . ( G ) Distribution of Q10 values of the slope across all models . ( H and I ) as panels ( F ) and ( G ) , but for the threshold of the f-I curves . ( J and K ) Sensitivity analysis of information transfer . For two basic noise models ( Poissonian and input-independent Gaussian ) , the temperature dependence of firing rate-based information transfer J is related to that of the slope of the f-I curve: Q10 ( 〈J〉 ) =[Q10 ( slope ) ]4 . The conductance parameters with highest impact were very similar to those of the changes in slope ( compare to panel F ) . Information transfer increased with temperature for all models . ( L ) Visualization of the RMSD for the parameter space spanned by the temperature dependencies of the Connor-Stevens model based on dimensional stacking . Axes order was chosen according to the impact ranking as presented in ( C ) ; color code as in ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02078 . 00710 . 7554/eLife . 02078 . 008Figure 2—figure supplement 1 . Temperature compensation in the Traub-Miles model . Temperature compensation of spike generation is also possible in a structurally different Traub-Miles model despite a realistic temperature dependence of its conductances ( gL , gK , gNa ) . Shown are the model at reference temperature ( 32°C for this model , black curve ) , and the model heated up by 10°C ( red curve ) or cooled down by 10°C ( blue curve ) , with temperature parameters that minimize the RMSD of the corresponding f-I curves within the physiologically realistic range ( identical to that explored for the Connor-Stevens model with peak conductances ( Q10 ( g¯ ) ∈[1 . 2 , 2] ) and transition rates of the ( in- ) activation variables Q10 ( {α , β} ) ∈[2 , 4] ) . Optimal parameters minimizing the RMSD were identified by a genetic algorithm . Relative changes in firing rate were on the order of those observed experimentally ( RMSD=0 . 34 for the red curve and RMSD=0 . 55 for the blue curve ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02078 . 008 To estimate the temperature dependence of a whole f-I curve , Q10 values are not ideal , as they are defined as the ratio of firing rates at two different temperatures , which will be infinitely large for inputs that only elicit spikes at the higher , but not the lower temperature . To circumvent this bias , we assessed the temperature dependence of a model neuron as the root-mean-squared difference between the firing rates at the two temperatures ( mean taken across input currents ) , normalized by the mean rate elicited at the colder temperature . We refer to this quantity as RMSD . It reflects the average relative change in firing rate with temperature . Note that a Q10 value of 1 . 5 is hence comparable to an RMSD of 0 . 5 ( 50% relative change ) . Across all model combinations , the relative change , RMSD , was distributed between 0 . 22 and 2 . 14 , with a median of 0 . 68 ( Figure 2D ) . This means that , intuitively , the median of the average change in firing rate of an f-I curve was 68% . The analysis showed that the effect of temperature on spike generation depended strongly on the specific temperature dependence of the ionic conductances . A fraction of models ( 18% ) exhibited temperature compensation with relative changes in firing rate comparable to those found experimentally ( RMSD <0 . 5 ) . This result shows that a low dependence of firing rate on temperature is feasible despite a substantial ( and hence realistic ) dependence of the individual conductances on temperature . Next , we asked which of the nine parameters ( i . e . , the temperature dependence of peak conductances and transition rates ) most affected the dependence of firing rate on temperature . To this end , we performed a systematic sensitivity analysis in the nine-dimensional space of all possible parameter combinations . We created—for each parameter—a distribution of local changes in RMSD induced by changes in that parameter . Specifically , this distribution captured changes in RMSD between all neighboring points along a specific parameter's dimension . Each distribution sampled the whole parameter space ( i . e . , all possible combinations of the other parameters ) , also see ‘Materials and methods’ . The impact of a given parameter on the temperature dependence of the f-I curve was then defined as the median of its specific distribution , directly relating the impact of a parameter to its quantitative effect on RMSD . Figure 2C depicts the impact values of all Q10 parameters on the RMSD . The sum of all absolute values of impacts is normalized to unity . The impact sign , that is whether an increase in a parameter on average led to an increase or decrease in the observable , is indicated by + and − , respectively . The analysis revealed that the largest impact on the RMSD was exerted by three parameters of potassium channels: temperature dependence of the delayed-rectifier potassium kinetics , Q10 ( n ) , and the A-type and delayed-rectifier potassium peak conductances , Q10 ( g¯A ) and Q10 ( g¯K ) , respectively . The impacts of both potassium channel peak conductances were negative ( i . e . , increases in their Q10 values decreased the RMSD ) , while the potassium activation Q10 ( n ) had a positive impact ( i . e . , increases in its temperature dependence increased the RMSD ) . To confirm that the results do not strongly depend on the specific choice of peak conductances in the Connor–Stevens model , we tested 24 alternative models with changes of ± 20% in the peak conductances of sodium , both potassium , and leak channels . The impact ranking across those models was highly similar to the ranking in the original Connor–Stevens model ( Figure 2E ) and we conclude that our results are robust . Moreover , we note that our results are not unique to the Connor–Stevens model . An analysis of a structurally different Traub–Miles model ( Traub et al . , 1991; Benda , 2002 ) showed that an equally low temperature dependence is possible ( Figure 2—figure supplement 1 ) . On a side note , a visualization of the RMSD across the complete nine-dimensional parameter space based on dimensional stacking is shown in Figure 2L , see ‘Materials and methods’ for details ( LeBlanc et al . , 1990; Taylor et al . , 2006 ) . Dimensional stacking maps the nine-dimensional space onto a two-dimensional representation with nine axes . Ordering of the axes is arbitrary and hence requires optimization to maximize visual information ( Taylor et al . , 2006 ) . Here , we introduce a new way to determine optimal axes order , defined directly by the ranking of impact scores . Parameters with highest impact on the RMSD are depicted on large-scale axes and parameters with low impact on small-scale axes . The success of the ordering is reflected in the visually structured areas of equal RMSD . As only a subset of all parameters had relevant influence on the RMSD , optimal axes ordering led to a clear visual structure . In contrast , for arbitrary axes ordering visual structure would be hard to recognize . As we saw , the temperature dependence of potassium channels plays a crucial role for temperature compensation . For a more detailed and intuitive understanding of the underlying mechanism , we next analyzed the transformation of the shape of f-I curves with temperature . Type I f-I curves , as they are found in the grasshopper , can be described by a square root function ( Izhikevich , 2007 ) . Temperature affects an f-I curve by shifting the curve horizontally ( i . e . , changing its threshold ) and by changing its slope ( which can also be termed gain ) . We used fits of the f-I curves by a square root model f ( I ) =A·I−I0 , based on the parameters A ( slope ) and I0 ( threshold ) . With heating , the slope always increased ( Figure 2G ) , while we found changes in both directions for the threshold ( Figure 2I ) . Temperature dependencies of the A-type potassium and leak peak conductances had the strongest influence on the threshold ( Figure 2H ) . In contrast , the slope was most sensitive to the temperature dependence of the delayed-rectifier potassium channel , Q10 ( n ) and Q10 ( g¯K ) , and the sodium channel inactivation , Q10 ( h ) ( Figure 2F ) . Beyond clarifying the specific effect of the aforementioned parameters on changes to the f-I curve , the analysis shows that temperature compensation ( i . e . , lower RMSD values ) was usually achieved by modest increases in threshold balancing the effects of an increase in slope ( Figure 2B ) . Changes in slope also have direct implications for the ability to infer information about the sound intensity from the firing-rate output of receptor neurons . We hence quantified how the capacity to transmit information from input I to firing rate f changes with temperature . To this end we use Fisher information . Considering the average information transferred for a given interval of firing rates [fmin , fmax] , information transfer scales with the slope of the f-I curve and its temperature dependence hence with Q10 ( A ) 4 ( ‘Materials and methods’ ) . Consequently , the same parameters that had the largest impact on the slope–potassium channel rate ( Q10 ( n ) ) and peak conductances ( Q10 ( g¯K ) , Q10 ( g¯A ) ) and sodium channel inactivation ( Q10 ( h ) ) —also influenced information transfer most ( Figure 2J ) . Overall , heating was advantageous for information transfer ( Figure 2K ) . Metabolic cost is increasingly recognized as an important constraint for neural function ( Attwell and Laughlin , 2001; Niven and Laughlin , 2008 ) and is likely to have shaped the design of neural systems—the more so if firing rates are large . In the grasshopper auditory periphery firing rates often exceed several hundreds of Hz , suggesting that metabolic cost may have played a role in the design of these cells . It is hence interesting to explore whether robustness to temperature changes compromises energy efficiency . To this end , we computed the energetic cost of spike generation and maintenance of the resting potential ( Figure 3A ) . Cost was quantified in terms of the total sodium current ( per action potential or per time , respectively ) . To assess the changes of energy consumption with temperature , energy use was characterized by its Q10 value ( i . e . , the ratio of energetic cost at 28° and 18° ) and averaged across input currents . 10 . 7554/eLife . 02078 . 009Figure 3 . Temperature dependence of the metabolic cost ( spiking and maintenance of the resting potential ) . ( A ) Illustration of the periods during which spiking cost and resting-potential cost were estimated in terms of the Na+ current ( two corresponding examples at the higher and lower temperature; red and blue curves , respectively ) . Energy consumption during spiking was averaged per spike and across the suprathreshold parts of the f-I curve . ( B ) Distribution of the temperature dependence of the spiking cost ( Q10 ( spiking cost ) ) . Top: all models; middle and bottom: distribution across the 25% of models with lowest and highest temperature dependence of firing rate ( RMSD ) , respectively . Spiking cost decreased at higher temperature for the majority of models . In particular , values of minimal energy consumption at the higher temperature were similar for the subgroups of models with lowest and highest temperature dependence . ( C ) Distribution of the temperature dependence of the resting cost ( Q10 ( resting cost ) ) , analog to ( B ) . Resting-potential cost decreased for the majority of models; top , middle and bottom panels comprising subgroups of models as in ( B ) . ( D ) Temperature dependence of firing rate and spiking energy consumption are determined by different sets of conductance parameters . While potassium-channel temperature dependencies have the largest impact on firing rate ( Figure 2C ) , the energy consumption per spike was predominantly determined by sodium-channel temperature dependence . Faster sodium inactivation ( Q10 ( h ) ) and lower peak sodium conductance ( Q10 ( g¯Na ) ) fostered energy efficiency at higher temperature . ( E ) Parameters that reduce resting energy at high temperature also reduce RMSD ( same sign of the impact values as in Figure 2C ) . ( F and G ) Impacts on energy consumption were robust against ±20% perturbations of the model's peak conductances at 18°C ( Figure 2E , ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02078 . 00910 . 7554/eLife . 02078 . 010Figure 3—figure supplement 1 . Alternative measures of metabolic cost and energy efficiency of spiking . ( A ) Schematic illustration of the measures used ( sodium-current-based and potassium-current-based cost , as well as energy efficiency estimated by the separability of sodium and potassium currents either in relation to the sodium current or the potassium current ) . ( B ) Impact of the model parameters' Q10 values for all four measures . In all cases , temperature dependence of sodium inactivation , Q10 ( h ) had the largest impact on metabolic cost or energy efficiency . In contrast , the temperature dependence of delayed-rectifier potassium activation , Q10 ( n ) ( which proved most influential for the temperature dependence of firing rate ) exerted a smaller impact on both potassium- and sodium-based energy consumption . ( C ) The distribution of energy measures across all spike generation models as well as models with particularly high and low RMSD ( subgroup of models within the bottom and top 25 percentile of RMSDs ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02078 . 01010 . 7554/eLife . 02078 . 011Figure 3—figure supplement 2 . Potassium-current based resting cost . ( A ) The distribution of the temperature dependence ( Q10 values ) of the resting cost based on potassium-current across all spike generation models as well as models with particularly high and low RMSD ( subgroup of models within the bottom and top 25 percentile of RMSDs ) . Overall , this type of resting cost increased with larger temperatures because A-type potassium currents were larger . ( B ) The temperature dependence of the leak conductance influenced this type of resting cost most . DOI: http://dx . doi . org/10 . 7554/eLife . 02078 . 011 For the majority of models , the energetic cost of an action potential decreased with heating ( 93% of models , Figure 3B ) . On average , temperature-compensated spike generation models ( 25% of models with lowest RMSD ) were slightly more costly than the most temperature-dependent models ( 25% of models with highest RMSD ) , see Figure 3B . Nevertheless , the minimum energy consumption was comparable in both groups . Resting cost was substantially lower than spiking cost , this trend increasing with larger firing rates . Sodium-current based resting cost tended to decrease with heating ( 77% of models ) . It was slightly lower for temperature-compensated models compared to strongly temperature-dependent models ( Figure 3C ) . The sensitivity analysis ( performed in analogy to the analysis of temperature dependence of firing rate ) revealed that the temperature-dependence of the sodium conductance ( Q10 ( h ) and Q10 ( g¯Na ) ) influenced energy consumption of spike generation the most ( Figure 3D ) . The impacts of conductance parameters on f-I curve temperature dependence and energy consumption were not significantly correlated in this case ( ρ = −0 . 23 , p=0 . 56 ) . In particular , the key parameters of largest influence on these features belonged to different channel groups: potassium channels in case of temperature compensation and sodium channels in case of energy efficiency of spike generation . We verified that the large influence of sodium channels was not biased by our sodium-current-based definition of metabolic cost . Three alternative measures—two quantifying energy efficiency based on the separability of sodium and potassium currents ( Alle et al . , 2009 ) , the other defined by the total potassium current—all confirmed the temperature dependence of sodium channel inactivation Q10 ( h ) as the most influential parameter for spiking cost ( Figure 3—figure supplement 1 ) . The sodium-current based resting cost was qualitatively influenced in a similar way as temperature dependence of firing rate ( Figure 3E ) : for all four relevant Q10 parameters a reduction of resting cost co-occurred with a reduction in temperature dependence of firing rate ( same sign of corresponding impacts , Figure 2C ) . For a potassium-current based resting cost the temperature dependence of leak channels had the dominant impact ( Figure 3—figure supplement 2 ) . In contrast to the sodium-current based resting cost , the potassium-current-based cost was larger at higher temperatures . For the majority of models , inactivation of A-type potassium channels was lower at the higher temperature ( due to a more negative resting potential ) and hence increased the A-type current . In summary , striving for temperature compensation does not have to compromise a neuron's energy efficiency . Both results ( for spiking and resting cost ) generalize beyond the specific choice of peak conductances in the Connor–Stevens model ( Figure 3F , G ) . In the previous paragraphs , we have shown that spike generation by itself can achieve a remarkably high invariance to temperature changes . The receptor neurons , however , have an additional processing stage involved in transferring the external input signal to a firing-rate response: the so-called transduction mediated by the mechanosensitive channels in the vicinity of the tympanal membrane . Transduction precedes spike generation and translates vibrations of the tympanal membrane caused by the sound pressure wave into receptor currents through these channels , which in turn drive spike generation ( Gollisch and Herz , 2005 ) . This mechanism may also contribute to temperature compensation . We therefore explored how temperature compensation can benefit from linking spike generation with the transduction process . Little is known about the temperature dependence of transduction in the grasshopper . Here , we combined the computational analysis of spike generation with the experimental findings for the sound-intensity to firing-rate relation in order to predict on a phenomenological level which features of the auditory transduction and its temperature dependence would improve temperature compensation . The experimentally measured receptor neuron responses to sound stimuli r=ρ ( IdB ) can be expressed as a cascade of mechanosensory transduction ( IC=θ ( IdB ) , with current IC and sound intensity IdB ) and spike generation ϕ ( IC ) :r=ϕ ( IC ) =ϕ ( θ ( IdB ) ) . Let us think of the Connor–Stevens spike generation model at the colder temperature ( illustrated in Figure 4A , blue curve ) . When combined with an ( upstream ) nonlinear translation of sound intensity to current ( Figure 4B , blue curve ) , it yields the full sound-intensity to firing-rate relation ( Figure 4C , blue curve ) , which corresponds to the quantity measured experimentally . Hence , if the receptor neuron response as a function of sound intensity is known from experimental data and we assume a specific spike generation model ( i . e . , a specific f-I curve ) , we can mathematically infer the transduction nonlinearity that gives the best match to the experimentally recorded sound-intensity to firing-rate relation by nonlinear regression ( Figure 4C , blue curves ) . We use the term reverse-engineering for this approach . It can be used at the higher temperature as well and enables us to derive for each of the more than 260 . 000 spike generation models the corresponding ideal transduction curve that best matches the experimentally recorded sound-intensity to firing-rate relation at this higher temperature ( Figure 4B , C , red curves ) . Comparing the reverse-engineered transduction curve at the colder temperature to the reverse-engineered curve at the higher temperature ( for each spike generation model ) , allows us to identify trends in the temperature dependence of mechanotransduction , which would foster a temperature robustness of the firing rate . 10 . 7554/eLife . 02078 . 012Figure 4 . Reverse engineering mechanosensory transduction functions that favor temperature compensation of firing rate . ( A ) Example of a model-based f-I curve , denoted r=ϕ ( IC ) , at two temperatures . ( B ) Example of a sigmoidal transduction function converting sound intensity to current , IC=θ ( IdB ) . ( C ) Representative receptor neuron responses , r=ρ ( IdB ) , at two temperatures ( dotted lines ) , as well as a receptor neuron response , r=ϕ ( θ ( IdB ) ) , ‘constructed’ from the f-I curve in A and the transduction function in B . For each model of spike generation , the optimal transduction function , IC=θ ( IdB ) , minimizing the error between the corresponding ‘constructed’ receptor neuron response and the representative receptor neuron response ( dotted line in C ) was derived . ( D ) The statistics of temperature dependence of the optimal transduction functions , IC , across all models . The width of the dynamic range depended most on temperature and increased with heating for nearly all optimized transduction functions . But the temperature dependence of the saturation current and the half-maximum sound intensity ( mainly when decreasing with temperature ) were also found to contribute to temperature compensation . Note that the ranges marked by the whiskers cover all data ( including outliers ) in this plot . DOI: http://dx . doi . org/10 . 7554/eLife . 02078 . 012 Specifically , we exploited the fact that experimental response curves ( firing rate to sound intensity ) were well fitted by a sigmoidal function ( Figure 1C ) and also assumed a sigmoidal shape for the transduction curves ( Hudspeth et al . , 2000; Fisch et al . , 2012 ) . Accordingly , we reverse-engineered the ideal transduction sigmoid at the warmer temperature for each spike generation model ( e . g . , those contributing to Figure 2C ) , so that the resulting sound-intensity to firing-rate relation best matched a representative receptor neuron response ( Figure 4C , for details on representative receptor neuron response , ‘Materials and methods’ ) . The model response curves r=ϕ ( θ ( IdB ) ) matched the experimental representative response curve very well ( R2 >0 . 98 for 99 . 5% of all models ) . Temperature dependence of the reverse-engineered transduction was then quantified based on the Q10 values ( i . e . , the relative changes with temperature ) of the three parameters that define each transduction sigmoid: saturation current , half-maximum sound intensity , and dynamic-range width ( Figure 4D ) . Evaluating the distribution of changes of the ideal transduction curves with temperature across all spike generation models , we found that the largest temperature dependence of these ‘matching’ transduction curves was to be expected for their dynamic-range width ( median Q10∼2 ) . In addition , changes in saturation current and half-maximum sound level also contributed to fostering temperature compensation of firing rate ( Figure 4D ) . These results show in particular that a suitable temperature dependence of the transduction process can support temperature compensation , even in cases where spike generation is less temperature robust . Neuronal processing is significantly challenged by variation in temperature due to the changes in chemical and physical processes . In many neurons across invertebrates and vertebrates , firing rate has been observed to at least double with increases of temperature , corresponding to Q10 values of two or above ( French and Kuster , 1982; Coro and Perez , 1990; Warzecha et al . , 1999 ) . It was hence surprising to observe that neurons in the auditory periphery of grasshoppers show Q10 values on the order of 1 . 5 and consequently are remarkably temperature compensated ( Hazel and Prosser , 1974; Boulant and Dean , 1986 ) . The temperature robustness of these neurons is hence comparable to that of ‘warm-insensitive’ neurons in the mammalian brain ( Curras and Boulant , 1989 ) . Temperature compensation has been studied in the context of neurons embedded in a network in a variety of systems ( see , e . g . , Wechselberger et al . , 2006; Tang et al . , 2010; Robertson and Money , 2012 ) . Temperature compensation in grasshopper receptor neurons , by comparison , must be based on cell-intrinsic processes . A similar temperature compensation that must be based on a single-cell mechanism has so far—to our knowledge—only been described experimentally for tarsal hairs in the locust ( Miles , 1985 ) . The cell-intrinsic mechanisms identified in our computational study attribute the observed robustness to a balancing of opposing processes . Phenomenologically , a rise in slope ( consistent across the whole parameter range explored ) is compensated for by an increase in the threshold of the f-I curve , minimizing the effect of temperature across a broader range of inputs . While alterations in threshold can be produced by changes in peak conductances of ion channels , they have also been described experimentally by heating in invertebrate systems ( Burkhardt , 1959; Abrams and Pearson , 1982; Kispersky et al . , 2012 ) in agreement with our observations . Biophysically , a heating-induced increase in the speed of repolarizing gating kinetics is opposed by an increase in peak potassium conductances which promote a more negative resting potential ( as can be derived from Equation 2 ) . Although the balancing is not perfect , average deviations on the order of not more than 50% can be easily achieved with strongly temperature-dependent conductances ( in particular , Q10∈[2;4] for all activation- and inactivation rates ) for ∼18% of the models . The temperature dependence of both delayed-rectifier and A-type potassium channels has a particularly large impact on temperature compensation . This matches experimental observations in neurons of the pancreas of mice ( Xu et al . , 2006 ) and molluscan neurons as well as previous simulations of an extended Hodgkin–Huxley model ( Rush and Rinzel , 1995 ) and is consistent with the effect of peak conductances on firing rate , for example ( Schreiber et al . , 2004 ) . Our results also hold for reference models quantitatively different from the original Connor-Stevens model—both for the total fraction of temperature-compensated models RMSD <0 . 5 ) and the strong influence of potassium channel dynamics on the temperature dependence of firing rate ( Figure 2E ) , ‘Materials and methods’ for details . These findings show that our results generalize beyond the specific quantitative choice of peak conductance parameters of the Connor–Stevens model . This is further supported by the fact that a structurally different Traub–Miles model could also exhibit a temperature dependence of firing rate as low as that described for the Connor–Stevens model . For auditory receptor neurons in the grasshopper energy efficiency of spike generation is likely to be a relevant factor , also see Niven and Farris ( 2012 ) . Firing rates in these cells approach 400 Hz and likely entail a high total cost of electrical signaling . Our results , however , show that temperature compensation need not impair energy efficiency of spike generation nor of maintenance of the resting potential . The rate of sodium channel inactivation ( Figure 3D ) proved to be most relevant in setting the energy consumption per spike generated , which is consistent with simulations and dynamic clamp experiments in various model systems ( Alle et al . , 2009; Hasenstaub et al . , 2010; Sengupta et al . , 2010 ) . We demonstrated that energy efficiency improved with heating for a wide range of temperature dependencies of ion channels , as was previously described for a model with fixed Q10 values ( Yu et al . , 2012 ) . A fast sodium inactivation limits the duration of the spike; this was consistent with the experimental data , as spike width decreased with heating . Most importantly , the key parameters regulating the energy efficiency of spiking were different from those regulating temperature compensation of firing rate ( Figure 2C , Figure 3D ) . The results could be confirmed for alternative measures of energy ( separability of sodium and potassium currents as well as the total potassium current ) . Apart from confirming the role of sodium channel inactivation , these analyses substantiated that the delayed-rectifier potassium channel kinetics ( which were most influential to the robustness of firing rate ) did not substantially contribute to metabolic costs based on potassium currents . Overall , our analysis focusses on a major source of metabolic cost: the flow of Na+ and K+ ions which on larger time scales can be compensated by the Na-K-ATPase . For completeness it should be noted , however , that in living cells metabolic costs can also arise from the flow of other ions not included in our study , like Ca2+ . Summarizing the considerations on metabolic cost , we find most noteworthy that from an evolutionary perspective , the relevant features—robustness of firing rate to temperature changes and reduction of metabolic cost—could both be achieved in parallel . Nevertheless , temperature compensation and energy efficiency would be of little use if the fundamental function of information transmission was impaired . Our sensitivity analysis revealed that higher temperatures are also advantageous for the transmission of information about sound intensity . This conclusion is based on our finding that the slope of the f-I curve increased with heating . The capacity to transmit information was affected most by the temperature dependence of the delayed-rectifier potassium conductance . These data , however , need to be interpreted with care . We cannot exclude that channel-type specific stochastic dynamics further influence information transfer in ways not captured by our approach . Implementing the specific stochastic dynamics for the whole parameter space of more than 260 . 000 models , however , goes beyond the scope of this study and merits future investigation . Auditory receptor neurons in the grasshopper constitute the bottom layer of a feedforward network: approximately 80 receptor neurons converge to ∼15 local neurons , which in turn project to ∼20 ascending neurons ( Vogel and Ronacher , 2007 ) . All auditory input passes through this peripheral network , which preprocesses information and extracts behaviorally-relevant features ( Clemens et al . , 2011 ) . The large investment into high firing rates and a comparatively high redundancy between neurons in this layer ( Machens et al . , 2001 ) also increases the need for energy-efficient spike generation . Optimization of receptor neurons in terms of temperature compensation hence seems a reasonable ‘strategy’ , as all effects of temperature on receptor neurons will be passed on to downstream neurons , where they may multiply . Although we currently do not know to which extent other parts of the auditory system are compensated , it is likely to ‘pay off’ to constrain the effects of temperature in the initial stages . Downstream neurons , in contrast , may be expected to adopt different strategies , as they can make use of different mechanisms: balancing of inhibition and excitation for robustness to temperature changes ( Robertson and Money , 2012 ) as well as an increase in population and temporal sparseness for energy efficiency and information transfer ( Clemens et al . , 2012 ) . While the considerations above refer to spike generation , little is known about the temperature dependence of the preceding transduction process . One hypothesis is that a change in the half-maximum sound intensity of the transduction process could foster temperature compensation in firing rate . Our computational analysis shows that , indeed , a slight shift of transduction to lower sound intensities with higher temperatures may be favorable . Such a shift would occur if the amplitude of the tympanal vibration increased with heating and a stimulus of given intensity hence opened more transduction channels . However , the computationally-derived changes are relatively moderate ( Figure 4D ) . This is consistent with the observation that the tympanal vibration in cicadas is relatively temperature independent ( Fonseca and Correia , 2007 ) . The other two parameters characterizing transduction ( saturation current and dynamic-range width ) reflect properties of the transduction channels ( i . e . , their peak conductance and activation range , respectively ) . The increase in dynamic-range width with heating can be interpreted as a decrease in gating force in a gating-spring model for the transduction as proposed for mechanosensory transduction in bullfrog saccular hair cells ( Howard and Hudspeth , 1988 ) . Again , increases in temperature are advantageous , because the gating-force magnitude is inversely related to transduction accuracy ( van Netten et al . , 2003 ) . Depending on the spike-generation process , increases or decreases in the saturation current foster temperature compensation . The former may directly arise from the temperature dependence of the transduction channels' maximal conductance . The latter may require additional heat-sensitive channels with a modulatory influence on the transduction process , such as thermosensitive transient receptor channels ( TRPA ) ( Kang et al . , 2012 ) , which in principle could down-regulate the saturation level of the transduction function via their increased calcium response ( Chadha and Cook , 2012 ) . Generally , the analysis shows that the transduction process can contribute to temperature compensation . While spike generation alone is sufficient to mediate robustness of the firing rate , a matched temperature dependence of the transduction process may allow for more flexibility in the ‘choice’ of spike generation parameters , including the possibility to meet additional constraints . Altogether , our data show that auditory receptor neurons in the grasshopper represent an example of remarkable cell-intrinsic temperature compensation in the absence of network effects . Our computational analysis clarifies that spike generation alone can achieve this high degree of invariance of firing rate to temperature changes . The identified mechanisms generalize to spike generation in other cell types . Moreover , additional nonlinear processing by static nonlinearities ( here interpreted as the transduction process involving the tympanal membrane , but on a wider scope also reflecting properties of synaptic transmission ) may foster temperature compensation , if well matched with the temperature dependence of spike generation . Overall , the dependence of neuronal processing on temperature merits further investigation , in particular as temperature fluctuations are an oftentimes underestimated variable in mammalian systems too . Experiments were performed on adult L . migratoria , obtained from a commercial supplier and held at room temperature ( 22–25°C ) . Intracellular recordings from auditory neurons within the metathoracic ganglion were conventionally conducted as described elsewhere ( Franz and Ronacher , 2002; Wohlgemuth and Ronacher , 2007 ) , using glass microelectrodes filled with a 3–5% solution of Lucifer yellow in 0 . 5 M LiCl . Neuronal responses were amplified ( BRAMP-01; npi electronic GmbH , Tamm , Germany ) and recorded by a data-acquisition board ( BNC-2090A; National Instruments , Austin , TX ) with 20 kHz sampling rate . To control for temperature , the preparation was placed directly on a Peltier element connected to a 2 V battery and a potentiometer . Temperature was monitored and recorded with a digital thermometer ( GMH 3210 , Greisinger electronic GmbH , Regenstauf , Germany ) connected to a NiCr-Ni-thermoelement ( GTF 300 , Type K , Greisinger electronic GmbH , Regenstauf , Germany ) . For each experiment , recordings were conducted first at a fixed higher tissue temperature ( in the range of 28–29°C ) , then the preparation was cooled down to a lower temperature ( in the range of 21–23° ) and recordings were repeated . To control for differences between the temperatures of the Peltier element and the tissue at the inner side of the tympanal membrane at the attachment site of receptor neurons , the dependence between those variables was measured directly and used for calibration ( Figure 1—figure supplement 1 ) . The calibration shows that at the higher Peltier temperature ( 30°C ) tissue temperature only reached 28°C ( in the steady state ) due to heat dissipation . After the cooling process the difference between Peltier and tissue temperature in the steady-state was less than 0 . 5°C . Moreover , cooling down proved to be slower in the tissue than at the Peltier element . In order not to underestimate Q10 values , we took a conservative approach: Electrophysiological recordings started 3–5 min after induction of the temperature change . Tissue temperature was derived from the calibration curve at the onset of a recording ( lasting 40 s ) . Although temperature may still have been subject to small changes during the recording , this procedure ensured that temperature changes ( i . e . , the difference between high and low temperature ) were—at most—slightly underestimated , favoring larger Q10 values . Consequently , the estimated Q10 values constitute an upper bound . In contrast , we cannot exclude that real Q10 values may even be slightly smaller , that is even more temperature compensated . After completion of the recordings , Lucifer yellow was injected into the recorded cell by applying a hyperpolarizing current . Subsequently , the thoracic ganglia were removed , fixed in 4% paraformaldehyde , dehydrated , and cleared in methylsalicylate . The stained cells were identified under a fluorescent microscope according to their characteristic morphology . Altogether , nine receptor neurons were recorded in eight preparations . To obtain spike rate vs intensity curves ( response curves ) , we used acoustic broad band stimuli ( 100 ms duration , 1–40 kHz bandwidth ) repeated five times each at 8 intensities , rising from 32 to 88 dB SPL . Acoustic stimuli were stored digitally and delivered by a custom-made program ( LabView 7 Express , National Instruments , Austin , TX ) . Following a 100 kHz D/A conversion ( BNC-2090A; National Instruments , Austin , TX ) , the stimulus was routed through a computer-controlled attenuator ( ATN-01M; npi electronic GmbH , Tamm , Germany ) and an audio amplifier ( Pioneer stereo amplifier A-207R , Pioneer Electronics Inc . , USA ) . Acoustic stimuli were broadcast unilaterally by speakers ( D2905/970000; Scan-Speak , Videbæk , Denmark ) located at ± 90° and 30 cm from the preparation . Sound intensity was calibrated with a half inch microphone ( type 4133; Brüel & Kjær , Nærum , Denmark ) and a measuring amplifier ( type 2209; Brüel & Kjær , Nærum , Denmark ) , positioned at the site of the preparation . Experimental spike times were extracted from the digitized recordings by applying a voltage threshold above background noise level . Mean spike rates were calculated for each intensity to obtain response curves ( spike rate r vs sound intensity IdB ) per neuron , stimulation side , and temperature . We fit a three-parameter sigmoid to each response curve , r=ρ ( IdB ) =rsat/ ( 1+exp ( −IdB−I50 , ρwρ ) ) , with saturation spike rate rsat , half-maximum sound intensity I50 , ρ , and dynamic-range width wρ . Unless noted otherwise , temperature dependence of a given observable x was quantified by the temperature coefficientQ10 ( x ) = ( x ( T+ΔT ) x ( T ) ) 10/ΔT . Q10 ( x ) is the factor by which x changes after a temperature increase of 10°C . Q10>1 and Q10<1 indicate an increase or decrease , respectively , of x with heating , while Q10=1 indicates perfect temperature invariance . For plots of Q10 values data points were presented as outliers when they fell outside the interval [q11 . 5·iqr , q3+1 . 5·iqr] , with the 25th and the 75th percentile defining q1 and q3 and an interquartile range iqr=q3−q1 . We also quantified the temperature-dependence of action-potential width at half-maximum amplitude , Q10 ( AP width ) , for every neuron during the stimulus period , separately at each stimulus . Figure 1C shows the distribution of Q10 ( AP width ) pooled across all stimulus amplitudes ( median 0 . 66 ) . Our results qualitatively agree with the finding of broader action potentials at lower temperatures reported for various vertebrate and invertebrate neurons ( Thompson et al . , 1985; Bestmann and Dippold , 1989; Janssen , 1992; Gabbiani et al . , 1999 ) . Further , our results agree quantitatively with those reported for locust motor neurons and locust L-neurons ( Burrows , 1989; Simmons , 1990 ) . We checked that our results on the temperature dependence of firing rate were not compromised by the effects of adaptation . To this end , we re-analyzed the experimental data , separately focusing on the early phase of stimulus presentation ( 10–40 ms post stimulus onset ) , and the late phase ( 70–100 ms post stimulus onset ) . Effects of adaptation were reflected in a ratio of the respective parameter values ( early-versus-late phase ) that differed from one . While individual characteristics of experimentally measured firing-rate curves were subject to adaptation ( e . g . , the slope at half-maximum sound level was steeper early on and shallower in the later part ) , the early-to-late ratios did not significantly change with temperature . The dependence on temperature T ( °C ) was introduced to the model at the level of reversal potentials E ( T ) , peak conductances g¯X ( T ) , and time constants of ( in- ) activation τx ( T ) . The Nernst equation defined the temperature dependence of reversal potentials:E=RTzFln[ion outside][ion inside]⇒E ( T0+ΔT ) =E ( T0 ) · ( 1+ΔTT0+273 . 15 ) . R denotes the universal gas constant , z the valence of the considered ion , and F the Faraday constant . T0 = 18°C sets the reference temperature , ΔT temperature differences . Temperature dependence of g¯X ( T ) is determined by the choice of the parameter Q10 , g¯X:g¯X→g¯X ( ΔT ) =g¯X·Q10 , g¯XΔT10 , X∈{Na , K , L , A} . For ΔT=0 each g¯X ( T ) takes the value of its original definition at 18°C . For a given gating variable , temperature dependence of its opening and closing rates are identical and read:αx→αx ( ΔT ) =αx·Q10 , xΔT10 , βx→βx ( ΔT ) =βx·Q10 , xΔT10 , x∈{n , m , h , a , b} . Consequently , τx→τx ( ΔT ) =τx/Q10 , x . Spike rates f in response to N = 12 current amplitudes IC defined the f-I curve for a given temperature ( spike detection threshold −30 mV ) . Temperature dependence of the f-I curve was quantified as the root-mean-squared difference between firing rates at the two temperatures across input currents , normalized by the average spike rate elicited at the lower temperature:RMSD=1N∑i=1N ( fi , Tcold ( IC , i ) −fi , Thot ( IC , i ) ) 21N∑i=1Nfi , Tcold ( IC , i ) , with Tcold = 18°C and Thot = 28°C . In agreement with the functional shape of type I spiking ( Ermentrout , 1996 ) , f-I curves for each Q10 parameter combination were fit to a square root model , r=ϕ ( IC ) =A·IC−I0 , where A denotes slope and I0 firing threshold of the f-I curve ( quality of fit R2 >0 . 97 for 99% of the models ) . For a spike generation process f ( I ) , Fisher information J ( I ) is a measure of how accurately a particular input current I can be decoded from the firing-rate response f ( I ) . It is formally defined asJ ( I ) =∫ ( ∂∂IlnP ( f\|I ) ) 2P ( f\|I ) df , with the conditional probability density of the spike rate given an input current , P ( f\|I ) , characterizing the output noise ( i . e . , spike-rate variability ) . We consider two empirical response models for the spike rate density: Poissonian and input-independent Gaussian , readingPP ( f∼\|I ) = ( f ( I ) ·b ) f∼·b ( f∼b ) ! ·exp ( −f ( I ) b ) andPG ( f∼\|I ) =12πσ2·exp ( − ( f∼−f ( I ) ) 22·σ2 ) , respectively . For the Poisson case , b denotes the time bin during which a certain spike count Nsp is observed . It is assumed to be sufficiently large so that Nsp/b is well approximated by the mean firing rate f ( I ) . σ2 denotes the variance of the Gaussian probability density . The corresponding Fisher information is given byJP ( I ) = ( f′ ( I ) ) 2f ( I ) andJG ( I ) = ( f′ ( I ) ) 2σ2 , respectively . To compare Fisher information across different temperatures , it was averaged across a fixed interval of output firing rates [fmin , fmax] . Accordingly , the input current interval [Imin , Imax] was computed for each model and temperature . Average Fisher information reads〈J〉= ( Imax−Imin ) −1∫IminImaxdI J ( I ) . For low noise , the average Fisher information is a lower bound to the neuron's capacity to transmit information ( Kostal et al . , 2013 ) , Clow=ln ( ∫dI J ( I ) 2πe ) . Exploiting the square-root shape of firing-rate curves , f ( I ) =A·I−I0 , and f′ ( I ) =A/2· ( I−I0 ) −1/2 , it follows that ( 1 ) 1ΔI≡1Imax−Imin=A2fmax2−fmin2 . For the Poisson probability density Fisher information is given by〈JP〉=1ΔI∫IminImaxdI A/4· ( I−I0 ) −3/2=1ΔI· ( −A/2 ) [ ( I−I0 ) −1/2]IminImax=1ΔI· ( −A2/2 ) [ ( AI−I0 ) −1]IminImax . With Equation 1 it can be expressed as〈JP〉=1ΔI· ( −A2/2 ) ( 1/fmax−1/fmin ) =A2fmax2−fmin2· ( −A2/2 ) ( 1/fmax−1/fmin ) =A4·12fmaxfmin ( fmax+fmin ) . For a Gaussian probability density we get〈JG〉=1ΔI∫IminImaxdIA2/ ( 2σ2 ) ·1I−I0=A2/ ( 2σ2 ) ΔI·[ln ( I−I0 ) ]IminImax=A2/ ( 2σ2 ) ΔI·ln ( Imax−I0Imin−I0 ) instead . Fisher information in this case reads〈JG〉=A4σ2 ( fmax2−fmin2 ) ·ln ( fmaxfmin ) . Because only the slope of the firing-rate curve , A , is temperature-dependent in 〈JP〉 and 〈JG〉 , the temperature dependence of Fisher information is given byQ10 ( 〈JP〉 ) =Q10 ( 〈JG〉 ) =[Q10 ( A ) ]4 . For the average value across a fixed output interval [fmin , fmax] Fisher information is invariant to shifts of the threshold . A heating-induced increase in the accuracy of a decoder hence requires an increase in slope of the firing-rate curve , that is Q10 ( A ) >1 . This is true for all spike-generation models considered . Sensitivity analysis was performed in the parameter space spanned by the nine temperature-dependence parameters ( each dimension sampled by four values ) . To quantify global impact of one parameter on a given observable ( like RMSD ) , we evaluated the distribution of point-wise changes in the observable along the dimension of a specific parameter . In total , for each parameter , 3·48 changes between neighboring points along the corresponding dimension need to be considered . These define a distribution of changes , whose median is indicative of the global impact of this parameter on the observable . The distribution's 25% and 75% percentiles are indicated as error bars ( see , e . g . , Figure 2C ) . For each observable , impact values were normalized to give unity when summed across all nine parameters . The sign of the impact provides an estimate of the qualitative influence of the parameter on the observable , that is whether an increase in the parameter value leads to an increase or decrease in the observable . We considered an impact reliable if both percentiles ( 25% and 75% ) had the same sign as the impact . Note that our impact evaluation constitutes a global sensitivity analysis , comparable to the Morris one-at-a-time method ( Morris , 1991 ) . In contrast to the latter approach , we use a full factorial ( grid ) set of inputs instead of a random one . Moreover , our measure is based on the median instead of the mean of a distribution of differences in the observable . Yet , the interpretation of a high absolute impact is comparable to that of a high ( absolute ) mean elementary effect ( the sensitivity measure in Morris 1991 ) , as is the interpretation of a large interquartile range of the difference distributions to a large standard deviation of the elementary effect . Alternatively to the coarse parameter grid search , we also used optimization by a genetic algorithm ( Mitchell , 1998 ) to validate the minimum RMSD . To this end , the turboGA function was used ( Matlab file exchange; settings: population size 1000 , 150 generations , 8 bit discretization , initial conditions uniformly random ) . For the Q10 parameter range used in the main article , the minimum RMSD identified by the genetic algorithm was very close to the minimum value on the grid ( 0 . 21 for the genetic algorithm , 0 . 22 on the grid ) . On average the coordinates of the grid-based minimum deviated 5% from the coordinates of the genetic algorithm-based minimum . Dimensional stacking is a method to visualize high-dimensional data , that is an observable f as a function of N parameters , p1 , … , pN , evaluated at a discrete set of parameter values . The method is described in detail , for categoric observables , in LeBlanc et al . ( 1990 ) and Taylor et al . ( 2006 ) . Mainly , the method maps the N-dimensional data to a two-dimensional representation by iteratively slicing the data in one dimension and stacking the slices in 2D ( Figure 2L ) . In this representation , the position of each pixel in the two-dimensional image corresponds to one parameter combination , and its color encodes the value of the observable . The image has N axes of different scales , each associated with one parameter . Visual informativeness of a dimensional stacking image crucially depends on the order in which the dimensions are stacked , that is , the axes order . The parameter dimensions associated with larger variability in the observable should be assigned larger-scale axes; those with lowest variability the small-scale axes . Sorting the axes with respect to their impact on the observable prior to dimensional stacking hence leads to a visually informative image , where color changes can be easily related to changes of the observable with individual parameters . For this study , we used the ranking of absolute impact scores ( described above ) to define the ‘optimal’ stack order , extending the optimization method described in LeBlanc et al . ( 1990 ) and Taylor et al . ( 2006 ) . The sensitivity analysis was performed with the Connor–Stevens model with original parameters for peak conductances g¯Na , g¯K , g¯A , and g¯L at the colder temperature ( Dayan and Abbott , 2005 ) . To test that our results are robust and do not strongly depend on this specific parameter choice , we additionally performed the whole sensitivity analysis for 24 models with peak conductances of the reference model perturbed by ±20% ( 8 models with one individual peak conductance lowered or raised by 20%; 16 models with all combinations of the four conductances either lowered or raised by 20% ) . The impacts for those models are summarized in Figure 2E and Figure 3F , G ( individual changes in peak conductances represented by black symbols , combined changes by gray ones ) . Note that for computational efficiency only three values per parameter ( instead of four ) were taken ( for perturbed models as well as the reference model , as presented in Figure 2E ) . The fraction of models with RMSD<0 . 5 ( across the temperature dependence parameter space ) was 18% in the original Connor–Stevens model . Variations in peak conductances did not change this finding much: for each perturbed model 15–19% of its temperature dependence combinations gave RMSD<0 . 5 . For completeness , we also checked that a structurally different vertebrate model with type I dynamics ( Traub-Miles , Traub et al . , 1991 ) , as defined in Benda 2002 ) was able to display temperature compensation despite a substantial temperature dependence of individual conductances ( same range of temperature parameters as in the Connor–Stevens models , Figure 2—figure supplement 1 ) . The search for the lowest temperature dependence within the parameter space ( sodium and potassium kinetics Q10 ( m ) , Q10 ( h ) , Q10 ( n ) , and the peak conductances of sodium , potassium and leak Q10 ( g¯Na ) , Q10 ( g¯K ) , Q10 ( g¯L ) ) was performed based on the genetic algorithm described above . As the model operates at 32° , we checked both heating and cooling the model by 10° . For a given pair of a receptor neuron response r=ρ ( IdB ) and a spike generation model r=ϕ ( IC ) , the transduction function ( current IC vs sound intensity IdB ) can be inferred . We assumed a sigmoidal shape of the transduction function , IC=θ ( IdB ) =IC , sat/ ( 1+exp ( −IdB−I50 , θwθ ) ) , with transduction saturation current IC , sat , half-maximum sound intensity I50 , θ , and dynamic-range width wθ . Further , we chose representative parameters for the receptor neuron response . To this end , the median cold temperature , T¯c , the median receptor neuron response parameters at cold temperature , p¯ρ∈{r¯sat , I¯50 , ρ , w¯ρ} , and the median temperature dependencies of the three receptor neuron response parameters , Q¯10 ( pρ ) , were determined from the experimental data . Using these , receptor neuron response parameters were inferred for temperatures of 18 and 28° ( the temperatures at which spike generation simulations were performed ) , according to pρ , T=p¯ρ· ( Q¯10 ( pρ ) ) T−T¯c10 . The resulting representative receptor neuron responses were used as objective functions for reverse engineering of the transduction curve . To infer the three parameters characterizing the optimal transduction curve for a given spike generation model r=ϕ ( IC ) , we computedr=ϕ ( θ ( IdB ) ) =A·IC , sat/ ( 1+exp ( −IdB−I50 , θwθ ) ) −I0 . The transduction parameters were chosen such that they minimized the root mean squared error between ϕ ( θ ( IdB ) ) and the representative receptor neuron response ρ ( IdB ) :θ ( IdB , optimal ) =arg minθ ( IdB ) {∫\|ρ ( IdB ) −ϕ ( θ ( IdB ) ) \|2dIdB} This fitting procedure was repeated for all hot and cold spike generation processes , and the temperature coefficients for the three transduction parameters were computed . The transduction parameters at the reference temperature ( 18°C ) were: IC , cold ( IdB ) =0 . 40μA/mm2 .	Warm-blooded animals—including mammals and birds—expend large amounts of energy in keeping their body temperature constant regardless of how hot or cold their environment is . By contrast , the body temperature of cold-blooded animals—including amphibians , reptiles , and insects—follows that of their surroundings . Cold-blooded animals must therefore have evolved a means to cope with the effects of changes in temperature , but exactly how they do this is not clear . Now , Roemschied et al . have obtained new insights into this process by studying the nerve cells in grasshoppers that allow them to hear sounds . The auditory system of the grasshopper comprises sensory receptor neurons that are located on the abdomen of the insect . Sound waves move the tympanal membrane , which causes ion channels within the cell membranes to open . This enables the neurons to produce an electrical signal known as a spike . Recordings from grasshoppers revealed that changing the outside temperature by up to 10°C affected the rate at which the neurons produced spikes by only about half the amount expected . Given that these neurons do not receive inputs from any other cells , this ability to withstand changes in temperature must be intrinsic to the neurons themselves . Consistent with this , computational modeling showed that while the activity of individual ion channels did indeed vary with changes in temperature , these changes in ion channel activity had little overall effect on the rate at which a neuron produced spikes . Whereas it has previously been assumed that compensation for changes in temperature occurs at the level of networks of neurons , the work of Roemschied et al . reveals that such compensation can occur in individual cells , and that it need not require a lot of energy to be expended .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2014	Cell-intrinsic mechanisms of temperature compensation in a grasshopper sensory receptor neuron
Results from genome-wide association studies ( GWAS ) can be used to infer causal relationships between phenotypes , using a strategy known as 2-sample Mendelian randomization ( 2SMR ) and bypassing the need for individual-level data . However , 2SMR methods are evolving rapidly and GWAS results are often insufficiently curated , undermining efficient implementation of the approach . We therefore developed MR-Base ( http://www . mrbase . org ) : a platform that integrates a curated database of complete GWAS results ( no restrictions according to statistical significance ) with an application programming interface , web app and R packages that automate 2SMR . The software includes several sensitivity analyses for assessing the impact of horizontal pleiotropy and other violations of assumptions . The database currently comprises 11 billion single nucleotide polymorphism-trait associations from 1673 GWAS and is updated on a regular basis . Integrating data with software ensures more rigorous application of hypothesis-driven analyses and allows millions of potential causal relationships to be efficiently evaluated in phenome-wide association studies . GWAS summary data , the non-disclosive results from testing the association of hundreds of thousands to millions of genetic variants with a phenotype , have been routinely collected and curated for several years ( Welter et al . , 2014; Li et al . , 2016; Beck et al . , 2014 ) and are a valuable resource for dissecting the causal architecture of complex traits ( Pasaniuc and Price , 2017 ) . Accessible GWAS summary data are , however , often restricted to ‘top hits’ , that is , statistically significant results , or tend to be hosted informally in different locations under a wide variety of formats . For other studies , summary data may only be available ‘on request’ from authors . Complete summary data are currently publicly accessible for thousands of phenotypes but to ensure reliability and efficiency for systematic downstream applications they must be harvested , checked for errors , harmonised and curated into standardised formats . GWAS summary data are useful for a wide variety of applications , including MR , PheWAS ( Millard et al . , 2015; Denny et al . , 2010 ) , summary-based transcriptome-wide ( Gusev et al . , 2016 ) and methylome-wide ( Richardson et al . , 2017; Hannon et al . , 2017a ) association studies and linkage disequilibrium ( LD ) score regression ( Bulik-Sullivan et al . , 2015; Zheng et al . , 2017b ) . MR ( Davey Smith and Ebrahim , 2003; Davey Smith and Hemani , 2014 ) uses genetic variation to mimic the design of randomised controlled trials ( RCT ) ( although for interpretive caveats see Holmes et al . , 2017 ) . Let us suppose we have a single nucleotide polymorphism ( SNP ) that is known to influence some phenotype ( the exposure ) . Due to Mendel’s laws of inheritance and the fixed nature of germline genotypes , the alleles an individual receives at this SNP are expected to be random with respect to potential confounders and causally upstream of the exposure . In this ‘natural experiment’ , the SNP is considered to be an instrumental variable ( IV ) , and observing an individual’s genotype at this SNP is akin to randomly assigning an individual to a treatment or control group in a RCT ( Figure 1a ) . To infer the causal influence of the exposure , one calculates the ratio between the SNP effect on the outcome over the SNP effect on the exposure . If there are many independent IVs available for a particular exposure , as is often the case , causal inference can be strengthened ( Johnson , 2012 ) . Here , we consider each SNP to mimic an independent RCT and we can adapt tools developed for meta-analysis ( Bowden et al . , 2017a ) to combine the results obtained from each of the SNPs , giving an overall causal estimate that is better powered ( Bowden et al . , 2017a ) . Crucially , MR can be performed using results from GWAS , in a strategy known as 2-sample MR ( 2SMR ) ( Pierce and Burgess , 2013 ) . Here , the SNP-exposure effects and the SNP-outcome effects are obtained from separate studies . With these summary data alone , it is possible to estimate the causal influence of the exposure on the outcome . This has the tremendous advantage that causal inference can be made between two traits even if they aren’t measured in the same set of samples , enabling us to harness the statistical power of pre-existing large GWAS analyses . Due to the flexibility afforded by the 2SMR strategy , MR can be applied to 1000s of potential exposure-outcome associations , where ‘exposure’ can be very broadly defined , from gene expression and proteins to more complex traits , such as body mass index and smoking . While MR avoids certain problems of conventional observational studies ( Davey Smith and Ebrahim , 2001 ) , it introduces its own set of new problems . MR is predicated on exploiting ‘vertical’ pleiotropy , where a SNP influences two traits because one trait causes the other ( Davey Smith and Hemani , 2014 ) . It is crucial to be aware of the assumptions and limitations that arise due to this model ( Haycock et al . , 2016 ) . The main assumptions ( Figure 1b ) are: the instrument associates with the exposure ( IV assumption 1 ) ; the instrument does not influence the outcome through some pathway other than the exposure ( IV assumption 2 ) ; and the instrument does not associate with confounders ( IV assumption 3 ) . The IV1 assumption is easily satisfied in MR by restricting the instruments to genetic variants that are discovered using genome-wide levels of statistical significance and replicated in independent studies . The other two assumptions are impossible to prove , and , when violated , can lead to bias in MR analyses . Violations of the IV2 assumption can be introduced by ‘horizontal’ pleiotropy where the SNP influences the outcome through some pathway other than the exposure . Such effects can manifest in various different patterns ( Figure 1c–f ) . When multiple independent instruments are available it is possible to perform sensitivity analyses that attempt to distinguish between horizontal and vertical pleiotropy and return causal estimates adjusted for the former ( Bowden et al . , 2016a; Bowden et al . , 2015; Hartwig et al . , 2017b ) . To improve reliability of causal inference , MR results should be presented alongside sensitivity analyses that make allowance for various potential patterns of horizontal pleiotropy . Further details on the design and interpretation of Mendelian randomization studies can be found in several existing reviews ( Davey Smith and Hemani , 2014; Haycock et al . , 2016; Swerdlow et al . , 2016; Holmes et al . , 2017; Zheng et al . , 2017a ) . A glossary of terms can be found in Supplementary file 1F . In this section we describe how to use MR-Base to conduct MR analyses ( Figure 2 ) . The data required to perform the analysis can be described as a ‘summary set’ ( Hemani et al . , 2017a ) , where the genetic effects for a set of instruments are available for both the exposure and the outcome . To create a summary set we select appropriate instruments , obtain the effect estimates for those instruments for the exposure and the outcome , and harmonise the effects so that they reflect the same allele . We can then perform MR analyses using the summary set . These steps are supported by the database of GWAS results and R packages ( ‘TwoSampleMR’ and ‘MRInstruments’ ) curated by MR-Base and the following R packages curated by other researchers: 'MendelianRandomization' ( Yavorska and Burgess , 2017 ) , 'RadialMR' ( Bowden et al . , 2017b ) , 'MR-PRESSO' ( Verbanck et al . , 2018 ) and 'mr . raps' ( Zhao et al . , 2018 ) . The statistical methods and R packages accessible through MR-Base are updated on a regular basis . Instruments are characterised as SNPs that reliably associate with the exposure , meaning they should be obtained from well-conducted GWAS , typically involving their detection in a discovery sample at a GWAS threshold of statistical significance ( e . g . p<5x10−8 ) followed by replication in an independent sample . The minimum data requirements for each SNP are effect sizes ( βx ) , standard errors ( σx ) and effect alleles . Also useful are sample size , non-effect allele and effect allele frequency . In order to generate the summary set , the effects of each of the instruments on the outcome need to be obtained . This typically requires access to the entire set of GWAS results because it is unlikely that the instrumental SNPs for the exposure will be amongst the top hits of the outcome GWAS . As with the exposure data , the outcome data must contain at a minimum the SNP effects ( βy ) , their standard errors ( σy ) and effect alleles . To generate a summary set , for each SNP we need its effect and standard error on the exposure and the outcome corresponding to the same effect alleles ( Hartwig et al . , 2016 ) . This is impossible to generate if the effect alleles for the SNP effects in the exposure and outcome datasets are unknown . MR-Base uses knowledge of the effect alleles , and where necessary the effect allele frequencies , to automatically harmonise the exposure and outcome datasets . The following scenarios are considered: The generated summary set can now be analysed using a range of methods ( summarised in Supplementary file 1B but new methods are added on a regular basis ) . The most basic way to combine these data is to use a Wald ratio where the estimated causal effect isβMR=βyβxand the standard error of the estimate isσMR=σyβx If there are multiple independent instruments for the exposure ( as is typically the case for complex traits with well-powered GWAS ) , then our analysis can potentially improve in two major ways: first , the variance explained in the exposure , and therefore statistical power will improve; second , we can evaluate the sensitivity of the estimate to bias arising from violations of the IV2 assumption by assuming different patterns of horizontal pleiotropy . Sensitivity analyses are performed automatically by MR-Base . It is recommended that the methods described above are applied to all MR analyses and presented in publications to demonstrate sensitivity to different patterns of assumption violations . MR-Base also automatically performs the following further sensitivity analyses and diagnostics In addition to the above , MR-Base also supports access to the following statistical methods implemented in other R packages: an extension of the IVW method that allows for correlated SNPs ( Yavorska and Burgess , 2017 ) , a method for the detection and correction of outliers in IVW linear regression ( MR-PRESSO , Verbanck et al . , 2018 ) , methods for fitting and visualising radial IVW and radial MR-Egger models ( Bowden et al . , 2017b ) and a method for correcting for horizontal pleiotropy using MR-RAPS ( 2SMR using robust adjusted profile scores ( Zhao et al . , 2018 ) . We created a repository for complete GWAS summary data , where complete refers to all SNPs reported in a GWAS analysis with no exclusions according to p-values for the association with the trait of interest ( e . g . datasets were not restricted to statistically significant SNPs ) . We included summary data from any array-based analysis , including targeted and untargeted arrays , with or without additional imputation for ungenotyped SNPs . The targeted arrays included immunochip and metabochip , as well as replication and fine-mapping studies with ≥10 , 000 variants . As of December 2017 , the repository was populated by curated and harmonised datasets from 1673 GWAS analyses , corresponding to approximately 11 billion SNP-trait associations in 4 million samples ( median sample size per study: 21 , 315 ) . Excluding replication and fine-mapping studies , the median number of SNPs per study was 6 . 1 million ( minimum = 79 , 129 , maximum = 22 , 434 , 434 ) ; 95% of studies reported ≥393 , 465 SNPs . The current database also includes nine studies with ≤64 , 494 SNPs that generally correspond to replication and fine-mapping studies . The analysed traits included 605 traits generated using the UK Biobank resource ( Millard et al . , 2017; Bycroft et al . , 2017; Churchhouse and Neale , 2017;GIANT consortium et al . , 2016; Jones et al . , 2016; Pilling et al . , 2016 ) , 575 metabolomic traits from two studies ( Shin et al . , 2014; Kettunen et al . , 2016 ) , 151 immunological traits from one study ( Roederer et al . , 2015 ) , and 342 other complex traits and diseases acquired from 123 GWAS publications ( Supplementary file 1A ) . The latter publications corresponded to 79 studies , including 39 consortia and three cohorts . Supplementary file 1A provides a detailed overview of the available studies with complete summary data in MR-Base at the time of writing but the number is updated on a regular basis . In addition to the ‘complete summary data’ , we also collected published GWAS associations that comprise only the significant hits of a GWAS after applying stringent p-value thresholds ( e . g . p<5 × 10−8 , a conventional threshold for declaring statistical significance in GWAS ) . These ‘top hits’ , which can be used to define genetic instruments for exposures in MR analyses ( see Materials and methods ) , include 29 , 792 SNPs obtained from clumping analysis of 1002 traits in the MR-Base database; 21 , 324 SNPs associated with 1628 complex traits and diseases in the NHGRI-EBI GWAS catalog ( Welter et al . , 2014 ) ; 187 , 318 SNPs associated with DNA methylation levels in whole blood at 33 , 256 genomic CpG sites , across five time points ( Gaunt et al . , 2016 ) ; 187 , 263 SNPs associated with gene expression levels at 27 , 094 gene identifiers , across 44 different tissues ( GTEx Consortium , 2015 ) ; 1088 SNPs associated with metabolite levels in whole blood for 121 different metabolites ( Kettunen et al . , 2016 ) ; and 56 SNPs associated with protein levels in 47 different analytes ( Deming et al . , 2016 ) . The repositories of GWAS results described above can be interrogated and exploited for 2SMR using the R packages curated by MR-Base and other researchers . The R packages currently curated by MR-Base include ‘TwoSampleMR’ ( https://github . com/MRCIEU/TwoSampleMR ) and ‘MRInstruments’ https://github . com/MRCIEU/MRInstruments ) . Users can check the MR-Base website for updates to the curated packages . Accessible R packages curated by other researchers are described in Supplementary file 1B . In an applied example , we conducted a MR study of the causal effect of LDL cholesterol on CHD risk , using summary data from the GLGC ( Willer et al . , 2013; Do et al . , 2013 ) and CARDIoGRAMplusC4D consortia ( Nikpay et al . , 2015 ) , respectively . There were 91 studies ( 214 , 370 subjects ) in the GLGC , 48 studies ( 195 , 813 subjects ) in CARDIoGRAMplusC4D and 17 studies that were common to both consortia ( including 59 , 970 subjects ) . We estimated that 31% of subjects in CARDIoGRAMplusC4D are also part of the GLGC and 28% of GLGC participants are also part of CARDIoGRAMplusC4D . The selected instruments ( Supplementary file 1C ) reportedly explained 2 . 4% of the variance in LDL cholesterol levels ( Willer et al . , 2013 ) , equivalent to to an F statistic of 85 in the GLGC . This indicates that the instrument is strong and therefore unlikely to be susceptible to weak instrument bias or bias from sample overlap ( Burgess et al . , 2011 ) . The random effects IVW estimate indicated that the odds ratio ( OR ) ( 95% confidence interval [CI] ) for CHD was 1 . 45 ( 1 . 30–1 . 62 ) per standard deviation increase in LDL cholesterol ( Figure 4 ) . There was , however , strong evidence for heterogeneity amongst SNPs ( Cochran’s Q value = 122 . 5 , p=4 . 72e-07 ) and funnel plot asymmetry ( Figure 4a and d ) , suggesting that at least some of the SNPs exhibit horizontal pleiotropy ( a violation of the IV2 assumption , as shown in Figure 1b ) . There was evidence for a negative intercept ( −0 . 013 [s . e . =0 . 005] , p=0 . 020 ) and stronger odds ratio ( 1 . 85 [1 . 48–2 . 32] ) in MR-Egger regression ( Figure 4b ) indicating some amount of directional horizontal pleiotropy . Similar results to the IVW estimate were provided by the weighted median ( 1 . 56 [1 . 43-1 . 70] ) and weighted mode ( 1 . 68 [1 . 56-1 . 80] ) estimators ( Figure 4 ) . In a leave-one-out analysis , we sequentially excluded one instrument ( SNP ) at a time to assess the sensitivity of the results to individual variants , finding that no single instrument was strongly driving the overall effect of LDL cholesterol on CHD ( Figure 4c ) . Inspection of the forest , funnel and scatter plots highlighted three outlier SNPs ( rs11065987 , rs1250229 and rs4530754 ) as potential sources of heterogeneity and the negative intercept in MR-Egger regression ( Figure 4 ) . For these three SNPs , the LDL-raising variant was associated with lower CHD risk , contrary to the results for the majority of the other LDL-raising variants . In such situations it is strongly advised to check for effect allele coding errors ( Hartwig et al . , 2016 ) . We confirmed that the SNPs were not palindromic ( such SNPs are particularly prone to coding errors in 2SMR ) and that the LDL and CHD risk variants were compatible with those reported in the GWAS catalog and the original study reports . The unusually strong cardio-protective effects of the three LDL raising variants are also compatible with horizontal pleiotropy , whereby the effects of the SNPs on CHD are independent of their effects on LDL cholesterol . To identify potential sources of horizontal pleiotropy , we performed a PheWAS of rs11065987 , rs1250229 and rs4530754 , using a threshold of p<2 . 04e-05 ( 0 . 05/2453 ‘trait lookups’ ) to select traits for further evaluation . Only rs11065987 , located upstream of the BRAP gene , was associated with non-lipid non-vascular-disease traits , including two markers of adiposity ( body mass index and hip circumference ) ; three blood pressure traits ( diastolic blood pressure , systolic blood pressure and mean arterial pressure ) ; five hematological markers ( hematocrit , haemoglobin concentration , packed cell volume , red blood cell count and platelet count ) ; five autoimmune diseases ( inflammatory bowel disease , primary biliary cirrhosis , Crohn's disease , rheumatoid arthritis and celiac disease ) ; four metabolites ( urate , kynurenine , erythronate and C-glycosyltryptophan ) ; serum cystatin C ( a marker of kidney function ) ; and tetralogy of Fallot ( Supplementary file 1D ) . In further MR analyses of these traits , we found that higher hematocrit , higher blood pressure , higher BMI and higher hip circumference were putatively associated with higher CHD risk ( p<0 . 05 ) ( Supplementary file 1D ) . However , of these traits , only the indirect effect of rs11065987 due to hematocrit ( -0 . 012 , SE=0 . 005 ) was in the same direction as the direct effect of rs11065987 on CHD ( −0 . 060 , SE = 0 . 011 ) , whereas the indirect effects mediated by hip circumference ( 0 . 004 , SE = 0 . 002 ) , BMI ( 0 . 007 , SE = 0 . 001 ) and LDL cholesterol ( 0 . 011 , SE = 0 . 001 ) were in opposite directions . Due to unreported effect alleles in the relevant GWAS , we were unable to assess the indirect effect of rs11065987 mediated by blood pressure . These results suggest that at least one cardio-protective mechanism for the LDL-raising variant of rs11065987 is due to a pleiotropic effect of hematocrit . However , this result should be interpreted with caution and requires replication in independent studies and further examination for potential violations of MR assumptions in sensitivity analyses . In order to gain insight into potential opportunities for repurposing or adverse effects of LDL cholesterol lowering - an established intervention stategy for CHD prevention - we conducted a hypothesis-free MR-PheWAS analysis ( Millard et al . , 2015; Haycock et al . , 2017 ) . Instrumented using 62 SNPs ( Supplementary file 1C ) , we obtained fixed effects IVW estimates of lower LDL cholesterol on 40 non-vascular diseases and 108 non-lipid complex traits in MR-Base ( Figure 5 ) . Using an unadjusted p-value of 0 . 05 to denote suggestive evidence for association , we identified 16 non-vascular traits associated with LDL cholesterol . Surpassing a 5% false discovery threshold were lower mortality measures , higher adiposity measures , and higher risk of type two diabetes . We emphasise that this analysis is shown here for purposes of demonstrating the utility of MR-Base for efficiently screening many traits for hypothesis generation , and any claims of causality must be followed up with rigorous examination of potential violations of MR assumptions in sensitivity analyses , replication in independent studies and triangulation with evidence from other study designs ( Lawlor et al . , 2016; Munafò and Davey Smith , 2018 ) . When the biases and limitations indicated above are avoidable , users should consider modifying their analyses . For example , users could use an automated approach for instrument selection in their primary analysis but use manually curated instruments in sensitivity analyses of prioritised results . Sometimes , however , biases may be unavoidable , in which case users should acknowledge their possible impact and relax their conclusions . For example , inferences about directions of causality , shared genetic architecture ( Burgess et al . , 2014 ) or null effects ( VanderWeele et al . , 2014 ) are usually much more reliable than inferences about magnitudes of causal effects , which are very sensitive to violations of assumptions . A number of resources are available for MR analysis or extracting and using GWAS summary data . The MendelianRandomization R package ( Yavorska and Burgess , 2017 ) is a standalone tool comprising several 2SMR methods and , in addition to the methods that we have implemented , we make it easy to import data from MR-Base into that R package . PhenoScanner ( Staley et al . , 2016 ) expands upon established GWAS catalogs ( Welter et al . , 2014; Li et al . , 2016 ) by storing a large number of complete summary level datasets and providing a web interface for specific SNP lookups . SMR ( Zhu et al . , 2016 ) has been developed to automate colocalisation analysis between eQTLs and complex traits . There has been a massive growth in the phenotypic coverage and statistical power of GWAS over the past decade ( Visscher et al . , 2017; Welter et al . , 2014 ) . Many approaches to studying complex traits and diseases can now be interrogated using GWAS summary level data . By harvesting and harmonising these data into a database and directly integrating this with analytical software for 2SMR , MR-Base greatly improves the efficiency and reliability of hypothesis-driven approaches . The database is a generic repository of GWAS summary data accessible via an API and future work will see extensions to support other methods for the investigation of complex trait genetic architecture , such as fine-mapping ( Benner et al . , 2016 ) , colocalisation ( Zhu et al . , 2016; Newcombe et al . , 2016; Giambartolomei et al . , 2014 ) and polygenic risk prediction ( Dudbridge , 2013; Euesden et al . , 2015 ) . The curation of data and methods achieved by MR-Base opens up new opportunities for hypothesis-free and phenome-wide approaches . The following resources are all part of the MR-Base platform: Code to reproduce the analysis in this paper: https://github . com/explodecomputer/mr-base-methods-paper . MR-Base comprises two main components: a database of GWAS summary association statistics and LD proxy information , and R packages and web applications for causal inference methods . The GWAS summary data are further structured into complete summary data ( i . e . all SNP-phenotype associations ) and ‘top hits’ , which comprises subsets of GWAS summary data ( typically the statistically significant results ) . The R packages include the TwoSampleMR ( https://github . com/MRCIEU/TwoSampleMR ) package , which supports data extraction , data harmonisation and MR analysis methods , and the MRInstruments ( https://github . com/MRCIEU/MRInstruments ) package , which is a repository for instruments . MR-Base also supports access to R packages curated by other researchers , including MendelianRandomization ( Yavorska and Burgess , 2017 ) ( https://cran . r-project . org/web/packages/MendelianRandomization/ ) , RadialMR ( Bowden et al . , 2017b ) ( https://github . com/wspiller/radialmr ) , MR-PRESSO ( Verbanck et al . , 2018 ) ( https://github . com/rondolab/MR-PRESSO ) and mr . raps ( Zhao et al . , 2018 ) ( https://github . com/qingyuanzhao/mr . raps ) . The available methods are updated on a regular basis . The database is accessible through an API ( http://api . mrbase . org ) and is therefore extendable to other causal inference methods not currently covered by the aforementioned packages . A web app ( http://app . mrbase . org ) was developed as a user-friendly wrapper to the R package using the R/shiny framework . The SNP lookup tool is available through the PheWAS web app ( http://phewas . mrbase . org/ ) . Scripts to perform the analyses presented in this paper are available at https://github . com/explodecomputer/mr-base-methods-paper ( Hemani , 2018; copy archived at https://github . com/elifesciences-publications/mr-base-methods-paper ) . GWAS summary data posted online tends not to follow standardised formats , therefore harmonisation of these disparate data sources is a manual process . We adopted a systematic approach to harmonize these data and developed an Elasticsearch database ( https://www . elastic . co/products/elasticsearch , v5 . 6 . 2 ) to store , structure and query the harmonised data . Insofar as it was possible we recorded all QC and harmonisation processes for each of the 1673 datasets to aid with reproducibility . The following steps were taken for each dataset: One of the main functions of the MR-Base database is to provide association data for requested SNPs from studies of interest to the user ( Figure 2 ) . Often , however , a requested SNP may not be present in the requested GWAS ( e . g . because of different imputation panels or because imputed SNPs were not available ) . In order to enable information to be obtained even when SNPs are missing , we provide an LD proxy function using 1000 genomes data from 503 European samples . For each common variant ( minor allele frequency [MAF]>0 . 01 ) we used plink1 . 90 beta three software to identify a list of LD proxies . We recorded the r2 values for each LD proxy and the phase of the alleles of the target and proxy SNPs . We limited the LD proxies to be within 250 kb or 1000 SNPs and with a minimum r2 = 0 . 6 . We have assembled a collection of strong SNP-phenotype associations from various sources that can be used as potential instruments in Mendelian randomization studies . Instruments are currently restricted to biallelic SNPs but in principle could be extended in future versions to accommodate multi-allelic SNPs or copy number variants ( CNVs ) . The potential instruments generally correspond to the ‘top hits’ from a GWAS , rather than the entire collection of GWAS summary statistics . As such , the traits included here can only be evaluated as potential exposures in a hypothesized exposure-outcome analysis ( complete summary data are required when evaluating traits as potential outcomes ) . All curated instruments are available through the MRInstruments R package ( https://github . com/MRCIEU/MRInstruments ) . This is a comprehensive catalog of reported associations from published GWAS ( Welter et al . , 2014 ) . To make the data suitable for Mendelian randomization , we converted odds ratios to log odds ratios and inferred standard errors from reported 95% confidence intervals or ( if the latter were unavailable ) from the reported p-value using the Z distribution . The GWAS catalog scales odds ratios to be greater than 1 and includes information on unit changes ( e . g . mg/dl increase ) for beta coefficients . In MR-Base we therefore assume that effect sizes are odds ratios if they are greater than 1 and are missing information on unit changes . We extracted information on the units of the SNP-trait effect; and identified effect and non-effect alleles by comparing the risk allele reported in the GWAS catalog to allele information downloaded from ENSEMBL , using the R/biomaRt package ( Durinck et al . , 2009 ) . R/biomaRt was also used to identify base pair positions ( in GRCh38 format ) and associated candidate genes . We inferred effect allele frequency from the risk allele frequency reported in the GWAS catalog . We excluded SNP-trait associations from the GWAS catalog if they were missing a p-value , beta ( estimate of the SNP-trait effect ) or a standard error for the beta . The MR-Base standardized version of the GWAS catalog ( 2017-03-20 release at the time of writing ) comprises 21 , 324 potential instruments for 1628 traits . There are , however , several caveats to using the GWAS catalog as a source of instruments . First , reported units of analysis are often unclear ( e . g . results are often reported as reflecting a ‘unit increase’ ) . Second , the GWAS catalog prioritises results from the largest reported analysis in the original study report ( typically the discovery study or a meta-analysis of discovery and replication studies ) . This makes instruments from the GWAS catalog susceptible to winner’s curse , which can compound the effect of weak instruments bias . We obtained a large set of SNPs associated with DNA methylation levels ( i . e . mQTLs ) using the ARIES dataset ( Gaunt et al . , 2016 ) . mQTLs were identified in 1000 mothers at two time points and 1000 children at three time points . Top hits were obtained from http://mqtldb . org with p<1e-7 . There are 33 , 256 unique CpG sites across the five time points with at least one independent mQTL . These mQTLs can be used as instruments for DNA methylation at CpG sites in Mendelian randomization analyses . The mQTLs can also be used to perform methylome-wide association studies ( MWAS ) , to evaluate the association between DNA methylation at each CpG site and a phenotype of interest ( implementable in MR-Base through the Wald ratio method when only a single mQTL for a CpG site is available ) . We used the GTEx resource ( GTEx Consortium , 2015 ) of published independent cis-acting expression QTLs ( cis-eQTLs ) to create a catalog of SNPs influencing up to 27 , 094 unique gene identifiers across 44 tissues . These eQTLs can be used as instruments for gene-expression in Mendelian randomization analyses . The eQTLs can also be used to perform transcriptome-wide association studies ( TWAS ) , to evaluate the association between expression of each gene and a phenotype of interest ( implementable in MR-Base using the Wald ratio method when only a single eQTL for a gene is available ) . SNPs influencing 121 metabolites measured using nuclear magnetic resonance ( NMR ) analysis in whole blood were obtained ( Kettunen et al . , 2016 ) , totalling 1088 independent QTLs across all metabolites . SNPs influencing 47 protein analyte levels ( Deming et al . , 2016 ) in whole blood were obtained , totaling 57 independent proteomic QTLs . The MR-Base repository of complete GWAS summary data , which contains all SNPs from a GWAS regardless of p-value , can also be used to define instruments . This involves extracting independent sets of SNP-phenotype associations that surpass user-specified p-value and clumping thresholds . However , as the MR-Base repository of complete summary data is based mostly on discovery studies , this strategy may be susceptible to false positive instruments ( where some of the selected genetic variants are not truly associated with the target exposure ) and winner’s curse . See discussion and Supplementary file 1E for potential implications on results of these limitations . The Elasticsearch database is behind a firewall and cannot be queried directly , in order to prevent misuse and to keep non-public data secure . An API ( http://api . mrbase . org ) is used to interface with the database , controlling access based on user permissions and using Google OAuth2 . 0 for user authentication . A user friendly interface to the API is provided through the TwoSampleMR R package . A complete guide to use the R package is available at https://mrcieu . github . io/TwoSampleMR/ and a list of the analytical functions that are currently implemented are shown in Supplementary file 1B . We used MR-Base to recapitulate the known ( Holmes et al . , 2017 ) causal effect of higher LDL cholesterol on CHD risk . To obtain the list of instruments for LDL cholesterol we searched for the GLGC entry ( Willer et al . , 2013 ) in the GWAS catalog dataset in the MRInstruments library . This returned 62 SNPs . Due to unclear effect size units in the GWAS catalog , we extracted these 62 SNPs from the MR-Base database to obtain effect sizes in standard deviation units . We then searched for these 62 SNPs in the CARDIoGRAMplusC4D GWAS dataset ( Nikpay et al . , 2015 ) in the MR-Base database . One of the SNPs was not available and an LD proxy was identified . We harmonised the dataset , setting the algorithm to assume all SNPs were coded with alleles on the forward strand . Disabling this option would have excluded 7 palindromic SNPs with allele frequencies close to 0 . 5 . The code to reproduce this analysis is below . library ( TwoSampleMR ) library ( MRInstruments ) data ( gwas_catalog ) library ( MRInstruments ) data ( gwas_catalog ) # Get published SNPs for LDL cholesterol ldl_snps <- subset ( gwas_catalog , grepl ( "LDL choles" , Phenotype ) & Author == "Willer CJ" ) $SNP# Extract from GLGC dataset exposure <- convert_outcome_to_exposure ( extract_outcome_data ( ldl_snps , 300 ) ) # Get outcome data from Cardiogram 2015 outcome <- extract_outcome_data ( exposure$SNP , 7 ) # Harmonise exposure and outcome datasets # Assume alleles are on the forward strand dat <- harmonise_data ( exposure , outcome , action=1 ) # Perform MR mr ( dat ) mr_heterogeneity ( dat ) # Label outliers and create plots dat$labels <- dat$SNP dat$labels[ ! dat$SNP %in% c ( "rs11065987" , "rs1250229" , "rs4530754" ) ] <- NA mr_plots ( dat ) A PheWAS of outliers in the LDL-CHD MR results was conducted to identify potential sources of horizontal pleiotropy . First , we searched the MR-Base database of complete summary data and the GWAS catalog for associations with outlier SNPs , using a threshold of p<2 . 04e-05 ( 0 . 05/2453 ‘trait lookups’ ) to select traits for further evaluation . When identical traits were duplicated across different GWAS analyses , we retained the GWAS with the largest sample size . Of the outlier SNPs analysed , only rs11065987 was associated with non-lipid-non-vascular-disease traits and was therefore the only SNP retained for further analyses . We then conducted MR analyses to estimate the effect of the selected traits on CHD , scaled to reflect the magnitude of the observed effect of rs11065987 on the trait . Instruments were based on SNPs associated with the traits at a p-value less than 5e-8 , with clumping to ensure independence between SNPs ( clumping r2 cutoff=0 . 001 and clumping window=10 , 000kb ) . The GWAS catalog was used to define instruments when the selected trait was unavailable in the MR-Base database of complete summary data . All instruments excluded rs11065987 . The effect of the traits on CHD was based on the slope from IVW linear regression , except where only a single SNP was available to instrument the trait , in which case the Wald ratio method was used . The variance of the slope from IVW linear regression was estimated using a random effects model , except where there was underdispersion in the causal estimates between SNPs , in which case a fixed effects model was used . We then compared the rs11065987-CHD effect ( the direct effect ) with the trait-CHD effect ( indirect effect of rs11065987 ) . Where effects were in opposite directions we concluded that it was less likely that the association between rs11065987 and CHD was mediated by the selected trait . Complete code for all analyses can be found here: https://github . com/explodecomputer/mr-base-methods-paper ( Hemani , 2018; copy archived at https://github . com/elifesciences-publications/mr-base-methods-paper ) .	Our health is affected by many exposures and risk factors , including aspects of our lifestyles , our environments , and our biology . It can , however , be hard to work out the causes of health outcomes because ill-health can influence risk factors and risk factors tend to influence each other . To work out whether particular interventions influence health outcomes , scientists will ideally conduct a so-called randomized controlled trial , where some randomly-chosen participants are given an intervention that modifies the risk factor and others are not . But this type of experiment can be expensive or impractical to conduct . Alternatively , scientists can also use genetics to mimic a randomized controlled trial . This technique – known as Mendelian randomization – is possible for two reasons . First , because it is essentially random whether a person has one version of a gene or another . Second , because our genes influence different risk factors . For example , people with one version of a gene might be more likely to drink alcohol than people with another version . Researchers can compare people with different versions of the gene to infer what effect alcohol drinking has on their health . Every day , new studies investigate the role of genetic variants in human health , which scientists can draw on for research using Mendelian randomization . But until now , complete results from these studies have not been organized in one place . At the same time , statistical methods for Mendelian randomization are continually being developed and improved . To take advantage of these advances , Hemani , Zheng , Elsworth et al . produced a computer programme and online platform called “MR-Base” , combining up-to-date genetic data with the latest statistical methods . MR-Base automates the process of Mendelian randomization , making research much faster: analyses that previously could have taken months can now be done in minutes . It also makes studies more reliable , reducing the risk of human error and ensuring scientists use the latest methods . MR-Base contains over 11 billion associations between people’s genes and health-related outcomes . This will allow researchers to investigate many potential causes of poor health . As new statistical methods and new findings from genetic studies are added to MR-Base , its value to researchers will grow .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "tools", "and", "resources" ]	2018	The MR-Base platform supports systematic causal inference across the human phenome
The dynamic organization of signaling cascades inside primary cilia is key to signal propagation . Yet little is known about the dynamics of ciliary membrane proteins besides a possible role for motor-driven Intraflagellar Transport ( IFT ) . To characterize these dynamics , we imaged single molecules of Somatostatin Receptor 3 ( SSTR3 , a GPCR ) and Smoothened ( Smo , a Hedgehog signal transducer ) in the ciliary membrane . While IFT trains moved processively from one end of the cilium to the other , single SSTR3 and Smo underwent mostly diffusive behavior interspersed with short periods of directional movements . Statistical subtraction of instant velocities revealed that SSTR3 and Smo spent less than a third of their time undergoing active transport . Finally , SSTR3 and IFT movements could be uncoupled by perturbing either membrane protein diffusion or active transport . Thus ciliary membrane proteins move predominantly by diffusion , and attachment to IFT trains is transient and stochastic rather than processive or spatially determined . Membrane proteins explore cellular space by diffusion or motor-driven transport on the cytoskeleton , but the relative contributions of these processes in specific cellular contexts have not been directly assessed . A prototypical system to study how proteins explore cellular space is the primary cilium , a surface-exposed compartment consisting of a microtubule-based axoneme ensheathed within a plasma membrane protrusion ( Satir and Christensen , 2007 ) . The primary cilium is an important signaling compartment ( Goetz and Anderson , 2010 ) : entire signaling cascades , from the most upstream membrane-embedded receptors to the ultimate effectors , are dynamically concentrated within the ciliary space ( Corbit et al . , 2005; Haycraft et al . , 2005; Rohatgi et al . , 2007 ) , and cilium dysfunction abolishes specific pathways such as Hedgehog signaling ( Huangfu et al . , 2003 ) . The assembly of cilia , and structurally related flagella , relies on intraflagellar transport ( IFT ) , a process in which moving IFT trains bring cargo , such as tubulin and axonemal precursors , to the tip of the axoneme by kinesin II motor-powered transport along axonemal microtubules ( Cole , 2008 ) . In reverse , polypeptides move by dynein 1b-powered IFT from the tip to the base of cilia . The requirement of the IFT machinery for ciliary assembly , and the bidirectional movements of IFT trains in mature cilia have been substantiated in every model system tested ( Pedersen and Rosenbaum , 2008 ) and it is generally accepted that soluble and membrane-embedded cargo explore the ciliary space by loading onto IFT trains at one end of the primary cilium and unloading at the other end ( Qin et al . , 2004; Pedersen et al . , 2006 ) . Yet , there is limited direct experimental evidence to support this model of IFT-mediated transport of ciliary membrane proteins . Pioneering studies using the flagellated green alga Chlamydomonas reinhardtii showed that externally applied beads , which artificially cluster mating receptors , are transported along the flagellar axoneme ( Bloodgood et al . , 1979; Bloodgood and Salomonsky , 1989 ) , but a role for IFT has not been tested in this context . While bulk imaging of some GFP-tagged ciliary membrane proteins ( e . g . , PKD2-GFP , OSM9-GFP ) discerned a few fluorescence foci moving processively in cilia ( Ou et al . , 2005; Huang et al . , 2007 ) , ciliary membrane proteins are generally distributed homogeneously inside cilia and information on their dynamics is therefore lacking . Furthermore , while IFT trains have been visualized as large polymeric assemblages by electron microscopy ( Pigino et al . , 2009 ) and as clusters of fluorescence by light microscopy ( Pedersen and Rosenbaum , 2008 ) , the IFT entities that ferry cargoes remain elusive . The goal of our work was to directly measure the contributions of IFT and diffusion to the movement of individual ciliary membrane proteins . We developed an experimental system to image single membrane receptors together with IFT trains in the primary cilium of live cells . By selectively disrupting membrane protein diffusion or motor-driven active transport , we show that cargo and IFT movements can be uncoupled from each other and that diffusion is sufficient for membrane proteins to explore the ciliary surface . To overcome the limitations of ensemble measurements of cargo proteins in primary cilia , we quantified the ciliary movements of individually labelled membrane proteins . We selected two ciliary membrane proteins: the G protein-coupled receptor ( GPCR ) , Somatostatin receptor 3 ( SSTR3 ) ( Händel et al . , 1999 ) , and the Hedgehog signaling intermediate Smoothened ( Smo ) , a non-GPCR 7-pass membrane protein ( Corbit et al . , 2005 ) . To facilitate cilia visualization , SSTR3 and Smo were tagged at the intracellular C-terminus with a fluorescent protein ( SSTR3-GFP and Smo-YFP ) . In order to detect single molecules we also fused a biotinylation acceptor peptide ( AP ) to the extracellular N-terminus of SSTR3 or Smo , and co-expressed the biotin ligase BirA in the ER of the stable cell lines ( Howarth and Ting , 2008 ) . Proteins were expressed stably and at low levels in IMCD3-Flp-In cells . Singly biotinylated SSTR3 and Smo molecules were labelled on the cell surface by extracellular addition of a low concentration ( 50 pM ) of monovalent streptavidin ( mSA; Howarth et al . , 2006 ) conjugated to Alexa647 fluorescent dye ( mSA-A647 ) ( Figure 1A ) . Under these conditions , SSTR3-GFP and Smo-YFP were distributed homogeneously in cilia , but each cilium contained between zero and five dots of mSA-A647 labelled molecules ( referred hereafter as single SSTR3 or Smo molecule ) ( Figure 1B ) ; cilia orientation was determined by co-expressing pericentrin-RFP ( PCNT; Gillingham and Munro , 2000 ) which localizes to the basal body at the base of the cilium ( Figure 1B ) . Single molecule imaging was performed at 2 Hz until fluorescence of mSA was lost due to bleaching ( 30–60 s ) . Time-space plots ( kymographs; Figure 1B , bottom; Figure 1–figure supplement 1 ) were generated from live cell time-lapse images to describe the movement of single molecules . 10 . 7554/eLife . 00654 . 003Figure 1 . Real-time imaging of single signaling receptors in cilia of live cells . ( A ) Schematic of single molecule labeling strategy . SSTR3 or Smo were fused at the extracellular N-terminus to an acceptor peptide ( AP ) for the biotin ligase BirA . Biotinylated AP-SSTR3 and AP-Smo molecules were sparsely revealed by Alexa647-conjugated monovalent streptavidin ( mSA-Alexa647 ) added at low concentrations ( 50 pM ) to the extracellular medium . ( B ) IMCD3 cells stably expressing AP-SSTR3-GFP ( SSTR3 , pseudo-colored red , left panel ) or AP-Smo-YFP ( SMO , pseudo-colored red , right panel ) were transfected with Pericentrin-RFP ( PCNT , pseudo-colored blue ) to mark the ciliary base and BirA to biotinylate AP-SSTR3-GFP . Biotinylated SSTR3 or Smo were detected with mSA-Alexa647 ( mSA , pseudo-colored green ) . The kymograph represents the movement of a single mSA-Alexa647 labeled AP-SSTR3-GFP or AP-Smo-YFP in live cells . The tip ( T ) and the base ( B ) of the cilium are indicated . Scale bars , 2 μm ( y ) , 4 s ( x ) . ( C ) Kymographs of simultaneous live cell imaging of TagRFP . T-IFT88 ( RFP-IFT88 , IFT train ) and single molecule SSTR3 ( SSTR3:mSA-A647 ) movement in untreated cells . The mobility of ciliary SSTR3 was assessed by half-cilium FRAP ( montage of heat-maps , bottom ) . Scale bars , 2 μm ( y ) , 5 s ( x ) . ( D ) Comparison of IFT88 foci track with single SSTR3 directional tracks in untreated cells . The processive movement of mSA labeled SSTR3 ( SSTR3:mSA-A647 , red dashed line ) and IFT88 foci tracks ( RFP-IFT88 , green dashed line ) are indicated . Little overlap is observed between IFT88 foci tracks and single SSTR3 tracks in untreated cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00654 . 00310 . 7554/eLife . 00654 . 004Figure 1—figure supplement 1 . Additional kymographs . Additional kymographs representing the movement of single mSA-Alexa647 labeled AP-SSTR3-GFP ( A ) and AP-Smo-YFP ( B ) . The tip ( T ) and the base ( B ) of the cilium are indicated . Scale bars , 2 μm ( y ) , 4 s ( x ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00654 . 00410 . 7554/eLife . 00654 . 005Figure 1—figure supplement 2 . Additional dual channel kymographs . Additional kymographs of simultaneous live cell imaging of TagRFP . T-IFT88 ( RFP-IFT88 , IFT complex ) and single molecule SSTR3 ( SSTR3:mSA-A647 ) . The processive movement of mSA labeled SSTR3 ( red dashed line ) and IFT88 foci tracks ( greed dashed line ) are indicated . Scale bars , 2 μm ( y ) , 4 s ( x ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00654 . 00510 . 7554/eLife . 00654 . 006Figure 1—figure supplement 3 . Half-cilium FRAP . ( A ) Representative example of a half-cilium FRAP . ( B ) Kinetics of fluorescence recovery of photobleached region ( red curve ) , unbleached region ( green curve ) , and the total cilium ( blue curve ) . ( C ) Averaged fitted curves of SSTR3-GFP half-cilium FRAP . DOI: http://dx . doi . org/10 . 7554/eLife . 00654 . 006 To our surprise , single molecules of SSTR3 or Smo moved in a saltatory manner that comprised mostly diffusive behavior interspersed with short periods of directional movements ( Figure 1B , Figure 1—figure supplement 1; Video 1 , and Video 2 ) . By using bright foci of RFP-tagged IFT88 as markers of motor-driven IFT trains , we found that the saltatory movement of SSTR3 was clearly different from the smooth , processive movement of IFT trains in the same cilia ( Figure 1C ) . Furthermore , not only were the movements of IFT trains and single SSTR3 molecules different , their tracks rarely overlapped even when single SSTR3 molecules displayed short directional movements ( Figure 1D , and Figure 1—figure supplement 2 ) . These observations indicate that SSTR3 and Smo moved primarily by diffusion in the ciliary membrane , and that very small IFT trains might actively move single SSTR3 molecules; these IFT nanotrains are below the sensitivity limit of our imaging . 10 . 7554/eLife . 00654 . 007Video 1 . Live cell imaging of a single ciliary SSTR3 molecule . IMCD3 cells stably expressing AP-SSTR3-GFP ( pseudo-colored red ) were transfected with BirA-ER to biotinylate AP-SSTR3-GFP . Biotinylated SSTR3 was detected with mSA-Alexa647 ( mSA , pseudo-colored green ) . The tip ( T ) and the base ( B ) of the cilium are indicated . Scale bar , 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 00654 . 00710 . 7554/eLife . 00654 . 008Video 2 . Live cell imaging of a single ciliary Smo molecule . IMCD3 cells stably expressing AP-Smo-YFP ( Smo , pseudo-colored red ) were transfected with BirA-ER to biotinylate AP-Smo-YFP . Biotinylated Smo was detected with mSA-Alexa647 ( mSA , pseudo-colored green ) . The tip ( T ) and the base ( B ) of the cilium are indicated . Scale bar , 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 00654 . 008 To determine how rapidly single SSTR3 and Smo molecules explore the entire ciliary space , in spite of their saltatory movements , we assessed their mobility by fluorescence recovery after photobleaching ( FRAP ) . We had previously uncovered a diffusion barrier at the ciliary base by showing that photobleaching Smo-YFP or SSTR3-GFP in the entire cilium led to no detectable recovery from its plasma membrane pool ( Hu et al . , 2010 ) . However , photobleaching Smo-YFP ( Hu et al . , 2010 ) or SSTR3-GFP ( Figure 1C , Figure 1—figure supplement 3 , and Video 3 ) in the distal half of the cilium led to fluorescence recovery in less than 60 s at the expense of signal diffusing from the proximal , unbleached half ( Hu et al . , 2010 ) . While this experiment did not distinguish between diffusion and active transport , it showed that the saltatory movements of SSTR3 and Smo are compatible with rapid exploration of the entire ciliary space . 10 . 7554/eLife . 00654 . 009Video 3 . Representative examples of half-cilium FRAP in different conditions . The distal half of the cilia were photobleached and the recovery rate of the SSTR3-GFP fluorescent signal in the bleached region was recorded at 1 s interval . Scale bar , 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 00654 . 009 Half-cilium FRAP allowed us to calculate an apparent diffusion coefficient of 0 . 25 μm2/s for SSTR3 , which reflects a mixture of active transport and diffusion ( ‘Materials and methods’ ) . Importantly , comparing the mobility of SSTR3 in the presence or absence of saturating amounts of mSA showed that our single molecule labeling strategy did not affect SSTR3 mobility ( Figure 2A–C ) . In addition , our tagging and labeling strategy does not disrupt the signaling properties of SSTR3 since AP-SSTR3-GFP underwent agonist-induced endocytosis from the plasma membrane , a common feature of GPCRs upon activation ( Figure 2D ) . These results lead us to conclude that our single molecule imaging reflects the dynamics of functional SSTR3 molecules . 10 . 7554/eLife . 00654 . 010Figure 2 . Functionality of the AP-SSTR3-GFP fusion protein . ( A ) Saturated labeling of biotinylated AP-SSTR3-GFP ( SSTR3-GFP ) with 20 nM of mSA-Alexa647 ( mSA ) for 1 hr . Scale bar , 2 μm . ( B ) Time series montage represents the mobility of ciliary AP-SSTR3-GFP saturatedly labeled with mSA-Alexa647 after photobleaching . Scale bar , 2 μm . ( C ) The diffusion coefficients quantified from control cells vs cells with saturated staining of mSA-Alexa647 ( mSA ) . More than 10 cilia were analyzed for each condition . Error bars , SD . p>0 . 05 . ( D ) Live cell imaging of HEK293T cells expressing AP-SSTR3-GFP . Cells were treated with 10 μM Octreotide or Somatostatin immediately before imaging . DOI: http://dx . doi . org/10 . 7554/eLife . 00654 . 010 To determine the proportion of time that SSTR3 and Smo spend undergoing active transport we used a statistical analysis of instant velocities , which overcomes the intrinsic bias in manually selecting processive segments from the single molecules kymographs . The combination of motor-driven velocities and diffusive velocities found in live cells was deconvolved by effectively removing motor-driven transport , thus isolating diffusive events . Since the IFT motors kinesin II and dynein 1b require ATP to power active transport , we depleted ATP using two independent approaches . First , we added antimycin and deoxyglucose to cells to interrupt mitochondrial respiration and anaerobic glycolysis . This treatment reduced cellular ATP levels by 90% ( Figure 3C ) , indicating that some ATP generation system ( s ) remained ( e . g . , phosphocreatine ) . Nevertheless , this treatment inhibited all IFT88 foci movements ( Figure 3A ) . However , the saltatory movements of single SSTR3 molecule in the same cilia were unaffected ( Figure 3B , and Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 00654 . 011Figure 3 . Single molecule imaging in ATP-depleted cells reveals the receptor population undergoing active transport in live cells . ( A ) Kymographs of GFP-IFT88 foci movements after Antimycin A and 2-deoxyglucose ( 2DG ) treatment . Scale bar , 2 μm . ( B ) Kymographs of simultaneous live cell imaging of tagRFP . T-IFT88 ( IFT88 ) and single molecule SSTR3 ( SSTR3 ) movement before and after 40 min of Antimycin A and 2DG treatment . Scale bar , 2 μm . ( C ) ATP levels quantified by luciferin-luciferase bioluminescence assay normalized to the levels in control-treated cells . To deplete intracellular ATP , IMCD3 cells were treated with 20 μM Antimycin + 10 mM 2DG for 40 min . Each treatment was measured in triplicate . ( D ) Instant velocity distributions of single SSTR3 movements in untreated vs ATP-depleted cells . Statistical analyses show a significant difference between velocity distributions of untreated and ATP-depleted cells for both the anterograde velocities ( p=0 . 03 ) and retrograde velocities ( p=0 . 03 ) . The live cell data ( blue ) was fitted to a mixed model combining the ATP-depleted data ( green ) and an additional Gaussian distribution ( red ) , with the latter found to contribute a fraction of 26 . 6 +/− 5 . 9% ( anterograde ) and 12 . 8 +/− 3 . 1% ( retrograde ) . n>1200 . DOI: http://dx . doi . org/10 . 7554/eLife . 00654 . 01110 . 7554/eLife . 00654 . 012Figure 3—figure supplement 1 . Additional kymographs . Additional kymographs representing the movement of single mSA-Alexa647 labeled AP-SSTR3-GFP in cells treated with Antimycin A and deoxyglucose . The tip ( T ) and the base ( B ) of the cilium are indicated . Scale bars , 2 μm ( y ) , 4 s ( x ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00654 . 012 To reveal the fraction of SSTR3 molecules undergoing active transport , we conducted a differential analysis of single SSTR3 molecule movements in cilia of control vs antimycin/deoxyglucose-treated cells . Since kinesin II and dynein 1b each move at characteristic velocities ( Scholey , 2008 ) , motor-driven movements of single SSTR3 molecules should be enriched in the distribution of instant velocities along the cilium in control cells vs antimycin/deoxyglucose-treated cells ( Figure 3D ) . Differences between distributions were analyzed by performing an effective histogram subtraction ( ‘Materials and methods’ ) . Anterograde and retrograde velocities were binned separately to create histograms with an optimum bin size that generated an unbiased estimate of the distributions ( Scott , 1979 ) . Using nonlinear least-squares regression , the live cell velocity histogram was fitted to a mixed distribution consisting of fraction f of a Gaussian with unknown mean and variance and fraction ( 1−f ) of the antimycin/deoxyglucose-treated cell distribution . This , and two other ( ‘Materials and methods’ ) , unbiased statistical tests were highly significant for both anterograde and retrograde SSTR3 velocities , and indicated that 27% of anterograde and 13% of retrograde movements of SSTR3 involved active transport ( Figure 3D ) . Second , we used the cholesterol-selective detergent digitonin . which selectively permeabilizes the plasma membrane but leaves the ciliary membrane intact ( Breslow et al . ) . This results in depletion of cytoplasmic ciliary contents including motors and ATP ( Figure 4A , top panel ) . As expected , this treatment reduced the ATP concentration to undetectable levels ( Figure 4E , and Figure 4—figure supplement 1 ) . Significantly , IFT foci movements were completely inhibited , while SSTR3 molecules retained their saltatory behavior ( Figure 4A , and Figure 4—figure supplement 2A ) . Again , three unbiased statistical tests found highly significant differences in the distribution of instant velocities between untreated and digitonin-treated cells . Applying the same effective subtraction of instant velocities as done for antimycin/deoxyglucose-treated cells , we found that 22% of anterograde and 24% of retrograde movements of SSTR3 involved active transport ( Figure 4B ) . Importantly , the mean of the Gaussians representing the differences of instant velocities between untreated and digitonin-permeabilized cells aligned closely with the mean velocities of IFT foci movements ( Figure 4D ) . This strongly suggests that the active transport events detected by our statistical subtraction are powered by the IFT motors kinesin II and dynein 1b . 10 . 7554/eLife . 00654 . 013Figure 4 . Single molecule imaging in digitonin-permeabilized cells reveals the receptor population undergoing active transport in live cells . ( A ) Kymographs of simultaneous live cell imaging of TagRFP . T-IFT88 ( RFP-IFT88 , IFT trains ) and single molecule SSTR3 ( SSTR3:mSA-A647 ) movements in digitonin semi-permeabilized cells . The immobilization of ciliary SSTR3 was confirmed by half-cilium FRAP ( montage of heat-maps , bottom ) . Scale bars , 2 μm ( y ) , 5 s ( x ) . ( B ) Instant velocity distribution of single SSTR3 movements along cilia in untreated and digitonin semi-permeabilized cells . The live cell data ( blue ) was fitted to a mixed model combining the permeabilized data ( green ) and an additional Gaussian distribution ( red ) , with the latter found to contribute a fraction of 21 . 8 +/− 12 . 6% ( anterograde ) and 24 . 3 +/− 5 . 5% ( retrograde ) . n>1200 . ( C ) Instant velocity distribution of single Smo movements along cilia in untreated and digitonin semi-permeabilized cells . The live cell data ( blue ) was fitted to a mixed model combining the permeabilized data ( green ) and an additional Gaussian distribution ( red ) , with the latter found to contribute a fraction of 32 . 1 +/− 5 . 9% ( anterograde ) and 34 . 1 +/− 1 . 9% ( retrograde ) . n>1200 . ( D ) Velocity distribution of GFP-IFT88 foci movements measured from kymographs . The mean velocities are shown in the plot . n>290 . ( E ) ATP levels quantified by luciferin-luciferase bioluminescence assay normalized to the levels in control-treated cells . To deplete intracellular ATP , IMCD3 cells were permeabilized with 60 μg/ml digitonin for 7 min . Each treatment was measured in triplicate . DOI: http://dx . doi . org/10 . 7554/eLife . 00654 . 01310 . 7554/eLife . 00654 . 014Figure 4—figure supplement 1 . Measurement of ATP levels . ATP levels quantified by luciferin-luciferase bioluminescence assay normalized to the levels in control-treated cells ( ATP level [%] ) . To deplete intracellular ATP , IMCD3 cells were permeabilized with 30 μg/ml ( Dig . 1× ) or 60 μg/ml ( Dig . 2× ) digitonin for 7 min . The efficiency of digitonin permeabilization was tested by Mab414 antibody staining . The percentages of unlabeled cells ( Intact cell [%] ) was similar to the corresponding ATP levels indicating that when a cell is permeabilized with digitonin , it looses all its ATP content . DOI: http://dx . doi . org/10 . 7554/eLife . 00654 . 01410 . 7554/eLife . 00654 . 015Figure 4—figure supplement 2 . Additional kymographs . Additional kymographs representing the movement of single mSA-Alexa647 labeled AP-SSTR3-GFP ( A ) and AP-Smo-YFP ( B ) in cells permeabilized with digitonin . The tip ( T ) and the base ( B ) of the cilium are indicated . Scale bars , 2 μm ( y ) , 4 s ( x ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00654 . 01510 . 7554/eLife . 00654 . 016Figure 4—figure supplement 3 . Digitonin permeabilization does not affect the mobility of SSTR3 in cilia . ( A ) Averaged fitted curves of SSTR3-GFP half-cilium FRAP in different conditions . ( B ) The diffusion coefficients quantified from the FRAP recovery curves in ( A ) . More than 10 cilia were analyzed for each condition . Error bars , SEM . p>0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 00654 . 016 While the proportions of active anterograde transport events uncovered by digitonin permeabilization and antimycin/deoxygluocose treatments are very similar , the proportions of active retrograde movements identified by the two treatments were different . A possible interpretation is that dynein 1b retains some ability to operate under load at low ATP levels while kinesin II is fully inactive when ATP levels are reduced 10-fold . Interestingly , FRAP analysis showed that the mobility of SSTR3-GFP was similar in untreated and digitonin-treated cells ( Figure 4A , Figure 4—figure supplement 3 , and Video 3 ) . This result strongly suggests that the combination of passive diffusion and active transport ( by nanotrains , see above ) are effective at enabling membrane proteins to explore the ciliary space . To assess whether the proportion of time spent by SSTR3 in active transport vs passive diffusion was similar for other proteins , we applied our permeabilization strategy to AP-Smo-YFP expressing cells ( Figure 4—figure supplement 2B ) . Again , the three statistical tests all showed a significant difference in the distribution of instant velocities between untreated and digitonin-treated cells and the Gaussians representing active transport amounted to 34% ( anterograde ) and 32% ( retrograde ) of the entire distributions ( Figure 4C ) . While the proportions of time undergoing active transport are slightly greater for Smo than for SSTR3 , they indicate that Smo also spends the majority of its time diffusing in the ciliary membrane rather than undergoing directed IFT . An alternative strategy to inhibit active transport is to use ciliobrevin ( Figure 5A ) , which selectively inhibits dynein but not kinesins ( Firestone et al . , 2012 ) . Treatment of cells with ciliobrevin resulted in the rapid and progressive inhibition of retrograde movement of IFT88 foci ( within approximately 2–3 min; Figure 5B ) , and after 30 min anterograde movements were also inhibited ( Figure 5A ) . Inhibition of anterograde IFT88 movement by ciliobrevin was unexpected since this direction of movement involves kinesin II , which is not targeted by ciliobrevin ( Firestone et al . , 2012 ) . It is possible that dynein is involved in the delivery of IFT complexes ( including kinesin II ) from the cytoplasm to the base of the cilium for anterograde transport , and that inhibition of dynein gradually reduces the replenishment of these complexes in the cilium . Ciliobrevin did not affect single SSTR3 molecule movements in the ciliary membrane ( Figure 5A , Figure 5—figure supplement 1 and Video 3 ) ; moreover , the same statistical tests that had been performed on the data from the digitonin and the antimycin/deoxyglucose treatments ( Figure 3D , and Figure 4B , C ) failed to uncover a significant difference between the distributions of SSTR3 velocities in control and ciliobrevin-treated cells ( Figure 5C ) . FRAP analysis also showed that the mobility of SSTR3-GFP ( Figure 5A , Figure 5—figure supplement 2 , and Video 3 ) was similar in untreated and ciliobrevin-treated cells . 10 . 7554/eLife . 00654 . 017Figure 5 . Ciliobrevin treatment abolishes IFT train movement but not active transport of SSTR3 . ( A ) Kymographs of simultaneous live cell imaging of TagRFP . T-IFT88 ( RFP-IFT88 , IFT complex ) and single molecule SSTR3 ( SSTR3:mSA-A647 ) movements in ciliobrevin treated cells ( >30 min ) . The mobility of ciliary SSTR3 was confirmed by half-cilium FRAP ( montage of heat-maps , bottom ) . Scale bars , 2 μm ( y ) , 5 s ( x ) . ( B ) Early time course of GFP-IFT88 foci movements after ciliobrevin D treatment . The velocity of retrograde tracks ( blue dashed line ) progressively decreased after 3 . 5 min , while anterograde foci ( red dashed line ) movements were unaffected until 5 min and then progressively reduced . ( C ) Instant velocity distributions of single SSTR3 movements in untreated vs ciliobrevin-treated cells . Statistical analyses show no significant difference ( p>0 . 9 ) between the distributions of velocities of untreated and ciliobrevin-treated cells . n>1200 . DOI: http://dx . doi . org/10 . 7554/eLife . 00654 . 01710 . 7554/eLife . 00654 . 018Figure 5—figure supplement 1 . Additional kymographs . Additional kymographs representing the movement of single mSA-Alexa647 labeled AP-SSTR3-GFP in cells treated with ciliobrevin . The tip ( T ) and the base ( B ) of the cilium are indicated . Scale bars , 2 μm ( y ) , 4 s ( x ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00654 . 01810 . 7554/eLife . 00654 . 019Figure 5—figure supplement 2 . Ciliobrevin treatment does not affect the mobility of SSTR3 in cilia . ( A ) Averaged fitted curves of SSTR3-GFP half-cilium FRAP in different conditions . ( B ) The diffusion coefficients quantified from the FRAP recovery curves in ( A ) . More than 10 cilia were analyzed for each condition . Error bars , SEM . p>0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 00654 . 019 Given prior models of IFT-mediated protein transport ( Ou et al . , 2005; Huang et al . , 2007 ) , it is surprising that ciliobrevin inhibited motor-driven movements of fluorescent IFT88 foci , but not those of single SSTR3 molecules . This raises questions about the relationship between IFT and active transport of SSTR3 , especially since directional tracks of single SSTR3 molecules rarely overlapped with movements of large IFT88 foci in control cells ( Figure 1D , and Figure 1—figure supplement 2 ) . As noted above , it is possible that the majority of directional SSTR3 movements are mediated by IFT nanotrains that are smaller than the IFT88 foci that we can resolve , and which might be less sensitive to dynein inhibition . Although ciliobrevin has been shown to inhibit dynein in gliding assays ( Firestone et al . , 2012 ) , it is also possible that it is less potent against dynein under specific loads such as those corresponding to very small IFT nanotrains . Another possibility is that directional SSTR3 movements occurred on other motors ( Hao et al . , 2011 ) that are not inhibited by ciliobrevin . Having found that IFT inhibition by three independent methods failed to affect SSTR3 movements , we hypothesized that the coupling between single SSTR3 molecules and IFT trains is weak and transient . This predicts that the movement of IFT trains is a constitutive process largely independent of binding to cargo such as SSTR3 . Since surface-exposed membrane proteins like SSTR3 are glycosylated , multivalent lectins such as wheat germ agglutinin ( WGA ) will cross-link and immobilize them in the ciliary membrane ( Golan et al . , 1986 ) . Indeed , addition of WGA completely stopped SSTR3 movements within minutes as measured by single molecule imaging and half-cilium photobleaching ( Figure 6A , Figure 6—figure supplements 1 and 2 , and Video 3 ) . However , WGA addition had only a small , albeit statistically significant effect on IFT movements ( Figure 6B–D , and Video 4 ) . Since WGA binds to nearly all membrane proteins , we can extrapolate these results to conclude that IFT88 movements can be largely uncoupled from binding to ciliary membrane protein cargo . 10 . 7554/eLife . 00654 . 020Figure 6 . Effect of immobilizing membrane proteins on IFT train movements . ( A ) Kymographs of simultaneous live cell imaging of mCherry-IFT88 ( IFT complex ) and single molecule SSTR3 ( SSTR3:mSA-A647 ) movements in WGA treated cells ( >7 min ) . The immobilization of ciliary SSTR3 was confirmed by half-cilium FRAP ( montage of heat-maps , bottom ) . Scale bars , 2 μm ( y ) , 5 s ( x ) . ( B ) Representative kymographs of GFP-IFT88 fluorescent foci movements before WGA treatment ( left , w/o WGA ) and after 10 min of WGA treatment ( right , WGA 10 min ) . The anterograde ( red dashed lines ) and the retrograde ( blue dashed lines ) movements are indicated in the bottom panels . Scale bars , 2 μm . ( C and D ) Bar charts representing the frequency ( C ) and velocity ( D ) of GFP-IFT88 fluorescent foci movements before and after 10 min of WGA treatment in IMCD3 cells . While the differences between WGA-treated and control cells are relatively small , with the exception of the frequency of anterograde trains they are statistically significant; N . S . : p>0 . 05; **p<0 . 05 . ( E ) Schematics of ciliary membrane protein dynamics . Single molecule imaging reveals that the majority of ciliary SSTR3 ( green ) undergo free diffusion , which allows them to explore the ciliary surface efficiently . Only a small portion of ciliary SSTR3 ( red ) movement is related to IFT . The interaction between SSTR3 and IFT appears to be transient and dynamic . DOI: http://dx . doi . org/10 . 7554/eLife . 00654 . 02010 . 7554/eLife . 00654 . 021Figure 6—figure supplement 1 . Additional kymographs . Additional kymographs representing the movement of single mSA-Alexa647 labeled AP-SSTR3-GFP in cells treated with WGA . The tip ( T ) and the base ( B ) of the cilium are indicated . Scale bars , 2 μm ( y ) , 4 s ( x ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00654 . 02110 . 7554/eLife . 00654 . 022Figure 6—figure supplement 2 . WGA treatment stops the mobility of SSTR3 in cilia . Averaged fitted curves of SSTR3-GFP half-cilium FRAP in different conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 00654 . 02210 . 7554/eLife . 00654 . 023Video 4 . Live cell imaging of GFP-IFT88 fluorescent foci movement before ( −WGA ) and after 10 min of WGA treatment ( +WGA ) in IMCD3 cells . Scale bar , 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 00654 . 023 By following the movements of single membrane proteins in the primary cilium , we find that previous models of cargo loading onto IFT trains are no longer tenable . While every evidence point to IFT trains assembling at one end of the cilium and then being transported by an IFT motor all the way to the other extremity of the cilium , keeping their state of assembly intact along the way , we find instead that cargo are highly labile components of IFT trains . Our finding that immobilization of membrane proteins ( cargoes ) only minimally affected IFT train formation and movement suggests that IFT trains assemble spontaneously rather than ‘on demand’ by binding cargo; this is similar in principle to the cargo-independent assembly of clathrin-coated pits ( Cocucci et al . , 2012 ) . In our model ( Figure 6E ) , IFT trains assemble spontaneously at a given frequency at the base and tip of cilia , forming conveyor belts along the length of the cilium onto which membrane cargo can hop on and off . While this model would at first glance appear to contradict prior findings of PKD2-GFP and OSM9-GFP foci moving processively along cilia ( Ou et al . , 2005; Huang et al . , 2007 ) , it is likely that this is due to differences between ensemble and single molecule imaging . For example , it is possible that each IFT train carries a constant number of cargo molecules but individual cargo molecules continually hop on and off; ensemble imagining would not detect these individual events which we could detect here at the single molecule level . Thus membrane protein transport in cilia may be reminiscent of the slow axonal transport pathway in which most proteins synthesized in the neuronal body move at an apparent slow rate along the axon because they infrequently hop onto motors ( Scott et al . , 2011 ) . Another interpretation of our data is that proteins undergo advective movements created by the constant movements of large IFT trains up or down the cilium . The influence of fluid motion to particle movements can be measured by the Péclet number Pe = UR/D where U is the speed of fluid motion , R is the radius of the structure and D the diffusion coefficient . If Pe < 1 , diffusion outpaces the speeds of fluid motion and the contribution of fluid motion to particle transport are minimal ( Goldstein et al . , 2008 ) . In the case of the cilium , U is of the order of IFT motor speeds ( 10−4 cm/s ) , R is ∼10−5 cm and D is 2 × 10−9 cm2/s for SSTR3 and 10 × 10−9 cm2/s for a typical protein such as tubulin . The calculated values of Pe are thus 0 . 5 for SSTR3 and 0 . 1 for tubulin , thereby suggesting that passive advection plays a very minor role in the movement of typical proteins inside cilia . Moreover , the passive advection model is incompatible with our finding that IFT foci movements can be interrupted without affecting active transport of SSTR3 ( Figure 5 ) . This experiment suggests that unresolvable IFT nanotrains transport SSTR3 when cells are exposed to ciliobrevin , and that large IFT trains visualized as readily detectable foci of fluorescence may be devoid of cargo and represent recycling IFT complexes . The hypothesis that IFT nanotrains , possibly reduced to single IFT complexes bridging a cargo to a motor , represent the transporting entities is in line with nearly all other example of motor-cargo adaptors ( Hirokawa et al . , 2009 ) . An alternative hypothesis is that ciliobrevin creates an artificial situation in cilia that forces cargo to utilize IFT nanotrains that are not normally found in cilia . A resolution of these possibilities will require advanced imaging approaches that are capable of detecting single molecules of IFT complexes . In summary , our studies indicate that ciliary membrane proteins such as SSTR3 and Smo explore the ciliary space by passive diffusion rather than only by motor-driven , directional transport . Indeed , the diffusion coefficient of SSTR3 ( D= 0 . 25 μm2/s; Figure 5—figure supplement 2B ) indicates that a single SSTR3 molecule could diffuse along the length of a 4 μm cilium in less than 40 s , while IFT-mediated transport would complete the same trip in 8 s . Interestingly , the dimensions of a cilium are similar to those of a bacterium in which proteins rapidly explore the cellular space by diffusion in the absence of cytoskeletal motors ( Elowitz et al . , 1999 ) . The importance of diffusion in exploration of the ciliary space does not rule out other roles for IFT in ciliogenesis , or the possible formation of IFT scaffolds for signaling cascades ( Pan and Snell , 2002; Wang et al . , 2006 ) . Further studies of other single molecule dynamics under signaling conditions , using techniques introduced here , will be needed to address these possibilities . Mouse SSTR3-GFP ( Berbari et al . , 2008 ) ( a gift from Dr Kirk Mykytyn , Ohio State University ) was fused at its N-terminal extracellular domain to a 13 amino acid acceptor peptide ( AP ) for the biotin ligase BirA to create AP-SSTR3-GFP . AP-Smoothened-YFP was provided by Carolyn Ott and and Jennifer Lippincott-Schwartz ( NIH ) . AP-SSTR3-GFP or AP-Smoothened-YFP was inserted into pEF5B/FRT-DEST to establish stable expression in FlpIn cells . pDisplay-BirA-ER ( Howarth and Ting , 2008 ) and the streptavidin constructs , pET21a-Streptavidin-Alive and pET21a-Streptavidin-Dead ( Howarth et al . , 2006 ) were provided by Dr Alice Ting ( MIT ) through Addgene . RFP-pericentrin PACT domain ( Gillingham and Munro , 2000 ) was a gift from Dr Sean Munro ( MRC Laboratory of Molecular Biology , Cambridge ) . Mouse inner medullary collecting duct ( mIMCD3 ) cells were maintained in Dulbecco’s modified Eagle medium/F12 ( DMEM/F12; Gibco , Grand Island , NY ) supplemented with 10% fetal bovine serum ( FBS ) and 2 mM L-Glutamine at 37°C in 5% CO2 . To induce ciliogenesis , cells were cultured in DMEM/F12 supplemented with 0 . 2% FBS for 24 hr . Lipofectamine 2000 ( Invitrogen ) was used to transfect plasmids into IMCD3 cells according to the manufacturer’s protocol . To block IFT , cells were either treated with 50 μM ciliobrevin D ( a gift from Dr James Chen , Stanford University ) for 30 min , or semi-permeabilized with highly purified digitonin purchased from EMD Millipore ( Billerica , MA ) ( #300410 ) . Serum-starved IMCD3 cells on coverslips were first placed on an ice-chilled metal block and washed twice with cold Digitonin Assay Buffer ( 20 mM Hepes , pH 7 . 4 , 115 mM KOAc , 1 mM MgCl2 , 1 mM EGTA ) . For cell permeabilization , coverslips were incubated for 7 min in cold Digitonin Assay Buffer supplemented with 30 μg/ml digitonin and protease inhibitors ( 10 μg/ml each of leupeptin , pepstatin A , bestatin , aprotinin , AEBSF , and E-64 ) . After permeabilization , coverslips were washed twice with Digitonin Assay Buffer and processed for subsequent imaging assays . For WGA treatment , cells were incubated with 75 μg/ml WGA for 7 min immediately prior to imaging . To test the signaling response of AP-SSTR3-GFP , HEK293T cells were transfected with AP-SSTR3-GFP and maintained in DMEM supplemented with 10% FBS for 24 hr and them treated with 10 μM Octreotide or Somatostatin ( SST ) immediately before live cell imaging . The level of intracellular ATP was determined by luciferin-luciferase bioluminescence assay following instructions from ATP determination kit ( A22066; Molecular Probes ) with homemade buffers . IMCD3 cells stably expressing AP-SSTR3-GFP were seeded on 35-mm dishes . To deplete intracellular ATP , cells were either incubated in PBS with 20 μM Antimycin A and 10 mM 2-deoxyglucose ( 2DG ) for 40 min or permeabilized with two different concentration of digitonin ( 30 μg/ml and 60 μg/ml ) . After treatments , cells were washed twice with PBS ( for Antimycin-2DG treatment ) or Digitonin Assay Buffer ( for digitonin treatment ) and placed in ice-cold ATP buffer ( 20 mM Tris , PH7 . 5 , 0 . 5% Nonidet P-40 , 25 mM NaCl , and 2 . 5 mM EDTA ) for 5 min . Cell lysates were then collected and centrifuged at 13 , 000×g for 15 min at 4°C and protein concentration was measured using Bradford Protein Assay reagent ( Bio-Rad , Hercules , CA ) . For each reaction , 0 . 5 μg protein was used to measure the ATP level . An IMCD3 cell line stably expressing AP-SSTR3-GFP or AP-Smo-YFP was generated using the FlpIn system ( Life Technologies , Grand Island , NY ) as described previously ( Jin et al . , 2010 ) . An IMCD3 host cell line containing a single flippase ( Flp ) recombination target ( FRT ) site was transfected with pEF5B/FRT-AP-SSTR3-GFP and the Flp recombinase expression plasmid , pOG44 . 48 hr after transfection , cells were selected in DMEM/F12 supplemented with 5 μg/ml blasticidin . Single cell clones were isolated and the expression level of AP-SSTR3-GFP or AP-Smo-YFP in cilia was assessed by fluorescence microscopy . We note that AP-Smo-YFP is constitutively localized to cilia in our cell line because it is expressed at a slightly higher level than the endogenous protein . Monovalent streptavidin was purified and conjugated to Alexa647 as described previously ( Howarth et al . , 2006 ) . To singly biotinylate SSTR3 or Smo , IMCD3 cells stably expressing AP-SSTR3-GFP or AP-Smo-YFP were transfected with pDisplay/BirA-ER ( BirA-ER is a biotin ligase targeted to the ER lumen ) and RFP-Pericentrin PACT domain ( a marker of the ciliary base ) , and maintained in DMEM/F12 supplemented with 0 . 2% FBS and 10 μM biotin ( #BIO200; Avidity ) for 24 hr . Excess biotin was removed from cells by washing with PBS . Alternatively , cell surface SSTR3 or Smo can be biotinylated by incubation with 0 . 3 μM recombinant BirA ligase , 10 μM biotin and 1 mM ATP as described previously ( Howarth and Ting , 2008 ) . Cells were then incubated with 50 pM mSA-Alexa647 in phenol red-free DMEM/F12 supplemented with 0 . 2% FBS and 25 mM HEPES ( imaging medium ) on a shaker at 20 rpm at room temperature for 1 hr . Next , excess mSA-Alexa647 was removed by washing cells with PBS , and cells were then incubated in imaging medium . Cells were seeded on 25 mm diameter cover-glass ( Electron Microscopy Sciences , Hatfield , PA ) 24 hr before transfection . After transfection , cells were serum-starved for 24 hr , and then the medium was replaced with imaging medium and cells were observed with a DeltaVision system ( Applied Precision , Issaquah , WA ) . The DeltaVision system was equipped with a PlanApo 60×/1 . 40 objective lens ( Olympus , Central Valley , PA ) , a CoolSNAP HQ2 camera ( Photometrics , Tucson , AZ ) and an EMCCD camera . To image mSA-Alexa647 labeled SSTR3-GFP , excitation light from the solid state illumination module ( InsightSSI ) was reduced to 10% intensity with a neutral density filter and the Cy5 channel was exposed for 0 . 4 s . Half-cilium FRAP was performed as described previously ( Hu et al . , 2010 ) . The GFP fluorescent signal from part of the primary cilium was photobleached with the 488 nm laser from Quantifiable Laser Module ( QLM ) , DeltaVision . The recovery rate of the GFP fluorescent signal in the bleached region was recorded at 1 s interval . The mSA-Alexa647 labeled single molecule SSTR3 and Smo were tracked using the SpotTracker plugin in ImageJ ( Sage et al . , 2005 ) . The ciliary base ( labeled by Pericentrin-RFP ) was used as a reference point to distinguish between anterograde ( particle moving away from the base ) and retrograde movement ( particle moving toward the base ) .	Primary cilia are tiny protrusions from the cell surface , which have a central role in processing sensory stimuli , such as light or odorants . Cilia are also involved in mediating the response to developmental signaling molecules , including Sonic Hedgehog , and may help to convert mechanical signals into electrical or chemical ones . Primary cilia are made up of an axoneme—a core structure that consists of microtubules extending along the length of the cilium—ensheathed by a membrane that contains a number of receptor proteins . These receptor proteins travel up and down the cilium , and it is generally assumed that an active process known as intraflagellar transport is responsible for their movement . This process is mediated by motor proteins called kinesins and dyneins , which carry cargo proteins along axonemal microtubules . However , it has been difficult to study the transport of individual receptor proteins directly because they are uniformly distributed over the membranes of the cilia . Now , Ye et al . have shown that intraflagellar transport is not the most important mode of transport for membrane proteins within primary cilia . By labelling individual receptors with a fluorescent dye and then filming their movements under a microscope , Ye et al . found that the receptors generally did not show the directed , linear motion that would be expected from intraflagellar transport . Instead , much of their movement occurred through passive diffusion , with occasional short bursts of directed motion . To investigate how rapidly receptor molecules could move through the cilium in this way , Ye et al . used a technique called fluorescence recovery after photobleaching ( FRAP ) . This involves using light to bleach the fluorescent dye attached to receptor molecules in part of the cilium , and then measuring how long it takes for the fluorescence to return as a result of other labelled molecules moving into the bleached area: the shorter this time , the faster the movement of the molecules . It took less than a minute for fluorescence to be restored within a primary cilium , indicating that passive diffusion with occasional active transport can move proteins rapidly through the structure . By using drugs to inhibit intraflagellar transport , Ye et al . confirmed that the majority of membrane protein transport within primary cilia occurs via diffusion . Further studies are now required to determine whether this is also the case for other molecules that travel along cilia , and whether intraflagellar transport may have a more important role in the assembly of these structures .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2013	Single molecule imaging reveals a major role for diffusion in the exploration of ciliary space by signaling receptors
In neuronal processes , microtubules ( MTs ) provide structural support and serve as tracks for molecular motors . While it is known that neuronal MTs are more stable than MTs in non-neuronal cells , the molecular mechanisms underlying this stability are not fully understood . In this study , we used live fluorescence microscopy to show that the C . elegans CAMSAP protein PTRN-1 localizes to puncta along neuronal processes , stabilizes MT foci , and promotes MT polymerization in neurites . Electron microscopy revealed that ptrn-1 null mutants have fewer MTs and abnormal MT organization in the PLM neuron . Animals grown with a MT depolymerizing drug caused synthetic defects in neurite branching in the absence of ptrn-1 function , indicating that PTRN-1 promotes MT stability . Further , ptrn-1 null mutants exhibited aberrant neurite morphology and synaptic vesicle localization that is partially dependent on dlk-1 . Our results suggest that PTRN-1 represents an important mechanism for promoting MT stability in neurons . In neurons , microtubules ( MTs ) provide structural support , provide tracks that molecular motors use to transport cargo from the cell body to the synapses , and promote the establishment and maintenance of neuronal polarity . The MT bundles in neuronal processes , especially axons , are exceptionally stable compared to those present in most cell types ( Conde and Cáceres , 2009 ) . Many proteins bind along the side or at the plus end of neuronal MTs to promote MT stability ( Conde and Cáceres , 2009 ) . Additionally , tubulin posttranslational modifications contribute to the structure and function of neuronal MTs ( Janke and Kneussel , 2010 ) . A long-standing question is what mechanisms prevent depolymerization from the MT minus ends . MTs are polarized , cylindrical structures assembled from α/β-tubulin heterodimers . Although tubulin dimers can be added and removed from the plus end of an MT , the minus end depolymerizes continuously if not stabilized ( Mimori-Kiyosue , 2011 ) . In most cells , minus ends are anchored at the centrosome by the γ-tubulin ring complex ( γ-TuRC ) . Ninein , another minus end-binding protein , both stabilizes MTs that have been released from the centrosome and anchors MTs at centrosomal and non-centrosomal sites ( Mogensen et al . , 2000 ) . To produce the MT bundles in neurites , MTs are nucleated at the centrosome and transported into neurites by MT motor proteins ( Yu et al . , 1993; Ahmad et al . , 1998; Wang and Brown , 2002 ) . Recently , Ori-McKenney et al . ( 2012 ) showed that , in the dendritic arbor of D . melanogaster neurons , minus ends are also both nucleated and stabilized by γ-tubulin localized to Golgi outposts . Still , in at least some cell types , γ-tubulin could not be detected in neurites ( Baas and Joshi , 1992 ) . Further , the centrosome is dispensable for promoting neuronal MT function in both D . melanogaster ( Basto et al . , 2006 ) and cultured hippocampal neurons ( Stiess et al . , 2010 ) . These studies imply that additional factors stabilize the minus ends of MTs released from the centrosome and nucleate MTs in neurites . The CAMSAP family of proteins has been identified as a group of conserved , MT minus end-binding proteins ( Baines et al . , 2009 ) . Patronin , the CAMSAP homolog in D . melanogaster , promotes MT stability by protecting minus ends released from the centrosome from depolymerization by kinesin-13 MT depolymerase ( Goodwin and Vale , 2010; Wang et al . , 2013 ) . In H . sapiens epithelial cells , CAMSAP3 ( Nezha ) stabilizes MT minus ends at adherens junctions and throughout the cytosol ( Meng et al . , 2008 ) . Along with the partially redundant CAMSAP2 , CAMSAP3 promotes proper MT organization and organelle assembly ( Tanaka et al . , 2012 ) . Importantly , both Patronin and CAMSAP3 have been shown to bind the MT minus end directly in vitro ( Meng et al . , 2008; Goodwin and Vale , 2010 ) . Meng et al . purified and fluorescence-labeled the C-terminal half of CAMSAP3 and sequentially added rhodamin-labeled and rhodamin-unlabeled MTs ( Meng et al . , 2008 ) . The CAMSAP3 fragment colocalized with the end of the MT with higher rhodamine fluorescence , which indicates that it was bound to the minus end ( Meng et al . , 2008 ) . Goodwin and Vale found that purified GFP–Patronin , which was attached to a coverslip bound and anchored rhodamine-MTs by a single end ( Goodwin and Vale , 2010 ) . Further , they used MT gliding assays in which either the plus-end motor kinesin or the minus-end motor dynein were added to the purified rhodamine-MT plus GFP–Patronin to show that the Patronin-bound end of the MT was the minus end ( Goodwin and Vale , 2010 ) . Taken together , this literature suggests that the CAMSAP family of proteins plays important roles in stabilizing MTs in vivo . We tested the hypothesis that CAMSAP proteins bind and stabilize MT minus ends in neuronal processes . We used C . elegans because neurite structure and function , along with subcellular protein localization , can be readily observed in vivo . Live imaging of the C . elegans CAMSAP homolog , PTRN-1 , in cells co-labeled with fluorescence-tagged MTs indicates that PTRN-1 localizes to MT-binding puncta throughout neuronal processes . Using a combination of live imaging and electron microscopy , we implicate a role for PTRN-1 in promoting MT stability and polymerization in neurites . Finally , we show that the loss of ptrn-1 function results in defective neurite branching and mislocalization of synaptic vesicles , indicating that it has an important role in neuron morphology and function . The loss of the DLK-1 pathway , which is known to function in synapse localization and neurite morphology ( Nakata et al . , 2005; Tedeschi and Bradke , 2013 ) , partially suppresses these defects . The C . elegans genome encodes a single homolog of the CAMSAP family of MT minus end-binding proteins , PTRN-1 . ( Figure 1—figure supplement 1A ) . PTRN-1a has a conserved domain structure with previously characterized CAMSAP proteins H . sapiens CAMSAP3 and D . melanogaster Patronin , consisting of a calponin homology domain near the N-terminus , a central region with predicted coiled-coil repeats , and a C-terminal CKK domain ( Figure 1—figure supplement 1B ) ( Meng et al . , 2008; Goodwin and Vale , 2010 ) . As the other PTRN-1 isoform , PTRN-1b , lacks the CKK domain , which is the domain required for MT binding in other CAMSAP proteins , we focused on the PTRN-1a isoform ( Meng et al . , 2008; Baines et al . , 2009; Goodwin and Vale , 2010 ) . Using a fosmid expressing mCherry from the ptrn-1 promoter ( Tursun et al . , 2009 ) , we observed ptrn-1 expression in many tissues throughout development , including neurons ( Figure 1—figure supplement 1C–H ) . We examined PTRN-1a subcellular localization in neurons using fluorescence-tagged PTRN-1a . Three fluorescence-tagged PTRN-1 constructs - PTRN-1a::YFP and PTRN-1a::tdTomato , which both used C-terminal tags , and GFP::PTRN-1 , in which PTRN-1 was tagged at the N-terminus – localized to small , closely spaced puncta throughout neurites ( Figure 1—figure supplement 2 , Figure 1 ) . We focused on the PVD neuron , which elaborates a branching dendrite arbor from two primary dendrites that run laterally along the animal , as well as a single axon that extends ventrally to make presynaptic connections in the ventral nerve cord ( VNC ) , thereby providing a useful system for visualizing multiple distinct processes ( Figure 1A ) . Expressing ptrn-1a ( cDNA ) ::tdTomato in a subset of cells including the PVD neuron , we observed irregularly spaced puncta of PTRN-1a::tdTomato throughout the PVD dendrites and axon ( Figure 1B , C , Figure 1—figure supplement 3A , B ) . A similar punctate localization was observed from PTRN-1 tagged with GFP at the N-terminus or with YFP at the C-terminus . We often observed continuous PTRN-1::tdTomato fluorescence in the primary dendrite adjacent to the cell body , and fewer , farther spaced puncta in the quaternary processes ( Figure 1B and data not shown ) . 10 . 7554/eLife . 01498 . 003Figure 1 . PTRN-1 localizes to puncta throughout neurites and colocalizes with MTs . ( A ) Schematic diagram of the central region of the PVD neuron . The cell body ( blue oval ) is in the posterior half of the animal . An elaborate dendritic arbor ( blue lines ) extends from the base of the head to the posterior of the animal , and the single axon ( magenta ) is extended into the ventral nerve chord ( VNC ) . Black lines represent the outline of the animal . ( B ) PTRN-1a::tdTomato localization in the PVD neuron . The cell body is outside of the image , close to the left edge . ( C ) Confocal micrographs from 10 animals showing PTRN-1a::tdTomato localization in the PVD primary dendrite directly posterior to the cell body . ( D–F ) . Colocalization of PTRN 1a::tdTomato ( magenta ) and EMTB::GFP ( green ) in the PVD neurites ( D ) , at the sarcolemma of the body wall muscle cells ( E ) , and in the cell interior of the body wall muscle cells ( F ) . Data were acquired from wyEx5968 and wyEx6022 transgenes coexpressed in the ptrn-1 ( tm5597 ) mutant . Closed arrowhead indicates the primary dendrite , the open arrowheads indicate tertiary dendrite , and arrow points to axon of the PVD neuron ( A , B , D ) . A , anterior; V , ventral . Scale bar: 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01498 . 00310 . 7554/eLife . 01498 . 004Figure 1—figure supplement 1 . PTRN-1 is broadly expressed . ( A ) The ptrn-1 ( F35B3 . 5 ) open reading frame , which encodes ptrn-1a and ptrn-1b . The tm5597 allele contains a 604 nt deletion , resulting in Met136>Thr , Ala137>STOP . The wy560 allele contains a 65 . 3-kb deletion spanning nucleotide 16 , 983 , 396-17 , 048 , 700 of LGX . ( B ) PTRN-1a has a conserved domain structure with CAMSAP proteins . CH , calponin homology; CC , coiled-coil; CKK , calmodulin-regulated spectrin-associated CKK domain . The PTRN-1a CKK domain is the most well-conserved portion of the protein , with 64% similarity to Patronin CKK . PTRN-1b lacks the CKK domain . ( C–H ) ptrn-1 is expressed in many tissues throughout development . Fluorescence of mCherry expressed from a fosmid encoding ptrn-1::GFP::SL2::mCherry in the ptrn-1 ( tm5597 ) mutant ( SL2: trans-splice leader 2 , which causes mCherry to be transcribed as part of the ptrn-1 transcript but translated independently ) . The head ( C ) and mid-body ( E ) of an adult , the posterior region of an L4 animal ( D and F ) , and the whole body of an L1 animal ( G and H ) are shown . A , anterior; R , right; V , ventral; neu , neuron; int , intestine; pha , pharynx; DTC , distal tip cell; VNC , ventral nerve chord; PLM , the PLM touch receptor neuron; hyp , hypodermis; mus , muscle . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01498 . 00410 . 7554/eLife . 01498 . 005Figure 1—figure supplement 2 . PTRN-1 exhibits punctate localization in neuronal processes and the body wall muscle cells . ( A ) PTRN-1a::YFP expressed from the Pptrn-1 promoter in the ptrn-1 ( tm5597 ) mutant . Pictured is a young adult animal . ( B-D ) Expanded view with a subset of confocal slices of region in the solid box ( B and C ) and dashed box ( D ) from A . ( B ) A single confocal slice at the sarcolemma of the body wall muscle cell . ( C ) A single confocal slice of the interior of the same body wall muscle cell as B . ( D ) Commissures from the ventral nerve chord ( VNC ) intersecting a sublateral neuronal process; a maximum projection of ∼6 μm is shown . ( E ) GFP::PTRN-1 expressed from the Punc-86 promoter in the ptrn-1 ( tm5597 ) mutant exhibits similar localization to PTRN-1a::YFP in neurites in the ventral nerve cord . The cell body is the HSN neuron . As PTRN-1b is produced by removing an intron for which the splice sites are within two ptrn-1a exons , the gfp::ptrn-1a ( cDNA ) construct is expected to produce both GFP::PTRN-1a and GFP::PTRN1b . The comparable localization between these two constructs indicates that the ptrn-1a isoform is sufficient to achieve punctate localization in neurites . For comparison , Chalfie and Thomson used electron microscopy of MTs in the VNC to show that the average distance between MT minus ends is approximately 1 . 7 μm . ( F ) Expanded view of boxed region from E . Solid arrow head: PVD cell body , open arrow: VNC . A: anterior , R: right , V: ventral . Scale bar: 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01498 . 00510 . 7554/eLife . 01498 . 006Figure 1—figure supplement 3 . PTRN-1 localizes to puncta throughout the neurites in the PVD and PHC neurons . ( A–C ) Confocal images of 8–10 ptrn-1 ( tm5597 ) mutants expressing PTRN-1a::tdTomato in a subset of neurons , including PVD and PHC , were straightened and aligned . Regions shown are the tertiary dendrite ( posterior the cell body ) ( A ) and the axon ( B ) of the PVD neuron , as well as the dendrite of the PHC neuron ( C ) . The PHC neuron with indicated approximate region of interest on the dendrite is diagramed in Figure 3A . A , anterior; D , dorsal . Scale bar: 5 μm . ( D and E ) Schematic diagrams of animals shown in Video 1 ( D ) and Video 2 ( E ) . Gray lines outline of the animal , black lines show PVD dendrites: the filled arrowhead indicates the PVD primary dendrite , the open arrowheads point to tertiary dendrites . Pink lines are PVD axons . The PVD cell body is out of frame to the right in both images . PTRN-1::tdTomato is also expressed in the body wall muscle of these animals , marked with the dashed blue lines . A , anterior; V , ventral; scale bar: 10 μm . We could not resolve the ends of the tertiary dendrites , but there are usually gaps between the tertiary branches from each secondary branch ( see Figure 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01498 . 00610 . 7554/eLife . 01498 . 007Figure 1—figure supplement 4 . EMTB::GFP binds MTs in the PVD neuron and the body wall muscles . ( A and B ) EMTB::GFP in the PVD neuron of wild-type ( A ) and ptrn-1 ( tm5597 ) mutant ( B ) animals . Note that fluorescence intensity becomes progressively dimmer from the primary to the quaternary dendritic processes , as compared to the cytosolic GFP shown in Figure 4—figure supplement 1 . ( C–F ) EMTB::GFP fluorescence in the body wall muscle cells of wild-type ( C and E ) and ptrn-1 ( tm5597 ) mutant ( D and F ) animals . ( C and D ) Confocal slice at the sarcolemma of the body wall muscle; ( E and F ) the interior of the body wall muscle cell . DOI: http://dx . doi . org/10 . 7554/eLife . 01498 . 00710 . 7554/eLife . 01498 . 008Figure 1—figure supplement 5 . Highly expressed PTRN-1a::tdTomato binds along MT filaments . PTRN-1a::tdTomato near the membrane ( A ) and in the interior ( B ) of a body wall muscle cell in an animal carrying a highly expressing ptrn-1a::tdTomato transgene . Compare to MT localization in Figure 1—figure supplement 4 . Scale bar: 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01498 . 008 We also examined PTRN-1a::tdTomato localization in the PHC sensory neuron , which has the simple bipolar morphology more typical of C . elegans neurons ( Figure 3A ) . In the PHC dendrite , PTRN-1::tdTomato localized to puncta with spacing similar to that of the PVD neuron ( Figure 1—figure supplement 3C ) , suggestive that the PTRN-1a localization pattern in neurites is similar across different neuron classes . 10 . 7554/eLife . 01498 . 009Figure 2 . PTRN-1 stabilizes MT foci in neurons and muscles . ( A–D ) PTRN-1a::tdTomato and EMTB::GFP at the sarcolemma ( A and C ) and cell interior ( B and D ) of body wall muscle cells after acute colchicine exposure ( C and D ) or M9 control ( A and B ) . ( E and F ) PTRN-1a ( ΔCKK ) ::tdTomato and EMTB::GFP at the sarcolemma ( E ) and cell interior ( F ) of body wall muscle cells after acute colchicine exposure . ( G–H ) Localization of PTRN-1a::tdTomato and EMTB::GFP in the PVD dendrite after acute colchicine exposure ( H ) or M9 control ( G ) . ( I ) PTRN-1a ( ΔCKK ) ::tdTomato and EMTB::GFP in the PVD primary dendrite after acute colchicine exposure . All data acquired from wyEx5968 with either wyEx6022 ( A–D and G and H ) , wyEx6092 ( I ) , or wyEx6165 ( E and F ) co-expressed in bus-17 ( e2800 ) mutant animals . Scale bar: 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01498 . 00910 . 7554/eLife . 01498 . 010Figure 2—figure supplement 1 . Acute colchicine exposure changes EMTB::GFP localization in body wall muscles . ( A and B ) EMTB::GFP localization in the body wall muscle after acute colchicine treatment in the bus-17 ( e2800 ) genetic background . ( A ) Confocal slice at the sarcolemma; ( B ) the interior of the body wall muscle cell . Note that this animal , unlike those shown in Figure 2 , does not have the wyEx6022 ptrn-1::tdTomato transgene , so the EMTB::GFP puncta from wyEx5968 that remain after the acute colchicine treatment are not caused by PTRN-1::tdTomato overexpression . DOI: http://dx . doi . org/10 . 7554/eLife . 01498 . 01010 . 7554/eLife . 01498 . 011Figure 2—figure supplement 2 . PTRN-1::tdTomato colocalizes with EMTB::GFP puncta after MT depolymerization by colchicine . bus-17 ( e2800 ) mutant animals co-expressing EMTB::GFP with either PTRN-1::tdTomato or PTRN-1a ( ΔCKK ) ::tdTomato were imaged after acute colchicine treatment ( example images shown in Figure 2 ) , and the Pearson's colocalization coefficient ( PCC ) between the two fluorescent proteins was calculated for body wall muscle cells ( A ) and PVD dendrites ( B ) . In A , plotted is the average PCC calculated from maximum projection images of confocal stacks from four ( PTRN-1a::tdTomato ) or three ( PTRN-1a ( ΔCKK ) ::tdTomato ) body wall muscle cells . In B , plotted is the average PCC calculated from linescans of PVD dendrites from four animals each for both PTRN-1a::tdTomato and PTRN-1a ( ΔCKK ) ::tdTomato . ( p<0 . 05 , one-tailed students t test ) . The number above each bar indicates the p value from a one-tailed , one-sample t test comparing the calculated PCC against 0 ( no colocalization ) ( Mcdonald and Dunn , 2013 ) . In the PVD neurites , PTRN-1a ( ΔCKK ) ::tdTomato exhibited a small but significant colocalization with EMTB::GFP , which was contrary to the hypothesis that the CKK domain is necessary for MT binding and stabilization . This colocalization might be due to the presence of other MT-protecting proteins or endogenous PTRN-1 . DOI: http://dx . doi . org/10 . 7554/eLife . 01498 . 01110 . 7554/eLife . 01498 . 012Figure 2—figure supplement 3 . PTRN-1a ( ΔCKK ) exhibits punctate localization in body wall muscle cells and neurons . ( A and B ) PTRN-1a ( Δ CKK ) ::tdTomato , in which the MT-binding CKK domain has been deleted , co-expressed with EMTB::GFP in ptrn-1 ( tm5597 ) mutant animals . Images show the body wall muscle cells at the sarcolemma ( A ) and in the cell interior ( B ) . Expanded boxes highlight regions where PTRN-1 ( ΔCKK ) ::tdTomato and EMTB::GFP do not colocalize . ( C ) PTRN-1a ( ΔCKK ) ::tdTomato expressed in the ptrn-1 ( tm5597 ) mutant exhibits punctate localization in the processes of the PVD neuron . A , anterior; V , ventral . Filled arrowheads point to PVD primary dendrite , open arrowheads point to PVD tertiary dendrites . Scale bar: 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01498 . 01210 . 7554/eLife . 01498 . 013Figure 3 . Immobility of PTRN-1a::tdTomato puncta contrasts with EBP-2::GFP movements in the PVD dendrite . ( A ) Live imaging was performed on a ptrn-1 ( tm5597 ) mutant animal co-expressing EBP-2::GFP ( green , from the wyEx4828 transgene ) , which labels growing plus-end of MTs , and PTRN-1a::tdTomato ( magenta , from the wyEx6022 transgene ) along a section of the tertiary dendrite of the PVD neuron . ( B–D ) . Kymographs of EBP-2::GFP ( B ) , PTRN-1a::tdTomato ( C ) , and overlay of EBP-2::GFP ( green ) with PTRN-1a::dtTomato ( magenta ) ( D ) from a 110 s video acquired from the PVD process shown in A . Time runs top to bottom . Arrows point to start of EBP-2::GFP movements . A , anterograde; R , retrograde . Scale bar: 5 μm , ∼22 s . DOI: http://dx . doi . org/10 . 7554/eLife . 01498 . 01310 . 7554/eLife . 01498 . 014Figure 3—figure supplement 1 . Immobility of PTRN-1a::tdTomato puncta contrasts with EBP-2::GFP movements in the PVD dendrite . Replicate 2 ( A–D ) and replicate 3 ( E–H ) of experiment in Figure 2 . ( A and E ) Live imaging was performed on ptrn-1 ( tm5597 ) mutant animals coexpressing EBP-2::GFP ( green , from the wyEx4828 transgene ) , and PTRN-1a::tdTomato ( magenta , from the wyEx6022 transgene ) along a section of the tertiary dendrite of the PVD neuron . ( B–D and F–H ) . Kymographs of EBP-2::GFP ( B , F ) , PTRN- 1a::tdTomato ( C and G ) , and overlay of EBP-2::GFP ( green ) with PTRN-1a::dtTomato ( magenta ) ( D and H ) from a 110 s video acquired from the PVD process shown in A and E . ( I–K ) Live imaging was performed as above on an animal lacking the ptrn-1a::tdTomato transgene . Time runs top to bottom . A , anterograde; R , retrograde . Scale bar: 5 μm , ∼22 s . DOI: http://dx . doi . org/10 . 7554/eLife . 01498 . 014 We used live imaging to examine the dynamics of the PTRN-1::tdTomato puncta in the PVD neurites ( Video 1 , Video 2 , Figure 1—figure supplement 3D , E ) . In 40-min videos , the majority of the PTRN-1::tdTomato puncta exhibit some slow movement . The puncta can be seen dividing , appearing and growing , dissolving , and merging . The movements of each punctum are not obviously correlated with those of the surrounding puncta . 10 . 7554/eLife . 01498 . 015Video 1 . PTRN-1::tdTomato movements in the PVD neuron . First example video of a L4 animal showing 40 min with 90 s/frame . See Figure 1—figure supplement 3D for diagram of neuron morphology . DOI: http://dx . doi . org/10 . 7554/eLife . 01498 . 01510 . 7554/eLife . 01498 . 016Video 2 . PTRN-1::tdTomato movements in the PVD neuron . Second example video of a L4 animal showing 40 min with 90 s/frame . See Figure 1—figure supplement 3E for diagram of neuron morphology . DOI: http://dx . doi . org/10 . 7554/eLife . 01498 . 016 To examine whether PTRN-1 binds MTs in neurites , we co-expressed PTRN-1a::tdTomato with EMTB::GFP , the MT-binding domain of ensconsin fused to GFP ( Masson and Kreis , 1993; Bulinski and Bossler , 1994; Faire et al . , 1999 ) . EMTB::GFP , which binds dynamically along the side of MTs , has been previously used to visualize MTs in vivo ( Bulinski et al . , 2001; Lechler and Fuchs , 2007; von Dassow et al . , 2009; Wühr et al . , 2010 ) . In C . elegans neurons , it generally exhibited continuous fluorescence throughout neuronal processes . In the PVD neuron , the dendritic arbor consists of processes that branch perpendicularly from each other , getting progressively thinner with each branching event ( Albeg et al . , 2011 ) . Accordingly , EMTB::GFP fluorescence in the PVD neuron was strong in the primary dendrites but weak and sometimes discontinuous in the tertiary and quaternary dendrites , likely because these narrow neurites contain few MTs ( Figure 1D , Figure 1—figure supplement 4A , B ) ( Albeg et al . , 2011 ) . We assessed the relationship between MTs and PTRN-1 by examining the fluorescence of these two fusion proteins in the body wall muscle cells , which have larger cell bodies than neurons , providing more space in which MTs are organized . EMTB::GFP-labeled MTs were strung throughout the cytosol of these cells ( Figure 1F , Figure 1—figure supplement 4E , F ) . They also formed parallel lines along the sarcolemma , from which emanated rows of evenly-spaced puncta visible in the next confocal slice or two closer to the membrane ( slices were 0 . 4 μm apart ) ( Figure 1E , Figure 1—figure supplement 4C , D ) . These puncta appeared to be MT ends . Hence , the disorganized MT strands are anchored at the regularly spaced loci on or near the plasma membrane . The angle and spacing of the MT lines at the sarcolemma suggest that they run parallel and perhaps adjacent to the dense bodies ( the C . elegans equivalent of Z-discs ) and M-lines . This microtubule organization in the body wall muscle cells resembles the pattern observed by fluorescence-labeled ELP-1 , the C . elegans EMAP ( Echinoderm Microtubule-Associated Protein ) -like protein ( Hueston et al . , 2008 ) . In mammalian muscle fibers , MTs filaments form both a grid-like organization aligned with the Z-discs , which is dependent on dystrophin ( Prins et al . , 2009 ) , and squiggles in the cytosol with less apparent organization ( Ralston et al . , 1999 ) . Whether fused to YFP or tdTomato , PTRN-1 localized to evenly spaced puncta at the sarcolemma and to irregularly spaced puncta throughout the interior of the body wall muscle cells ( Figure 1—figure supplement 2A–C , Figure 1E , F ) . The PTRN-1a::tdTomato puncta within the muscle cytosol always co-localized with the EMTB::GFP cytosolic threads ( Figure 1F ) , confirming that PTRN-1 localizes to MTs . Further , PTRN-1a puncta colocalized with EMTB::GFP puncta at the sarcolemma ( Figure 1E ) , a finding which suggests that , like its homologs in fruitflies and humans ( Meng et al . , 2008; Goodwin and Vale , 2010; Tanaka et al . , 2012 ) , PTRN-1a localizes to MT ends . It is unclear how PTRN-1::tdTomato is localized at either the sarcolemma or in the muscle cell interior . Interestingly , in mammalian muscle , MTs are nucleated from the immobile Golgi elements strung throughout the cytoplasm ( Oddoux et al . , 2013 ) . Although CAMSAP proteins preferentially bind to the minus ends of MTs , when they are overexpressed , CAMSAPs have also been shown to bind along the side of MTs ( Meng et al . , 2008; Baines et al . , 2009; Goodwin and Vale , 2010 ) . Similarly , highly expressed PTRN-1a::tdTomato localized along the side of MTs in the body wall muscle cells ( Figure 1F ( bottom left of main panel ) , Figure 1—figure supplement 5 ) . We next sought to determine whether PTRN-1 localizes to sites where MTs are stabilized . We treated animals with the MT depolymerizing drug colchicine for 1 hr and examined the effect on EMTB::GFP localization . Because the C . elegans cuticle is largely impermeable to colchicine , we performed this experiment in the bus-17 ( e2800 ) genetic background , which has increased permeability to drugs , including colchicine ( Leung et al . , 2008; Gravato-Nobre et al . , 2005; Bounoutas et al . , 2009 ) . Acute colchicine treatment dramatically altered the distribution of EMTB::GFP such that fibers were no longer visible . Small EMTB::GFP foci remained at the cell membrane and in the cell interior , along with a haze of fluorescence throughout the cytosol ( Figure 2—figure supplement 1A , B ) . Since EMTB::GFP normally binds to the sidewalls of MTs ( Faire et al . , 1999 ) , this change in its distribution confirms that the colchicine treatment led to MT depolymerization , as expected . The localization of PTRN-1a::tdTomato appeared to be unaffected by the acute colchicine exposure , and the PTRN-1a::tdTomato puncta co-localized with the EMTB::GFP puncta both at the sarcolemma and in the cell interior ( Figure 2A–D , Figure 2—figure supplement 2 ) . These data show that , in the body wall muscle cells , the localization of PTRN-1a puncta is not dependent on MTs . They further imply that PTRN-1a localizes to sites where MTs are stabilized , even under conditions that cause complete depolymerization of all other MTs in the cell . Finally , because PTRN-1a::tdTomato and EMTB::GFP puncta exhibit a regular , repeating pattern at the sarcolemma , the fact that these rows are unaffected by the acute colchicine treatment indicates that these puncta represent sites of MT anchorage . We used acute colchicine treatment to examine whether PTRN-1a likewise localizes to sites of MT stabilization in neurons . In the PVD neuron , acute colchicine exposure caused the continuous EMTB::GFP staining to dissolve into closely spaced puncta ( Figure 2G , H ) . As in the body wall muscle cells , the PTRN-1a::tdTomato localization in the PVD neurites appeared unaffected by the MT depolymerization , and the remaining EMTB::GFP colocalized with the PTRN-1a::tdTomato puncta ( Figure 2G , H , Figure 2—figure supplement 2 ) . We interpret these data as suggestive that PTRN-1a localizes to sites where MTs are stabilized in the neurites . As previous studies have shown that the CKK domain of CAMSAP proteins is involved in MT binding ( Baines et al . , 2009; Goodwin and Vale , 2010 ) , we analyzed PTRN-1a ( ΔCKK ) ::tdTomato to determine whether PTRN-1a itself stabilizes MTs . PTRN-1a ( ΔCKK ) ::tdTomato exhibited similar localization in body wall muscle cells as full-length PTRN-1a::tdTomato , though these PTRN-1a ( ΔCKK ) ::tdTomato puncta sometimes did not colocalize with EMTB::GFP ( Figure 2—figure supplement 3A , B ) . Performing acute colchicine treatment on animals co-expressing EMTB::GFP and PTRN-1a ( ΔCKK ) ::tdTomato in the body wall muscle cells , we found that GFP-stained MT filaments were transformed into GFP puncta , but these puncta did not colocalize with PTRN-1a ( ΔCKK ) ::tdTomato ( Figure 2E , F , Figure 2—figure supplement 2 ) . There appeared to be fewer puncta in the muscle cells after the acute colchicine treatment , particularly at the sarcolemma . This may indicate that PTRN-1a ( ΔCKK ) ::tdTomato localization in the muscle cells is dependent on MTs . The EMTB::GFP foci present after acute colchicine treatment in this strain might be stabilized by endogenous PTRN-1 and/or other MT binding proteins . Finally , although PTRN-1a ( ΔCKK ) ::tdTomato localized to puncta in the PVD processes ( Figure 2—figure supplement 3C ) , after acute colchicine exposure , PTRN-1a ( ΔCKK ) ::tdTomato puncta exhibited reduced colocalization with EMTB::GFP ( Figure 2I , Figure 2—figure supplement 2 ) . Taken together , these data indicate that one of the MT ends is more stable than the rest of the MT in vivo , possibly due to end-binding proteins , and PTRN-1a::tdTomato itself stabilizes MT foci . Live imaging of EBP-2 ( EB1 ) , an MT-binding protein that specifically associates with growing plus ends , has been used to visualize MT polymerization in C . elegans neurites ( Mimori-Kiyosue et al . , 2000; Maniar et al . , 2012 ) . To examine the relationship between PTRN-1 and dynamic MT plus ends , we performed live imaging on PVD tertiary dendritic processes co-expressing EBP-2::GFP with PTRN-1a::tdTomato ( Figure 3 , Figure 3—figure supplement 1 ) . As reported for other neurons , each EBP-2::GFP punctum appeared , migrated in a single direction , and disappeared ( blue arrowheads , Figure 3B , Figure 3—figure supplement 1B , E ) . In some cases , multiple EBP-2::GFP movements emanated from the same position in the course of a video , suggestive of a local factor that promotes MT polymerization from these loci . There were also motionless EBP-2::GFP puncta that were present whether the neuron expressed the EBP-2::GFP transgene alone or with the PTRN-1a::tdTomato ( Figure 3B , Figure 3—figure supplement 1C , F , I–K ) . PTRN-1a::tdTomato , in contrast , was localized almost exclusively to immobile puncta for the duration of these 110 s videos ( Figure 3C , Figure 3—figure supplement 1C , G ) . Many of the EBP-2::GFP movements appeared to emanate from PTRN-1a::tdTomato puncta , though the close spacing of the PTRN-1a::tdTomato puncta makes quantification of this observation impracticable ( Figure 3D , Figure 3—figure supplement 1D , H ) . To investigate the requirement for PTRN-1 in neurite MT dynamics , we performed live imaging of EBP-2::GFP movements in wild-type vs ptrn-1 mutant animals . We obtained two ptrn-1 ( null ) alleles: tm5597 , which carries an intragenic deletion that introduces an early nonsense mutation , and wy560 , a 65 kb deletion that spans the entire ptrn-1 locus ( Spilker et al . , 2012 ) . Strains carrying either of the ptrn-1 ( null ) alleles exhibited grossly wild-type growth , development , and neuronal morphology ( Figure 1—figure supplement 1 , Figure 4—figure supplement 1 , and data not shown ) . As the PHC dendrite has been used previously to monitor EBP-2::GFP movements ( Yan et al . , 2013 ) , we used this system to examine EBP-2::GFP movements in the ptrn-1 mutants ( Figure 4A ) . In wild-type animals , EBP-2::GFP comets move predominantly toward the cell body ( Yan et al . , 2013 ) , consistent with the known minus-end-out polarity of MTs in dendrites ( Burton , 1985 ) . In both of the ptrn-1 mutant strains , we observed fewer total EBP-2::GFP movements than in the wild-type strain ( Figure 4B ) , but the direction of EBP-2::GFP movements was like that of wild-type ( Figure 4C ) . Expression of ptrn-1a::tdTomato in the PHC neuron of ptrn-1 ( tm5597 ) mutant animals rescued the decreased number of EBP-2::GFP movements ( Figure 4B ) , indicating that the requirement for PTRN-1 in promoting EBP-2::GFP movements is cell-autonomous . These data implicate PTRN-1 in promoting MT polymerization in the dendrite but not directly organizing MT polarity . This loss of dynamic MTs in the ptrn-1 mutants could be indicative of a reduction in the total number of MTs in the neurite , which would be suggestive of a role for PTRN-1 in MT nucleation or stabilization . These data do not quantify the stable MT population , however , so an alternate explanation for the reduction in EBP-2::GFP movements is that there is an increase in neurite MT stability in the ptrn-1 mutants . 10 . 7554/eLife . 01498 . 017Figure 4 . PTRN-1 promotes MT polymerization in neurites . ( A ) Schematic diagram of the PHC neuron . The anterior process is the axon; the posterior process is the dendrite . Live imaging was used to monitor EBP-2::GFP movements in the boxed region of the PHC dendrite . A: anterior , V: ventral . ( B and C ) Quantification of EBP-2::GFP anterograde and retrograde movements in the PHC dendrite of wild-type ( WT ) vs ptrn-1 ( tm5597 ) and ptrn-1 ( wy560 ) mutant animals , and vs the ptrn-1 ( tm5597 ) mutant carrying the Pdes-2::ptrn-1::tdTomato transgene , which is expressed in a subset of neurons as well as the body wall muscle . ( B ) Total EBP-2::GFP movements in each strain normalized against the wild-type control . ( C ) Fraction of EBP-2::GFP movements in each strain that moved in the retrograde direction . Mean ± SEM . ( n = 3 experiments , each with at least 10 animals/genotype , p<0 . 05 , p<0 . 01 , ANOVA with Bonferroni post test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01498 . 01710 . 7554/eLife . 01498 . 018Figure 4—figure supplement 1 . Neuronal morphology is grossly unaffected by loss of ptrn-1 . ( A and B ) The PVD neuron visualized by cytosolic GFP in wild-type ( A ) versus ptrn-1 ( tm5597 ) mutant ( B ) animals . ( C and D ) The PHC neuron visualized by cytosolic GFP in wild-type ( C ) versus ptrn-1 ( tm5597 ) mutant ( D ) animals . In the schematic diagrams , the PHC neuron is represented in black . ( E and F ) The DD/VD-type motorneurons visualized by cytosolic mCherry in wild-type ( E ) versus ptrn-1 ( tm5597 ) mutant ( F ) animals . ( G and H ) The PLM neuron , along with other touch receptor neurons , visualized by cytosolic GFP in wild-type ( G ) versus ptrn-1 ( tm5597 ) mutant ( H ) animals . Arrow points to PLM cell body , open arrowhead points to PLM commissure . Fluorescence in the head is from the co-injection marker . A , anterior; V , ventral . Scale bar: 50 μm ( A , B , G , H ) , 5 μm ( C–F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01498 . 018 To determine whether PTRN-1 promotes neurite MT stabilization , we examined the interaction between ptrn-1 and the MT destabilizing drug colchicine . Although the impenetrability of the C . elegans cuticle at the L4 and older stages necessitated the use of the bus-17 mutation for the acute colchicine treatment described above , wild-type animals reared from hatching in a low dose of colchicine exhibit defects in MT organization and neuronal function , indicating that the drug reaches the neurons in this longer timeframe ( Chalfie and Thomson , 1982 ) . Furthermore , rearing animals in a low dose of colchicine has been shown to suppress neurite morphology defects caused by several dominant alleles of β-tubulin mec-7 , supporting the hypothesis that these alleles caused increased MT stability ( Savage et al . , 1994; Kirszenblat et al . , 2013 ) , Therefore , this method of administering colchicine can reveal pharmacogenetic interactions between colchicine and genes that affect neuronal MT stability . Although the neurite morphology of many C . elegans neurons in the ptrn-1 ( tm5597 ) mutant all appeared grossly wild-type under normal growth conditions ( Figure 4—figure supplement 1 ) , growth in a low dose of colchicine caused dramatic ectopic sprouting from the sides of neurites in the ptrn-1 ( tm5597 ) mutant but not wild-type animals ( Figure 5 ) . In the DD/VD-type motorneurons , the cell bodies are situated in the ventral nerve cord ( VNC ) , and a single unbranched commissure per cell connects processes in the VNC with those in the dorsal nerve cord . In wild-type animals grown in the presence of colchicine , these processes generally appear to be largely morphologically normal ( Figure 5A , C ) . In the ptrn-1 ( tm5597 ) mutant grown in colchicine , in contrast , we observed ectopic branching of the commissures , as well as additional neurites sprouting from processes in the VNC ( Figure 5B , C ) . This ectopic branching could be rescued by tissue-specific ptrn-1 expression either pan-neuronally or exclusively in the DD/VD neurons ( Figure 5C ) . This synthetic interaction between colchicine and ptrn-1 is suggestive that ptrn-1 promotes MT stabilization in the DD/VD-type neurons . 10 . 7554/eLife . 01498 . 019Figure 5 . PTRN-1 supports MT stability in neurites . ( A-C ) Wild-type ( A ) and ptrn-1 ( tm5597 ) mutant ( B ) animals were grown in 0 . 13 mM colchicine to the L4 stage , and cytosolic RFP was used to visualize the DD/VD neurons . Scale bar: 10 μm . ( C ) Fraction of animals with ectopic sprouting from the DD/VD neurons , scored based on severity as described in ‘Materials and methods’ ( n = at least 80 animals/genotype , p<0 . 001 , Chi-squared test with Šidák correction ) . ( D–F ) Wild-type ( D ) and ptrn-1 ( tm5597 ) mutant ( E ) animals were grown in 0 . 035 mM colchicine to the L4 stage , and the PLM neuron was visualized with myr::GFP . Scale bar: 5 μm . ( F ) Fraction of animals exhibiting ectopic sprouting from the PLM neuron , scored based on severity ( n = at least 60 animals/genotype , p<0 . 001 , Chi-squared test with Šidák correction ) . In schematic diagrams , the light gray lines represent the outline of the animal , DD and PLM neurons are black , and other neurons ( in D and E only , one short , unbranched process near the PLM cell body of each image ) are dark gray . For tissue specific rescue , DD/VD ( Punc-47L ) , Pan-neu: pan-neuronal ( Prab-3 ) , Hyp: hypodermal ( Pdpy-7 ) , Int: intestinal ( Pvha-6 ) , neu: a subset of neurons including PLM ( Punc-86 ) . ( G ) Touch sensitivity of wild-type vs ptrn-1 ( tm5597 ) mutant animals . Mean ± SEM . ( n = 3 experiments , each with 8–12 animals/genotype , ns not significant ( p=0 . 46 ) , t test ) . ( H ) Touch sensitivity of wild-type vs ptrn-1 ( tm5597 ) mutant animals grown in the indicated concentrations of colchicine . Mean ± SEM . ( n = 2 experiments , each with 10 animals/genotype , p<0 . 001 , p<0 . 05 , t test for each drug concentration ) . ( I ) Average fluorescence of GFP expressed from the Pmec-7 ( β-tubulin ) promoter in the PLM cell body of wild-type vs ptrn-1 mutant animals . Mean ± SEM . ( n = 2 experiments , each with at least 13 animals/genotype , p<0 . 01 , t test ) . ( J and K ) Transmission electron microscopy of the PLM neuron in wild-type ( J ) and ptrn-1 ( tm5597 ) mutant ( K ) young adult animals , sectioned near the rectum . Note MTs with abnormally smaller diameters ( Asterisks ) , and a MT sheet shaped like an ‘S’ ( Arrow head ) . Scale bar: 100 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 01498 . 01910 . 7554/eLife . 01498 . 020Figure 5—figure supplement 1 . PTRN-1 protects the ALM touch receptor neuron against ectopic neurite sprouting during growth in colchicine . Wild-type ( A ) and ptrn-1 ( tm5597 ) mutant ( B ) animals were grown in 0 . 035 mM colchicine to the L4 stage , and the ALM neuron was visualized with myr::GFP . Scale bar: 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01498 . 020 Of note , we also tested the hypothesis that the colchicine-induced branching in the ptrn-1 mutant could be due to increased in vivo colchicine levels compared to wild-type . The most likely cause of such an effect would be ptrn-1-dependent defects in the hypodermis or intestine , tissues involved in drug uptake in C . elegans ( Leung et al . , 2008 ) . We therefore created transgenic ptrn-1 mutants expressing PTRN-1a in the hypodermis and intestine . This transgene had little or no effect on colchicine-induced ectopic branching in the DD/VD neurons ( Figure 5C ) , indicating that the drug–gene interaction leading to ectopic neuronal branching is unlikely to be due to increased drug permeability . Whereas most C . elegans neurons contain four to six 11-protofilament ( pf ) MTs , the six touch receptor neurons ( TRNs ) have a strikingly different MT organization . These mechanoreceptor neurons have 15-pf MTs produced from tubulin genes expressed predominantly in the TRNs ( Chalfie and Thomson , 1979; Hamelin et al . , 1992; Fukushige et al . , 1999 ) . The TRNs have 25–50 MTs per neurite cross-section in young adult animals . Both the morphology and function of these cells are particularly sensitive to perturbations in MT stability ( Chalfie and Thomson , 1982 ) . We focused on the two PLM neurons , which each elaborate both an anterior-directed process and a posterior-directed process from the cell bodies located at the base of the tail ( Figure 4—figure supplement 1G , H ) . A single commissure extends from each of the PLM anterior processes to the VNC , where PLM makes presynaptic connections with other neurons , making these anterior-directed processes axon-like . The posterior-directed processes make neither presynaptic nor postsynaptic connections . Under normal growth conditions , the morphology of the PLM neurons in the ptrn-1 ( tm5597 ) mutant was largely wild-type ( Figure 4—figure supplement 1 ) , with several more subtle defects described below . Growth in a low dose of colchicine , however , resulted in extensive ectopic sprouting from the PLM axon in the ptrn-1 ( tm5597 ) mutant but not in the wild-type strain ( Figure 5D–F ) . Similar ectopic sprouting was observed in the ALM neuron , another TRN in the anterior half of the animal ( Figure 5—figure supplement 1 ) . Tissue-specific ptrn-1 expression either in all neurons or in subset of neurons that includes PLM but not in both the hypodermis and the intestine rescued this ectopic sprouting of the touch receptor neurons ( Figure 5F ) . The PLM axon is extended during embryogenesis . As the C . elegans eggshell is impermeable to colchicine ( Bounoutas et al . , 2009 ) , the ectopic sprouting occurs after neurogenesis has been completed . Therefore , the ectopic sprouting reflects a defect in neurite maintenance rather than neurite outgrowth during development . Taken together , these data indicate that the loss of ptrn-1 enhances sensitivity to colchicine cell-autonomously in neurons containing either 11-pf or 15-pf MTs . This enhanced sensitivity to colchicine likely reflects reduced MT stability in the ptrn-1 mutant , suggesting that PTRN-1 promotes MT stabilization . TRNs mediate the behavioral response to light touch ( Chalfie , 2009 ) . We assessed the functionality of the TRNs in the absence of ptrn-1 function by quantifying light touch sensitivity in ptrn-1 ( tm5597 ) mutant vs wild-type animals . We found no significant difference in light touch response between the ptrn-1 ( tm5597 ) mutant and wild-type animals grown in the absence of colchicine , though our assay may have lacked sufficient sensitivity to parse subtle differences ( Figure 5G ) . Growing the animals in several concentrations of colchicine , however , we found that the ptrn-1 ( tm5597 ) mutant lost light touch sensitivity at a lower concentration of colchicine than the wild-type strain ( Figure 5H ) . MT destabilization has long been known to negatively regulate tubulin production ( Cleveland , 1988 ) . In the C . elegans TRNs , MT destabilization induced by genetic or pharmacological manipulations results in not only decreased levels β-tubulin mec-7 mRNA but also a general decrease in protein production , including GFP driven by the mec-7 promoter ( Savage et al . , 1994; Bounoutas et al . , 2011 ) . Similarly , the ptrn-1 ( tm5597 ) mutant exhibited decreased GFP fluorescence from a Pmec-7::gfp transgene relative to the wild-type strain ( Figure 5I ) , further implicating PTRN-1 in promoting MT stability . MT density and protofilament composition in the neurites of the TRNs have been well characterized by electron microscopy ( Chalfie and Thomson , 1979 , 1982; Cueva et al . , 2007; Cueva et al . , 2012 ) . To better understand the function of PTRN-1 , we used electron microscopy to compare the PLM MTs in the ptrn-1 ( tm5597 ) mutant to wild-type . In cross sections of the PLM neuron in wild-type animals reared at 25°C , there are 25–50 15-pf MTs and occasionally one or two 11-pf MTs ( Figure 5J ) ( Chalfie and Thomson , 1979; Cueva et al . , 2012; our unpublished data ) . We examined cross sections of the PLML/R axons from two ptrn-1 ( tm5597 ) mutant young adult animals sectioned at the rectum and found they had 13 , 15 , 2 , and 14 MTs , respectively . The majority of these MTs had the characteristically large diameter of 15-pf MTs , but several MTs had smaller diameters indicative of 11-pf MTs ( Figure 5K ) . Furthermore , in one of the PLML cross sections , we observed an irregular MT structure that persisted through three serial sections ( 150 nm ) that was ‘S’ shaped instead of circular ( Figure 5K ) . We speculate that such a structure might have formed from two circular MTs opening and then joining . The reduction in MT number found in all four cells , as well as the ‘S’ shaped MT structure in one , implicates a role for PTRN-1 in maintaining the integrity of MTs in neuronal processes . Given the requirement for ptrn-1 in MT stability in the PLM neuron , we examined the effect of ptrn-1 deficiency on PLM morphology in greater detail . Roughly 20% of ptrn-1 ( tm5597 ) mutant L4 animals exhibited defective extension of the PLM commissure ( Figure 6A–C ) . In wild-type animals , this commissure is extended during the L1 larval stage from the axon to the VNC posterior of and close to the vulva , and it is present in every L4 animal ( Figure 6A ) . In ptrn-1 ( tm5597 ) mutants with this defect , we generally observed one or several ventrally directed buds along the region of PLM axon , where the commissure is normally positioned ( Figure 6B ) . The defective PLM commissure extension observed in the ptrn-1 ( tm5597 ) mutant was fully rescued by ptrn-1a cDNA expressed in the PLM neuron , indicating that the role for PTRN-1 in commissure formation is cell-autonomous ( Figure 6C ) . 10 . 7554/eLife . 01498 . 021Figure 6 . PTRN-1 promotes synapse localization and neurite morphology in the PLM neuron . ( A and B ) myrGFP was used to visualize the PLM commissure in wild-type ( A ) and ptrn-1 ( tm5597 ) mutant ( B ) animals . Arrows point to commissure or commissure bud . ( C ) Fraction of animals with a PLM commissure connecting the axon to the ventral nerve cord . Mean ± SEM . ( n = 3 experiments , each with at least 30 animals/genotype , p<0 . 01 , **p<0 . 001 , one-way ANOVA with Bonferroni post test ) . ( D and E ) mCherry::RAB-3 at the synaptic patch of the PLM neurons of wild-type ( D ) and ptrn-1 ( tm5597 ) mutant ( E ) animals . ( F and G ) mCherry::RAB-3 in the posterior process of the PLM neurons in wild-type ( F ) and ptrn-1 ( tm5597 ) mutant ( G ) animals . ( H ) Fraction of wild-type and ptrn-1 mutant animals with visible accumulation of mCherry:RAB-3 at the synaptic patch and the posterior process of the PLM neuron . The Punc-86 promoter is expressed in a subset of neurons including the TRNs; the Pmec-3 promoter is expressed in the TRNs . Animals with two visible mCherry:RAB-3 patches in the PLM synaptic region were counted as having synaptic accumulation , and animals with one or no visible mCherry::RAB-3 patches were considered to have loss of synaptic accumulation . Mean ± SEM . ( n = 2 experiments , each with 30 animals/genotype , p<0 . 05 , *p<0 . 001 , two-way ANOVA with Bonferroni post test ) . ( I ) Fraction of wild-type and ptrn-1 ( tm5597 ) mutant animals with visible accumulation of SNB-1::GFP at the synaptic patch and the posterior process of the PLM neuron . Synaptic patch accumulation was scored as in H . Mean ± SEM . ( n = 2 experiments , each with at least 20 animals/trial . p<0 . 001 , two-way ANOVA with Bonferroni post test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01498 . 021 We next examined the localization of synaptic material in the PLM neuron . Each PLM neuron has a presynaptic specialization in the VNC at the end of the commissure , where synaptic vesicles ( SVs ) and associated proteins such as the small GTPase RAB-3 and SNB-1/synaptobrevin are localized ( Chalfie et al . , 1985; Schaefer et al . , 2000 ) . In wild-type animals , mCherry::RAB-3 localized to two patches along the VNC that correspond to the synaptic patches of the two PLM neurons ( Figure 6D , H ) . In the ptrn-1 mutants , there was an incompletely penetrant loss of mCherry::RAB-3 at the synaptic patch region , and in many ptrn-1 mutant animals , at least one of the PLM neurons had no visible accumulation of mCherry::RAB-3 at the synaptic patch ( Figure 6E , H ) . The ptrn-1 ( wy560 ) strain had a higher penetrance of this defect than the ptrn-1 ( tm5597 ) strain ( Figure 6H ) . The ptrn-1 ( wy560 ) allele is a deletion that removes not only the entire ptrn-1 locus but also seven surrounding ORFs , including the Muscleblind homolog mbl-1 . As mbl-1 has been shown to promote the accumulation of synaptic material at the presynaptic region of other C . elegans neurons ( Spilker et al . , 2012 ) , we speculate that this difference in penetrance might be due to mbl-1 deficiency in the wy560 allele . In addition to the loss of mCherry::RAB-3 from the synaptic patches , we observed a fully penetrant ectopic accumulation of mCherry::RAB-3 in the PLM posterior process in both ptrn-1 mutant strains ( Figure 6F–H ) . The localization of SNB-1::GFP in the PLM neuron of ptrn-1 ( tm5597 ) was similar to that of RAB-3::mCherry ( Figure 6I ) . These data implicate a requirement for ptrn-1 in proper SV localization . To determine whether the requirement for ptrn-1 in SV localization is cell-autonomous , we used two different promoters to drive ptrn-1a::yfp cDNA expression in the PLM neuron of the ptrn-1 mutants . Both constructs rescued the defects in mCherry::RAB-3 localization ( Figure 6H ) , indicating that PTRN-1 functions cell autonomously in the PLM neuron to promote proper SV localization . What mechanism underlies the aberrant commissure formation and SV mislocalization in the ptrn-1 mutant ? DLK-1 is a conserved mitogen-activated protein kinase kinase kinase ( MAPKKK ) that functions in a variety of situations in neurons , including neurite outgrowth , synapse development , and axon regeneration ( Tedeschi and Bradke , 2013 ) . In C . elegans , the DLK-1 pathway is required to mediate the response to MT destabilization in the PLM neuron , and it also promotes proper synapse localization ( Nakata et al . , 2005; Bounoutas et al . , 2011 ) . Indeed , hyperactivation of the DLK-1 pathway causes a defect in the PLM commissure similar to that observed in the ptrn-1 ( tm5597 ) mutant , albeit with a higher penetrance ( Grill et al . , 2007 ) . We examined PLM commissure formation and SV localization in a dlk-1 ( ju476 ) ; ptrn-1 ( tm5597 ) double mutant . In this double mutant , we found that dlk-1 completely suppressed both the commissure extension defect and the loss of mCherry::RAB-3 from the synaptic patch ( Figure 7A ) , indicating that dlk-1 is required to mediate these aspects of the ptrn-1 mutant phenotype . The PMK-3 p38 MAPK functions downstream of DLK-1 ( Nakata et al . , 2005 ) . We observed similar suppression of these ptrn-1 phenotypes in a pmk-3 ( ok169 ) ; ptrn-1 ( tm5597 ) double mutant ( Figure 7—figure supplement 1 ) , corroborating the role for the dlk-1 pathway in mediating the ptrn-1 PLM commissure formation and SV localization defects . 10 . 7554/eLife . 01498 . 022Figure 7 . Aberrant phenotype of the PLM neuron in the ptrn-1 mutant is mediated partially by the DLK-1 pathway . ( A ) The wyIs97 ( Punc-86::myrGFP , Punc-86::mCherry::rab-3 ) transgene was used to simultaneously visualize commissure formation and SV localization . Only animals with an intact PLM commissure were counted for mCherry:RAB-3 localization at the synaptic patch . Values represent mean ± SEM . ( n = 3 experiments , each with at least 30 animals/genotype , p<0 . 01 , p<0 . 001 , two-way ANOVA with Bonferroni post test ) . ( B–F ) Animals were grown in 0 . 13 mM colchicine ( C and D ) or 0 . 5 mM colchicine ( E and F ) to the L4 stage , and the PLM neuron was visualized with myr::GFP . In schematic diagrams , the PLM neurons are black , and other neurons in the image are dark gray . ( B ) Fraction of animals exhibiting ectopic sprouting from the PLM neuron , scored based on severity ( n = at least 120 animals/genotype , p<0 . 001 , Chi-squared test for each drug concentration ) . ( G and H ) Quantification of EBP-2::GFP anterograde and retrograde movements in the PHC dendrite . ( G ) Total EBP-2::GFP movements in each strain normalized against the wild-type control . ( H ) Fraction of EBP-2::GFP movements in each strain that moved in the retrograde direction . Mean ± SEM . ( n = 3 experiments , each with at least 9 animals/genotype , p<0 . 05 , ns not significant , one-way ANOVA with Bonferroni post test . The data for the ptrn-1 single mutant is the same as that shown in Figure 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01498 . 02210 . 7554/eLife . 01498 . 023Figure 7—figure supplement 1 . Loss of pmk-3 partially suppresses the aberrant phenotype of the PLM neuron in the ptrn-1 mutant . wyIs97 ( Punc-86::myrGFP , Punc-86::mCherry::rab-3 ) transgene was used to simultaneously visualize commissure formation and SV localization . Only animals with an intact PLM commissure were counted for mCherry:RAB-3 localization at the synaptic patch . Values represent mean ± SEM . ( n = 2 experiments , each with at least 30 animals/genotype , p<0 . 01 , *p<0 . 001 , two-way ANOVA with Bonferroni post test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01498 . 023 We next asked whether the synthetic interaction between ptrn-1 and colchicine that resulted in neurite sprouting from the PLM axon is also mediated through the DLK-1 pathway . Indeed , dlk-1 ( ju476 ) completely suppressed the neurite sprouting observed in ptrn-1 ( tm5597 ) during growth on colchicine ( Figure 7B–D ) . Interestingly , the wild-type strain reared in a higher dose of colchicine exhibited neurite sprouting similar to that seen in the ptrn-1 ( tm5597 ) mutant at the lower colchicine concentration , and this sprouting was likewise suppressed by dlk-1 ( ju476 ) ( Figure 7B , E , F ) . Similar effects of dlk-1 were observed on colchicine-induced neurite sprouting in the ALM neurons ( data not shown ) . DLK-1 was not required to mediate all of the abnormal phenotypes caused by ptrn-1 loss of function , however: the ectopic accumulation of mCherry::RAB-3 in the PLM posterior process was similar in the dlk-1 ( ju476 ) ; ptrn-1 ( tm5597 ) double mutant to that of the ptrn-1 ( tm5597 ) mutant strain ( Figure 7A ) . Therefore , the accumulation of SVs in the PLM posterior process is separable from the loss of SVs at the synaptic patch , and it is mediated by a mechanism other than the DLK-1 pathway . Finally , we asked whether the reduction in EBP-2::GFP movements in the ptrn-1 mutant is also dependent on dlk-1 . Interestingly , we observed an increase of roughly two-fold in EBP-2::GFP movements in the PHC dendrite in the dlk-1 ( ju476 ) single mutant relative to wild-type ( Figure 7G ) . Loss of dlk-1 had no effect on the orientation of EBP-2 movements ( Figure 7H ) . There was a trend for the dlk-1 ( ju476 ) ; ptrn-1 ( tm5597 ) double mutant to have increased EBP-2::GFP movements relative to the ptrn-1 single mutant , though this difference was not statistically significant ( Figure 7G ) . However , the dlk-1 ( ju476 ) ; ptrn-1 ( tm5597 ) double mutant exhibited fewer movements than the dlk-1 single mutant ( Figure 7G ) . This intermediate phenotype in the dlk-1 ( ju476 ) ; ptrn-1 ( tm5597 ) double mutant indicates that the DLK-1 pathway is not the only mechanism required for the ptrn-1 mutant phenotypes , and it is suggestive that DLK-1 functions partially in parallel to PTRN-1 to influence EBP-2::GFP movements in the PHC neuron . The mechanisms preventing depolymerization from the MT minus ends within neuronal processes are a long-standing mystery . Previous studies have shown that CAMSAP family proteins directly bind MT minus ends ( Meng et al . , 2008; Goodwin and Vale , 2010 ) . Our data suggest that PTRN-1 binds to MT minus ends and protects them from depolymerization in neuronal processes . Multiple lines of evidence support this conclusion . First , PTRN-1 localized to puncta throughout the neuronal processes that directly bind and stabilize MTs . Of note , electron microscopy reconstruction of neurites in the VNC showed that MT ends are staggered , with an average distance between minus ends of roughly 1 . 7 μm ( Chalfie and Thomson , 1979 ) . Second , our electron microscopy data revealed that the ptrn-1 ( tm5597 ) mutant strain had fewer total MTs and some MTs with abnormal structure in the PLM neuron . Third , the ptrn-1 mutant exhibited a pharmacogenetic enhancement with colchicine in respect to neurite morphology . Fourth , the ptrn-1 mutant exhibited a reduction in the number of neurite MT polymerization events as determined by counting EBP-2::GFP movements . Finally , the SV mislocalization and aberrant neurite branching observed in the ptrn-1 mutant were suppressed by dlk-1 , which is known to influence the effects of MT destabilization in C . elegans . The D . melanogaster CAMSAP protein Patronin localizes to MT minus ends throughout the cytoplasm in interphase S2 cells ( Goodwin and Vale , 2010 ) . Acute MT depolymerization in this system resulted in puncta of mCherry-tubulin that co-localized with the GFP–Patronin foci , similar to our findings in C . elegans neurons and muscle cells in vivo . Importantly , by allowing MT repolymerization , Goodwin and Vale established that the GFP–Patronin foci represented MT nucleation centers . It is unclear from our studies whether PTRN-1 localizes to MT nucleation sites in neurites , though the colocalization between PTRN-1 and the beginning of EBP-2 movements is suggestive that it might be . H . sapiens CAMSAP3 ( Nezha ) was originally identified as a component of epithelial cell adherens junctions , where it is anchored by cadherins and p120-catenin ( Meng et al . , 2008 ) . We showed that PTRN-1 puncta in neuronal processes were unaffected by drug-induced MT depolymerization , and PTRN-1 localization was independent of the CKK domain thought to bind MTs ( Baines et al . , 2009; Goodwin and Vale , 2010 ) . These data suggest that , like CAMSAP3 at adherens junctions , PTRN-1 is localized in an MT-independent manner . If PTRN-1 were the sole mechanism protecting minus ends in neurites , we would expect the phenotype of mutants carrying ptrn-1 null alleles to include severe defects in neuronal morphogenesis . We observed , however , relatively mild defects under standard growth conditions . Because of these data , we speculate that PTRN-1 functions in parallel with other mechanisms that promote MT stability in C . elegans neurites . These are likely to include tubulin posttranslational modifications and MT-associated proteins ( Poulain and Sobel , 2010 ) . Of particular interest , tubulin detyrosination protects MTs from depolymerization by kinesin-13 family motors in fibroblasts and neurites ( Peris et al . , 2009; Ghosh-Roy et al . , 2012 ) . Because D . melanogaster Patronin protects MTs from kinesin-13-mediated depolymerization ( Goodwin and Vale , 2010; Wang et al . , 2013 ) , these modifications are an attractive candidate for how MTs are stabilized in the absence of ptrn-1 . Defective regulation of α-tubulin acetylation , another prevalent posttranslational tubulin modification in neurites , causes abnormal neurite morphology and function in both mice and C . elegans ( Creppe et al . , 2009; Topalidou et al . , 2012 ) , though its effect on MT stability is uncertain and may be circumstance-dependent ( Janke and Kneussel , 2010 ) . In C . elegans , electron microscopy studies have shown that the loss of α-tubulin acetyltransferases causes a decrease in MT abundance , increase in MTs with irregular protofilament number , and appearance of MTs in which the protofilament lattice had opened into semicircular or sheet-like structures ( Cueva et al . , 2012; Topalidou et al . , 2012 ) . Further , the loss of the α-tubulin acetyltransferase MEC-17 in the PLM neuron causes an increase in dynamic MTs and , in older adult animals , loss of synaptic material at the synaptic region accompanied by accumulation of synaptic material in the posterior process ( Neumann and Hilliard , 2014 ) . Consistent with the notion that ptrn-1 functions in parallel with other mechanisms to promote MT stability , the ptrn-1 mutant strain grown in a low dose of colchicine exhibited aberrant neurite outgrowth . A higher dose of colchicine caused similar ectopic branching in the PLM neuron of the wild-type strain , indicating that MT destabilization is sufficient to elicit this phenotype . What is the mechanism by which MT destabilization leads to the neurite outgrowth from axons ? When collateral branches form along an axon , the budding branch and surrounding axon have fewer , shorter MTs than regions of the axon with no collateral branching ( Yu et al . , 1994 ) . Further , pharmacological or genetic manipulations that decrease MT stability have been shown to cause neurite outgrowth along the length of mature neurites ( Bray et al . , 1978; Yu et al . , 2008 ) . Perhaps the combination of ptrn-1 mutation with colchicine results in fewer , shorter MTs in these neurites , and this MT status promotes ectopic sprouting . Alternatively , the response to MT destabilization resulting from loss of PTRN-1 function may be more akin to the regeneration response to an axonal lesion , since this also results in ectopic branching and growth cone formation in motor and sensory neurons ( Hammarlund et al . , 2009; Yan et al . , 2009 ) . Taken together , our data indicate that PTRN-1 represents one of the elusive factors that stabilize the MT minus ends in neurites , promoting both the stable and dynamic MTs during development and maintenance of the nervous system . Through regulation of MTs , PTRN-1 supports proper SV localization and the balance between neurite stability and remodeling . C . elegans strains were cultured on E . coli OP50 as described ( Brenner , 1974 ) . Data were collected from L4 stage animals except where otherwise noted , and all experiments were performed at 25°C because this elevated temperature enhanced the neuronal defects in the ptrn-1 mutants ( data not shown ) , except for those shown in Figure 7B–F , which were performed at 20°C . The ptrn-1 ( tm5597 ) allele was obtained from the National Bioresource Project in Japan and backcrossed three times . The ptrn-1 ( wy560 ) was isolated from RB771 provided by the CGC ( Spilker et al . , 2012 ) , which is funded by NIH Office of Research Infrastructure Programs ( P40 OD010440 ) . The following additional strains were used in this study: N2 Bristol , TV13426 ptrn-1 ( tm5597 ) , TV15320 ptrn-1 ( tm5597 ) ; wyEx6181 [ptrn-1::gfp::SL2::mCherry in fosmid WRM0615Ab03; Podr-1::gfp] , TV14056 ptrn-1 ( wy560 ) ; wyEx5730 [Pptrn-1::ptrn-1a::yfp; Podr-1::gfp] , TV15195 ptrn-1 ( tm5597 ) ; wyEx6022 [Pdes-2::ptrn-1a::tdTomato; Podr-1::gfp] , TV15770 ptrn-1 ( tm5597 ) ; wyEx6022; wyEx5968 [Pdes-2::EMTB::gfp; Podr-1::rfp] , TV15399 ptrn-1 ( tm5997 ) ; wyEx6216 [Pmyo-3::ptrn-1a::tdTomato; rol-6 ( d ) ] , TV15773 ptrn-1 ( tm5597 ) ; wyEx6165 [Pmyo-3::ptrn-1a ( ΔCKK ) ::tdTomato; rol-6 ( d ) ]; wyEx5968 , TV15790 ptrn-1 ( tm5597 ) ; wyEx6092 [Pdes-2::ptrn-1a ( ΔCKK ) ::tdTomato; Podr-1::gfp; Pdes-2::bfp]; wyEx5968 , TV14687 wyEx5968 , TV15383 ptrn-1 ( tm5597 ) ; wyEx5968 , ptrn-1 ( tm5597 ) ; [Punc-86::gfp::ptrn-1a; Podr-1::rfp] , TV15774 bus-17 ( e2800 ) ; wyEx5968 , bus-17 ( e2800 ) ; wyEx6022; wyEx5968 , TV15772 bus-17 ( e2800 ) ; wyEx6092; wyEx5968 , TV15776 bus-17 ( e2800 ) ; wyEx6165; wyEx5968 , ptrn-1 ( tm5597 ) ; wyEx4828 [Pdes-2::ebp-2::gfp; Podr-1::gfp]; wyEx6022 , TV11781 wyEx4828 , TV14069 ptrn-1 ( tm5597 ) ; wyEx4828 , TV13424 ptrn-1 ( wy560 ) ; wyEx4828 , TV16422 ptrn-1 ( tm5597 ) ; wyIs602; wyEx4828 , TV12310 wyIs371 [ser-2prom3::myrGFP , Prab-3::mCherry , Podr-1::rfp] , TV15768 wyIs371; ptrn-1 ( tm5597 ) , TV1204 wyIs75 [Pexp-1::gfp , Punc-47L::rfp] , TV15151 wyIs75; ptrn-1 ( tm5597 ) , TV15314 wyEx6177 [pPD117 . 01 Pmec-7::gfp , Podr-1::gfp] , TV15317 ptrn-1 ( tm5597 ) ; wyEx6177 , TV1838 wyIs97 [Punc-86::myr-gfp , Punc-86::mCherry::rab-3] , TV13422 wyIs97; ptrn-1 ( tm5597 ) , TV12134 wyIs348 [Pmec-17::mCherry::rab-3 , Pmec-17::CD4::spGFP1-10] , TV13423 wyIs348; ptrn-1 ( tm5597 ) , TV15322 wyIs348; ptrn-1 ( tm5597 ) ; wyEx6023 [Punc-86::gfp::ptrn-1] , TV13430 wyIs348; ptrn-1 ( wy560 ) , TV14063 wyIs348; ptrn-1 ( wy560 ) ; wyEx5782 [Pmec-3::ptrn-1::yfp] , NM0664 jsIs37 [Pmec-7::snb-1::gfp] , TV14346 jsIs37; ptrn-1 ( tm5597 ) , TV15777 dlk-1 ( ju476 ) ; wyIs97 , TV15778 dlk-1 ( ju476 ) ; wyIs97; ptrn-1 ( tm5597 ) , TV16093 pmk-3 ( ok169 ) wyIs97 , TV16240 pmk-3 ( ok169 ) wyIs97; ptrn-1 ( tm5597 ) , TV16396 wyIs75; wyEx6577 [Podr-1::gfp] , TV16394 wyIs75; ptrn-1 ( tm5597 ) ; wyEx6575 [Podr-1::gfp] , TV16398 wyIs75; ptrn-1 ( tm5597 ) ; wyEx6579 [Prab-3::ptrn-1a , Podr-1::gfp line 1] , TV16399 wyIs75; ptrn-1 ( tm5597 ) ; wyEx6580 [Prab-3::ptrn-1a , Podr-1::gfp line 2] , TV16395 wyIs75; ptrn-1 ( tm5597 ) ; wyEx6576 [Punc-47L::ptrn-1 , Podr-1::gfp line 1] , TV16397 wyIs75; ptrn-1 ( tm5597 ) ; wyEx6578 [Punc-47L::ptrn-1 , Podr-1::gfp line 2] , wyIs75; ptrn-1 ( tm5597 ) ; [Pdpy-7::ptrn-1 , Pvha-6::ptrn-1 , Podr-1::gfp lines 1-4] , wyIs97; wyEx6582 [Podr-1::rfp] , TV16401 wyIs97; ptrn-1 ( tm5597 ) ; wyEx6582 [Podr-1::gfp] , TV16400 wyIs97; ptrn-1 ( tm5597 ) ; wyEx6589 [Prab-3::ptrn-1 , Podr-1::rfp] , TV16402 wyIs97; ptrn-1 ( tm5597 ) ; wyEx6583 [Punc-86::ptrn-1 , Podr-1::rfp] , wyIs97; ptrn-1 ( tm5597 ) ; [Pdpy-7::ptrn-1 , Pvha-6::ptrn-1 , Podr-1::rfp lines 1-4] . Expression vectors were made in the pSM vector , a derivative of pPD49 . 26 ( A Fire , unpublished data ) with added cloning sites ( S McCarroll and CI Bargmann , unpublished data ) using standard techniques . Plasmids were coinjected with markers Podr-1::gfp , Podr-1::rfp , or rol-6 ( d ) . Images were acquired with a Zeiss LSM510 confocal microscope using a Plan-Apochromat 63x/1 . 4 objective . Data were analyzed using ImageJ software . Visual inspection and quantification of the penetrance of fluorescence localization were performed using a Zeiss Axioplan 2 microscope with a 63x/1 . 4NA objective and Chroma HQ filter sets for GFP , YFP , and RFP . Animals were immobilized in 2 . 5 mM levamisol +0 . 225 mM BDM ( 2 , 3-butanedione monoxime ) or 2 mM levamisole for confocal imaging or fluorescence microscopy , respectively ( Sigma , St Louis , MO ) . Colocalization was assessed using the Colocalization Finder in ImageJ . Dynamic imaging was performed on an inverted Zeiss Azio Observer Z1 microscope using a Plan-Apochromat 63x/1 . 4 objective . L4 stage animals were anesthetized in 0 . 1% tricane +0 . 01% tetramisole for 15–30 min , then mounted on a 5% agarose pad on a slide for imaging ( Sulston and Horvitz , 1977 ) . All videos were acquired with a Quantum 512C camera . Videos of animals co-labeled with EBP-2::GFP and PQN-34a::tdTomato were 110-s videos with roughly 2 frames per second . Videos used to quantify EBP-2::GFP movements were 25-s videos with 8 frames per second . Kymographs were generated with ImageJ . Video 1 and Video 2 were acquired over 40 min with 90 s/frame . Z-stacks were acquired at each time point and maximum projections are shown . For prolonged colchicine treatment , animals were grown from eggs on NGM plates containing colchicine as described ( Chalfie and Thomson , 1982 ) . For acute colchicine treatment , L4-stage animals were soaked in a drop of 10 mM colchicine in M9 or M9 alone for 1 hr . Animals were alive and thrashing at the end of the treatment . Mechanosensory assays were performed as described ( Hobert et al . , 1999 ) . Briefly , L4 animals were tapped 10 times each with an eyelash , alternating between the anterior and posterior half of the body . The reaction was scored as either a ‘response’ , if the animal reversed direction of movement , or ‘no response’ , if it did not . The fraction of touches resulting in a response were averaged for each animal to give the ‘fraction touch sensitive’ . Animals from 3 to 6 plates/genotype were scored blind to genotype according to the following categories . For DD neurons , the categories were None: no ectopic branches or sprouting from any of the DD commissures or from the nerve cords , Mild: at least one branch or sprout from a commissure , Moderate: at least four ectopic branches , or three branches total from two different commissures , Severe: at least eight branching events , often with large growth cones projecting multiple filopodia . For the PLM axon , the categories were None: no ectopic branches , Mild: at least one branch or at least 4 large bulges along the axon , Severe: At least three branches , often many more accompanied by expanses of distended axon in the region containing the ectopic branches . For the tissue specific rescue experiments , four independent lines were initially assessed per each expression construct . We observed no rescue from any of the intestine-plus-hypodermis lines , and so all four lines were used for the experiments shown in Figure 5C , F . For the lines expressing ptrn-1 in the neurons , we observed some lines provided stronger rescue than others , likely due to differences in expression level or mosaicism of the transgene . We therefore included the 1–3 lines with the most rescue per construct in the experiments shown in Figure 5C , F . Young adult wild-type N2 and ptrn-1 ( tm5597 ) animals were prepared as described ( Cueva et al . , 2012 ) . Briefly , animals were frozen in an EMPACT2 high-pressure freezer system , and a Leica AFS freeze substitution apparatus ( Vienna , Austria ) was used to preserve in 2% glutaraldehyde plus 1% osmium tetroxide and embed in Epon/Araldite . A Leica Ultracut S microtome equipped with a diamond knife was used to cut 50-nm serial sections , which were collected on Formvar-coated copper slot grids . The grids were poststained to enhance contrast in 3 . 5% uranyl acetate ( 30 s ) and Reynold’s lead citrate preparation ( 3 min ) . The grids were imaged on a transmission electron microscope ( JEOL TEM 1230 , Tokyo , Japan ) , and images were acquired with an 11 megapixel bottom-mounted cooled CCD camera ( Orius SC1000 , Gatan , Pleasanton , CA ) .	Microtubules are tiny tubular structures made from many copies of proteins called tubulins . Microtubules have a number of important roles inside cells: they are part of the cytoskeleton that provides structural support for the cell; they help to pull chromosomes apart during cell division; and they guide the trafficking of proteins and molecules around inside the cell . Most microtubules are relatively unstable , undergoing continuous dis-assembly and re-assembly in response to the needs of the cell . The microtubules in the branches of nerve cells are an exception , remaining relatively stable over time . Now Richardson et al . and , independently , Marcette et al . , have shown that a protein called PTRN-1 has an important role in stabilizing the microtubules in the nerve cells of nematode worms . By tagging the PTRN-1 proteins with fluorescent molecules , Richardson et al . were able to show that these proteins were present along the length of the microtubules within the nerve cells . Further work showed that the PTRN-1 proteins stabilize the microtubule filaments within the branches of these nerve cells and also hold them in position . Richardson et al . also found that worms that had been genetically modified to prevent them from producing PTRN-1 failed to traffic certain molecules to the synapses between nerve cells . Moreover , these mutants also had problems with the branching of their nerve cells; however , these defects were relatively mild , which suggests that other molecules and proteins act in parallel with PTRN-1 to stabilize microtubules in nerve cells . Further work should be able to identify these factors and elucidate how they work together to stabilize the microtubules in nerve cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "neuroscience" ]	2014	PTRN-1, a microtubule minus end-binding CAMSAP homolog, promotes microtubule function in Caenorhabditis elegans neurons
Proteins are necessary for cellular growth . Concurrently , however , protein production has high energetic demands associated with transcription and translation . Here , we propose that activity of molecular chaperones shape protein burden , that is the fitness costs associated with expression of unneeded proteins . To test this hypothesis , we performed a genome-wide genetic interaction screen in baker's yeast . Impairment of transcription , translation , and protein folding rendered cells hypersensitive to protein burden . Specifically , deletion of specific regulators of the Hsp70-associated chaperone network increased protein burden . In agreement with expectation , temperature stress , increased mistranslation and a chemical misfolding agent all substantially enhanced protein burden . Finally , unneeded protein perturbed interactions between key components of the Hsp70-Hsp90 network involved in folding of native proteins . We conclude that specific chaperones contribute to protein burden . Our work indicates that by minimizing the damaging impact of gratuitous protein overproduction , chaperones enable tolerance to massive changes in genomic expression . Optimal allocation of cellular resources is a central concept in cell biology ( Basan et al . , 2015; Hui et al . , 2015 ) . Protein biosynthesis consumes a huge amount of energy: an estimated 30–50% of the energy consumption of dividing cells is dedicated to translation of the proteome ( Buttgereit and Brand , 1995; Russell and Cook , 1995 ) . Therefore , surplus protein production incurs a substantial fitness cost . As the ratio of unneeded protein reaches 30% of total protein in bacteria , ribosomes are destructed and growth is completely inhibited ( Dong et al . , 1995 ) . Protein burdens ( i . e . protein overexpression costs ) are most relevant shortly after an environmental change , and are subsequently reduced once the translation has adjusted to their novel steady-state level ( Shachrai et al . , 2010 ) . Deciphering the key molecular mechanisms that shape protein burden is an important challenge for systems biology . Moreover , this problem has biotechnological relevance as well . Protein engineering efforts towards microbial production of a single heterologous protein are often problematic , as full induction of the engineered constructs frequently yields bacteria with limited or no growth ( Kurland and Dong , 1996 ) . Gene expression costs are frequently not due to the detrimental activity of unnecessary proteins , as reduced viability was observed with the overexpression of proteins with no apparent cellular activities ( Andrews and Hegeman , 1976; Dong et al . , 1995; Kurland and Dong , 1996; Stoebel et al . , 2008; Scott et al . , 2010 ) . Most notably , a recent systematic study in baker’s yeast ( Saccharomyces cerevisiae ) measured the copy number limit of gene overexpression across all protein coding genes ( Makanae et al . , 2013 ) . Dosage sensitive genes were generally highly expressed , and replacement of the open reading frame of these genes with a green fluorescent protein ( GFP ) left the fitness cost largely unaltered . Studies in bacteria ( Stoebel et al . , 2008 ) and yeast ( Kafri et al . , 2016 ) demonstrated that growth impairment results from the process of protein production and not due to accumulating the unneeded protein product per se . Protein production of an unneeded protein consumes nutrients and has a high energetic demand . Associated costs may arise at the level of transcription due to waste of nucleotides incorporate into RNA or occupation of RNA polymerases . Translation of the unneeded proteins may be especially costly , as it wastes amino acids , charged tRNAs and occupies ribosomes . It has been shown that these two major limiting factors of protein production vary across environments , depending on the availability of nutrients ( Kafri et al . , 2016 ) . Transcription dominates protein burden in low phosphate , while translation dominates costs in low nitrogen conditions . Hypothetically , unneeded proteins may also overload the cellular systems involved in protein folding and degradation . Yet , the role of chaperone networks in contributing to protein burden has remained unexplored . In this work , we show that accumulation of an unneeded protein in yeast ( S . cerevisiae ) has a relatively mild impact on fitness when nutrients are in excess and no internal or external stresses are present . However , impairment of specific molecular chaperones rendered yeast cells sensitive to gratuitous protein overproduction . Recent works showed that the fitness costs associated with expressing unneeded fluorescent proteins do not result from protein toxicity or impaired metabolic processes , indicating that it is the outcome of a limitation in the protein production process itself ( Makanae et al . , 2013; Kafri et al . , 2016 ) . In this paper , we employ yEVenus , a rapidly folding and non-toxic YFP ( yellow fluorescent protein ) variant ( Sheff and Thorn , 2004 ) to study protein burden . Using this protein has several advantages for our study: the amino acid composition of yEVenus and the yeast proteome are highly similar to each other ( Pearson’s correlation , r = 0 . 6477 , p<0 . 01 ) and it is codon optimized specifically for yeast studies . Accordingly , toxicity of yEVenus due to misfolding is negligible . We expressed yEVenus in S . cerevisiae from single , low and high-copy-number plasmids , respectively , ( Gietz et al . , 1988 ) all of which are under the control of a strong constitutive promoter ( pHSC82 , see Materials and methods ) . The control strain carried the same vector backbone without the yEVenus open reading frame . Fitness of each genotype was determined by measuring colony size on synthetic selection medium agar plates ( for further details , see Materials and methods ) . Cost is defined as the reduction in fitness of yEVenus overexpressing genotype relative to fitness of control cells in the same synthetic drop-out medium . When expressed from a single copy plasmid , yEVenus had no detectable fitness cost , while it caused a small , but significant 2 . 5% fitness decline expressed from a high-copy ( 2 µ ) plasmid ( Figure 1A ) . A denaturing polyacrylamide gel electrophoresis analysis ( PAGE ) indicated that when expressed from the high-copy plasmid , yEVenus constitutes ~3 . 7% of the total cellular proteome ( Figure 1B ) . The above results indicate that accumulation of an unneeded protein in the cell has a relatively mild impact on fitness when nutrients are in excess and no internal or external stresses are present . However , such robustness to protein burden may be restricted to certain conditions: many genetic and environmental factors could potentially shape the associated fitness costs . To identify genes modulating protein burden , we performed a genetic interaction screen using the synthetic genetic array ( SGA ) approach ( Tong and Boone , 2006 ) with the query strain carrying the yEVenus multi-copy plasmid . The screen involved construction of high-density arrays of double mutants by crossing the query mutation ( yEVenus overexpression plasmid ) against an array of ~5000 viable null mutants . We simultaneously measured yEVenus fluorescence intensity and fitness in all genotypes studied . Using a robotized replicating system , fitness was estimated by measuring colony size on solid agar media . Digital images were processed to calculate colony sizes , and potential systematic biases in colony growth were eliminated ( see Materials and methods ) . Deviation of the double-mutant fitness from the product of the corresponding single-mutant fitness values was used to assess genetic interaction scores ( ε , Figure 1C , Supplementary file 1 ) . Biomass-normalized fluorescence level had no major impact on the distribution of genetic interaction scores ( Figure 1D ) . This pattern was not due to any major deviation from wild type cell size ( Figure 1—figure supplement 1A ) . This indicates that genetic interactions reflect a change in the fitness cost , but not in the extent of protein overexpression . As the aim of this study was the identification of genes that mitigate the fitness costs of yEVenus overexpression , we focused on negative genetic interactions ( ε < 0 ) , i . e . when the double mutant has a lower fitness than would be expected from the product of the single-mutant fitness values . At an ε = - 0 . 05 cutoff value ( and using a p<0 . 05 cutoff based on bootstrap analysis ) , 184 genes showed such interactions with yEVenus . By definition , lack of these genes substantially increased the fitness cost of yEVenus overexpression ( Figure 1E ) . A functional enrichment analysis revealed that these genes are preferentially involved in translation , transcriptional control ( e . g . transcription termination and elongation ) , mitochondria-related processes , and protein folding ( Table 1 , Figure 1—figure supplement 1B ) . Remarkably , deletion of genes encoding specific chaperones caused a 2–4 fold increment in the fitness costs of yEVenus overexpression ( Figure 1E ) . Enrichment of the above functional categories was not found in the set of genes showing positive genetic interactions with yEVenus overexpression . It is worth noting however a specific case , where positive genetic interaction was especially strong . Deletion of RPI1 , a specific repressor of the Ras-cAMP pathway removed protein burden ( Supplementary file 1 , Figure 1—figure supplement 1C ) . The underlying molecular mechanisms need further investigation . Protein synthesis is frequently limited by the availability of free ribosomes ( Vind et al . , 1993 ) . Therefore , excess proteins occupy free ribosomes , which could be better used for the translation of native proteins . Therefore , impairment of genes involved in translation should increase protein burden . We investigated this issue further by measuring fitness in the presence of a translation inhibitor chemical agent . Cycloheximide binds the ribosome and inhibits eEF2 mediated translocation during translation ( Obrig et al . , 1971 ) . In agreement with expectation , partial inhibition of translation elongation by cycloheximide treatment elevated protein burden ( Figure 2A ) . Similarly , inactivation of genes involved in transcriptional elongation ( HPR1 , DST1 , CDC73 , ELP3 ) significantly increased protein burden . To validate this result , we tested the effect of a transcriptional elongation inhibitor on protein burden . Mycophenolic acid interferes with nucleotide biosynthesis ( Costa and Arndt , 2000 ) , through inhibiting IMP dehydrogenase ( IMPDH ) . It thereby reduces the endogenous GTP/UTP and stalls RNA polymerases . Treatment of cells with sub-inhibitory concentration of this chemical agent significantly enlarged protein burden ( Figure 2B ) . Another source of protein burden may arise due to wastes of cellular resources , including ATP and amino acids needed for protein synthesis . Indeed , inactivation of amino acid metabolism genes ( AAT2 , BAT2 , CYS3 , PRS3 , LEU3 ) influenced protein burden ( Supplementary file 1 ) , suggesting that protein burden depends on the availability of amino acids in the environment . It was indeed so: depletion of amino acids in the growth medium increased the fitness cost ( Figure 2C ) . Moreover , genes with mitochondria-related functions , including mitochondrial translation ( e . g . MRPS9 , MRPL22 ) , mitochondrial DNA replication and growth ( e . g . MMM1 ) , mitochondrial distribution and morphology ( e . g . MDM38 ) are on the gene list identified by the SGA analysis ( Supplementary file 1 ) . Taken together , results of genetic and chemical perturbations of specific cellular subsystems demonstrate that impairment of transcription , translation and amino acid availability increase protein burden . Finally , one may argue that growth rate reduction per se - irrespective of the exact nature of the environmental or genetic perturbation - may imply elevated protein burden upon overexpression . However , this is unlikely to be so , for three reasons . First , the functional roles of genes that showed genetic interactions were far from being random ( Table 1 ) . Second , and more generally , the correlation between the fitness of the deletion strains and the strength of the genetic interaction was very weak ( Figure 1F ) . Finally , despite major differences in growth rates of yeast grown on glucose , galactose or raffinose as sole carbon sources , the relative fitness costs of protein burden remained unchanged ( Figure 2D ) . The genetic interaction screen revealed that molecular chaperones are overrepresented in the list of genes that influence protein burden . Most notably , the list includes several members of the Hsp40-70-110 complex ( FES1 , SSE1 and YDJ1 ) , and an Hsp70-90 scaffold protein ( STI1 ) . These Hsp70-associated proteins are functionally highly related ( Rizzolo et al . , 2017 ) , and all play critical roles in the ATPase activation and the nucleotide exchange regulation of the Hsp70 class Ssa chaperones . Accordingly , impairment of these proteins decreases the activity of Ssa chaperones and thereby perturbs the recognition and clearance of misfolded proteins . As a consequence , aggregated proteins accumulate in the cell ( Mayer , 2013; Clerico et al . , 2015 ) . Based on these findings we hypothesized that molecular chaperones have a critical role in buffering protein burden . Several further observations support the hypothesis . First , we tested the impact of temperature stress on protein burden , not least because genes ( e . g . CPR7 , YDJ1 ) involved in the GO term ‘response to heat’ were on the list of negative genetic interactions . Subjecting yeast cells to high temperature causes a severe proteotoxic stress as it induces protein misfolding of nascent proteins and perturbs proteome homeostasis ( Trotter et al . , 2002 ) . As expected , protein burden significantly increased with rising temperature ( Figure 3A ) . Reassuringly , these results are insensitive to the exact promoter employed for the expressional control of yEVenus ( Figure 3—figure supplement 1A , Figure 3—figure supplement 1B , Figure 3—figure supplement 1C , Supplementary file 2 ) . Second , as mistranslation during protein synthesis promotes misfolding and protein aggregation ( Lee et al . , 2006; Yang et al . , 2010; Paredes et al . , 2012 ) , reduction of translation fidelity should also exacerbate the fitness deficit caused by protein overproduction . Reassuringly , CTK2 and CTK3 , two genes involved in controlling the fidelity of translation elongation ( Röther and Strässer , 2007 ) were on the list of genes showing negative genetic interaction with yEVenus overexpression ( Supplementary file 1 ) . Third , induction of protein misfolding by a chemical agent enhanced protein burden . We studied the cellular response to misfolded proteins generated by azetidine-2-carboxylic acid ( AZC ) stress ( Shichiri et al . , 2001 ) . AZC is a toxic analog of proline , and incorporation of this chemical agent into proteins causes misfolding ( Trotter et al . , 2002; Albanèse et al . , 2006 ) . Application of sub-lethal dosages of AZC elevated the fitness costs associated with yEVenus overproduction ( Figure 3B ) . The fourth piece of evidence comes from monitoring cellular aggregation . An established method ( Kaganovich et al . , 2008 ) was utilized to measure the misfolding propensity of a fluorescently-tagged reporter protein ( VHL-mCherry ) . Active quality-control machinery in the wild type yeast prevents misfolding of the reporter protein , leading to its uniform distribution in the cell . However , when the protein folding machinery is impaired or becomes overloaded , the reporter protein misfolds and becomes spatially sequestered . As the fluorescent tag of the reporter protein remains fully functional , protein aggregation spots within the cell become easily visible as fluorescent foci ( Kaganovich et al . , 2008 ) . In wild type cells , protein misfolding propensity did not increase significantly upon protein burden ( Figure 3C ) . This is in line with expectation , as the fitness cost of protein overexpression in wild type was only around 2 . 5% ( Figure 1A ) . The situation was very different when genotypes impaired in protein folding ( Δfes1 , Δsse1 , Δsti1 , Δydj1 , Δpfd1 , Δgim5 , Δcpr7 ) were considered , all of which showed negative genetic interactions with yEVenus overexpression . In these genotypes , protein burden elevated the propensity of protein misfolding ( Figure 3C , Figure 3—figure supplement 1D ) . The above results indicate a crucial role of the Hsp70-associated molecular chaperones in mitigating protein burden . Why should this be so ? One possibility is that the unneeded proteins bind to key regulators of the Hsp70-associated chaperones which otherwise would be used to navigate folding of native proteins within the cell . To investigate the feasibility of this idea , we performed a GFP co-immunoprecipitation ( co-IP ) assay to identify weak in vivo physical interactions between yEVenus and native cellular proteins . In order to extract cellular proteins without disturbing physical interactions , we used an established protocol ( Visweswaraiah et al . , 2011 ) specifically designed for the identification of weak protein-protein interactions . Total protein extracts from mid-exponential growth phase were immunopurified ( IP ) using anti-GFP antibody coupled magnetic beads and the IP-purified proteins were then subjected to LC-MS/MS analysis ( see Materials and methods ) . Relative abundance of individual proteins in the samples was estimated by retrieving peptide counts of the individual proteins . After applying several filtering steps ( see Materials and methods ) , we identified 34 proteins that bind to yEVenus ( Supplementary file 3 ) . Altogether , the list of putative interacting partners includes five proteins with chaperone-related functions ( Supplementary file 3 ) . Notably , Sti1p and Ydj1p not only binds to yEVenus , but were identified also in the genetic interaction assay . Both proteins are involved in the activation of Ssa proteins , key components of the Hsp70 complex . The above results indicate that as a globular protein , yEVenus binds weakly , but significantly to certain molecular chaperones and to Sti1p in particular ( Supplementary file 3 ) . This raises the possibility that the protein burden is linked to perturbation of the native physical interactions of Sti1p by yEVenus . To investigate this issue , we performed a reciprocal co-IP assay with the aim to identify quantitative changes in physical interactions of Sti1p in response to protein burden . Accordingly , we used a strain that expresses a C-terminally epitope-tagged Sti1p ( Sti1p-3xFLAG ) and investigated it both under low and high protein burden . Total protein extracts from mid-exponential growth phase were immunopurified ( IP ) using anti-FLAG antibody coupled beads and the IP-purified proteins were then subjected to LC-MS/MS analysis , as previously ( Materials and methods ) . The analysis focused on 18 proteins , all of which have been described to physically interact with Sti1p in prior studies ( Cherry et al . , 2012 ) . Our method confirmed half of these 18 protein interactions under low protein burden , that is when yEVenus was expressed from a single-copy plasmid ( Figure 3D , Supplementary file 4 ) . Remarkably , we observed a significant drop in protein-binding affinity of Sti1p with as many as 8 out of the nine detected interaction partners under high protein burden ( Figure 3D , Supplementary file 4 ) . Most notably , protein-binding affinity decreased by 70% , 75% and 55% in the cases of Ssa1 , Ssa2p and Hsp90p , respectively . This is all the more significant , as these proteins are exceptionally important and well-characterized interaction partners of Sti1p ( Chen and Smith , 1998; Song and Masison , 2005; Balchin et al . , 2016 ) . Finally , protein-binding between yEVenus and Sti1p was detectable under high protein burden only ( Supplementary file 4 ) . We speculate that promiscuous binding of Sti1p with certain globular proteins ( such as yEVenus ) has no functional consequences unless the cellular dosage of the partner protein exceeds a critical threshold . Collectively , these data suggest that protein burden promotes a partial disassociation of interaction partners from Sti1p , putatively leading to partial disassociation of the Hsp70-Hsp90 chaperone complex . Our work demonstrates that even gross accumulation of an unneeded gratuitous protein in the cell has a relatively mild impact on fitness when no internal or external stresses are present ( Figure 1A ) . However , such robustness to protein burden was restricted to specific conditions only . We explored the molecular mechanisms underlying robustness to protein overproduction . Our main findings are as follows . First , deletion of genes involved in translation , transcriptional control , and mitochondria-related processes rendered yeast cells hypersensitive to protein overexpression . Our observation that translational and transcriptional perturbations modulate protein burden was validated further by chemical and environmental stress screens , and is also consistent with prior studies ( Kafri et al . , 2016 ) . Therefore , protein burden varied substantially across genetic backgrounds and environmental stresses . We note that mutants with impaired mitochondria exhibit reduced respiratory growth , and therefore they have to rely on less efficient modes of ATP production . However , beyond ATP production , mitochondria are involved in the synthesis of certain amino acids as well ( Ahn and Metallo , 2015; Zong et al . , 2016 ) . Therefore , future works should elucidate the exact molecular mechanisms underlying the elevated protein burden in cells deficient in mitochondrial functions . Second , prior studies suggested that expression of an unneeded protein effectively decreases the fraction of proteome allocable to ribosomes and useful biosynthetic proteins , thereby causing a growth defect ( Scott et al . , 2010 ) . In principle , mutations could therefore modulate protein burden by simply increasing the proteome fraction of the unneeded protein . However , the fractional contribution of yEVenus to the total proteome was not elevated in gene knock-out strains ( Figure 1D , Figure 1—figure supplement 1A ) . This indicates that allocation models that rely on transcription and translation only cannot fully account for protein burden . Third , and most significantly , an interacting chaperone network shapes protein burden ( Figure 4 ) . The Hsp70 complex is a key player in the maintenance of normal proteostasis . The soluble Ssa proteins ( members of the Hsp70 family ) recognize and associate transiently with exposed hydrophobic patches of misfolded proteins in the cytosol and prevent protein aggregation ( Mayer , 2013; Clerico et al . , 2015 ) . Deletion of specific activators ( YDJ1 , STI1 , FES1 or SSE1 ) of Ssa proteins substantially elevated protein burden , and resulted in protein aggregation . Indeed , Ssa protein’s capacity to bind and release client proteins heavily depends on these activators ( Wegele et al . , 2003 ) . In particular , the nucleotide exchange factors ( Sse1p and Fes1p ) are responsible for client-release and thereby support the refolding or the proteasomal degradation of misfolded proteins ( Gowda et al . , 2013 ) . It is worth noting that due to partial functional redundancy of Ssa proteins ( Hasin et al . , 2014 ) , the corresponding SSA genes did not emerge in the screen . In agreement with expectation , temperature stress , elevated mistranslation rate and a chemical misfolding agent all substantially enhanced protein burden . We conclude that molecular chaperones have an important role in buffering protein burden . Finally , we found evidence that yEVenus - a typical , globular fluorescent protein binds to Sti1p , one of the key regulators of the Hsp70-Hsp90 complex ( Song and Masison , 2005; Wolfe et al . , 2013 ) . We hypothesize that Sti1p may be especially prone to promiscuous protein binding , as it has an over 2-fold higher fraction of unstructured residues than the proteome average ( data not shown ) . Approximately , half of Sti1p putative physical interacting partners ( Cherry et al . , 2012 ) are involved in the maintenance of normal proteostasis . The list includes members of the Hsp70-Hsp90 complex , Hsp104 disaggregase , proteasome subunits and ubiquitin-associated proteins . Therefore , one might expect that perturbation of Sti1p interactions by a highly abundant , weakly interacting protein ( Figure 3D ) would have serious fitness consequences in times of proteotoxic stress . Future works should elucidate this hypothesis further and specifically the role of promiscuous peptide binding in protein burden . Our work has important implications for future studies . The distribution of genomic expression generally follows a highly skewed power-law like distribution with a small number of exceptionally highly expressed genes ( Ueda et al . , 2004; Lu and King , 2009 ) . Highly expressed genes contain various cost-minimizing gene architectures ( Frumkin et al . , 2017 ) . Such genes are under especially severe selective constraints , possibly to avoid misfolding and consequent formation of protein aggregates ( Geiler-Samerotte et al . , 2011 ) . Even though highly expressed proteins are not particularly prone to misfolding , they may still indirectly influence protein aggregation in the cell . Specifically , our work raises the possibility that highly expressed proteins bind to key components of the chaperone network which otherwise would be used to navigate folding of other native proteins within the cell . As a consequence , the availability of active chaperone molecules decreases , leading to increased propensity for damaging protein aggregation , especially in times of proteotoxic stress . It is important to emphasize that yEVenus is a codon optimized fluorescent protein ( Sheff and Thorn , 2004 ) , and is not particularly prone to misfolding and consequent toxicity ( Kafri et al . , 2016 ) . Therefore , this hypothesis is conceptually distinct and complementary to the issue of whether aggregation-prone proteins impose a fitness cost through toxicity ( Plata et al . , 2010; Geiler-Samerotte et al . , 2011 ) . More generally , several molecular chaperones can buffer the damaging effects of protein mutations ( Csermely , 2001; Queitsch et al . , 2002; Cowen and Lindquist , 2005; Paaby and Rockman , 2014 ) . Chaperone overload by highly expressed proteins may influence this process . In a similar vein , it appears that protein burden depends on genetic variation and environmental conditions as well . Therefore , the cellular capacity to tolerate major fluctuations in genomic expression heavily depends on the genetic makeup: the associated fitness costs should vary extensively across microbial species occupying different environmental niches . Finally , we anticipate that our genome-wide approach uncovering the determinants of protein burden will help the design of improved host strains for the efficient overproduction of recombinant proteins . All strains used in this study were derived from the Y7092 Saccharomyces cerevisiae parental strain ( SGA query strain: MAT alpha; can1delta::STE2pr-Sp_his5 , lyp1delta , his3delta1 leu2delta0 , ura3delta0 , met15delta0 ) . The fluorescent yEVenus protein was transformed into the parental Y7092 strain on a high copy number plasmid ( YEplac181 , [Gietz et al . , 1988] ) by a standard protocol ( Gietz and Schiestl , 2007 ) . The transformants were selected on leucine dropout synthetic complete medium ( SC-MSG , 1 g/l monosodium glutamate ( Sigma-Aldrich , Germany ) , 1 . 7 g/l Yeast Nitrogen Base ( BD , Germany ) , supplemented by amino-acid mix without leucine ) . To measure the fitness cost of protein overexpression , yEVenus , a non-toxic protein with no enzymatic activity and optimized codon usage was selected ( Sheff and Thorn , 2004 ) . The corresponding gene was integrated into a high copy expression vector . Heterologous promoters frequently perturb the transcription of other genes , by binding/titrating essential transcription factors , causing a skewed distribution of transcription factors . To minimize this problem , expression of yEVenus was driven by the native promoter of Hsc82p . Hsc82p is one of the most abundant cellular proteins in yeast ( Borkovich et al . , 1989; Ghaemmaghami et al . , 2003 ) . In contrast to many other chaperones ( such as Hsp82p ) , it is expressed constitutively and shows only minor variation across stress conditions . The high copy hc-Venus plasmid was constructed in three steps . First , the genomic HSC82 gene of the Saccharomyces cerevisiae strain BY4741 including its promoter sequence was amplified from genomic DNA using restriction site containing oligonucleotides ( B_HSC_promoter , B_HSC82_terminator ) . The product was cut with BamHI and PstI endonucleases , and was ligated to BamHI and PstI digested YEplac181 ( Gietz et al . , 1988 ) plasmid , generating the hc HSC82 construct . The promoter region was also PCR amplified with B_HSC_promoter primer and HSC-promoter-HSP-orf-reverse primer , which product was BamHI digested and ligated into a BamHI and StuI digested hc_HSC82 plasmid . The resulting plasmid ( pHSC_promoter plasmid ) was designed to facilitate the insertion of virtually any ORF using its NheI and PstI restriction sites . The yEVenus ORF along with the ADH1 terminator was amplified from the pKT0090 plasmid ( Sheff and Thorn , 2004 ) using NheI-Venus_ATG and Adh1_term_primer_pst1 oligonucleotides . The given PCR product was NheI and PstI digested and ligated to the identically digested pHSC_promoter plasmid . The generated plasmid ( hc_Venus ) was used to express yEVenus in S . cerevisae , under the control of the strong constitutive HSC82 promoter . For the selection of the plasmid , LEU2 marker was used in a leucine dropout synthetic medium . The control strains carry the original backbone plasmid ( YEplac181 ) without the fluorescent protein . To investigate the effect of plasmid copy number variation on protein burden , was inserted both into the BamHI-PstI digested single ( YCplac111 , [Gietz et al . , 1988] ) and low copy plasmid ( pRS315 , [Sikorski and Hieter , 1989] ) . Finally , to ensure that the key results are insensitive to the exact promoter used for controlling the expression of yEVenus , we constructed four extra isogenic plasmids with different , naturally occurring promoters in the yeast genome . These promoters drive the expression of cytosolic proteins ( Gpp1p , Tal1p , Pdc1p , and Tdh3p ) , all which are as highly abundant as the constitutively expressed Hsp90p ( HSC82 , source: PeptideAtlas 2013 dataset [Wang et al . , 2012] ) . Specifically , the pHSC82 region was eliminated from the hc_Venus plasmid after SacI-NheI digestion . Next , the promoter regions of GPP1 , TAL1 , PDC1 , and TDH3 were amplified from wild type genomic DNA using restriction site-containing oligonucleotides ( frw_SacI , rev_NheI ) . Finally , the PCR products ( pGPP1 , pTAL1 , pPDC1 , and pTDH3 ) were inserted into the SacI-NheI digested hc_Venus plasmid backbone . Fluorescence level showed only minor variation across the five high copy plasmid constructs ( Figure 3—figure supplement 1A ) . To quantify the yEVenus protein within the proteome , whole cell extracts were prepared from wild type cells , in the presence and absence of the yEVenus plasmid . Single colonies were inoculated into leucine dropout SC-MSG liquid medium , and were grown until saturation at 30°C . The saturated cultures were diluted and grown to mid-exponential phase ( OD600 = 0 . 8 ) , and 108–109 cells were used to extract total protein using established protocol ( Visweswaraiah et al . , 2011 ) . Whole cell extract ( WCE ) concentration was determined by using Bicinchoninic Acid Kit ( Sigma-Aldrich ) , according to the manufacturer's instructions . Whole cell extracts from the control and overexpression strain were separated on a 4–20% gradient Tris-Glycine gel ( Lonza , Germany ) under denaturing ( SDS , sodium dodecyl sulfate ) conditions , along with a dilution series ( 100–800 ng ) of a standard protein ( 1 mg/ml bovine serum albumin , BSA , Sigma-Aldrich ) . Densitometry analysis of the protein bands on SDS-polyacrylamide gel was conducted by ImageJ software ( Schneider et al . , 2012 ) . A standard curve was established by plotting the pixel numbers of BSA dilution series bands versus BSA concentrations . The yEVenus band ( 27 kDa ) intensity was corrected by subtracting the intensity of the equal-sized protein band in the control strain . Based on the standard curve , the pixel number of the yEVenus band ( 27 kDa ) was converted into concentration , and the ratio of the quantified yEVenus protein to the loaded whole cell extract was calculated . To identify genes mediating yEVenus burden , we performed a synthetic genetic array ( SGA ) screen ( Tong and Boone , 2006 ) . The query mutation ( in our case the yEVenus carrying plasmid ) was crossed to an ordered array of ~5000 viable , non-essential gene deletion mutants ( MATa; YKO collection , Open BioSystem , Dharmacon Inc , Lafayette , Colorado , United States , [Giaever et al . , 2002] ) . The method applies a series of replica pinning steps onto solid medium in an automated manner , using the following series of steps: ( a ) selection for MATa/α diploids ( SC-MSG medium ( 1 g/l monosodium glutamate , 1 . 7 g/l Yeast Nitrogen Base , supplemented by amino-acid mix ) with G418 ( 200 µg/ml , Sigma-Aldrich ) was used ) , ( b ) induced sporulation by reducing carbon and nitrogen levels in the nutrient , ( c ) selection for MATα meiotic progeny ( can1∆::MFA1pr-HIS3 , lyp1∆ ) using canavanine ( 50 mg/L , Sigma-Aldrich ) and thialysine ( S- ( 2-Aminoethyl ) -L-cysteine hydrochloride , 50 mg/L , Sigma-Aldrich ) containing medium , ( d ) selection for the query mutation ( leucine dropout medium ) , and finally selection for the gene deletions ( G418 containing medium; KanMX4 cassette confers resistance against G418 ) . Finally , the array of meiotic progeny harboring both mutations ( yEVenus plasmid and gene deletion ) was scored for fitness ( see below ) . To evaluate genetic interactions , an array of ‘single’ mutants was also constructed , where the query strain harbors the control high copy plasmid ( YEplac181 ) , without the fluorescent protein ORF . The HIS3 ( YOR202W ) deletion strain ( his3::KanMX4 ) was used as wild type control , for the following reasons: ( 1 ) fitness of this strain is indistinguishable from the BY4741 parental wild type strain ( Qian et al . , 2012 ) ; ( 2 ) it possesses the same selection marker ( required for the SGA method ) as all other single gene deletion strains; ( 3 ) it carries the KanMX4 cassette in the nonfunctional his3Δ1 allele . We developed a robust high-throughput and precise workflow for fitness measurements based on colony size . Solid media were prepared using 2% agar ( 2% was previously found to be optimal for reproducible colony size measurement , data not shown ) . The ordered arrays of strains at 384-density were replicated onto solid medium with a robotized replicating system . The system consists of a Microlab Starlet liquid-handling workstation ( Hamilton Bonaduz AG , Switzerland ) , equipped with a 384-pin replicating-tool ( S&P Robotics Inc , Toronto , Ontario , Canada ) and a custom-made sterilization station for the replicating-tool . After 48 hr of acclimatization to the medium at 30°C , plates were replicated again onto the same medium and photographed after 48 hr of incubation at 30°C . Digital images were processed to calculate colony sizes . We took special care to control for potential systematic biases in colony growth , such as uneven media composition , changes in physical parameters of incubation , or competition for nutrients between neighboring colonies ( Szamecz et al . , 2014 ) . Colonies located next to the edges/corners of the plates and colonies with low circularity ( i . e . circularity <0 . 8 ) were removed from further analysis . Genotype fitness was estimated by the mean fitness of six replicate colonies . The replicate number used is comparable to ( eight replicates for Kuzmin et al , in preparation ) or even higher than the number of replicates other studies ( four replicates for ( [Hoke et al . , 2008; Baryshnikova et al . , 2010; Costanzo et al . , 2010] ) used to estimate fitness based on colony size . Genetic interactions score was calculated as ε = fab − ( fa ×fb ) , where fa and fb are quantitative fitness measures of the two single ( deletion or yEVenus overexpression ) mutants , while fab is the fitness of the double mutant ( deletion and yEVenus overexpression ) . Negative ( ε <0 ) and positive ( ε >0 ) interaction scores indicate that the fitness defect of the double mutant is higher and lower than expected by the multiplicative model , respectively . We applied the confidence threshold of \|ε\|>0 . 05 and p<0 . 05 to define significantly interacting gene pairs . p-values were calculated using the bootstrap method ( Efron and Tibshirani , 1994 ) , resampling fa , fb , and fab separately . We tested the null hypothesis that ε = 0 . Based on the systematic genetic-genetic interaction screen , the list of genes showing negative interaction with the yEVenus overexpression ( i . e . their deletions increased the fitness effect of overexpression ) were retrieved and tested for Gene Onthology term enrichment with topGO ( version 2 . 28 ) ( Alexa et al . , 2006 ) and org . Sc . sgd . db ( version 3 . 3 . 0 , [Carlson , 2016] ) packages in R programming environment ( Core Team , 2017 ) . To focus on the important GO terms , we restricted our search to the GOSlim categories maintained by the SGD project ( Cherry et al . , 2012 ) . A GO category was termed as enriched significantly , if the genes annotated to a particular GO term were significantly overrepresented ( Fisher's exact test , odds ratio >1 , p<0 . 05 , FDR-corrected p<0 . 1 ) in the given gene set using the complete list of screened genes as background . Genotype fitness was estimated under control ( no-stress ) and different stress environments , as above . Unless otherwise indicated , all conditions used leucine dropout SC-MSG medium . The following non-lethal stress conditions were used: translation inhibition ( 0 . 0018–0 . 18 µg/ml cycloheximide , AppliChem GmbH , Germany ) , transcription inhibition ( 0 . 30 µg/ml mycophenolic acid ( MPA ) , Santa Cruz Biotechnology , Germany ) , heat stress ( 37°C and 40°C ) , proteotoxic stress ( 1–2 . 5 mM azetidine-2-carboxylic acid ( AZC ) , Santa Cruz Biotechnology ) , amino acid limitation ( auxotrophic amino acids were supplied at normal concentration to the medium , while the non-auxotrophic amino acids were serially diluted ( i . e . 0x - 2x of the regular concentration ) ) . Fitness cost of yEVenus protein overproduction ( proxy for protein burden ) is defined by 1 - WV/WC , where WV and WC indicate absolute fitness values ( i . e . colony sizes ) of the genotypes with yEVenus and control plasmids , respectively . The fluorescence level of the final SGA array strains was evaluated by measuring yEVenus signal in liquid medium . Briefly , the array of colonies were inoculated into liquid leucine dropout SC-MSG medium , and kinetic runs were initiated in a Synergy 2 fluorescence plate reader ( Biotek , Winooski , Vermont , United States ) for 48 hr , using the following filters: 500/27 ( excitation ) , 528/20 ( emission ) . During the kinetic run , the absorbance ( OD600 ) and yEVenus fluorescence ( λex515 nm / λem528 nm ) of the growing cultures were monitored simultaneously , with time points taken every 1 . 5 min . For each time points , the OD600 normalized yEVenus fluorescence ( FLOD ) was calculated . The fluorescence of a given strain was assessed by calculating the median of the five highest FLOD values . In order to quantitatively measure and compare the level of protein aggregation in the double mutants to the corresponding single mutants ( i . e . deletion ) , an established method ( Kaganovich et al . , 2008 ) was applied . This method examines the condition of the protein quality-control machinery of the cell , based on the aggregation of a fluorescently tagged ( mCherry , λex587nm/λem610nm ) human protein ( von-Hippel-Lindau , VHL ) . This human protein is prone to misfolding in the absence of its cofactor ( elongin BC ) , which is not present in S . cerevisiae . Fully functional quality-control machinery can stop aggregation of VHL-mCherry , leading to disperse cytosolic localization of the fluorescence . On the other hand , an overload of the control machinery promotes VHL protein aggregation , while leaving the fluorescent tag functional . In this case , the red fluorescence appears as a puncta inside the cell , due to the sequestration of aggregated proteins into dedicated compartments . All mutants carrying the plasmid ( pGAL-VHL-mCherry-Ura ) were grown until saturation in leucine and uracil dropout SC-MSG medium , containing 2% raffinose as carbon source . To induce VHL-mCherry production , the saturated cultures were diluted into leucine and uracil dropout SC-MSG medium , containing 1% raffinose and 2% galactose . After 14 hr of induction , cell fluorescence was detected by high content microscopy , using the following filter sets: excitation: 560–580 nm , emission: 590–640 nm . Images were acquired by employing an Operetta high-content screening microscope ( PerkinElmer , Waltham , Massachusetts , United States ) . Samples were grown and images were acquired in black optical 96-well plates ( Greiner Bio-One , Austria ) using a 60x high-numerical aperture objective . Five image stacks were made in each well , each of which consists of 7 z-stacks ranging from −1 . 5 µm to 1 . 5 µm relative to the focal plane with 0 . 5 µm step size . The following custom developed image and data analysis pipeline was used . First , an image filter was applied to amplify spots and project a z-stack . Images were corrected for illumination inhomogeneities ( Smith et al . , 2015 ) , single cells were segmented and 118 cellular features were measured based on morphology , shape and intensities . Machine learning-based phenotypic analysis was performed ( Horvath et al . , 2011; Piccinini et al . , 2017 ) using supervised learning and the ratio of phenotypic classes was determined . The ratio of cells containing aggregation loci was calculated using at least 2000 cells . To reveal the in vivo physically interacting protein partners of yEVenus , whole cell extracts were prepared from wild type cells in both the presence and absence of the yEVenus overexpression plasmid , and then a GFP co-immunoprecipitation ( GFP co-IP ) assay was performed . First , single colonies were inoculated into leucine-dropout SC-MSG liquid medium , and were grown until saturation at 30°C . The saturated cultures were diluted and grown to mid-exponential phase ( OD600 = 0 . 8 ) , and 108–109 cells were collected , flash frozen and used to extract total protein using an established protocol ( Visweswaraiah et al . , 2011 ) . Protein concentration of the whole cell extract ( WCE ) was determined by using Bradford Protein Assay ( Bio-Rad , Hercules , California , USA ) , according to the manufacturer's instructions . Total protein extracts ( 2 mg ) were immunopurified ( IP ) using 40 µl anti-GFP antibody-coupled 50 nm superparamagnetic beads ( µMACS GFP Isolation Kit , Miltenyi Biotec , Germany ) . The unbound material was removed by washing the beads with 2 ml ( equal to 50x beads volume ) detergent-free buffers as follows: three times with 1x TBS and once with 25 mM ABC ( NH4HCO3 ) buffer . The immunopurified proteins were desalted ( Hubner et al . , 2010 ) after on-bead-digestion with trypsin ( Promega , Germany ) . The LC-MS/MS analysis was performed by using a nanoflow RP-HPLC on-line coupled to a linear ion trap-Orbitrap ( Orbitrap-Elite , Thermo Fisher Scientific , Germany ) mass spectrometer as in a previous study ( Kobayashi et al . , 2015 ) with the following modification: the 20 most abundant , multiply charged ions were selected from each MS survey for MS/MS analysis . Raw data were converted into peak lists using Proteome Discoverer ( v 1 . 4 , Thermo Fisher Scientific ) . First , we performed a search against the Swissprot and Uniprot databases ( Pundir et al . , 2017 ) , taking into consideration of the sequence of yEVenus . Search parameters and acceptance criteria were set as previously published ( Kobayashi et al . , 2015 ) . Close homologues were only reported if at least three unique peptides matched to the protein . Spectral counting was used to estimate relative abundance of individual proteins in the samples: peptide counts of the individual proteins were normalized to the total number of peptide identifications in each sample ( Horvath et al . , 2017 ) . Proteins ( i ) with reproducible detection ( \|log2fold-change\| < 0 . 67 between biological replicates ) , ( ii ) with at least two identified peptides , ( iii ) with at least 5% coverage and ( iv ) with a median-normalized protein binding affinity score above a previously defined cutoff value ( 2 according to [Li et al . , 2016] ) were considered as proteins that specifically associate with yEVenus . Protein-binding affinity to yEVenus was estimated by calculating the peptide count fold change of yEVenus IP ( wild type strain with yEVenus plasmid ) samples relative to the negative control IP samples ( wild type strain with control plasmid ) . Reciprocal co-immunoprecipitation ( co-IP ) was performed in order to investigate physical interaction partners of Sti1p . First , a PCR-based C-terminal epitope-tagging of Sti1p was performed using established protocols ( Funakoshi and Hochstrasser , 2009 ) . Briefly , the transformation cassette was amplified from the pFA6a-TEV-6xGly-3xFlag-HphMX plasmid ( a gift from Tim Formosa , Addgene plasmid # 44083 ) with primers containing homology to the C-terminal of STI1 . Transformants were selected on YPD containing 300 µg/ml hygromycin ( Santa Cruz Biotechnology ) . Correct clones were verified by colony-PCR and subsequent capillary sequencing of the C-terminal of STI1 . Next , the single copy ( low protein burden ) or high copy ( high protein burden ) yEVenus plasmid was transformed into the Sti1p-FLAG-tagged strain . Finally , the yEVenus expressing strains were subjected to co-IP assay . Whole cell extraction ( WCE ) , immunoprecipitation ( IP ) and washing steps were performed as above , with the following modification: to reduce the effect of protein burden , a more stringent washing step was applied using the manufacturer’s ( μMACS DYKDDDDK Isolation Kit , Miltenyi Biotec ) ‘Wash 1’ buffer ( 150 mM NaCl , 1% Igepal CA-630 , 0 . 5% sodium deoxycholate , 0 . 1% SDS , 50 mM Tris-HCl , pH 8 . 0 ) . The LC-MS/MS and raw data analysis were the same as above . Close homologues were only reported if at least three unique peptides matched to the protein . The effect of protein burden on Sti1p interacting partners was investigated by comparing the protein binding affinity of these partners under low and high protein burden . Binding affinity scores below the cutoff value indicate weaker , non-specific associations of proteins with Sti1p . Protein-binding affinity to Sti1p was estimated by calculating the peptide count fold change of Sti1p IP ( IP with specific antibody to FLAG ) samples relative to the negative control IP ( IP without specific antibody ( protein A ) ) samples both under low and high protein burden . Proteins i ) with at least two identified peptides; ii ) with at least 5% coverage and iii ) with a median-normalized protein-binding affinity score above a previously defined cutoff value ( two according to [Li et al . , 2016] ) were considered as proteins that specifically associate with Sti1p under low protein burden .	Proteins are vital for almost every process that keeps cells alive . They are made from chains of small molecules called amino acids , which need to fold into three-dimensional structures for the protein to become active . Specific molecules called chaperones help the proteins to fold properly . However , to produce proteins a lot of energy is needed . Therefore , this process is tightly coordinated with the needs of the cells to conserve energy . If too much protein is made , it can put a burden on cells and harm the organism , even when it is a protein with no apparent cellular activities . This can be a problem under stressful conditions , for example , when cells are exposed to heat or lack nutrients . For researchers who want to engineer cells to produce different or additional proteins , this poses a great challenge , as the modified cells often grow slowly or not at all . Until now , it was not known why proteins are harmful when produced in excess . To investigates this , Farkas , Kalapis et al . modified the cells of baker’s yeast to overproduce an unneeded protein . The yeast cells were then exposed to different environmental stresses , such as too much heat or lack of nutrients , and scanned for any damage . Moreover , any potential protein burden was also measured in a collection of different cells in which each lacked one dispensable gene . The results showed that when enough nutrients where present , producing too much of the protein only mildly affected cell growth . However , when exposed to different stressors , the cells grew more slowly . When Farkas , Kalapis et al . then blocked specific chaperones , the proteins could no longer fold properly and consequently , the cells became very sensitive to when the protein was produced in bulks . This study shows that chaperones or environmental stress can shape protein production costs . A next step will be to investigate how sensitive other species are to protein burden , and what the underlying molecular mechanisms might be . A better understanding of how environmental and genetic factors affect the way the organisms deal with excess proteins may help to improve engineered protein-production systems in the future .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "genetics", "and", "genomics" ]	2018	Hsp70-associated chaperones have a critical role in buffering protein production costs
Zinc is widely recognized as essential for growth and proliferation , yet the mechanisms of how zinc deficiency arrests these processes remain enigmatic . Here we induce subtle zinc perturbations and track asynchronously cycling cells throughout division using fluorescent reporters , high throughput microscopy , and quantitative analysis . Zinc deficiency induces quiescence and resupply stimulates synchronized cell-cycle reentry . Monitoring cells before and after zinc deprivation we found the position of cells within the cell cycle determined whether they either went quiescent or entered another cell cycle but stalled in S-phase . Stalled cells exhibited prolonged S-phase , were defective in DNA synthesis and had increased DNA damage levels , suggesting a role for zinc in maintaining genome integrity . Finally , we demonstrate zinc deficiency-induced quiescence occurs independently of DNA-damage response pathways , and is distinct from mitogen removal and spontaneous quiescence . This suggests a novel pathway to quiescence and reveals essential micronutrients play a role in cell cycle regulation . Zinc ( Zn2+ ) is the second most abundant transition metal in biology and is widely recognized as an essential micronutrient to all living organisms ( Kaur et al . , 2014 ) . Zn2+ was first reported to be essential for growth of Aspergillus niger in 1869 and subsequently demonstrated for plants , animals , and humans ( Prasad , 1993 ) with the first cases of human Zn2+ deficiency and the associated growth and developmental disorders described in 1961 ( Prasad et al . , 1961 ) . Zn2+ deficiency has since been recognized as a global health problem , and the World Health Organization ( WHO ) estimates a staggering one third of the world’s population does not consume adequate Zn2+ and is therefore at risk for associated side effects and comorbidities ( https://www . who . int/whr/2002/chapter4/en/index3 . html ) ( Roohani et al . , 2013 ) . While the clinical manifestations of Zn2+ deficiency are diverse and can be organism specific , one defining feature is universal: Zn2+-deficient cells fail to divide and proliferate normally , leading to organismal growth impairment ( Vallee and Falchuk , 1993 ) . Despite recognition of the fundamental role of Zn2+ for proliferation , the mechanisms of how Zn2+ deficiency leads to cell-cycle arrest at the cellular and molecular level remain poorly defined . Eukaryotic cell proliferation is governed by the cell-division cycle , a series of highly choreographed steps that involve gap ( G1 ) , DNA replication ( S-phase ) , gap ( G2 ) , and mitosis ( M ) phases . Regulated transitions between proliferative and quiescent ( i . e . reversible non-proliferative ) states are essential for maintaining genome integrity and tissue homeostasis , ensuring proper development , and preventing tumorigenesis . Given the essentiality of Zn2+ for growth and proliferation , a fundamental question is whether Zn2+ serves as a nutrient , like amino acids , whether it affects the rate of cell cycle progression , or whether it is required at a specific phase of the cell cycle . Pioneering work by Chesters et al sought to define precisely when Zn2+ is required in the mammalian cell cycle . By chelating Zn2+ at different timepoints after release from serum starvation-induced quiescence , they found that Zn2+ was important for thymidine incorporation and thus DNA synthesis , leading to the conclusion that Zn2+ was required for the G1 to S transition ( Chesters et al . , 1989 ) . Subsequent studies confirmed that treatment of mammalian cells with high concentrations of metal chelators ( DTPA and EDTA ) seemed to compromise DNA synthesis ( Chesters et al . , 1990; Chesters and Boyne , 1991; Watanabe et al . , 1993; Prasad et al . , 1996 ) . However , later studies by Chesters et al suggested that after cells passed the restriction point in mid-G1 there was no further Zn2+ requirement for DNA synthesis in S phase , but rather Zn2+ was needed to transition from G2/M back into G1 ( Chesters and Petrie , 1999 ) . The restriction point is classically defined as the point at which cells commit to completing the cell cycle , regardless of the presence of external growth factors such as mitogens and/or serum ( Pardee , 1974 ) . Thus , while these early studies suggested that Zn2+ was important for progression of the mammalian cell cycle , the precise role of Zn2+ and whether it is required at a specific stage have remained enigmatic . There are three limitations of these early studies on the role of Zn2+ in cell proliferation . First , because the analyses were carried out on populations of cells , the cells were synchronized by artificial means ( serum starvation or hydroxyurea treatment ) and the cell cycle phase was inferred based on release from the cell cycle block . Recently , it has become clear that synchronization can induce stress response pathways that are specific to the type of arrest ( Ly et al . , 2015; Matson and Cook , 2017; Min and Spencer , 2019 ) . Further , cells induced into quiescence by different mechanisms ( serum starvation , loss of adhesion , contact inhibition ) exhibit overlapping but distinct transcriptional profiles , suggesting that different synchronization approaches impact cell cycle analysis upon emergence from quiescence ( Coller et al . , 2006 ) . Second , population level analyses such as immunoblotting and qPCR mask cellular heterogeneity and subpopulations of cells with different cell fates and cell cycle dynamics ( Matson and Cook , 2017; Spencer et al . , 2013 ) . Recent application of imaging and measurement tools for single cell analysis has uncovered distinct subpopulations of cells with different cell cycle dynamics ( Spencer et al . , 2013 ) , and revealed key orders of molecular events in the decision between proliferation and quiescence ( Spencer et al . , 2013; Cappell et al . , 2016; Heldt et al . , 2018; Moser et al . , 2018 ) . Third , many of the previous investigations into the role of Zn2+ in the mammalian cell cycle have relied on high concentrations of chelators ( DTPA , EDTA or TPEN ) to induce Zn2+ deficiency . However , these studies did not explicitly define how these perturbations changed the intracellular labile Zn2+ pool , nor did they characterize how the chelators affected cell viability . Indeed , the concentrations of chelators used have been shown to induce apoptosis in a number of different cell types ( Sunderman , 1995; Johnson et al . , 2000; Hyun et al . , 2001; Kolenko et al . , 2001; Canzoniero et al . , 2003; Corniola et al . , 2008; Lee et al . , 2008; Makhov et al . , 2008; Carraway and Dobner , 2012; Mendivil-Perez et al . , 2012; Zhu et al . , 2017 ) . In this study , we revisit the fundamental and unresolved question of how Zn2+ deficiency blocks cell proliferation using a combination of fluorescent reporters , high throughput microscopy , and quantitative image analysis . It is often overlooked that in addition to serving as a reservoir for mitogens , serum is also the major source of essential micronutrients including Zn , Fe , Cu , and Mn , and thus complete removal of serum also eliminates exogenous supply of these and other essential nutrients . By controlling Zn2+ in the medium , while maintaining mitogens at levels that normally sustain proliferation , we induced subtle perturbations of labile Zn2+ in the cytosol from 1 pM to 210 pM and tracked asynchronously cycling cells over multiple rounds of cell division . We found that Zn2+ deficiency induces cellular quiescence , but not death , and Zn2+ resupply stimulates synchronized cell cycle reentry . By following the entry of single cells into quiescence over time after Zn2+ deprivation , we found that depending on where cells were in the cell cycle , they either entered quiescence immediately after mitosis , or entered the cell cycle but stalled in S phase . Further , we determined that cells stalled in S phase were defective in DNA synthesis and had increased levels of DNA damage , consistent with previous bulk analysis studies ( Ho and Ames , 2002; Ho et al . , 2003; Yan et al . , 2008 ) suggesting a critical role for Zn2+ in maintaining genome integrity during replication . Finally , we found that Zn2+deficiency-induced quiescence does not require p21 , suggesting a mechanism distinct from spontaneous quiescence ( Spencer et al . , 2013; Moser et al . , 2018 ) , and follows a different pattern than mitogen withdrawal . Ultimately , our study provides new insights into when Zn2+ is required during the mammalian cell cycle and the consequences of insufficient Zn2+ levels . To revisit the question of how Zn2+ deficiency blocks cell growth and proliferation , we leveraged tools to visualize , track and measure molecular markers using fluorescent reporters in naturally cycling cells at the single cell level . Mammalian cells are generally recognized to contain hundreds of micromolar total Zn2+ ( Krezel and Maret , 2006 ) , which they are able to concentrate from the extracellular environment . The concentration of Zn2+ in human serum is about 12–15 µM ( Hess , 2017 ) and cell culture medium typically contains 1–40 µM Zn2+ ( Glassman et al . , 1980 ) , much of which is supplied by the serum . To rigorously control Zn2+ availability in our growth medium , we treated serum and insulin ( major sources of Zn2+ ) with Chelex 100 to scavenge Zn2+ . We then generated a minimal medium ( MM ) containing a low percentage of serum ( 1 . 5% ) still sufficient for proliferation that contained 1 . 46 µM Zn2+ as determined by Inductively Coupled Plasma Mass Spectrometry ( ICP-MS ) , which was significantly lower than the 2 . 2 µM Zn2+ measured in MM not treated with Chelex 100 ( Figure 1—figure supplement 1 ) . To further manipulate Zn2+ , MM was either supplemented with 30 µM ZnCl2 to generate a Zn2+ replete medium ( ZR ) or 2–3 µM of a Zn2+ chelator , tris ( 2-pyridylmethyl ) amine ( TPA ) to generate a Zn2+-deficient medium ( ZD ) ( Huang et al . , 2013 ) . TPA can chelate other metal ions ( Figure 1—figure supplement 2 ) , but with significantly lower affinity , except for Cu which binds TPA with high affinity . Using the CF4 Cu fluorescent probe ( Xiao et al . , 2018 ) , we determined that the low concentration of TPA used in this study does not perturb the labile Cu pool . On the other hand , the Cu chelator neocuproine does deplete Cu , indicating that CF4 is capable of detecting depletion of the Cu pool in MCF10A cells ( Figure 1—figure supplement 2 ) . To establish the effect of ZD , MM , and ZR media on cell viability , we grew cells in the respective medium for 30 hr and measured viability using trypan blue . We also compared TPA to N , N , N' , N'-tetrakis- ( 2-pyridylmethyl ) ethylenediamine ( TPEN ) , another Zn2+ chelator that has been widely used in studying the effect of Zn2+-deficiency on cell proliferation ( Beyersmann and Haase , 2001; Haase and Maret , 2003; Kaltenberg et al . , 2010 ) . Even at the low end of TPEN concentrations reported in the literature ( 3 µM ) , greater than 70% cell death was observed at 24 hr , compared to 3 µM TPA with ~15 % cell death . When noted , an even milder ZD condition of 2 µM TPA was used and this condition resulted in only ~1% cell death ( Figure 1—figure supplement 3 ) . Our results are consistent with several studies that found TPEN induces apoptosis ( Sunderman , 1995; Johnson et al . , 2000; Hyun et al . , 2001; Kolenko et al . , 2001; Canzoniero et al . , 2003; Corniola et al . , 2008; Lee et al . , 2008; Makhov et al . , 2008; Carraway and Dobner , 2012; Mendivil-Perez et al . , 2012; Zhu et al . , 2017 ) . To determine how defined Zn2+ medium conditions influenced intracellular labile Zn2+ levels in the cytosol , we created an MCF10A cell line stably expressing a genetically encoded FRET-based sensor for Zn2+ ( ZapCV2 Fiedler et al . , 2017 ) , grew the cells in ZD , MM , and ZR media , and measured the resting FRET ratio in individual cells . Cells grown in ZD had a significantly lower average FRET ratio than cells grown in MM , while those grown in ZR conditions had significantly higher FRET ratios ( Figure 1A ) . The FRET ratio correlates with the amount of labile Zn2+ in the cytosol , and in situ calibration suggests the respective Zn2+ levels are approximately 1 , 80 , and 210 pM for ZD , MM , and ZR media , respectively ( Figure 1—figure supplement 4 ) , indicating that exogenous nutritional Zn2+ levels positively influence intracellular free Zn2+ levels . To assess how the three nutritional Zn2+ regimes influenced cellular proliferation , we counted cells as a function of time . Naturally cycling cells expressing H2B-mCherry were imaged for 60 hr and cells in each frame were segmented and counted using a custom automated analysis as described in Supplementary file 1 . ZD growth conditions exhibited significantly reduced cell counts over time compared to MM and ZR conditions , with cell proliferation effectively halted after about 15 hr ( Figure 1B ) . Cells grown in ZR conditions reached higher cell counts , demonstrating that increased Zn2+ in the medium promotes cellular proliferation . Having established that our Zn2+-deficient conditions did not result in increased cell death , the decreased cell counts in ZD conditions could result either from a longer time between cell divisions , or an increased fraction of cells that enter a non-proliferative quiescent state . To differentiate these possibilities , we tracked individual cells over time and counted the number of mitosis events and the time between these events . Mitosis events were identified by a combination of the change in intensity of H2B-mCherry and change in size of the nucleus as described in Supplementary file 1 . The number of mitosis events in ZD media decreased over time with few mitosis events detected after about 15 hr ( Figure 1C ) . Cells grown in MM and ZR conditions underwent mitosis events throughout the observation period , with a comparable inter-mitotic time ( peaking around 13 hr , Figure 1—figure supplement 5 ) . The inter-mitotic time could not be measured in ZD media because few cells underwent multiple rounds of cell division . Resupply of Zn2+ by adding back either MM or ZR media after 24 hr in ZD conditions restored cell proliferation ( Figure 1B and C ) , revealing that the cells were cell-cycle competent . Combined , these results suggest that mild Zn2+ deprivation reduces cell proliferation , not by induction of cell death , but by inducing cell cycle arrest . To further examine how ZD conditions halted cell division and characterize the state of cells in ZD medium , we examined single cell fate using a fluorescent reporter of CDK2 activity ( Spencer et al . , 2013 ) . Following cell division , CDK2 activity is low and the fluorescent reporter is localized in the nucleus , but as the cell cycle proceeds CDK2 activity increases and the reporter is progressively translocated into the cytosol ( Spencer et al . , 2013 ) . Thus , the ratio of cytosolic/nuclear fluorescence can be used as a readout for CDK2 activity and serves a ‘molecular timer’ for progression through the cell cycle . As described previously , CDK2 activity defines subpopulations of cells with different cell fates in a population of naturally cycling cells ( Spencer et al . , 2013; Cappell et al . , 2016; Moser et al . , 2018; Arora et al . , 2017; Yang et al . , 2017 ) . When the CDK2 ratio remains low after mitosis ( CDK2low ) , a cell is classified as quiescent whereas when CDK2 activity increases above a defined threshold within 4 hr after mitosis ( CDK2inc ) , a cell is born committed to cell cycle entry ( Figure 2A ) . A third classification has been observed , in which a cell is born with low CDK2 activity ( low CDK2 ratio ) but eventually ramps up activity and commits to the cell cycle ( CDK2emerge ) . To define how Zn2+ availability in the media affects cell cycle commitment , we used an MCF10A cell line stably expressing the fluorescent CDK2 reporter and H2B-mTurquoise2 and imaged cells in ZD , MM or ZR media for 60 hr . Individual cells were segmented , tracked , and analyzed for CDK2 activity using a custom MATLAB pipeline ( Supplementary file 1 ) . In Zn2+-sufficient medium , cells cycled naturally throughout the observation window , as evidenced by the observation of mitosis events , inter-mitotic time , and cyclical decrease in CDK2 activity after mitosis , followed by increase marking cell cycle commitment ( Figure 2B , Figure 1—figure supplement 5 ) . When cell traces from each condition were aligned computationally to mitosis , we observed all three cell fate classifications ( CDK2inc , CDK2emerge , CDK2low ) with a similar percentage in each category in MM and ZR media ( CDK2inc56% vs . 60% , CDK2emerge49% vs 41% , CDK2low17% vs 14% for MM and ZR , respectively ) . However , in ZD medium the number of mitosis events ceased after about 20 hr , the CDK2 activity either stayed low or rose to an intermediate level , and cells rarely underwent multiple rounds of the cell cycle ( Figure 2B ) . Computational alignment to mitosis and cell classification revealed that few cells were born with CDK2 activity sufficient to commit to the next cell cycle ( 9% CDK2inc ) , a significant decrease compared to MM and ZR , and there was a substantial increase in the percentage of cells with low CDK2 activity following mitosis CDK2low ( 41% ) . One of the most striking differences between ZD and MM or ZR conditions was the fate of cells that attempted to re-enter the cell cycle ( i . e . whose CDK2 activity increased ) following mitosis . As shown in Figure 2C , in ZD medium CDK2emerge cells increased CDK2 activity after a period of transient quiescence but plateaued at an intermediate CDK2 activity ( a ratio of about 1 . 2 ) , did not achieve maximal CDK2 activity , and failed to divide . Similarly , CDK2inc cells increased CDK2 activity to an intermediate level before they dropped to a low level ( Figure 2C ) . Analysis of mitosis events revealed that these cells did not divide before entering the CDK2low state . These results suggest a significant increase in the number of cells that go quiescent after mitosis and the emergence of a new cell fate , where cells attempt to re-enter the cell cycle but stall part-way through under conditions of Zn2+ deficiency . We also measured the elemental profile in response to 24 hr of growth in Zn-deficient conditions as compared to MM ( Figure 2—figure supplement 1 ) . In addition to Zn , there were significant decreases in Fe and Cu levels and significant increases in Mg , K , P , S , and Mn in ZD compared to MM growth conditions . These findings are consistent with the fact that large transcriptional changes have been identified in quiescent cells and suggest major remodeling of metal homeostasis following a period of Zn deficiency-induced quiescence . To further define how Zn2+ deficiency influences cell fate and characterize the consequences of the altered CDK2 activity profile in ZD cells , we examined a downstream CDK2 substrate , retinoblastoma protein ( Rb ) . When CDK2 levels are low , Rb binds to and inhibits E2F family transcription factors , blocking cell cycle progression ( Giacinti and Giordano , 2006 ) . As CDK2 activity increases , Rb gets hyper-phosphorylated which releases the inhibition , enabling E2F to transcribe cell cycle genes . A previous study showed that cells born with elevated CDK2 activity also had hyper-phosphorylated Rb ( pRb ) , as determined by immunofluorescence , whereas cells born with CDK2low had low levels of phosphorylated Rb ( Spencer et al . , 2013 ) . We employed a similar protocol to measure phosphorylated Rb and DNA content by Fluorescence Activated Cell Sorting ( FACS ) . After 24 hr of growth in MM or ZR media , the majority of cells had hyper-pRb with either 2N , intermediate , or 4N DNA content , suggesting cells were actively cycling through G1 , S , G2 , and M ( Figure 2D ) . The small fraction of cells with hypo-pRb and 2N DNA content , correspond to the small fraction of cells with CDK2low and represent quiescent cells . Treatment of cells with 2 μM TPA for 24 hr revealed that most cells had 2N DNA content and hypo-pRb , consistent with most cells being in a quiescent state . However , some cells had hyper-pRb , indicating elevated CDK2 activity and an attempt to progress through the cell cycle , although there was a decrease in the fraction of cells with 4N DNA content indicating a deficiency in DNA replication . With 3 μM TPA the majority of cells had hypo-pRb with 2N DNA content , consistent with a quiescent state . A small population of cells had hypo-pRb and 4N DNA content , suggesting that after DNA replication , the cells entered quiescence without undergoing mitosis , consistent with the CDK2inc population of cells in Figure 2C that slips back to a CDK2low state . Combined , these results indicate that Zn2+ is required for cell cycle progression and there is heterogeneity in the cellular response to Zn2+ deprivation; some cells are born with low CDK2 activity and immediately enter quiescence , while others are born with elevated CDK2 activity ( CDK2inc ) or increase CDK2 activity after some delay ( CDK2emerge ) . Further , our results suggest that the milder the Zn2+ deficiency , the more cells attempt to progress through the next cell cycle following mitosis . However , there is a clear requirement for Zn2+ to successfully progress past S-phase to G2/M , which we explore below . The experiments in Figure 2 revealed heterogeneity in cell fate in response to Zn2+ deficiency . Given that the cells were cycling asynchronously prior to Zn2+ deprivation , we wondered whether the cell fate was determined by a cell’s position in the cell cycle at the time of Zn2+ withdrawal . To address this , we imaged cells expressing the CDK2 sensor in MM for 8 hr to track cell cycle progression prior to Zn2+ deprivation and follow entry of cells into quiescence . We binned cell traces according to when cells divided within specific 4 hr windows relative to TPA addition ( hr 0 ) , from 4 hr before TPA addition ( −4 to 0 ) up to 16 hr after TPA addition ( Figure 3A , gray shaded boxes ) . For cells that divided within 4 hr prior to TPA addition , a small but elevated percent of cells went quiescent ( 15% vs . 7% in MM ) , suggesting that cells need Zn2+ when exiting mitosis and progressing into G1 . Still , when cells divided within 4 hr of Zn2+ deprivation , the majority of cells re-enter the cell cycle either immediately ( CDK2inc ) or with a slight delay ( CDK2emerge ) . What was striking about this population of cells , was that only a small fraction was able to complete the next round of cell division compared to cells in MM conditions ( Figure 3A top two panels ) , and instead most cells stalled with an intermediate CDK2 ratio . Thus , even if cells are born with elevated CDK2 activity and pass the classical restriction point defined by a need for extracellular growth signals such as mitogens ( Pardee , 1974 ) , they rarely progress past an intermediate CDK2 activity in the absence of Zn2+ . In cells that divided in subsequent windows of time after TPA addition , there was a progressive decrease in the percentage of cells born with elevated CDK2 activity and classified as CDK2inc ( 43% , 30% , 17% , and 5% ) , indicating that longer Zn2+ deprivation increases the probability of cells entering quiescence after mitosis ( Figure 3A top to bottom ) . There was a small increase in the percent of quiescent cells in unperturbed MM over time due to increased cell density and quiescence induced by contact inhibition . Notably , when Zn2+ was removed prior to cell division and cells attempted to enter another round of the cell cycle , they stalled at an intermediate CDK2 activity , consistent with an inability to progress past S-phase in the absence of Zn2+ . To more precisely examine the timing of events ( Zn2+ removal , mitosis and cell fate ) and compare to the behavior to mitogen removal , we plotted heat maps of individual cell traces over time ( Figure 3B ) . We computationally grouped cell traces by their CDK2 activity at the end of the growth period , demonstrating 1 ) the two major cell fates for ZD cells: quiescence in red and stalled at intermediate CDK2 in green-turquoise and 2 ) that if cells divided more than 8 hr after TPA addition they entered quiescence immediately after cell division ( Figure 3B top panel ) . If cells divided within 8 hr of TPA addition , the majority re-entered the cell cycle and stalled at an intermediate CDK2 activity ( green-turquoise ) , consistent with an inability to progress past S-phase . As the time between TPA addition and cell division increased from 0 to ~8 hr , a greater proportion of cells experienced a prolonged low CDK2 activity period ( longer red streak ) before re-entering the cell cycle . These results suggest that if cells experience a short window of Zn2+ deficiency they have a greater chance of re-entering the cell cycle; but as the length of time increases , cells experience prolonged bouts of low CDK2 activity , suggesting that Zn2+ plays a role in processes involved in ramping up CDK2 activity to promote cell cycle entry . In contrast , the majority of cells in MM continued to cycle throughout the measurement window . To compare Zn2+ deficiency-induced quiescence to mitogen withdrawal , we aligned CKD2 traces computationally to the time of mitosis , with the first mitosis event at the top of the heat map ( 3B , bottom panel ) . Previously , this analysis demonstrated that if mitogens are withdrawn from newly born CDK2inc cells , they completed one additional round of the cell cycle ( Spencer et al . , 2013 ) , indicating that achieving a certain threshold of CDK2 activity marks cell-cycle commitment , regardless of mitogen availability . When aligned in a similar manner ( Figure 3B , bottom ) , our traces reveal that the CDK2 activity window that defines cell cycle commitment with respect to mitogen removal does not apply to Zn2+ removal . Instead , as other data representations suggest , even if cells pass the restriction point for mitogens , they stall at an intermediate CDK2 activity in the absence of sufficient Zn2+ , suggesting that Zn2+ deficiency-induced quiescence acts through a different pathway compared to mitogen withdrawal . Because we found a large population of cells which stalled at an intermediate CDK2 activity under Zn2+-deficient conditions and speculated that these cells were stalled in S phase , we wanted to measure whether these cells were capable of DNA synthesis . We grew cells as in Figure 3 , measured 5-ethynyl-2’-deoxyuridine ( EdU ) incorporation during a 15 min window of labeling at the end of this growth period , and stained for DNA content using propidium iodide/RNAse . Plotting EdU intensity against DNA content for single cells revealed the expected distribution of cell cycle phases , where EdU negative cells ( Edu- ) with 2N DNA were in quiescence ( G0 ) or G1 , EdU- cells with 4N DNA were in G2 or M , and EdU positive cells ( Edu+ ) transitioning between 2N and 4N were in S phase ( Figure 4A , cell cycle phases shown in boxes on right plot ) . In MM and ZR , the Edu vs . DNA content density plot followed a classical arch distribution , as has been found previously for MCF10A in full growth medium ( Gookin et al . , 2017 ) . In ZD conditions , cells did not exhibit high EdU intensity , indicating normal DNA synthesis was impaired , and a large portion of cells were EdU- with 2N DNA content , consistent with quiescence . Interestingly , in ZD conditions , many cells between 2N and 4N DNA content exhibited some Edu staining above that of cells classified as EdU- , suggesting that some cells are able to enter S phase , begin DNA synthesis , but at a reduced rate in the 15 min interval . The EdU intensities for cells grown in 2 µM TPA were slightly higher than those for cells grown in 3 µM TPA , demonstrating that 2 µM TPA is a milder ZD condition and impairment of DNA synthesis was less severe . These data confirm our findings from Figure 3 , where in addition to quiescence , a cell fate with intermediate CDK2 activity exists . This state of intermediate CDK2 activity is indeed S phase , as indicated by cells undergoing DNA synthesis , albeit at a reduced rate . Thus , though these cells have crossed a G1/S transition , Zn2+ is required in S phase for DNA synthesis to proceed at a normal rate and for cells to complete DNA synthesis and progress to G2/M . Cells that experience mild DNA damage can temporarily exit the cell cycle and enter a quiescent state in order to avoid passing a damaged genome on to the next generation . Given that long-term ( multi-day ) growth in Zn2+-deficient medium has previously been shown to increase DNA damage in a population of cells by the comet assay ( Ho and Ames , 2002; Ho et al . , 2003; Yan et al . , 2008 ) , we wondered whether mild Zn2+ deficiency could induce DNA damage on a shorter timescale and whether this DNA damage could be a cause of quiescence induced by Zn2+ deficiency . Further , because we found that ZD conditions caused a defect in DNA synthesis , we wanted to assess whether cells stalled in S phase were experiencing replication stress . Because we observed heterogeneity in cell cycle fates upon Zn2+ withdrawal , we sought to measure DNA damage at the single cell level and correlate it with a cell cycle marker . Previously , Arora and coworkers determined that whether naturally cycling cells enter quiescence and how long they spend in quiescence is in part explained by levels of double-stranded break ( DSB ) DNA damage inherited from mother cells , as measured by tracking fluorescent 53BP1 foci ( a known marker of DSBs ) ( Arora et al . , 2017 ) . Around the same time , Barr and coworkers demonstrated that both DSBs and single strand breaks ( SSBs , measured by RPA2 foci ) during S phase contribute to induction of quiescence ( Barr et al . , 2017 ) . We used a similar approach , quantifying 53BP1 and RPA2 puncta in individual cells to identify the presence of DSB and SSB , respectively . We correlated DNA damage markers with the cell cycle using phospho-Rb status in thousands of individual cells exposed to 24 hr of either ZD , MM , or ZR growth conditions , where cells that were hypo-pRb were classified as quiescent , while hyper-pRb cells were classified as cycling ( G1 , S , G2 or M ) . For cells classified as quiescent ( hypo-pRb ) , there was no significant difference in the fraction of cells with 53BP1 foci ( DSBs ) and a small but statistically significant increase in RPA2 foci ( SSB ) ( 4% versus <1% in MM and ZR ) in ZD versus MM and ZR media . These results suggest that DNA damage is likely not the primary mechanism of quiescence induction in Zn2+-deficient cells . However , in cycling cells , there was a significant increase in 53BP1 foci ( 87% in ZD versus ~45% in MM and 41 ZR ) and RPA2 foci ( ~17% in ZD versus <1% in MM and ZR ) . DNA damage was measured after 24 hr in the respective medium , and for cells in ZD , this time point corresponds to about 60% of the cells in a low CDK2 activity/quiescent state and 40% of the cells stalled at a state with intermediate CDK2 activity , consistent with S-phase ( Figures 3B , 24 hr time point ) . Given the increase in DNA damage in cycling cells , but relatively subtle change in DNA damage in quiescent cells , we speculate that the cells with increased DNA damage correspond to the cells stalled in the cell cycle at S phase ( Figure 3 ) , and that Zn2+ deficiency induces a defect in DNA synthesis ( Figure 4 ) that contributes to the inability of these cells to progress to G2/M . Our results also suggest that Zn2+ deficiency can induce quiescence independent of induction of DNA damage because those cells that have gone quiescent by 24 hr do not exhibit a profound increase in DNA damage . p21 is a cyclin dependent kinase inhibitor that binds to and inhibits the activity of cyclin-CDK complexes , thus regulating cell cycle progression . p21 is upregulated in response to contact inhibition and growth factor withdrawal and contributes to cell cycle arrest upon these perturbations ( Perucca et al . , 2009 ) . Furthermore , p21 is a transcriptional target of p53 and has been shown to be upregulated in response to DNA damage , which results in cell cycle arrest and presumably enables DNA repair prior to cell cycle re-entry . Thus , p21 has emerged as an important regulator of the proliferation-quiescence decision ( Moser et al . , 2018; Barr et al . , 2017 ) . This is underscored by the observation that in p21 null cells , the incidence of spontaneous quiescence is reduced ( Spencer et al . , 2013; Arora et al . , 2017 ) . Given that Zn2+ deficiency influences cell cycle progression by inducing quiescence and that it also results in increased DNA damage , we sought to determine whether quiescence resulting from Zn2+ deficiency requires p21 . We grew WT and p21-/- MCF10A cells expressing the CDK2 reporter and measured the fraction of cells in each classification ( CDK2low/emerge/high ) under ZD , MM , and ZR conditions ( Figure 6A ) . p21 knockdown was validated as described in the methods section ( Figure 6—figure supplement 1 ) . In WT cells grown in MM or ZR , 17% or 16% ( respectively ) of cells were classified as quiescent ( CDK2low ) , while in p21-/- cells , only 8% or 5% cells were classified as quiescent , consistent with previous results suggesting that spontaneous quiescence in naturally cycling cells is induced by endogenous DNA damage and is dependent on p21 ( Arora et al . , 2017; Barr et al . , 2017 ) . In ZD conditions quiescence occurred at a similar rate in p21-/- and WT cells ( 42% vs . 56% ) , suggesting that quiescence caused by Zn2+ deficiency does not explicitly require p21 ( Figure 6A ) . This is perhaps not surprising , given that the cell population that was quiescent after 24 hr of growth in Zn2+-deficient medium did not exhibit increased DNA damage ( Figure 5 ) . Combined , our data indicate that Zn2+-deficiency induces quiescence via a p21-independent pathway . Because p21 is also upregulated to maintain quiescence ( Coller et al . , 2006 ) , we measured p21 levels in WT MCF10A cells using immunofluorescence after 40 hr of growth . In ZD medium , cells that maintained low CDK2 activity had higher levels of p21 , similar to cells in MM medium ( Figure 6B ) . These data suggest that although p21 is not required for entry into quiescence , in WT cells p21 is upregulated when cells are quiescent , likely to maintain their quiescent state by suppressing CDK2 activity . To determine whether Zn2+ deficiency-induced quiescence is reversible , we grew cells for 24 hr in ZD medium followed by 36 hr with either MM or ZR media . Zn2+ resupply , by adding either MM or ZR media , caused CDK2 activity to increase and the resumption of mitosis events ( red dots ) , indicating active cell division ( Figure 7A ) . Notably , when individual elements contained in MM were resupplied , only Zn2+ , but not Fe2+ or Cu2+ rescued ZD-induced quiescence , suggesting Zn2+ alone is sufficient to induce cell-cycle re-entry , despite the reduction in Fe and Cu measured in quiescent cells by ICP-MS ( Figure 7—figure supplement 1 , Figure 2—figure supplement 1 ) . It appeared that a smaller subset of cells remained quiescent when rescued with ZR as opposed to when rescued with MM ( see CDK2low traces in highlighted windows in Figure 7A ) . To quantify this , we generated CDK2 activity probability density plots for each hr after resupply with either MM or ZR ( Figure 7B and Figure 7—figure supplement 1 ) . After 1 hr of Zn2+ resupply , the majority of cells were quiescent in all three conditions ( CDK2low , activity mean ~0 . 5 ) . After 8 hr of resupply , cells not resupplied ( ZD ) remained in a CDK2low state , while resupplied cells emerged from quiescence and entered the cell cycle , as indicated by the cell populations shifting towards higher CDK2 activity , with a mean around 1 . 25 . The ZR resupplied cells had a higher probability of being in this higher CDK2 state and a corresponding lower probability of being in the CDK2low state compared to cells resupplied with MM , suggesting a positive correlation between the amount of Zn2+ in the medium and cell cycle re-entry after a period of deficiency . Accurate duplication of the genome and separation of chromosomes into daughter cells through the process of the cell-division cycle is one of the most essential functions individual cells must execute . Regulated exit of the cell cycle to quiescence is an important quality control pathway that reduces metabolic and biochemical activities and protects cells against stress and toxic metabolites ( Tümpel and Rudolph , 2019 ) . Given the importance of understanding the factors that regulate cell cycle entry , progression , and exit to quiescence , decades of research have sought to define the underlying mechanisms of the proliferation-quiescence ‘decision’ . Much of our understanding of the mammalian cell cycle has derived from studies in which populations of cells were forced to exit the cell cycle upon induction of stress , such as serum starvation or amino acid deprivation . These studies have revealed many of the key regulators of the cell cycle and introduced the concept of a restriction point , a point at which cells commit to the cell cycle , and become mitogen-independent . However , recent application of single cell technologies for tracking cellular and molecular markers and the fate of individual cells have led to key revisions of the textbook model of the mammalian cell cycle , including the discovery that in naturally cycling cells there are multiple proliferation decisions ( Matson and Cook , 2017; Spencer et al . , 2013; Cappell et al . , 2016; Heldt et al . , 2018 ) . Furthermore , it is now appreciated , but still poorly understood that quiescence is not a single dormant state , but rather an assemblage of heterogeneous states , that is actively maintained ( Matson and Cook , 2017; Coller et al . , 2006; Yao , 2014 ) . Broadly speaking , control of the decision to proliferate or arrest is fundamental to various aspects of tissue architecture maintenance , differentiation , DNA damage repair , wound healing , and normal vs . cancerous cell growth ( Matson and Cook , 2017; Heldt et al . , 2018; Hanahan and Weinberg , 2011 ) . Thus , understanding how individual triggers act to induce quiescence is important for ultimately identifying targets for improving nutritional deficiencies or disease states . Building on the emerging conceptual framework of studying the cell cycle in asynchronous populations of cells , in this work we apply a combination of fluorescent reporters , high throughput imaging , and quantitative image analysis to revisit the important but unresolved question of how Zn2+ deficiency blocks cell proliferation . While it has long been recognized that Zn2+ is required for cell proliferation , how Zn2+ deficiency influences the cell cycle on a single cell level has not been elucidated and whether Zn2+ affects the critical proliferation-quiescence cell fate decision has not been determined . Zn2+ is an essential metal that serves as a critical cofactor in approximately 10% of the proteins encoded by the human genome ( Andreini et al . , 2006 ) , including over 700 Zn2+-finger containing transcription factors , DNA polymerase , superoxide dismutase , and proteins involved in DNA repair ( Lambert et al . , 2018 ) . Thus , it is required for several key cellular processes involving these proteins , including processes relevant for the cell cycle such as transcription , antioxidant defense , DNA synthesis and repair . It is often overlooked that in addition to serving as a reservoir for mitogens , serum is also the major source of essential micronutrients including Zn , Fe , Cu , and Mn , and thus complete removal of serum also eliminates exogenous supply of these and other essential nutrients . We sought to isolate the effect of Zn2+ by examining Zn2+ deprivation in an asynchronous population of cells still containing sufficient mitogen levels . We subjected cells to mild Zn2+ perturbations that altered the labile pool of cytosolic Zn2+ between 1 and 210 pM and avoided induction of cell death . Upon Zn2+ deprivation , cells lost the ability to proliferate and the majority experienced one of two cell fates: entry into quiescence or stall in S phase . This resultindicates that unlike mitogens there is not a single restriction point for Zn2+ , perhaps because it would be too hard to evolve a checkpoint for every nutrient , or perhaps because Zn2+ is essential for so many cellular processes . Tracking individual cells before and after Zn2+ deprivation revealed the temporal development of these two distinct fates , demonstrating that Zn2+ is required at multiple places in the cell cycle and the depth of Zn2+ deficiency relative to mitosis determines the cell fate . If cells underwent mitosis within 8 hr of Zn2+ withdrawal , they were likely to re-enter the cell cycle and complete G1 before stalling in S phase , with a markedly prolonged S-phase ( plateau at intermediate CDK2 activity in Figure 3A ) . These cells synthesized DNA at reduced rates and accumulated increased amounts of DNA damage , indicating that sufficient Zn2+ is necessary for maintaining DNA integrity . These results are consistent with previous work which showed at a population level that long term ( multi-day ) Zn2+ deficiency led to increased DNA damage as measured by a comet assay ( Ho and Ames , 2002; Ho et al . , 2003 ) . Here , we find that Zn2+ deficiency increases both DSBs and SSBs in cycling hyper-phosphorylated Rb cells ( ~1 . 9 fold increase in DSBs and ~170 fold increase in SSBs compared to cells in MM or ZR ) . It is well established that Zn2+ is a critical cofactor in a number of genome caretaker proteins , such as XPA and RPA of the nucleotide excision repair pathway , PARP involved in base excision repair , HERC2 involved in DSB recognition , CHFR involved in checkpoint regulation , and transcription factors such as p53 and BRCA that are involved in DNA damage response and repair ( Danielsen et al . , 2012; Ahel et al . , 2008; Witkiewicz-Kucharczyk and Bal , 2006 ) . While in vitro studies indicate these proteins would not function without Zn2+ , little is known about the metal-binding properties in cells , how the myriad proteins of the zinc proteome acquire their Zn2+ , and whether these proteins are sensitive to subtle perturbations of the labile zinc pool . Ho and coworkers demonstrated that long-term Zn2+ depletion decreased the binding of transcription factors such as p53 and AP1 to DNA ( Ho and Ames , 2002 ) , and resulted in differential expression of genes involved in the cell cycle as well as DNA damage response and repair ( Yan et al . , 2008 ) . However , nothing was known about how rapidly cells sense Zn2+ deficiency and how consequences of Zn2+ deficiency , such as increased DNA damage , correlate with cell fate . By following the fate of cells over time in response to Zn2+ withdrawal , we show that individual cells sense depletion of the labile Zn2+ pool quickly ( within 4 hr ) , and that cells in S-phase experience decreased rates of DNA synthesis and increased DNA damage , inducing cell cycle exit , despite these cells having sufficient mitogens to have passed the restriction point . If cells divided >8 hr after Zn2+ withdrawal , the vast majority were born with low CDK2 activity and immediately entered a quiescent state . Intriguingly , these cells didn’t exhibit substantial increases in DNA damage and p21 was not required for entry into this quiescent state , indicating the mechanism of quiescence-induction is distinct from that of spontaneous quiescence . Entry into ‘spontaneous’ quiescence observed in naturally cycling cells was found to be associated with increased levels of DNA damage from the previous cell cycle was dependent on p21 activity ( Arora et al . , 2017; Barr et al . , 2017 ) . The observation that cells were born with low CDK2 activity suggests that Zn2+ deficiency is sensed in the mother cell during the previous cell cycle . This new micronutrient deficiency-induced quiescent state occurs in the presence of mitogens and is not induced via replication stress/DNA damage . Importantly , quiescent cells and cells stalled in S phase after a period of Zn2+ deficiency could be rescued by resupply of Zn2+ and were competent to re-enter the cell cycle . Interestingly , we found that while the TPA chelator didn't perturb the labile pool of other metals such as Cu , the Zn2+-deficient quiescent state was characterized by notable changes in other essential metals , such as Cu , Fe , and Mn , suggesting significant changes to metal homeostasis in this quiescent state . Although quiescence was rescued exclusively by Zn ( and not by Cu or Fe ) , it would be intriguing to study the mechanisms of metal homeostasis remodelling in a future study . By addressing the question of how mild Zn2+ deficiency reduces cell proliferation with modern tools at the single cell level in naturally cycling cells , we revealed that Zn2+ status influences multiple checkpoints in the cell cycle , and uncovered a new quiescent state induced by micronutrient deficiency that is distinct from spontaneous quiescence and mitogen withdrawal . These findings are important for understanding how this cell cycle control might be perturbed in disease states such as cancer where Zn2+ levels and localization have been shown to be perturbed ( Hanahan and Weinberg , 2011; Prasad et al . , 2009; Pan , 2017; Chandler et al . , 2016 ) . Future proteomic and transcriptomic studies have the potential to reveal the relevant Zn2+ dependent proteins that sense Zn2+ status , thus uncovering the mechanism of entry into this quiescent state and the key Zn2+-dependent proteins responsible for maintaining genome integrity during DNA replication . MCF10A cells were obtained from ATCC and maintained in full growth DMEM/F12 medium ( FGM ) supplemented with 5% horse serum , 1% Pen/strep antibiotics , 20 ng/mL EGF , 0 . 5 μg/ml hydrocortisone , 100 ng/ml cholera toxin , and 10 μg/ml insulin in a humidified incubator at 37°C and 5% CO2 , as described previously by the Brugge lab ( Debnath et al . , 2003 ) . Cells were passaged with trypsin-EDTA . For imaging and growth experiments , cells were grown in 50:50 Ham’s F12 phenol red free/FluoroBrite DMEM with 1 . 5% Chelex 100-treated horse serum , 1% Pen/strep antibiotics , 20 ng/mL EGF , 0 . 5 μg/ml hydrocortisone , 100 ng/ml cholera toxin , and 10 μg/ml Chelex 100-treated insulin . Chelex 100 was used to chelate excess Zn2+ from horse serum and insulin to generate the defined minimal medium ( MM ) . ICP-MS was used the measure the elemental content of Chelex 100-treated medium compared to untreated medium and revealed that Chelex 100 treatment significantly reduced Zn and Ni in the medium ( Figure 1—figure supplement 1 ) . We did not add back Ni ( or other elements ) , but we consistently use Chelex 100–treated MM for all experiments so the concentration of elements other than Zn are the same in all media conditions . ZR medium was generated by supplementing MM with 30 µM ZnCl2 and ZD media was generated by adding 2–3 µM TPA . Cell lines were routinely tested and confirmed to be mycoplasma negative by PCR . ICP-MS was used to measure total Na , Mg , Al , K , Ca , P , S , Mn , Fe , Co , Ni , Cu , and Zn content in defined MM with and without Chelex 100 treatment of horse serum and insulin . Media samples were diluted 1:10 in Chelex 100- treated water with 1% TraceSELECT Ultra nitric acid ( free of trace elements ) . Cells were grown for 24 hr in either 3 µM TPA , 2 µM TPA , or MM and the same elements as above were analyzed ( four replicates per condition ) . Before digestion , cells were counted in triplicate . Cells were pelleted , rinsed with PBS , and cell pellets containing ~6–8 million cells were digested in 100 µL TraceSELECT Ultra nitric acid . Digestion was performed by vortexing constantly for 3 min followed by a 30 min incubation at 70°C . Samples were then diluted to 1 mL in Chelex 100 -treated water . All ICP-MS samples were processed at the Dartmouth Trace Element Analysis Core . Cell ICP-MS results were normalized to average cell counts . MCF10A cell lines expressing PB-NES ZapCV2 ( Fiedler et al . , 2017 ) and PB-H2B-mCherry were generated using the PiggyBac Transposon system via electroporation-mediated transfection . Stable cell lines used for long-term imaging were generated with G418 and puromycin antibiotic selection followed by FACS enrichment of dual positive fluorescent cells . The MCF10A cell line stably expressing DHB-Venus ( CDK2 sensor ) and H2B-mTurquoise2 and the MCF10A p21-/- were provided by the laboratory of Dr . Sabrina Spencer , CU Boulder , and generated as described previously ( Spencer et al . , 2013 ) . Validation of p21 knockout in this cell line was performed using immunofluorescence with a p21 antibody , as described below ( See Figure 6—figure supplement 1 ) . Cells were counted with a Countess II Automated Cell Counter ( Thermo Fisher Scientific , Waltham , MA ) and plated at a density of 2 , 500–3 , 500 cells/well ~ 24 hr before imaging in minimal medium in glass bottom 96-well plates ( P96-1 . 5H-N , Cellvis , Mountain View , CA ) . This starting density was chosen to avoid significant contact inhibition during the imaging period . In Figures 1 , 2 and 5 , cells were plated in 100 µL MM and 2x media of each specific nutritional regime ( ZD , MM , ZR ) was added immediately prior to imaging . In Figure 3 , cells were plated in 100 µL MM and 2x ZD medium was added after 8 hr . In Figure 7 , cells were plated in 100 µL , 2x medium of specified conditions was added immediately prior to imaging; after 24 hr , 100 µL of medium was removed , and 2x medium of specified conditions was resupplied ( effectively half of the medium was replaced with MM or 2x ZR media ) . For resupply of individual elements in Figure 7—figure supplement 2 , ZnCl2 , FeCl2 , or CuCl2 were added at concentrations found in a 50% MM change , 0 . 73 µM , 0 . 8 µM , and 0 . 16 µM , respectively ) . Images were collected using a Nikon Ti-E inverted microscope microscope with a Lumencor SPECTRA X light engine ( Lumencor , Beaverton , OR ) and Hamamatsu Orca FLASH-4 . 0 V2 cMOS camera ( Hamamatsu , Japan ) . Images were collected in time lapse series every 12 mins with a 10 × 0 . 45 NA Plan Apo objective lens ( Nikon Instruments , Melville , NY ) . During imaging , cells were in a controlled environmental chamber surrounding the microscope ( Okolab Cage Incubator , Okolab USA INC , San Bruno , CA ) at 37°C , 5% CO2 and 90% humidity . Total light exposure time was ≤600 ms per timepoint . Filter sets used for live cell imaging and immunofluorescence ( described below ) were as follows: CFP Ex: 440 Em: 475/20; GFP Ex: 470 , Em: 540/21; YFP Ex: 470 , Em: 540/21; CFPYFP FRET Ex: 395 , Em: 540/21; mCherry Ex: 555 , Em: 595/40; Cy5 Ex: 640 , Em: 705/22 . A detailed description of our custom MATLAB R2018A ( Mathworks ) pipeline for automated cell segmentation and tracking , as well as methods for calculating the FRET ratio and CDK2 ratio are provided in Supplementary file 1 and the tracking code is available for download here: https://biof-git . colorado . edu/biofrontiers-imaging/palmer-zinc-cell-cycle . Briefly , mitosis events were identified when the nuclear signal generated by fluorescent H2B split into two distinct objects . The FRET ratio ( FRET intensity/CFP intensity ) was calculated in a cytosolic region outside the nuclear mask and the CDK2 ratio was calculated as the cytosolic intensity/nuclear intensity of the fluorescent CDK2 sensor . Cell pellets were washed with PBS , fixed with 4% paraformaldehyde and permeabilized with methanol at −20°C . Cells were stained for one hr with Phospho-Rb Ser 807/811 ( D20B12 XP , Cell Signaling Technology , Danvers , MA ) at 1:500 dilution prior to Alexa Fluor 488 secondary antibody ( ab150073 , Abcam , Cambrdige , MA ) staining at 1:500 dilution for one hr . Antibodies were diluted in PBST with 1% BSA . PI DNA staining was performed using FxCycle PI/RNAse ( Thermo Fisher Scientific F10797 ) . FACS for phospho-Rb and PI was performed on a BD FACSCelesta instrument and analyzed with BD FACSDiva v8 software ( BD Biosciences , San Jose , CA ) . GFP: Ex 488 , Em 530/30; mCherry Ex 561 , Em 610/20 . FACS enrichment of stable cell lines was performed using BD FACSAria Fusion with the following optics: CFP: Ex 445 , Em 470/15; YFP: Ex 488 , Em 530/30; and mCherry Ex 561 , Em 610/20 . Cells were fixed with 4% paraformaldehyde , washed with PBS , and permeabilized with 0 . 2% Triton X-100 solution . Blocking was performed in 3% BSA for one hr at 4°C . Primary antibody staining occurred overnight at 4°C , with the following antibody concentrations: p21 , 1:200; pRB , 1:100 , RPA2 , 1:250; 53BP1 , 1:200 . Following primary staining , cells were washed with PBS and secondary antibody staining was performed for one hr with either AlexaFluor 488 , AlexaFluor 568 , or AlexaFluor 647 each at a 1:500 dilution . Antibodies were diluted in PBS with 3% BSA . Hoechst staining for 15 mins at 0 . 1 µg/mL diluted in PBS was used to identify nuclei . For experiments comparing CDK2 activity vs . p21 intensity , the fluorescence of the CDK2 sensor was preserved upon fixation . Images were acquired on a Nikon Ti-E inverted microscope ( as described in Live Cell Imaging ) with a 40 × 0 . 95 NA Plan Apo Lamda ( Nikon ) . Primary antibodies used were: p21 Waf1/Cip1 ( CST , 2947S , Lot 10 ) , pRB Ser807/811 ( CST , D20B12 XP , Lot8 ) , RPA2 ( Abcam , ab2175 , Lot GR3224197-5 ) , 53BP1 ( BD Biosciences , 612523 , Lot 6217571 ) and secondaries were: AlexaFluor 488 ( Abcam , ab150073 , Lot GR328726-1 ) , AlexaFluor 568 ( Life Sciences A10042 , Lot 1134929 ) , or AlexaFluor 647 ( Thermo Fisher Scientific , A21244 , Lot 1156625 ) . Analyses were performed with custom MATLAB scripts and run in MATLAB R2018a . Briefly , nuclei were segmented using a combination of adaptive thresholding and the watershed algorithm to segment clumps of cells . The nuclear mask was used to calculate the mean nuclear pRB or p21 intensities for each cell . For cells containing the CDK2 sensor , the nuclear mask was used to draw a ring three pixels wide around the nucleus for computing the CDK2 activity ( defined as the ratio of cytoplasmic intensity/nuclear intensity ) of the cell . For DNA damage experiments , the centroid of each cell ( as defined by the nuclear mask ) was used to construct a 140 × 140 pixel square around each cell; this enabled the use of MATLABs ‘adaptthresh’ function for constructing an accurate foci mask for each cell . The foci mask was further refined by filtering out any objects that were not the appropriate size ( area 10–100 px2 for RPA2 and 10–200 px2 for 53BP1 ) or shape ( eccentricity <0 . 6 , where 0 is a perfect circle and one is straight line ) . Cells were scored as being positive for damage when they had one or more foci present . Live cells kept at controlled environmental conditions ( 37°C , 5% CO2 , 90% humidity ) were labeled with EdU according to the manufacturer’s instructions ( Thermo Fisher Scientific , C10356 ) . Briefly , cells were labeled with 10 µM EdU for 15 mins , followed by fixation and permeabilization , as described for immunofluorescence . Cells were labeled for 30 mins via a click reaction with AlexaFluor 647 azide using CuSO4 as catalyst . FxCycle PI/RNAse was used to quantify DNA content ( 2N vs 4N ) . Images were acquired on a Nikon Ti-E inverted microscope system with a 40 × 0 . 95 NA Plan Apo Lambda ( Nikon ) . Analysis was performed with custom MATLAB scripts and run in MATLAB R2018a . The analysis workflow was the same as for immunofluorescence , with the exception that high residual background EdU staining necessitated background subtraction before quantifying intensities . Background subtraction was performed by dividing the image into 11 × 11 blocks and then using the lowest 5th percentile intensity value for each block as background . Quantification of DNA content was performed by computing the integrated intensity of each cell . Sensor calibrations of MCF10A cells stably expressing PB-NES-ZapCV2 were performed using the Nikon Ti-E . Cells were grown for 24 hr in either ZD ( 3 μM TPA ) , MM , or ZR media . For collection of Rrest , cells were imaged for CFP-YFP FRET ( 200 ms exposure ) and CFP ( 200 ms exposure ) every 30 s for several mins . To collect Rmin , 50 µM TPA in MM was added and cells were again imaged for several mins . Cells were then washed three times with phosphate , calcium , and magnesium free HEPES-buffered HBSS , pH 7 . 4 , for removal of TPA . Finally , for collection of Rmax , cells were treated with this HBSS buffer with 119 nM buffered Zn2+ solution , 0 . 001% saponin , and 5 µM pyrithione , as previously described ( Carter et al . , 2017 ) . Average Rrest and Rmin were calculated by averaging across the timepoints collected . The maximum FRET ratio achieved after Zn2+ addition was used as Rmax . Images were background corrected by drawing a region of interest in a dark area of the image and subtracting the average fluorescence intensity of the background from the average intensity of each cell . FRET ratios for each cell ( n = 8 per condition ) were calculated with the following equation: ( FRETintensity of cell - FRETintensity of background ) / ( CFPintensity of cell - CFPintensity of background ) . Dynamic range ( DR ) of the sensor in each condition was calculating as Rmax/Rmin . Fractional saturation ( FS ) of the sensor in each condition was calculated as follows: ( Rrest-Rmin ) / ( Rmax-Rmin ) . Finally , Zn2+ concentrations were estimated by [Zn2+]=KD ( ( Rrest - Rmin ) / ( Rmax - Rrest ) ) 1/n , where KD = 5300 pM and n = 0 . 29 ( Hill coefficient ) . In Figure 1A , FRET ratios between cells grown under ZD , MM , or ZR were compared using One-way ANOVA with post-hoc Tukey HSD , performed using KaleidaGraph v4 . 02 . Alpha was 0 . 05/confidence level was 0 . 95 . Data were plotted in MATLAB v R2017b . On each box , the central mark is the median and the edges of the box are the 25th and 75th percentiles . The whiskers extend to the most extreme data points , excluding outliers , which are plotted individually with + marks . In Figure 5A , differences in DNA damage across conditions were assessed using one-way ANOVA with Sidak’s multiple comparisons test ( six comparisons , alpha = 0 . 05 ) . Analyses and graphing were performed using GraphPad Prism 8 . 2 . 0 . For plots , p<0 . 05 , p<0 . 01 , **p<0 . 0001 . For Figure 1—figure supplement 1 , two-tailed paired t-tests were used to compare MM to MM+Chelex for each element analyzed . Alpha was 0 . 05/confidence level was 0 . 95 . For plots p<0 . 05 and exact values for significant values are listed . Analysis and graphing were performed using GraphPadPrism 8 . 2 . 1 . For Figure 2—figure supplement 1 , one-way ANOVAs were used to determine if ZD3 or ZD2 conditions had elemental levels significantly different than MM levels . Dunnett’s multiple comparisons test was used with MM as control sample . Alpha was 0 . 05/confidence level was 0 . 95 . Analysis and plotting was performed using GraphPad Prism 8 . 2 . 1 . In plots * indicates when ZD3 or ZD2 was significantly different from MM at p<0 . 05 . Exact p values are listed in a chart with the figure . 200 , 000 MCF10A cells were plated in MM in glass bottom six well plates ( P06-1 . 5H-N Cellvis , Mountain View , CA ) and grown for 24 hr . Experimental conditions were as follows: cells left in MM , cells treated with 50 µM neocuproine for 30 min , or cells treated with 3 µM TPA for 2 hr . At the end of the treatment period , 2 µM CF4 Cu probe ( reconstituted in MeOH ) ( Xiao et al . , 2018 ) was added and cells were incubated for 20 min at 37°C . Prior to imaging , each dish was rinsed 3X with either MM , MM containing 50 µM Neocuproine , or MM containing 3 µM TPA . Finally , images were collected with a 10 × 0 . 45 NA Plan Apo Lambda ( Nikon ) using a 514 laser with 525–555 nm emission filter on a Nikon Spinning Disc Confocal microscope . The camera was a 2X Andor Ultra 888 EMCDD ( Oxford Instruments , UK ) . The light source was a HBO Arc lamp . During imaging , cells were in a controlled environmental chamber surrounding the microscope ( Okolab Cage Incubator , Okolab USA INC , San Bruno , CA ) at 37°C , 5% CO2 and 90% humidity . Image J was used to threshold images to segment cells to generate an ROI for each cell . CF4 intensity for each cell ROI was measured . Background intensity was subtracted from each cell . For each condition , n > 200 cells from one well .	For an animal to grow , its cells have to divide . Cell division can only take place if the cell meets certain conditions: for example , the cell’s DNA must not be damaged . To ensure that cells only divide when these conditions are met , the cell goes through a series of stages or phases with checkpoints known as the cell cycle . The ability to control whether cells divide is essential for an animal to correctly form organs and tissues , or to heal wounds . Zinc is a metal that animals get in their diet , and when zinc levels are low , animals usually grow more slowly because their cells stop dividing . Around 2 billion people worldwide do not get enough zinc in their diet . Amongst other processes , zinc is necessary for DNA repair , which could explain why low levels of zinc stop the cell from dividing . However , the evidence for why zinc is required for cell division is contradictory . Although it seems clear that zinc is necessary for cells to progress through the cell cycle , it was unknown whether it is needed at a specific stage or whether it influences a cell’s decision to divide . Lo et al . have used microscopy to examine the effects that different levels of zinc had on individual cells grown in the lab . The results suggest that cells can monitor the levels of zinc in their environment , and respond to low levels by shutting down growth and cell division . This happens independently of the stage of the cell cycle a cell finds itself in , explaining the discrepancies between older studies . Additionally , the results show that although zinc is required for DNA repair , this process is not what triggers cells to stop dividing in the absence of zinc . In fact , low levels of zinc seem to be stopping cell division through a previously unknown mechanism . Lo et al . ’s findings illustrate the high sensitivity of cells to changes in zinc availability and highlight the importance of zinc in our diet . Further study of how cells determine zinc levels and how those levels affect the cell cycle may help explain zinc’s role in health and disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2020	Single cell analysis reveals multiple requirements for zinc in the mammalian cell cycle
It is well documented that transposable elements ( TEs ) can regulate the expression of neighbouring genes . However , their ability to act in trans and influence ectopic loci has been reported rarely . We searched in rice transcriptomes for tissue-specific expression of TEs and found them to be regulated developmentally . They often shared sequence homology with co-expressed genes and contained potential microRNA-binding sites , which suggested possible contributions to gene regulation . In fact , we have identified a retrotransposon that is highly transcribed in roots and whose spliced transcript constitutes a target mimic for miR171 . miR171 destabilizes mRNAs encoding the root-specific family of SCARECROW-Like transcription factors . We demonstrate that retrotransposon-derived transcripts act as decoys for miR171 , triggering its degradation and thus results in the root-specific accumulation of SCARECROW-Like mRNAs . Such transposon-mediated post-transcriptional control of miR171 levels is conserved in diverse rice species . Transposable elements ( TEs ) constitute a large fraction of eukaryotic genomes . Given their mutagenic potential and largely unknown functions , they were often considered as genomic parasites that are silenced by host epigenetic mechanisms ( Fultz et al . , 2015; Girard and Hannon , 2008 ) . However , there is increasing evidence that TEs contribute to various chromosomal functions , to the evolution of genomes by increasing genetic variation , and to the direct regulation of genes ( Lisch , 2013 ) . Several studies have revealed that TEs in plants endow genes with both coding and regulatory sequences ( Lisch , 2013 ) . For example , the Arabidopsis transcription factors FHY3 and FAR1 , involved in light signalling , are derived from the transposase of the Mutator-like DNA transposon ( Hudson et al . , 2003 ) . The domestication of hAT and Mutator-like transposases contributed to the evolution of the DAYSLEEPER and MUSTANG gene families , respectively . DAYSLEEPER was shown to play a critical role in plant development ( Bundock and Hooykaas , 2005; Cowan et al . , 2005; Knip et al . , 2013; Knip et al . , 2012 ) . More recently , a protein derived from the transposase of the Pif/Harbinger transposon family was shown to be an inhibitor of POLYCOMB REPRESSIVE COMPLEX 2 ( Liang et al . , 2015 ) . TEs residing outside protein-coding regions of genes can influence their expression by interfering with promoters , providing enhancers , or altering RNA processing and/or epigenetic regulation . For example , TEs residing in introns or UTRs may alter the availability of splicing sites and/or splicing efficiencies . They can also shift polyadenylation signals or supply binding sites for miRNA and RNA-binding proteins ( Feschotte , 2008 ) . In contrast to the numerous examples of local influence on gene regulation in cis , examples of TEs mediating the regulation of distant genes are rare . For example , the Arabidopsis ddm1 mutant , which is impaired in epigenetic suppression of transposon-derived transcription , accumulates 21-nt small RNAs derived from Athila retrotransposons . These small RNAs impair the levels and the translation of the mRNA of OLIGOURIDYLATE BINDING PROTEIN1 ( UBP1 ) activated by abiotic stress ( McCue et al . , 2012 ) . Interestingly , although more than 20 genes in Arabidopsis have putative binding sites for transposon-derived small RNAs that would allow regulation analogous to that of UBP1 , such a network of interactions has not been documented so far ( McCue et al . , 2013 ) . TE transcripts often contain features resembling micro RNA ( miRNA ) genes ( Li et al . , 2011 ) or sequences that are potential targets of miRNAs ( Creasey et al . , 2014 ) . miRNAs are a class of small non-coding RNAs that , directed by their sequences , selectively repress gene expression by translation inhibition or cleavage of mRNA ( Rogers and Chen , 2013 ) . Interestingly , miRNAs can interact both with their target mRNAs and also with other RNAs containing similar binding sites . Such ‘rival’ RNAs , which were seen as competing endogenous RNAs ( ceRNAs ) ( Kartha and Subramanian , 2014; Salmena et al . , 2011; Tay et al . , 2014 ) , can be derived from pseudogenes or long non-coding RNAs and also from protein-coding mRNAs . They may also appear in the form of circular RNAs . Although studies of their transgenic overexpression support activity as miRNA sponges , a possible biological role has not been demonstrated so far by their loss of function ( Thomson and Dinger , 2016 ) . Although , transposon-derived transcripts were not considered previously to be a source of ceRNAs , we decided to search rice transcriptomes for indications of this activity . Approximately 35% of the rice genome consists of TEs , which is significantly higher than in Arabidopsis ( 14% ) ( International Rice Genome Sequencing Project , 2005; Arabidopsis Genome Initiative , 2000 ) . More important and similar to maize ( Erhard et al . , 2009; Hollick et al . , 2005; Parkinson et al . , 2007 ) , rice mutants impaired in epigenetic silencing of transposons , such as DICER-LIKE 4 ( DCL4 ) or RNA-DEPENDENT RNA POLYMERASE 6 ( RDR6 ) , show severe developmental abnormalities ( Liu et al . , 2007; Song et al . , 2012 ) , whilst the corresponding mutants of Arabidopsis have no apparent morphological aberrations ( Xie et al . , 2005 ) . This dissimilarity suggests that restrained TE-derived transcription is important for rice development ( Liu et al . , 2007; Song et al . , 2012; Wei et al . , 2014 ) . Here we have specifically investigated TE-derived transcripts as potential regulators of rice development . We found that numerous TEs display patterns of transcriptional activity that are associated with particular plant tissues . Remarkably , a significant proportion of TE-derived transcripts correlate with the mRNA levels of genes transcribed in the same tissues and the two classes of transcripts often share patches of homology . Notably , the sequences of these patches appear to be significantly enriched for miRNA-binding sites . Therefore , we investigated whether some of the transposon-derived transcripts act as ceRNAs . Experiments to test this hypothesis led to the identification of a novel domesticated retrotransposon that is highly expressed in rice roots and that acts as a ceRNA post-transcriptionally controlling the level of miR171 . This particular ceRNA is also a target mimic of miR171 , which potentially enhances its sponging activity towards miR171 . Tissue-specific adjustment of miR171 levels is essential to the proper development of roots and this appears to be regulated by a retrotransposon-derived ceRNA . We demonstrated that mutations in its miR171-binding site result in an abnormal root system . To examine tissue-specific abundance of TE-derived transcripts in rice , we accessed publicly available RNA sequencing ( RNA-seq ) datasets for various tissues of rice ( Figure 1A ) . We considered only the datasets of Japonica rice , cultivar Nipponbare and applied the same data-processing pipeline to raw sequencing results generated in different laboratories ( details in the Materials and methods ) . This way we achieved consistent results and samples representing particular tissues were clustered together ( Figure 1A ) . Such combined dataset yielded 2961 transcribed TEs ( filtered for maximal RPKM ( Reads Per Kilobase per Million reads ) >1 ) . Remarkably , the TEs were transcribed in most rice tissues and their transcriptomes exhibit clear tissue specificity ( Figure 1A and Figure 1—figure supplement 1A ) . The rice expression patterns differ from those of Arabidopsis , where TEs are activated in a non-selective way and only in seed endosperm and the vegetative cells of pollen grains ( Figure 1—figure supplement 1B ) ( Slotkin et al . , 2009 ) . Thus , in rice , the two-dimensional correlation matrix of TE transcriptomes showed distinct TE groups reflecting their RNA abundance in various tissues and at different developmental stages ( Figure 1—figure supplement 1A ) . In contrast , Arabidopsis TEs exhibit more uniform expression patterns ( Figure 1—figure supplement 1B ) . We detected TE-derived transcripts in rice tissues that do not contribute to the germ line ( e . g . endosperm , leaves and roots ) . These transposon activities , even when resulting in insertions , would not be transmitted to the next generation . In the case of such apparently unproductive TE activity , reactivated TEs may possibly be regulatory or , as in Arabidopsis , may be the RNA substrates of small RNAs involved in TEs silencing ( Creasey et al . , 2014 ) , or may simply reflect insignificant transcriptional noise . For regulatory activity that influences gene expression , we assumed that TE transcripts would share homologies with the transcripts of co-expressed genes . Indeed , multiple alignments revealed homology patches in approximately 64% of co-transcribed TEs , while silent TEs matched only by 41% ( Figure 1—figure supplement 1C ) . Moreover , sequence comparison of expressed TEs and corresponding genes showed a strong bias towards sense direction , by which the aligned regions of TEs were much longer than for TEs matching gene transcript in the antisense direction ( Figure 1—figure supplement 1D , left panel ) . Intriguingly , orientation bias was observed when the alignments were sought against processed mRNAs of genes but not when introns were included ( Figure 1—figure supplement 1D ) . Next , we sought sequence features differentially enriched in the co-expressed TEs and found enrichment for miRNA-binding sites ( Figure 1B ) . This raised the possibility that some TE transcripts interfere with miRNA-mediated gene regulation , possibly by competing for miRNA binding . As two different RNAs interacting with the same miRNA would need to be co-expressed in the same tissue , we compared tissue-specific expression of TEs and matching genes . We found 763 and 400 TE-gene pairs in sense and antisense orientation , respectively , including 282 sense and 111 antisense pairs with correlation coefficient above 0 . 5 ( Figure 1C ) . Such correlated expression patterns between mRNAs of genes and TE-derived transcripts in sense orientation was most evident in roots ( Figure 1—figure supplement 1E ) . Collectively , the results of our examination of tissue-specific transcriptomes are consistent with the hypothesis that TEs regulate gene expression by miRNA sequestration . To test this hypothesis , we rigorously re-analysed 61 root-specific rice transcriptomes and selected a particular TE , which we named MIKKI ( ‘decoy’ in Korean ) , for further investigation ( Figure 2A ) . First , we validated RNA-seq results of root-specific transcription of MIKKI by RT-qPCR ( Figure 2B ) . To distinguish the spliced transcript from precursor mRNA ( pre-mRNA ) , we designed primers across exon junction or within the intron , respectively ( Figure 2B , left and right panel ) . The RT-qPCR results confirmed that the mature MIKKI transcript is highly abundant in roots , present at low levels in leaves , and almost absent in panicles ( Figure 2B , left panel ) . A similar expression pattern was observed for unspliced RNA but at much lower levels ( Figure 2B , right panel ) . These results are consistent with tissue-specific regulation of MIKKI at the transcriptional level . MIKKI is a TE-derived locus which includes Osr29 Long Terminal Repeat ( LTR ) retrotransposon . Based on sequence divergence between the two LTRs , an Osr29 element transposed about 3 . 7 million years ago ( mya , Figure 2C and Figure 2—figure supplement 1A ) . We also found sequences of three further retrotransposons , BAJIE , Osr30 and Osr34 , inserted subsequently into Osr29 ( Figure 2C ) . Advanced degeneracy prevented estimate of the insertion times of Osr30- and Osr34-related sequences; however , the generation time of the solo LTR derived from the BAJIE family was estimated to be approximately 1 . 2 mya ( Figure 2C and Figure 2—figure supplement 1B ) . The MIKKI gene product was predicted to encode just a partial reverse-transcriptase ( RTase ) protein and no other protein domains were found ( Figure 2—figure supplement 1C ) . Given that several amino acid residues essential for catalytic activity of RTase are mutated in MIKKI’s RTase ( Figure 2—figure supplement 1D ) , it seems unlikely that MIKKI’s RTase domain would be active . Thus , we concluded that MIKKI is not expected to have regulatory role at the protein level . Most important , the mature transcript of such a rearranged Osr29 ( MIKKI ) was found to contain an imperfect binding site for miR171 , generated by a splicing event joining BAJIE solo LTR sequences to specific sequences of Osr29 ( Figure 2C , E and Figure 2—figure supplement 1E ) . miR171 is one of the miRNAs conserved across the plant kingdom and previous studies revealed that Arabidopsis miR171 ( ath-miR171 ) is abundant in flowers but sparse in roots ( Figure 2—figure supplement 2C–E; Llave et al . , 2002 ) . Rice miR171 ( osa-miR171 ) displays a similar expression pattern ( Figure 2D , left panel and Figure 2—figure supplement 2B ) . Thus , miR171 levels seem to be similar in particular tissues of these two distant species , highest in reproductive organs and lowest in roots . It is well documented that ath-miR171 targets mRNAs encoding SCARECROW-Like ( SCL ) transcription factors for cleavage and , thus , SCL transcript levels display patterns opposite to miR171 ( Llave et al . , 2002 ) . The same SCL transcript distribution was observed in rice ( Figure 2D , right panel ) , implying the regulation of SCL transcript stability also by osa-miR171 . Moreover , the sequence identity of rice and Arabidopsis SCL mRNAs across the miR171-binding region is also consistent with the evolutionary conservation of miR171-mediated cleavage of SCL transcripts ( Figure 2—figure supplement 2F ) . Indeed , analyses of the RNA degradome in rice panicles ( Wu et al . , 2009 ) revealed specific cleavage of OsSCL21 mRNAs at the osa-miR171 binding region ( Figure 2E , left panel ) . We also examined whether the MIKKI transcript is also targeted by osa-miR171 but found no signals indicative of site-directed cleavage of MIKKI transcripts at the putative miR171-binding site ( Figure 2E , right panel ) . Importantly , there are two mismatches in the miR171-binding region of MIKKI at positions 11th and 14th . Conservation of nucleotides at these sites is known to be essential for target RNA cleavage ( Jeong et al . , 2013; Liu et al . , 2014; Llave et al . , 2002 ) . It is , therefore , possible that the mismatches around the cleavage sites in MIKKI transcripts attenuate the cleavage activity of osa-miR171 . Altogether , these data are consistent with the possibility that MIKKI is a target mimic of osa-miR171 in rice roots . To examine the target mimicry of the MIKKI transcript towards miR171 , we overexpressed MIKKI in both rice and Arabidopsis ( Figure 3—figure supplement 1A and B , top panel ) , and applied RT-qPCR analyses . miR171 levels were downregulated in independent transgenic lines generated from both plant species ( Figure 3A , top panel and Figure 3—figure supplement 1B , middle panel ) and the transcript levels of target genes were markedly upregulated ( Figure 3A , bottom panel and Figure 3—figure supplement 1B , bottom panel ) . Previous studies revealed abnormalities in floral organs of Arabidopsis plants in which ath-miR171 levels were decreased by overexpression of artificial target mimics ( Ivashuta et al . , 2011; Todesco et al . , 2010 ) . Consistent with these observations , plants ectopically overproducing MIKKI transcripts also displayed severe defects in reproductive organs and low fertility ( Figure 3B , C and Figure 3—figure supplement 1C ) . To address directly the developmental role of the MIKKI retrotransposon , we generated the MIKKI mutants mikki-1 and mikki-2 using CRISPR-Cas9 ( Miao et al . , 2013 ) . To ensure the targeting specificity of guide RNAs ( gRNA ) , we designed them to target the unique junction region between Osr29 and BAJIE ( Figure 3—figure supplement 2A ) . Transgenic plants were examined by sequencing for mutations in this region and two independent alleles were found ( Figure 3—figure supplement 2A and B ) . The mikki-1 allele had a 2 bp deletion at the splice donor site that resulted in retention of the intron . Intron retention disrupted the miR171-binding site and generated multiple premature stop codons ( Figure 3—figure supplement 2A ) . This transcript is likely recognized by a nonsense-mediated mRNA decay pathway and rapidly turned over ( Shoemaker and Green , 2012 ) . Indeed , the RT-qPCR analyses revealed thousand-fold reduction of MIKKI transcripts in the mikki-1 mutant ( Figure 3—figure supplement 2C ) . The mikki-2 allele has an 8 bp deletion in the region containing the miR171-binding site ( Figure 3—figure supplement 2A ) . This deletion did not alter RNA levels but was predicted to lose target recognition by osa-miR171 ( Figure 3—figure supplement 2A and C ) . Next , we performed RT-qPCR on the wildtype ( wt ) and the mutants . The levels of osa-miR171 were high in both mikki-1 and mikki-2 ( Figure 3D , top panel and Figure 3—figure supplement 2E ) . This correlated with a decrease in RNA levels of OsSCL21 targeted by osa-miR171 ( Figure 3D , bottom panel ) . In Arabidopsis , mutation of SCLs leads to defects in root development ( Wang et al . , 2010 ) . From a Korean rice seed bank we obtained two independent mutant alleles of OsSCL21 that showed the highest transcript levels among OsSCLs targeted by miR171 ( Figure 3—figure supplement 3 ) . Similar to Arabidopsis , the roots of both osscl21 mutants were shorter than wt ( Figure 3—figure supplement 2D ) . Subsequently , we examined the development of mikki-1 and mikki-2 roots . Root lengths were affected in both mutants , resembling mutants in the OsSCL21 gene ( Figure 3E and Figure 3—figure supplement 2D ) . Histological analyses were also performed to observe the cellular consequences of MIKKI mutation . Both mutants showed reduced cell elongation above meristematic region , while the cell widths were similar to wt ( Figure 3F and G ) . These data are consistent with the hypothesis that MIKKI negatively regulates osa-miR171 levels in rice roots , acting through a ceRNA containing target mimic site for osa-miR171 . Next , we asked whether post-transcriptional regulation by a ceRNA with target mimicry is the major regulatory mechanism governing tissue-specific levels of osa-miR171 . For this , we determined the levels of the primary transcript of osa-miR171 ( pri-osa-miR171 ) in MIKKI overexpression and mutant plants ( Figure 4A , B and Figure 4—figure supplement 1A ) . The abundance of pri-osa-miR171 was similar in different rice tissues and was not affected by the alteration of MIKKI transcript levels or mutation , implying that mature osa-miR171 is regulated post-transcriptionally by the activity of MIKKI . In contrast to rice , the tissue-specific distribution of primary transcripts of miR171 in Arabidopsis was the same as the mature miRNA , which is consistent with transcriptional regulation and thus an entirely different regulatory mechanism ( Figure 4C ) . MIKKI is present and has a conserved structure in AA-genome Oryza species ( Figure 4—figure supplement 2A and B ) , suggesting strong selective advantage of this particular transposon . MIKKI is present in the genomes of Oryza sativa ssp . indica , O . rufipogon , O . nivara , O . barthii , and O . glaberrima ( Figure 4—figure supplement 2A ) . Furthermore , insertion of the BAJIE-derived solo LTR and the resulting intron with a miR171-binding site at the splice junction are perfectly conserved ( Figure 4—figure supplement 2B ) , implying that the formation of a splicing-dependent miR171 binding site retained in these related species . We examined MIKKI splicing in five of these species using available RNA-seq data ( Zhai et al . , 2013; Zhang et al . , 2016; Zhang et al . , 2014 ) and detected identical splicing patterns of the critical intron 4 ( Figure 4—figure supplement 3A ) . Moreover , we found that the MIKKI homolog of Indica rice displays a developmental expression pattern similar to Japonica rice ( Figure 4—figure supplement 3B ) . We also examined tissue-specific levels of primary transcripts and mature miR171 in monocotyledonous Brachypodium ( Figure 4—figure supplement 1B and C ) . As in rice , the primary transcripts of miR171 were high in all tissues examined , suggesting analogous post-transcriptional control of miR171 levels . However , so far we have not identified an MIKKI-related element in the genome of Brachypodium . Since the transcription of TEs is usually controlled by epigenetic mechanisms , we examined DNA methylation and selected histone modifications associated with MIKKI in different rice tissues ( Figure 5 ) . In roots , where MIKKI is actively transcribed , its upstream region was enriched in lysine 4 tri-methylation of histone H3 ( H3K4me3 ) and depleted of suppressive H3K9me2 ( Figure 5B ) . DNA methylation levels were also lower than in panicles ( Figure 5C ) . Analysis on the public RNA-seq data generated from rice OsDCL3a RNAi knock-down lines also showed derepression of MIKKI ( Figure 5—figure supplement 1 ) . These epigenetic signatures were well correlated with tissue-specific transcription of MIKKI . In summary , we propose a model by which MIKKI influences rice root development via the regulation of osa-miR171 levels by tissue-specific expression of a ceRNA encoding target mimicry of miR171 ( Figure 6 ) . In Arabidopsis , transposable elements are mostly silenced by epigenetic mechanisms preventing their transcription during development of the sporophyte . In contrast , in plant species such as maize or rice , transposon-derived transcripts are detected during specific developmental transitions or in various organs ( Li et al . , 2010; Tamaki et al . , 2015 ) . Since most of the tissues examined do not contribute to the formation of gametophytes and thus to transgenerational inheritance , the benefits of transposon transcription remain unclear . The prevalent view is that their transcripts are a source of mobile small RNAs that , if transported into germline progenitor cells , would contribute to silencing of transposons there and thus prevent their transgenerational accumulation ( Calarco et al . , 2012; Creasey et al . , 2014; Slotkin et al . , 2009 ) . An alternative explanation for the developmental regulation of transposon-derived transcription , however , is that their transcripts have particular functions in a specific tissue or organ . To examine this latter possibility , we systematically analysed tissue-specific transcriptomes of rice transposons by re-analysing available raw RNA sequencing data . These analyses uncovered a surprisingly high fraction of TE-derived transcripts in specific tissues or developmental stages of rice plants . We also observed that transposon-derived transcripts are enriched in miRNA binding sites . It has been proposed that Arabidopsis miRNAs trigger generation of transposon-derived small RNAs that contribute subsequently to transposon silencing . These small RNAs may later spread to other cell types ( Creasey et al . , 2014; Martínez et al . , 2016 ) . This scenario is consistent with the transposon defence hypothesis described above and similar mechanisms could certainly operate in rice . Alternatively , transposon-derived RNAs containing binding sites for miRNAs could also act as ceRNAs and there are experimental examples in Arabidopsis supporting this possibility . The first and the physiologically important example of an Arabidopsis ceRNA was non-coding RNA INDUCED BY PHOSPHATE STARVATION 1 ( IPS1 , Franco-Zorrilla et al . , 2007 ) . The IPS1 locus is transcriptionally activated upon phosphate starvation and encodes RNA that binds to miR399 . miR399 also targets the mRNA PHO2 gene that encodes an E2 ubiquitin-conjugating-like enzyme affecting the phosphate content of shoots . Importantly , the miR399-binding region in IPS1 RNA has a 3-nt bulge in the cleavage site and , similar to MIKKI , it is resistant to cleavage . This unproductive binding of miR399 to IPS1 RNA triggers miRNA degradation ( Yan et al . , 2012 ) , thus reducing its level . The mechanism of such a miRNA decoy was termed ‘target mimicry’ . It has been suggested that the small RNA-specific nucleases in Arabidopsis reduce miRNA levels when target mimics are overexpressed ( Ramachandran and Chen , 2008; Yan et al . , 2012 ) . However , the mechanism of miRNA degradation has not been fully elucidated and it is not known how miRNA is recognized for degradation when associated with RNA encoding target mimics but not when associated with mRNAs encoding its bona fide targets . Subsequently , considerable efforts have been made to identify miRNA target mimics in genomes and transcriptomes of plants ( Fan et al . , 2015; Meng et al . , 2012 ) and mammals ( Clark et al . , 2014; Helwak and Tollervey , 2014; Imig et al . , 2015 ) . In addition , further examples of their biological activities have received experimental support ( Franco-Zorrilla et al . , 2007; Wang et al . , 2015 ) ; H . -J . Wu et al . , 2013 ) . Notably , experimental support for the activity of ceRNAs , including those from plants containing ‘target mimicry’ sites , is based on transgenic overexpression of micro RNAs or transcripts containing their target mimics . Unfortunately , these assays significantly alter the natural stoichiometry of such regulatory systems . As a consequence , the role of ceRNAs in nature has been queried , given that their relatively low abundance may be insufficient to significantly alter the levels of very dynamically regulated miRNAs ( Thomson and Dinger , 2016 ) . In addition , mathematical modelling of miRNA target competition supports the notion that target mimics and miRNAs must be at particular levels for maximal effect of target mimicry ( Bosia et al . , 2013; Figliuzzi et al . , 2013; Yip et al . , 2014; Yuan et al . , 2015 ) . Since MIKKI transcripts in rice roots are in ample excess over OsSCL mRNAs ( Figure 2B and D ) , the proportions of the components of the MIKKI-miR171-OsSCLs module appear to be naturally sufficient for highly effective regulatory activity . Moreover , we have directly examined the biological role of MIKKI mRNA as a ceRNA containing target mimics of osa-miR171 by using site-directed mutation of its miRNA binding site or mutations in the splicing site that result in its depletion . These experiments , circumventing artificial changes in the stoichiometry of the interacting components , revealed a regulatory function of MIKKI transcripts in the proper development of rice roots . The tissue-specific posttranscriptional control of miR171 in rice contrasts with restriction of miR171 availability in Arabidopsis roots , which seems to be determined by transcription of miR171 . Activation of miR171 is thought to be by AtSCL proteins binding directly to promoters ( Xue et al . , 2014 ) . Given that the promoter sequences of miRNA-encoding genes are generally less conserved than miRNA-coding regions ( Zhu et al . , 2015 ) , corresponding genes in different plant species may be the subject of different transcriptional regulation . The levels of miR171 in different plant tissues are decisive for plant development and posttranscriptional control , implemented by a domesticated retrotransposon , reinforces differential organ distribution of mature miR171 . This mechanism seems to be highly conserved among distantly related rice species , suggesting a strong selective advantage . The fact that osa-miR171 and MIKKI transcript levels were not affected by OsSCL21 mutation ( Figure 3—figure supplement 3D and E ) further supports the hypothesis of independent and unique post-transcriptional controlling mechanism emerged . Interestingly , miR171 in the model grass Brachypodium also seems to be regulated posttranscriptionally but a potential ceRNA has not yet been identified . It has been stated frequently that transposition bursts make a large contribution to host plant genome structure and function ( Lisch , 2013 ) . However , examples of transposon-mediated control of plant development through the regulation of distantly located genes have not been reported . Although TEs are an ample source of miRNAs and can potentially act as potent target mimic , they have not been examined for such function . Our study provides the first example of the TE-derived target mimic and thus a novel mechanism for TEs acting as trans-acting regulators of genes . However , for an effective target mimic several conditions should be met , including high transcript levels , good binding affinity and advantageous stoichiometry to miRNA target genes , therefore it is currently still difficult to predict how general this regulatory mechanism will be . We have discovered many more putative TE target mimics in the rice genome ( listed in Supplementary file 1 ) . However , genetic interference with these hypothetical regulatory loops is difficult due to multicopy components and thus genetic redundancy , as is the case for most transposons . Potentially , the increasingly efficient site-specific alteration of genomes by CRISPR-Cas9 or a population genetics approach using natural variation may improve functional accessibility to the transposon-derived fractions of plant genomes and reveal the extent of their regulatory input . Husks of rice seeds were removed and the seeds were surface sterilized in 20% bleach and germinated in ½-strength Murashige and Skoog media . The 2-week-old seedlings were transferred to soil and grown to maturity in a greenhouse . Root and leaf samples were harvested from 2-week-old seedlings and panicles collected immediately after heading . The rice strains used in this study were Oryza sativa ssp . japonica cv . Nipponbare , O . sativa ssp . japonica cv . Hwayoung and O . sativa ssp . indica cv . IR64 . The mutant lines of osscl21 were identified from a T-DNA tagged population established at Kyunghee University , Korea and genotyped for the selection of homozygotes . ( http://cbi . khu . ac . kr/RISD_DB . html/ ) . For rice transcriptome analysis , raw FASTQ files of the following RNA sequencing datasets were downloaded: GSE16631 , DRA000385 , SRP008821 , DRA002310 , SRP028376 and GSE50778 . The adapter-trimmed clean reads were mapped to the reference genome of MSU7 using TOPHAT 2 . 0 , with most of the options set to the default but with some optimization ( -g 300 ) . Cufflinks was used to call the RPKM . All the downstream analyses and plotting , for example , heat-maps of transcript levels and correlation matrix , box/violin plots , and statistical analysis were performed in R studio . The transcriptome data of different Oryza species were obtained from PRJNA264484 , PRJNA264480 , PRJNA264485 , SRP070627 , GSE41797 and analysed as described above . Arabidopsis transcriptome data were obtained from PRJNA314076 and analysed as for the rice transcriptome using the TAIR10 reference genome . For the degradome and small RNA analysis , raw FASTQ files from GSE18251 and GSE16350 were downloaded and mapped uniquely to the MSU7 reference genome using BOWTIE2 . The resulting BAM file was visualized by IGV . Arabidopsis small RNA-seq data were obtained from GSE28591 and analysed as for the degradome . HTSeq was used to calculate the read counts for each miRNA . Total RNA was extracted using the RNeasy Plant mini kit ( Qiagen , Hilden , Germany ) following the manufacturer’s recommendations . Reverse transcription reactions were performed using the VILO RT kit ( Invitrogen , California , USA ) with a random hexamer for priming . Real-time quantitative PCR was carried out using the Roche Light-Cycler ( Roche , Basel , Switzerland ) in a volume of 10 µl and analysed by the ΔΔCt method . All data in this study are the average of three biological replicates performed in technical triplicate ± standard deviation and normalized against eEF1α . An Ncode miRNA first-strand cDNA synthesis kit ( Invitrogen , California , USA ) was used for miRNA quantification; normalization was against miR166 , which is expressed constitutively in rice ( Figure 2—figure supplement 2B ) . Sequences of primers used are listed in Supplementary file 2 . A total of 15 µg of RNA was separated on 15% urea-TBE gels ( Thermo Fisher Scientific , Massachusetts , USA ) , transferred to Hybond N + nylon membranes ( Amersham Biosciences , Buckinghamshire , UK ) and fixed chemically using EDC ( Sigma-Aldrich , Missouri , USA ) . The membranes were prehybridized for 1 hr and hybridized for at least 16 hr in DIG Easy Hyb buffer ( Roche , Basel , Switzerland ) at 37°C . Membranes were washed twice with 2X SSC ( saline sodium citrate ) , 0 . 1% SDS . Immunological detection of DIG-labeled probe was performed using DIG wash and block buffer set ( Roche , Basel , Switzerland ) and DIG luminescent detection kit ( Roche , Basel , Switzerland ) . Luminescent signal was detected with Amersham Imager 600 ( Amersham Biosciences , Buckinghamshire , UK ) . Roots from 4-day-old rice plants were fixed in FAA ( formaldehyde , acetic acid , ethanol ) solution overnight in cold room , wax embedded using Leica ASP300 tissue processor ( Leica Biosystems , Wetzlar , Germany ) and sectioned by 4 µm using microtome ( Leica Biosystems , Wetzlar , Germany ) . After dewaxing and ethanol washing , samples were stained with Calcofluor White ( Sigma-Aldrich , Missouri , USA ) to visualize cell wall . Images were taken with a Zeiss LSM 700 confocal microscope ( Leica Biosystems , Wetzlar , Germany ) . Leaf and root samples were collected from Oryza sativa ssp . japonica cv . Nipponbare plants grown for 2 weeks under short-day conditions ( 10 hr light/14 hr dark ) . Panicles were harvested immediately after heading from plants grown in the greenhouse . Samples were crosslinked with 1% formaldehyde , flash-frozen , and ground in liquid nitrogen . Chromatin was fragmented by sonication and immunoprecipitated using the following antibodies: H3K4me3 ( ab8580; abcam , Cambridge , UK ) and H3K9me2 ( ab1220; abcam , Cambridge , UK ) . The immunoprecipitated DNA was quantified by qPCR and normalized against levels of input and the reference genes indicated . All the results are presented as means ± standard deviation ( sd ) of three biological replicates performed in technical triplicate . Sequences of primers used are listed in Supplementary file 2 . Genomic DNA was extracted from rice tissues using the DNeasy plant mini kit ( Qiagen , Hilden , Germany ) . The Epitect bisulfite kit ( Qiagen , Hilden , Germany ) was used for bisulfite conversion of unmethylated cytosines . The primer design and data analysis used kismet ( Gruntman et al . , 2008 ) . At least 15 clones from each sample were analysed . Sequences of primers used are listed in Supplementary file 2 . A cDNA fragment of MIKKI from the start to the stop codon was amplified , cloned into the pUN1901 and pGPTVII binary vectors , and transformed into rice and Arabidopsis , respectively ( Walter et al . , 2004; Wang et al . , 2004 ) . For rice transformation , embryo-derived 2-week-old calli were immersed in agrobacterium-containing media . Transgenic rice plants were obtained after antibiotic selection and differentiation of plantlets . The detailed procedure was as described previously ( Nishimura et al . , 2006 ) . Arabidopsis transformation was by the floral dip method as described previously ( Clough and Bent , 1998 ) . The oligonucleotide of the designed guide RNA was inserted into the pOs-sgRNA entry vector and shuttled to the pH-Ubi-cas9-7 destination vector by the LR recombination reaction . The resulting binary vector was transformed into rice as described above ( Miao et al . , 2013 ) . To detect mutation by CRISPR-Cas9 , the region containing the targeted region from genomic DNA was amplified , cloned into the pGEM T-easy vector ( Promega , Wisconsin , USA ) and sequenced . Selected mutant lines were cultured to the next generation to segregate away the T-DNA and individuals homozygous for mutant allele were selected . Sequences of primers used are listed in Supplementary file 2 . Genome sequences of the selected Oryza species were obtained from Ensembl Plants ( http://plants . ensembl . org/ ) . Local BLAST analysis was performed manually using the MIKKI genomic sequence , followed by multiple sequence alignment in ClustalW2 and visualization by FigTree v . 1 . 4 . 2 and boxshade v . 3 . 21 . LTR retrotransposon age was estimated as described previously ( Ma and Bennetzen , 2004 ) . Briefly , for Osr29 , the divergence was calculated from sequence degeneracy of two LTRs . Age of insertion was computed using the equation: T = D/2 t , where T is the time since insertion , D is the divergence and t is the substitution rate of 1 . 3 × 10−8 per site per year as proposed previously ( Ma and Bennetzen , 2004 ) . For BAJIE solo LTR , the sequence was compared to the consensus of BAJIE LTR sequences . We assumed that the consensus BAJIE LTR sequence represents the youngest copy and used the equation of T = D/t . MIKKI , LOC_Os06g02304; OsSCL8 , LOC_Os02g44360; OsSCL21 , LOC_Os04g46860; OsSCL29 , LOC_Os06g01620 .	An organism’s genome contains all of the DNA the individual needs to survive and grow . Transposons are pieces of DNA that can move around the genome . They make up almost half of human DNA and over 85% of the DNA of major crop plants like maize , barley and wheat . When transposons move they can cause harmful changes in the regions where they insert into the DNA and so cells have mechanisms in place to tightly control the activities of the transposons . However , some transposons cause changes to DNA that are beneficial to the organism . Thus , the relationship between transposons and their host organisms is an example of a delicate but mostly peaceful coexistence . Although the cellular mechanisms controlling transposons are quite well known , the extent to which the transposons affect the ability of organisms to survive , develop and reproduce is poorly understood . A family of proteins known as the SCARECROW-like transcription factors are important for the roots of plants to develop properly . In other organs such as the leaves or flowers these proteins can cause developmental defects , so the plants carefully control where the proteins are made . Thus , plants normally produce a molecule called miR171 in leaves and flowers , but not in roots , that inhibits protein production by binding to and destabilising the RNA molecules that act as templates to make these proteins . Cho and Paszkowski have now identified a transposon that produces an RNA molecule with similarities to the RNA templates used to make the SCARECROW-like transcription factors . The experiments show that this transposon RNA is found in very high amounts in roots and mimics the transcription factor RNA so well that miR171 binds to it . This inactivates miR171 in roots to allow the SCARECROW-like transcription factors to be produced . These findings reveal a new mechanism by which transposons may regulate how plants develop and provide possible new approaches for boosting the growth of rice and other crop plants . Similar regulatory interactions between transposons and their host DNA may also be present in animals and other organisms .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology" ]	2017	Regulation of rice root development by a retrotransposon acting as a microRNA sponge
Diatoms are photosynthetic microorganisms of great ecological and biogeochemical importance , forming vast blooms in aquatic ecosystems . However , we are still lacking fundamental understanding of how individual cells sense and respond to diverse stress conditions , and what acclimation strategies are employed during bloom dynamics . We investigated cellular responses to environmental stress at the single-cell level using the redox sensor roGFP targeted to various organelles in the diatom Phaeodactylum tricornutum . We detected cell-to-cell variability using flow cytometry cell sorting and a microfluidics system for live imaging of oxidation dynamics . Chloroplast-targeted roGFP exhibited a light-dependent , bi-stable oxidation pattern in response to H2O2 and high light , revealing distinct subpopulations of sensitive oxidized cells and resilient reduced cells . Early oxidation in the chloroplast preceded commitment to cell death , and can be used for sensing stress cues and regulating cell fate . We propose that light-dependent metabolic heterogeneity regulates diatoms’ sensitivity to environmental stressors in the ocean . Diatoms are considered amongst the most successful and diverse eukaryotic phytoplankton groups , and are estimated to contribute 20% of global net primary production ( Armbrust , 2009; Nelson et al . , 1995; Malviya et al . , 2016 ) . They form massive blooms and are thus central to the biogeochemical cycling of important elements such as carbon , nitrogen , phosphorus , iron and silica , in addition to their important role at the base of marine food webs ( Armbrust , 2009; Nelson et al . , 1995; Morel and Price , 2003; Strass and Nöthig , 1996; Tréguer et al . , 2018 ) . As other phytoplankton , diatoms need to constantly acclimate to physicochemical gradients in a fluctuating environment . They are exposed to stress from different biotic and abiotic origins such as grazing , viruses , bacteria , allelopathic interactions , light availability , and nutrient limitations ( Bidle , 2015; Bidle , 2016; Ianora et al . , 2006; Thamatrakoln et al . , 2013; Thamatrakoln et al . , 2012; Kimura and Tomaru , 2015; van Tol et al . , 2017; Vardi et al . , 2006 ) . Importantly , induction of programmed cell death ( PCD ) in response to different stressors has been suggested as an important mechanism contributing to the fast turnover of phytoplankton and the rapid bloom demise ( Bidle , 2015; Bidle , 2016; Vardi et al . , 2007 ) . Recent studies suggested that diatoms can differentially respond to diverse environmental cues based on compartmentalized redox fluctuations that also mediate stress-induced PCD ( Graff van Creveld et al . , 2015; Rosenwasser et al . , 2014; Volpert et al . , 2018 ) . Reactive oxygen species ( ROS ) are known to play an important role in sensing stress and additional signals across kingdoms , from bacteria to plants and animals ( Vardi et al . , 1999; Mittler et al . , 2011; Suzuki et al . , 2012; D'Autréaux and Toledano , 2007; Dietz et al . , 2016 ) . They are produced as byproducts of oxygen-based metabolism in respiration and photosynthesis , by ROS generating enzymes , and due to various stress conditions ( Graff van Creveld et al . , 2015; D'Autréaux and Toledano , 2007; Sheyn et al . , 2016; Foyer and Noctor , 2016; Luo et al . , 2014; Vardi et al . , 2002; Waring et al . , 2010 ) . To maintain redox balance and avoid oxidative damage , cells harbor various ROS scavenging enzymes and small antioxidant molecules that regulate and buffer ROS levels , such as glutathione ( GSH ) , ascorbate and NADPH . ROS can cause rapid post-translational modifications of pre-existing proteins through oxidation , affecting their activity faster than changes in gene expression ( D'Autréaux and Toledano , 2007 ) . The specificity of the ROS signal is derived from the specific chemical species , its concentration , sub-cellular localization , temporal dynamics , and available downstream ROS-sensitive targets ( Graff van Creveld et al . , 2015; D'Autréaux and Toledano , 2007; Sheyn et al . , 2016; Noctor and Foyer , 2016; Owusu-Ansah and Banerjee , 2009; Tsukagoshi et al . , 2010 ) . Therefore , ROS production and redox metabolic networks can be used to sense and integrate information of both the metabolic state of the cell and its microenvironment . H2O2 is a relatively mild and stable ROS that can accumulate in cells due to various stress conditions , thus often serves as a signaling molecule ( Vardi et al . , 1999; Mittler et al . , 2011; Suzuki et al . , 2012; D'Autréaux and Toledano , 2007; Dietz et al . , 2016; Exposito-Rodriguez et al . , 2017 ) . It has a preferential activity towards cysteine residues , and can remodel the redox-sensitive proteome network ( Rosenwasser et al . , 2014; Dietz et al . , 2016; Noctor and Foyer , 2016 ) . In addition , H2O2 can diffuse across membranes ( depending on membrane properties ) and through aquaporins channels ( Bienert and Chaumont , 2014 ) . Combined properties as lower toxicity , diffusibility and selective reactivity make H2O2 suitable for studying signaling in various biological systems ( Dietz et al . , 2016; Noctor and Foyer , 2016 ) . Since many environmental stressors induce ROS generation ( Vardi et al . , 1999; Sheyn et al . , 2016; Waring et al . , 2010; Exposito-Rodriguez et al . , 2017; Asada , 2006; Diaz et al . , 2018; Diaz et al . , 2013 ) , application of H2O2 can reproduce the downstream cellular response ( Graff van Creveld et al . , 2015; Rosenwasser et al . , 2014 ) . Importantly , H2O2 application in marine diatoms led to oxidation patterns similar to other environmental stressors such as nutrient limitations , toxic infochemicals and high light , demonstrating that it induces similar physiological responses within the cell ( Figure 1 ) ( Graff van Creveld et al . , 2015; Rosenwasser et al . , 2014 ) . Although the H2O2 concentrations used in these studies were higher than measured in bulk seawater , which is typically in the nanomolar range ( Zinser , 2018 ) , local concentrations at the microenvironment of phytoplankton can be significantly higher due to patchiness in time and space ( Stocker , 2012; Chrachri et al . , 2018; Seymour et al . , 2017 ) . Local and temporal production of H2O2 by the cell itself or by its neighbors could lead to intracellular concentrations which are orders of magnitude higher than measured in bulk seawater in the field , especially during dense blooms or in aggregates and biofilms ( Diaz et al . , 2018; Diaz et al . , 2013; Stocker , 2012; Chrachri et al . , 2018; Seymour et al . , 2017 ) . H2O2 application in the model diatom P . tricornutum also led to the induction of cell death , in a dose-dependent manner , with characteristics of PCD that included externalization of phosphatidylserine , DNA laddering , and compromised cell membrane ( Graff van Creveld et al . , 2015 ) . Moreover , early oxidation of the mitochondrial GSH pool preceded subsequent cell death at the population level following exposure to H2O2 or diatom-derived infochemicals ( Graff van Creveld et al . , 2015; Graff van Creveld et al . , 2016 ) . In the current work , we investigated phenotypic variability within diatom populations in response to oxidative stress . Classically , phenotypic variability has been studied primarily in established bacterial , yeast and mammalian model systems with little ecological relevance ( Balaban , 2011; Raj and van Oudenaarden , 2008 ) . However , examination of phenotypic heterogeneity in phytoplankton cells in response to stress is still underexplored , and research has been carried out primarily at the population level , masking heterogeneity . Cell-to-cell variability could result in different cellular strategies employed by the population to cope with environmental stress , and may provide important insights into cell survival during bloom succession . We established single-cell approaches to measure in vivo ROS dynamics in the model diatom P . tricornutum using flow cytometry and microfluidics live-imaging microscopy . We used P . tricornutum strains expressing redox-sensitive GFP ( roGFP ) targeted to different sub-cellular compartments ( Rosenwasser et al . , 2014 ) and exposed cells to oxidative stress and high light conditions . The oxidation of roGFP is reversible , and can be quantified using ratiometric fluorescence measurements ( Meyer , 2008 ) . The roGFP oxidation degree ( OxD ) reports the redox potential of the GSH pool ( EGSH ) , which represents the balance between GSH and its oxidized form ( Meyer , 2008 ) . Therefore , roGFP OxD provides an important metabolic readout of the redox state of the cell , and represents the oxidation state of native proteins in the monitored organelles ( Rosenwasser et al . , 2014; Meyer , 2008 ) . We uncovered a previously uncharacterized phenotypic heterogeneity in the response of a marine diatom to oxidative stress and high light . Furthermore , we revealed a specific link between oxidation patterns in the chloroplast and cell fate regulation . Previous works demonstrated that P . tricornutum exhibits organelle-specific oxidation patterns in responses to diverse environmental stress conditions ( Figure 1 ) ( Graff van Creveld et al . , 2015; Rosenwasser et al . , 2014; Graff van Creveld et al . , 2016 ) . However , these findings were based on bulk measurements averaging the phenotypes within the population . In order to investigate heterogeneity within the population , we measured the response to oxidative stress in P . tricornutum strains expressing roGFP targeted to the chloroplast , nucleus and mitochondria at single-cell resolution using flow cytometry . At steady-state conditions without perturbations , roGFP OxD distribution in the population had a single distinct peak , representing a reduced state at all examined compartments ( Figure 2A , E , I and Figure 2—figure supplements 1–2 ) . Next , we examined the response of cells to oxidative stress by treatment with H2O2 at concentrations that led to oxidation patterns similar to other environmental stressors ( Figure 1 ) , and that also led to death of part of the population ( Graff van Creveld et al . , 2015 ) . Chloroplast-targeted roGFP ( chl-roGFP ) exhibited a distinct bimodal distribution following treatments of 50–100 µM H2O2 , revealing two distinct subpopulations of ‘oxidized’ and ‘reduced’ cells ( Figure 2B–D and Figure 2—figure supplement 3A–C ) . These subpopulations emerged within the first few minutes after H2O2 addition ( Figure 2B–D ) . The existence of these subpopulation is masked in bulk analysis ( Figure 2—figure supplement 1 ) , demonstrating the importance of single-cell measurements . In the ‘oxidized’ subpopulation , roGFP completely oxidized ( ~100% ) in response to H2O2 , reaching a similar distribution of the fully oxidized positive control ( 200 µM H2O2 ) ( Figure 2B–D and Figure 2—figure supplement 3G ) . However , in the ‘reduced’ subpopulation roGFP reached lower values of 30–43% OxD within 2–24 min post treatments , and then gradually recovered ( Figure 2B–D and Figure 2—figure supplement 3G ) . Only a minor fraction of the cells displayed intermediate oxidation , suggesting that these subpopulations represent discrete redox states . Interestingly , a larger fraction of cells was within the ‘oxidized’ subpopulation at 20–25 min post treatment compared to later time points , indicating that some cells were able to recover during this time ( Figure 2M ) . The proportion between these subpopulations stabilized after 46–51 min post treatment , and was H2O2-dose dependent , as more cells were within the ‘oxidized’ subpopulation at higher H2O2 concentrations ( Figure 2M ) . The quick emergence of stable co-existing ‘oxidized’ and ‘reduced’ subpopulations exposed underlying heterogeneity within the diatom population , resulting in a differential response to oxidative stress . This clear bi-stable pattern was unique to the chloroplast-targeted roGFP . Nuclear-targeted roGFP displayed a continuous distribution in response to H2O2 treatments , and no distinct subpopulations could be observed ( Figure 2F–H ) . Within minutes post treatment , nuclear roGFP exhibited fast oxidation even in response to a low H2O2 concentration of 50 µM , which had only a mild effect on the chloroplast ( Figure 2B , F ) . At that concentration , nuclear roGFP oxidation was followed by a gradual and much slower recovery , which lasted >5 hr post treatment ( Figure 2F ) . At higher concentrations , the entire population was oxidized within 3 min post treatment , and most cells remained stably oxidized >5 hr post treatment ( Figure 2G–H ) . The mitochondria-targeted roGFP exhibited a heterogeneous redox response , as seen in the 80 µM and 100 µM H2O2 treatments starting at ~24 min post treatment ( Figure 2K–L ) . However , distinct subpopulations were not clearly separated until later stages , and were not detected consistently in different experiments ( Figure 2K–L and Figure 2—figure supplement 4 ) . Therefore , we chose to focus on the chl-roGFP strain , which revealed two discrete subpopulations . Next , we examined the possible link between early chloroplast EGSH oxidation and subsequent H2O2-dependent cell death . We quantified cell death 24 hr post H2O2 treatment using flow cytometry measurements of Sytox green staining , which selectively stains nuclei of dead and dying cells . The fraction of ‘oxidized’ cells 1–2 hr post treatment was correlated with the fraction of dead cells at 24 hr ( Figure 3A , R2 = 0 . 89 , p=2 . 2·10−16 ) , suggesting that early oxidation in the chloroplast in distinct subpopulations may predict cell fate at much later stages . To investigate directly the link between early chl-roGFP oxidation and subsequent cell death , we used fluorescence-activated cell sorting ( FACS ) to sort cells based on chl-roGFP oxidation and measured their survival . Single cells of the ‘oxidized’ and ‘reduced’ subpopulations were sorted into fresh media at different time-points following the addition of 80 µM H2O2 , and colony forming units ( CFU ) were counted to assess survival 3 . 5–9 weeks later ( Figure 3B and Figure 3—figure supplement 1 ) . The CFU assay provides a direct link between chl-roGFP oxidation and the ability of individual cells to proliferate , and in addition enables to generate clonal populations which were also used for downstream analyses ( Figure 3—figure supplement 1 ) . When sorted 30 min post treatment , the ‘oxidized’ subpopulation exhibited a high survival rate ( 92 . 3 ± 1 . 4% ) that was similar to the ‘reduced’ subpopulation ( 94 . 1 ± 1 . 1% , p=0 . 24 , paired t-test ) and to sorted untreated control ( 96 ± 0 . 9% , p=0 . 29 , Dunnett test , Figure 3B ) . However , at later time-points the survival of the ‘oxidized’ subpopulation gradually diminished , and was significantly lower than both the ‘reduced’ and control sorted cells ( p<0 . 001 for all comparisons , paired t-test for comparisons with ‘reduced’ cells , Dunnett test for comparisons with control ) . When sorted 60 min post treatment almost half of the ‘oxidized’ cells recovered ( 45 . 1 ± 2% ) , but when sorted 100 min following treatment only 12 . 7 ± 2 . 1% survived ( p<0 . 001 , Tukey test , Figure 3B ) . These results suggest that after a distinct exposure time , cell death is induced in an irreversible manner in the ‘oxidized’ subpopulation . In contrast , the ‘reduced’ subpopulation from the same culture and treatment exhibited a high survival rate similar to the control at all time-points examined , demonstrating its resilience to the stress ( Figure 3B , P≥0 . 86 for all comparisons , Dunnett test ) . Corroborating these findings , cell death measurements using Sytox staining of sorted subpopulations also showed higher mortality in the ‘oxidized’ cells compared to the ‘reduced’ and control cells , which remained viable ( Figure 3—figure supplement 2 ) . Taken together , these results demonstrate that the ‘oxidized’ subpopulation was sensitive to the oxidative stress which led to induction of cell death in those cells , while the ‘reduced’ subpopulation was able to survive . In addition , we detected a distinct phase of ‘pre-commitment’ to cell death , ranging approximately 30–60 min in most cells , during which despite the strong oxidation in the chloroplast the fate of the ‘oxidized’ subpopulation is still reversible upon removal of the stress , and they are still able to survive . After this initial phase , the ‘oxidized’ cells were not able to survive even when the stress was removed from the system by sorting into fresh media . In order to track oxidation dynamics and subsequent cell fate of individual cells , we established a microfluidics platform for in vivo long-term epifluorescence imaging , under controlled flow , light and temperature conditions customized specifically for diatom cells ( Figure 4—figure supplement 1 and Videos 1–2 ) . We introduced cells expressing chl-roGFP into a custom-made microfluidics device , let the cells settle , and introduced treatments of either 80 µM H2O2 or fresh media ( control ) continuously for 2 . 5–3 hr , after which the treatment was washed by fresh media ( see Materials and methods , Figure 4—figure supplement 1A ) . In addition , the use of microfluidics enabled imaging of the basal OxD state of single cells prior to treatment , as well as the introduction of Sytox green at the end of the experiment to visualize cell death . We detected the distinct ‘oxidized’ and ‘reduced’ subpopulations following 80 µM H2O2 treatment , similar to the flow cytometry experiments ( Figure 4C , F , and Video 1 ) . However , no clear differences were observed in their OxD prior to treatment ( Figure 4—figure supplement 2 and Video 1 ) . The separation between the subpopulations emerged within 20 min of exposure to 80 µM H2O2 , and remained stable over the course of the experiment with the ‘oxidized’ subpopulation maintaining a high OxD above 80% ( Figure 4F , Figure 4—figure supplement 2B and Video 1 ) . The ‘reduced’ subpopulation exhibited an immediate response to H2O2 comparable with flow cytometry measurements , from 25–45% OxD before treatment to 30–65% OxD during the first 20 min post 80 µM H2O2 treatment ( Figure 4F , Figure 4—figure supplement 2B , and Video 1 ) . Following this initial oxidation , the ‘reduced’ cells recovered gradually over the next hours , reducing to 5–25% OxD 8 hr post treatment , even below the initial basal state ( Figure 4F , Video 1 ) . A gradual slow reduction was also observed in control cells over the course of the experiment ( Figure 4E , Video 2 ) , which may represent acclimation to the experimental setup or a diurnal redox alteration . Control cells did not oxidize in response to addition of fresh media ( Figure 4E and Figure 4—figure supplement 2A ) , excluding the possibility that the oxidation observed in 80 µM H2O2 treated cells was due to shear stress during treatment . We detected a clear correlation between initial oxidation in the chloroplast in response to oxidative stress and subsequent cell fate ( Figure 4A–G ) . Cells that exhibited high chl-roGFP oxidation within the first 40 min also died at a much later stage , while cells that maintained a lower OxD were able to recover ( Figure 4G ) . In addition , cells of the ‘reduced’ subpopulation and of control treatment were able to proliferate , further demonstrating their viability under these conditions ( Videos 1–2 ) . Logistic regression modeling of cell death as a function of chl-roGFP OxD at this time-point revealed a threshold of ~74% OxD , which discriminated with high accuracy ( 98 . 8% ) between cells that subsequently died and cells that survived the treatment ( 0 . 8% false positive , 1 . 7% false negative; Figure 4G and Figure 4—figure supplement 3 ) . These results corroborate the flow-cytometry analysis , and demonstrate that under these conditions early chloroplast EGSH response is linked to subsequent cell fate determination at the single-cell level . The differential chloroplast oxidation of the observed subpopulations could be due to genetic variability or due to phenotypic plasticity within the population . To differentiate between the two scenarios , we sorted chl-roGFP individual cells of the ‘oxidized’ and ‘reduced’ subpopulations 30 and 100 min post 80 µM H2O2 treatment as well as untreated control cells , and regrew them to generate clonal populations derived from cells exhibiting specific phenotypes . The clonal progeny cultures were subsequently exposed to 80 µM H2O2 and their chl-roGFP oxidation was measured ( Figure 3—figure supplement 1 ) . The two distinct subpopulations were detected in all the clones measured , and the fraction of the ‘oxidized’ subpopulation was again correlated with cell death ( Figure 5 and Figure 5—figure supplement 1 ) . Therefore , the different subpopulations observed did not originate from genetic differences , but rather represent phenotypic variability within isogenic populations . One possible source for phenotypic variability in genetically homogenous populations can be explained by differences in the cell cycle phase , as the cell cycle is linked to metabolic changes including redox oscilations ( Papagiannakis et al . , 2017; Burhans and Heintz , 2009; Mathis and Ackermann , 2016; Diaz-Vivancos et al . , 2015 ) . Therefore , we sorted the ‘oxidized’ and ‘reduced’ subpopulations 30 min following 80 µM H2O2 treatment into a fixation solution , and stained the fixed cells with 4′ , 6-diamidino-2-phenylindole ( DAPI ) to quantify DNA content for cell cycle analysis . The sorted ‘oxidized’ subpopulation had a higher fraction of cells at G1 ( 86 . 9 ± 1 . 8% ) compared to control untreated cells ( 76 . 8 ± 0 . 7% , p=0 . 0024 , Tukey test , Figure 6 ) . The ‘reduced’ subpopulation on the other hand had a smaller fraction of G1 cells ( 68 . 7 ± 2 . 2% ) compared to both control ( p=0 . 011 , Tukey test ) and ‘oxidized’ cells ( p=0 . 0001 , paired t-test ) , and exhibited a larger fraction of G2/M cells ( Figure 6 ) . These results demonstrate that although cell cycle phase alone cannot explain the differences between the subpopulations , it is linked to the chloroplast EGSH response to oxidative stress and may represent an important factor that affects H2O2 sensitivity in the population . Photosynthesis is a major source for reductive power as well as ROS in algal cells , and exposure to dark was shown to increase sensitivity to oxidative stress in another marine diatom ( Volpert et al . , 2018 ) . Therefore , we hypothesized that light regime will affect the bimodal pattern of chl-roGFP following oxidative stress , and investigated the effects of short exposure to darkness during daytime . Cells were treated with 0–100 µM H2O2 and were immediately moved to the dark for 90 min , after which they were moved back to the light ( dark treated , Figure 7A–C and Figure 7—figure supplement 1 ) . These cells were compared to cells that were kept in the light during this time ( light treated ) . The transition to the dark caused an immediate oxidation of the basal chl-roGFP OxD ( without H2O2 treatment ) , reaching a peak within 15 min ( Figure 7A , D ) . Then , while still under dark , chl-roGFP gradually reduced while maintaining a continuous distribution ( Figure 7A , D ) . Upon shifting back to the light , chl-roGFP reduced within 2 min back to its basal state prior to dark exposure ( Figure 7A , D ) . The dark mediated oxidation was specific to the chloroplast and was not detected in the nucleus ( Figure 7—figure supplement 2 ) , demonstrating organelle specificity of these redox fluctuations . The transition to the dark eliminated the bimodal pattern of chl-roGFP oxidation in response to H2O2 ( Figure 7C ) . No distinct subpopulations were observed while cells were under darkness even in cells treated with low H2O2 doses ( Figure 7B and Figure 7—figure supplement 1B–C ) . The transition to the dark increased H2O2 sensitivity in the entire population , and following treatment of 80 µM H2O2 and transition to the dark chl-roGFP fully oxidized in the entire population and remained stably oxidized even after transition back to the light ( Figure 7C ) . The bimodal pattern was regained only upon transition back to the light , and only at lower doses of 30 µM and 50 µM H2O2 , in which some or most cells were able to recover following this transition ( Figure 7B and Figure 7—figure supplement 1C ) . In accordance with the higher chl-roGFP oxidation , ‘dark’ treated cultures also exhibited higher mortality compared to ‘light’ treated cells ( p≤0 . 0064 for all pairs in 50–100 µM H2O2 treatments , t-test , Figure 7E ) . Therefore , we conclude that the mechanism generating the bimodal response in the chloroplast is light dependent and plays an important role in cell fate regulation in diatoms . Next , we investigated the effects of high light ( HL ) , a key environmental stressor in the marine ecosystem , which was shown to induce ROS generation in the chloroplast ( Waring et al . , 2010; Exposito-Rodriguez et al . , 2017; Asada , 2006 ) . We hypothesized that HL may cause H2O2 accumulation in the chloroplast , leading to a similar phenotypic variability as detected in response to H2O2 treatment . Cells were exposed to HL treatment of 2 , 000 µmol photons m−2 s−1 in order to evaluate the effect of HL exposure equivalent to full sunlight in nature ( Long et al . , 1994 ) , and as was used in previous studies in P . tricornutum ( Lepetit et al . , 2013 ) . Indeed , 1 . 5 hr after exposure to HL an ‘oxidized’ subpopulation started to emerge ( Figure 8A and Figure 8—figure supplement 1B ) . The fraction of ‘oxidized’ cells gradually increased over time of HL exposure ( Figure 8A–B and Figure 8—figure supplement 1A–F ) . These subpopulations were not detected in control cells that were kept under low light ( LL , Figure 8A and Figure 8—figure supplement 1G–H ) , nor in cells expressing nuclear targeted roGFP that were exposed to HL ( Figure 8—figure supplement 2A–B ) , demonstrating the specificity of the redox signal to the chloroplast . To measure the survival of the subpopulations that emerged under HL conditions , individual cells from the ‘oxidized’ and ‘reduced’ subpopulations were FACS sorted after different HL exposure times into agar plates for the single-cell CFU survival assay ( see Materials and methods , Figure 8C–D ) . CFU survival of the ‘oxidized’ subpopulation gradually decreased over time of HL exposure and was significantly lower than both control and ‘reduced’ cells ( p<0 . 001 for all comparisons . Tukey test was used for comparison with control , ANCOVA was used for comparison with ‘reduced’ and for the interaction with exposure time , Figure 8D ) . While most ‘oxidized’ cells survived ≤3 hr HL exposure ( 77 . 3 ± 4 . 2% CFU survival ) , longer exposure of >6 hr to HL led to only 3 . 1 ± 0 . 9% CFU survival ( Figure 8D ) . In contrast , the ‘reduced’ subpopulation in the same HL treatment exhibited high CFU survival of 92 . 7 ± 3 . 6% , and maintained high CFU survival at all time-points examined similar to the control ( p=0 . 93 , Tukey test , Figure 8D ) . Interestingly , when cells were exposed to 6 . 3 hr HL and then moved to 1 hr LL the separation between the subpopulations became clearer , with almost no cells with intermediate oxidation states ( Figure 8A and Figure 8—figure supplement 1E–F ) , resembling the response to H2O2 treatment ( Figure 2—figure supplement 3B ) . To conclude , these findings demonstrate that HL can generate heterogeneity in ROS accumulation in the chloroplast within diatom populations , leading to differences in survival probability and likely affecting sensitivity to additional stressors in the marine environment . Our current understanding of the mechanisms that mediate acclimation to environmental stressors in marine microorganisms , including diatoms , has been derived primarily from observations at the population level , neglecting any heterogeneity at the single-cell level . Co-existence of distinct subpopulations that employ diverse cellular strategies can be significant for the survival of this globally important phytoplankton group , as was shown in other microorganisms ( Balaban , 2011; Schreiber et al . , 2016; Sengupta et al . , 2017 ) . Here , we established a novel system for studying phenotypic variability in the marine diatom P . tricornutum using flow cytometry and a microfluidics system for live imaging microscopy . Based on a metabolic readout of chloroplast EGSH oxidation , we uniquely identified two distinct subpopulations that emerged as an early response to H2O2 and high light , demonstrating the importance of phenotypic variability in cell fate regulation in diatoms . We propose that in diatoms , chloroplast EGSH is involved in sensing specific environmental stress cues and in regulating cell fate ( Figure 9 ) . The chloroplast is a major source for generation of both ROS and reductive power to generate and recycle NADPH , thioredoxin and GSH ( Dietz et al . , 2016 ) . In plants , chloroplast-generated ROS were demonstrated to be involved in retrograde signaling from the chloroplast and in hypersensitive response cell death ( Dietz et al . , 2016; Exposito-Rodriguez et al . , 2017; Van Aken and Van Breusegem , 2015; Liu et al . , 2007 ) . In diatoms , specific stress cues can lead to ROS accumulation and EGSH oxidation in the chloroplast , as was shown in response to nitrogen limitation , the diatom-derived toxic infochemicals cyanogen bromide , and HL ( Figure 8A and Figure 1 ) ( Graff van Creveld et al . , 2015; Rosenwasser et al . , 2014 ) . Specifically , HL can lead to ROS accumulation in the chloroplast by generation of singlet oxygen ( 1O2 ) in PSII , and by photoreduction of O2 to superoxide ( O2- ) in PSI , which in turn is rapidly converted to H2O2 by superoxide dismutase ( SOD ) ( Waring et al . , 2010; Exposito-Rodriguez et al . , 2017; Asada , 2006 ) . Redox fluctuations in the chloroplast can serve as a rapid mechanism to perceive specific environmental cues , by regulating key metabolic pathways on the post-translational level . Analysis of the redox-sensitive proteome in P . tricornutum revealed over-representation of chloroplast-targeted proteins , that were also oxidized to a greater degree under H2O2 treatment as compared to other subcellular compartments , further supporting the existence of a redox-based signaling network in the chloroplast ( Rosenwasser et al . , 2014; Woehle et al . , 2017 ) . The role of chloroplast EGSH perturbations in sensing specific stress cues gains further support by the early chl-roGFP oxidation , which preceded the ‘point of no return’ , after which cell death was irreversibly activated in the ‘oxidized’ subpopulation ( Figure 3B ) . This ‘pre-commitment’ phase provides an opportunity for cells to recover if conditions change during a narrow time frame of ~30–60 min following oxidative stress in most cells ( Figure 3B ) , before the cell has accumulated damage beyond repair or a PCD cascade was fully activated . This ‘pre-commitment’ phase was shown previously in diatoms , as exogenous application of the antioxidant GSH rescued cells from otherwise lethal treatments of infochemicals or H2O2 only during the first hour ( Graff van Creveld et al . , 2015; Volpert et al . , 2018 ) . These findings shed light on the timeline of events in PCD progression in diatoms . To date , the role of the chloroplast in mediating PCD has remained elusive , although mitochondria-generated ROS are known to play a key role in PCD in plants and animals ( Van Aken and Van Breusegem , 2015; Lam et al . , 2001 ) . This knowledge gap is even greater in unicellular marine algae , for which the molecular basis for the PCD machinery is largely unknown ( Bidle , 2016 ) . In P . tricornutum , early mitochondrial oxidation was shown to precede subsequent cell-death in response to various stress conditions at the population level , but the link with chloroplast EGSH was less clear and depended on the specific stress cue ( Graff van Creveld et al . , 2015 ) . In another diatom , chloroplast EGSH was shown to mediate changes in oxidative stress sensitivity upon light-dark transitions ( Volpert et al . , 2018 ) . A recent model in plants suggested possible mitochondria-chloroplast cooperative interactions in the execution of ROS-mediated PCD ( Van Aken and Van Breusegem , 2015 ) . In addition , there are evidence for energetic coupling of chloroplasts and mitochondria in diatoms , and they use extensive energetic exchanges between these organelles to regulate the ATP/NADPH ratio ( Bailleul et al . , 2015 ) . Taken together with the results presented here , we suggest that redox dynamics of both the mitochondria and the chloroplast are involved in cell fate regulation in diatoms . We propose that cells that accumulate ROS above a certain threshold are likely to induce cell death with PCD-like hallmarks ( Figure 9 ) , as was shown in response to H2O2 ( Graff van Creveld et al . , 2015 ) and as observed in the death of the ‘oxidized’ subpopulation ( Figures 3B and 4F-G , and Figure 8D ) . Cells that do not cross this threshold are able to recover and acclimate , as in the ‘reduced’ subpopulation ( Figures 3B and 4F-G , and Figure 8D ) . Based on the data from the microfluidics setup , which allowed cell tracking throughout the entire dynamics during exposure to H2O2 , we propose that such a ‘death threshold’ could be detected by early chl-roGFP oxidation ( Figure 4G ) . This ‘death threshold’ may be dependent on additional factors , such as ecological context or the specific stressor . The balance between the cellular metabolic state , antioxidant capacity , and the magnitude of the applied stress determines whether a cell will cross the ‘death threshold’ , leading to a differential response within the population . Harsher stress conditions will have a stronger effect on the population , leading to more cells crossing the threshold and exhibiting early oxidation and subsequent cell death , as shown with increasing H2O2 doses ( Figure 2M ) or prolonged exposure to HL ( Figure 8B ) . The source for the cell-to-cell variability observed in our system is yet to be further explored , but the results provide insights into factors that may drive it . Since clonal populations originating from single-cell isolates maintained the bi-stable chloroplast response , the variability does not result from genetic differences but rather from phenotypic plasticity ( Figure 5 ) . The combination of factors such as life history ( Graff van Creveld et al . , 2016; Murik et al . , 2014 ) , cell cycle phase ( Papagiannakis et al . , 2017; Mathis and Ackermann , 2016 ) , cell age ( Levy et al . , 2012; Radzinski et al . , 2018 ) , metabolic activity ( Şimşek and Kim , 2018; Campbell et al . , 2018 ) , heterogeneous microenvironment ( Stocker , 2012 ) and biological noise ( Balaban , 2011; Raj and van Oudenaarden , 2008 ) results in a distribution of different metabolic states within the population ( Ackermann , 2015; Takhaveev and Heinemann , 2018 ) , which could lead to differential sensitivity to oxidative stress . In yeast for example , redox-based heterogeneity was linked to proliferation and aging ( Radzinski et al . , 2018 ) . It remains to be investigated whether the emergence of the subpopulations represents heterogeneity that occurs following exposure to stress , or rather a pre-existing variability within the population . Nevertheless , differences in cell-cycle phase distribution between the subpopulations likely represent pre-existing disparities , supporting the latter . However , differences in cell cycle do not completely explain the variability , and are likely to be a contributing factor rather than the source , for example by antioxidants oscillations ( Papagiannakis et al . , 2017; Burhans and Heintz , 2009; Mathis and Ackermann , 2016; Diaz-Vivancos et al . , 2015 ) . In addition , the microfluidics experiments showed no clear differences in chl-roGFP OxD between the subpopulations prior to the treatment ( Figure 4—figure supplement 2B ) , demonstrating that the possible pre-existing heterogeneity is not reflected in the chloroplast EGSH basal level , but rather is based on a different parameter that is yet to be identified . Unlike in previous studies conducted on heterogeneity , the mechanism that generates variability in our system is light-dependent , as the bi-stable chl-roGFP pattern was abolished when the cells were under darkness and the entire population became more sensitive to oxidative stress ( Figure 7A–C ) . The antioxidant capacity of a diatom cell depends on photosynthesis-generated NADPH , which is also used for GSH recycling . The transition to the dark may have compromised the biosynthesis and recycling of GSH , therefore enhancing sensitivity to oxidative stress , as was shown in another diatom ( Volpert et al . , 2018 ) . Taken together , we propose that the source for heterogeneity could be variability in the flux of photosynthesis-derived reductive power , regulating the recycling rates of antioxidants . The light-dependent emergence of distinct subpopulations and the increased sensitivity under dark suggest important implications for environmental scenarios . Fluctuating light conditions are frequent in natural environments due to mixing , shading , the diel cycle , and tide ( in coastal and intertidal regions ) , and may greatly affect diatoms’ susceptibility to diverse abiotic stresses and biotic interactions with pathogens . Phenotypic variability can provide an important strategy to cope with fluctuating environments in microbial populations ( Ackermann , 2015 ) . Co-existence of subpopulations with different susceptibilities to specific stressors can be viewed as a ‘bet-hedging’ strategy of the population , enabling at least a portion of the population to survive unpredicted stress events and subsequently leads to a growth benefit at the population level ( Schreiber et al . , 2016; Sengupta et al . , 2017; Levy et al . , 2012; Ackermann , 2015 ) . In diatoms , phenotypic variability in cell size , shape and susceptibility to stress conditions was suggested ( Armbrust , 2009; Graff van Creveld et al . , 2015; Volpert et al . , 2018; De Martino et al . , 2011 ) , but until now the experimental setups were not designed to study individuality in stress response . Future studies are required to investigate the possible tradeoff involved in maintaining high antioxidant capacity . For example , resilience to oxidative stress may come with a cost in the ability to sense environmental cues with high precision , as high ROS buffering capacity may mask milder ROS cues ( Woehle et al . , 2017 ) . Redox-based phenotypic variability may provide a rapid and adjustable strategy to cope with unpredicted stress conditions as compared to relying only on genetic diversity . The novel approaches developed here provide new insights into individuality in marine phytoplankton , and enable studying dynamic processes at the single-cell level in diatoms and possibly other ecologically relevant microorganisms . The mechanisms that underlie differential sensitivity to oxidative stress are yet to be explored . The findings presented here show promising ecological implications for light-dependent heterogeneity , and future studies will unravel its ecological significance in the marine environment . P . tricornutum accession Pt1 8 . 6 ( CCMP2561 in the Provasoli-Guillard National Center for Culture of Marine Phytoplankton ) was purchased from the National Center of Marine Algae and Microbiota ( NCMA , formerly known as CCMP ) . Cultures were grown in filtered seawater ( FSW ) supplemented with F/2 media ( Guillard and Ryther , 1962 ) at 18°C with 16:8 hr light:dark cycle and 80 μmol photons m−2 sec−1 light intensity supplied by cool-white LED lights ( Edison , New Taipei , Taiwan ) . Strains expressing roGFP were obtained as described previously ( Graff van Creveld et al . , 2015; Rosenwasser et al . , 2014 ) . Cultures were kept in exponential phase ( <2·106 cells·ml−1 ) for at least 1 week and were sequentially diluted at least three times prior to experiments , experiments were performed in ~0 . 5–1·106 cells·ml−1 . All cultures were counted and diluted a day before the experiment to ensure the same cell concentrations between samples . Cell concentration was measured using Multisizer 4 COULTER COUNTER ( Beckman Coulter ) . roGFP oxidation was measured over time following the addition of H2O2 or in untreated control using the ratio between two fluorescence channels , i405 and i488 , by fluorescence microscopy ( described below ) and by flow cytometry using BD LSRII analyzer , BD FACSAria II and BD FACSAria III . The roGFP ratio ( i405/i488 ) increases upon oxidation of the probe ( Schwarzländer et al . , 2008 ) . The oxidation degree of roGFP ( OxD ) was calculated according to Schwarzländer et al . ( Schwarzländer et al . , 2008 ) :OxDroGFP=R-Rredi488oxi488redRox-R+R-Rred Where R is the roGFP ratio of i405/i488 , Rred is the ratio of fully reduced form ( 15–50 min post treatment with 2 mM Dithiothreitol , DTT ) , Rox is the ratio of the fully oxidized form ( 7–30 min post treatment with 200 μM H2O2 ) , and i488ox and i488red are the i488 of the maximum oxidized and maximum reduced forms respectively . For sorting purposes , roGFP ratio was used , as exact OxD cannot be calculated prior to sorting , both parameters give a similar partition between the subpopulations ( data not shown ) . In flow cytometry measurements , i405 was measured using excitation ( ex ) 407 nm , emission ( em ) 530/30 nm or 525/25 nm , and i488 was measured using ex 488 nm , em 530/30 nm . Only roGFP positive cells ( roGFP+ ) which had a clear roGFP fluorescence signal separated from WT auto-fluorescence ( AF ) were included in the analysis . roGFP+ gate was determined either by roGFP relative expression level , which was measured by multiplication of i405 and i488 ( Figure 2—figure supplement 2D ) , or based on i405 and i488 intensity in sorting experiments , as relative roGFP expression could be calculated only post acquisition ( see Figure 8—figure supplement 1 for example ) . Dynamic range of roGFP was calculated by the ratio of Rox/Rred ( Table 1 ) . For H2O2 treatments , H2O2 was added at time 0 from a freshly prepared 20 mM stock to P . tricornutum cultures to a final concentration of 5–200 µM . Flow cytometry measurements were done under ambient light ( 5 . 5–14 μmol photons m−2 sec−1 ) and temperature ( 20–22°C ) conditions , unless stated otherwise . For dark treatment samples were covered with aluminum foil . High light of 1 , 700–2 , 200 μmol photons m−2 sec−1 was applied using a LED lamp ( deviations are due to uneven illumination depending on the distance from the lamp center ) . To avoid heating by the HL lamp , samples were kept at 16°C using a chilled stage ( LCI , Korea ) . For cell death analysis , samples were stained with a final concentration of 1 μM Sytox Green ( Invitrogen ) , incubated in the dark for 30–60 min at RT and analyzed using an Eclipse flow cytometer ( ex 488 nm , em 525/50 nm ) . Unstained samples were used as control to discriminate background signal . In microfluidics experiments , Sytox was dissolved in fresh media ( FSW + F/2 ) to a final concentration of 1–2 µM , and inserted into the system at 21 . 5–23 hr post treatment , without changing the flow rate ( 1 µl/min ) . Fresh stain was continuously flowing through the system for at least 1 . 5 hr during which cells were imaged for Sytox staining ( ex 470/40 nm , em 525/50 nm ) as described below . For Sytox staining analysis , images of 30–106 min incubation time were used based on highest staining and best focus . To measure survival and generate clonal populations originating from different subpopulations , cells expressing chl-roGFP of the ‘oxidized’ and ‘reduced’ subpopulations were sorted according to their roGFP ratio at different times post 80 µM H2O2 treatment or HL using BD FACS AriaII and BD FACS AriaIII . The ‘oxidized’ and ‘reduced’ subpopulations gates were based on visible separation and avoiding cells near intermediate values . Untreated roGFP+ control cells were sorted based on clear separation of roGFP fluorescence from WT AF as described above ( the gate upstream of the subpopulations’ gates ) , regardless of roGFP oxidation . To avoid oxidation due to darkness within the FACS machine , sorting times were minimized and the ‘oxidized’ subpopulation was sorted first followed immediately by sorting of the ‘reduced’ subpopulation . However , in longer sorting sessions as for the Sytox and cell cycle analyses , some cells within the sorted ‘oxidized’ subpopulation may have been oxidized due to the combined effect of H2O2 and exposure to the dark within the FACS . For Sytox analysis of cell death post sorting , 10 , 000 cells/well were sorted into fresh media ( FSW + F/2 ) in triplicate biological repeats . For single cell survival 1 cell/well was sorted into 96-well plates containing either ‘agar’ ( 1 . 5% agarose +FSW/2 + F/2 + antibiotics ) or ‘liquid’ ( FSW + F/20 ) fresh media . For the HL CFU assay , 384 cells were sorted onto one-well agar plates . Cells grown in liquid were further diluted and then spotted on agar plates . 3 . 5–9 weeks post sorting colonies were counted manually ( H2O2 experiments ) or scanned using Amersham Typhoon 5 Biomolecular Imager and quantified using ImageQuant ( HL experiments ) , assuming each colony originates from a single surviving cell . CFU survival was calculated as CFU number divided by the number of sorted cells . Since survival was highly similar in liquid and in agar the results of these two methods were combined together . Each method was done in biological triplicates per medium type per experiment . In Figure 3B data is shown for two independent experiments for time points 30 and 100 min and one experiment for the 60 min time point . For generation of clonal populations , single cells sorted into liquid medium were used . Clones were exposed to 80 µM and 100 µM H2O2 ~ 3–6 weeks post sorting , and their chl-roGFP oxidation was measured using flow cytometry . A total of 18 ‘control’ , 29 ‘oxidized’ and 32 ‘reduced’ clones were examined in two independent experiments . Microfluidics chip design was based on Shapiro et al . ( Shapiro et al . , 2016 ) and was modified for P . tricornutum cells . Each chip contained 4 channels of 2 cm length X 0 . 2 cm width X 150 µm height with one circular widening of 0 . 4 cm diameter , with a total volume of ~12 . 7 µl per channel ( Figure 4—figure supplement 1B–C ) . The microfluidics chip was etched into a silicone elastomer ( SYLGARD 184 , Dow Corning ) using soft lithography . Silicone elastomers were prepared by mixing the two components in a 10:1 ratio and were poured onto the dust-free wafer , de-aired in a desiccator to eliminate air bubbles , and incubated overnight at 60°C for curing to generate the Polydimethylsiloxane ( PDMS ) microfluidics chips . Inlet and outlet holes were punched at both ends of each channel using a 1 mm biopsy punch ( AcuDerm , FL , USA ) . The PDMS chip was placed on the clean surface of a new glass microscope 60 × 24 mm cover slip using plasma bonding with a BD-20AC corona treater ( Electro-Technic Products ) followed by heating of 100°C for >15 min to ensure covalent bonding of the PDMS and the glass . Light and epifluorescence microscopy imaging was performed using a fully motorized Olympus IX81 microscope ( Olympus ) equipped with ZDC component for focus drift compensation , 20X air objective ( numerical aperture 0 . 5 ) , and Lumen 200PRO illumination system ( Prior Scientific ) . Images were captured using a Coolsnap HQ2 CCD camera ( Photometrics , Tuscon , AZ , USA ) . The microfluidics chip was mounted on a motorized XY stage ( Prior Scientific , MA , USA ) with a temperature-controlled inset ( LCI , Korea ) set to 18°C ( Figure 4—figure supplement 1D ) . The outlet tubes were connected to syringe pumps ( New Era Pump Systems , NY , USA ) set to withdrawal mode , using negative pressure for flow generation . The inlet tubes were connected to Eppendorf reservoirs , containing the fluid to be inserted into the system . Experiment layout is shown in Figure 4—figure supplement 1A . Chambers were washed with at least 500 µl of pure ethanol , then double-distilled water and then fresh media prior to the introduction of cells . Cells were introduced into the system and settled on the glass bottom . Flow rate was kept at 1 µl/min for the duration of the experiment , except during cell introduction ( 100 µl/min ) , cell settlement ( up to 20 µl/min with occasional stops ) , and treatment introduction ( 10 µl/min for the initial 10 min for rapid replacement of media ) . Following settlement and at least 1 hr after cells were introduced to the system , cells were imaged for roGFP measurements ( roGFP i405: ex 405/20 nm , em 525/50 nm; roGFP i488: ex 470/40 nm , em 525/50 nm ) , chlorophyll auto-fluorescence ( ex 470/40 nm; em 590 lp ) and bright field ( BF , without a condenser ) . Each chip contained four chambers that were imaged sequentially: chl-roGFP control , chl-roGFP 80 µM H2O2 treated , WT 80 µM H2O2 treated and WT control ( WT strain can be used to monitor auto-fluorescence changes and leakage during experiments ) . In each chamber , 5–6 different fields were imaged every 20 min over the course of >24 hr to avoid photo-toxicity . Ambient light was provided during light period using the microscope’s BF illumination without a condenser , light intensity ranging between 34 ( at the very edge , outside the imaging region ) to 80 ( center ) μmol photons m−2 sec−1 . No images were obtained during the night to avoid disturbance to the diurnal cycle . After imaging the basal state of the cells , treatments of either 80 µM H2O2 dissolved in fresh media ( FSW + F/2 ) or fresh media control were introduced to the system continuously for ~2 . 5–3 hr , after which they were gradually washed away by fresh media . To quantify cell death , Sytox green was introduced into the system at 21 . 5–23 hr post treatment ( see above ) and was imaged using the roGFP i488 channel with a shorter exposure time . The Sytox signal was separated from the roGFP i488 based on its stronger fluorescence and localization to the nucleus . Only a small fraction of cells within the control treatment were Sytox positive ( 0 . 0054% ) , and in addition some control and ‘reduced’ cells proliferated during the experiments ( Videos 1–2 ) , indicating that cells remained viable in this experimental setup . Image analysis was performed using a designated MATLAB based script ( see overview in Figure 4—figure supplement 4 ) that is available on GitHub: https://github . com/aviamiz/ITRIA ( Mizrachi , 2018; copy archived at https://github . com/elifesciences-publications/ITRIA ) . Images were imported using bio-formats ( Linkert et al . , 2010 ) . Then , image registration for XY drift correction was done using the Image Stabilizer plugin ( Li , 2008 ) for FIJI ( Fiji Is Just ImageJ ) and using MIJI ( Prodanov and Tinevez , 2012 ) to access FIJI from MATLAB . Then images were normalized by bit-depth . Background subtraction was done based on mean value of a user-defined region of interest ( ROI ) that did not include cells . All fluorescence channels ( i405 , i488 and chlorophyll ) were thresholded by a user-defined value to generate masks of positive expression . The roGFP relative expression level was calculated pixel-by-pixel by multiplication of the i405 and i488 , only at pixels that were co-localized in the i405 and i488 masks . Then , roGFP relative expression ( i405 * i488 ) was thresholded in order to include only pixels with high enough signal , based on a user-defined threshold . The roGFP ratio and OxD were calculated pixel-by-pixel as described above , pixels that were not included in the roGFP expression mask were excluded and set to NaN ( not a number ) . For values of maximum oxidation and reduction of roGFP , cells were imaged in the same microfluidics imaging setup following treatments of 200 µM H2O2 and 2 mM DTT respectively ( see ‘roGFP calculations’ ) . Cell segmentation was based on i405 ( chl-roGFP strain ) or chlorophyll ( WT strain ) masks and fluorescence intensity using watershed transformation . Cells were filtered based on area , major and minor axis length , and eccentricity in order to exclude clumps of cells and doublets . Cell tracking was adapted and modified from a MATLAB code kindly provided by Vicente I . Fernandez and Roman Stocker ( Smriga et al . , 2016; Shapiro et al . , 2014 ) . In short , particles were tracked based on minimizing the distance between particle centroids in adjacent frames within a distance limit . Sytox analysis was based on a user defined threshold and co-localization of the Sytox with the extended cell region within the cell segmentation mask . Images from the same experiment were analyzed using the same values for all thresholds and parameters , except for Sytox analysis in which the threshold was adjusted manually to validate correct assignment of cell-fate and to avoid effects of focus differences . Cells that were not detected in the frame used for Sytox analysis or were not tracked for at least six consecutive frames were excluded from further analysis . The 74% OxD threshold used for early discrimination between cells that subsequently died or survived ( Figure 4E–G and Figure 4—figure supplement 2 ) was based on logistic regression modeling of cell fate at the end of the experiments as a function of chl-roGFP OxD 40 min post 80 µM H2O2 treatment ( Figure 4—figure supplement 3 ) . The observed roGFP OxD of more than 100% oxidation in some cells could result from increased auto-fluorescence leakage to the i405 channel at later times post treatment ( see Figure 2—figure supplements 5–6 ) . Cell cycle analysis was based on Huysman et al . ( 2010 ) and modified for sorted cells . 30 , 000 cells of ‘oxidized’ and ‘reduced’ sub-populations were sorted 30 min post 80 µM H2O2 treatment into 260 µl 80% ethanol kept at 5 ˚C , reaching a final concentration of 70% ethanol . Control untreated cells and synchronized cells for reference ( 20 hr dark , as previously described; Huysman et al . , 2010 ) were sorted based on positive roGFP fluorescence . Cells were then gently mixed and kept at 4 ˚C until further processing . Then , 500 µl of 0 . 1% bovine serum albumin in phosphate-buffered saline ( PBS ) was added to improve pellet yield , and samples were centrifuged at 4000 rcf at 4 ˚C for 10 min to discard supernatant . Cells were then washed with PBS , re-suspended and stained with 4' , 6-diamidino-2-phenylindole ( DAPI , Sigma ) at a final concentration of 10 ng/ml . Samples were analyzed using BD LSRII analyzer , with ex 355 nm and em 450/50 nm . Synchronized cells were used as a reference to validate the gates for G1 and G2/M phases ( data not shown ) . S phase was not clearly detected in this analysis . All statistical analyses were done in R . ANOVA or ANCOVA were used for multiple comparisons , and then Dunnett test or Tukey test were performed were applicable . For comparisons of two samples , t-test was used , and paired t-test was used where applicable . Values are represented as mean ± SEM unless specified otherwise . Box-plot was generated using the web tool BoxPlotR http://shiny . chemgrid . org/boxplotr/ ( Spitzer et al . , 2014 ) using Tukey whiskers , which extend to data points that are less than 1 . 5 x Interquartile range away from 1 st/3rd quartile . All relevant data supporting the findings of the study are available in this article and its Supplementary Information , or from the corresponding author upon request . All data generated or analysed during this study are included in the manuscript and supporting files . Source data files have been provided for Figures 1– 8 . MATLAB script used for image analysis is available at GitHub , as referenced in the methods section: https://github . com/aviamiz/ITRIA .	Microscopic algae , such as diatoms , are widely spread throughout the oceans , and are responsible for half of the oxygen we breathe . At certain times of the year these algae grow very rapidly to form large “blooms” that can be detected by satellites in space . These blooms are generally short-lived because the algae are either eaten by other marine organisms , run out of nutrients , or die as a result of being infected by viruses or bacteria . However , some diatom cells survive the end of the bloom and go on to generate new blooms in the future , but it is still not clear how . As the bloom collapses , diatoms experience many stressful conditions which can cause active molecules known as reactive oxygen species , or ROS for short , to accumulate inside cells . Normally growing cells also produce low amounts of ROS , which regulate various processes that are important for maintaining a cell’s health . However , high amounts of ROS can cause damage , which may lead to a cell’s death . Now , Mizrachi et al . investigated why some algae survive while others die in response to stressful conditions , focusing on the amount of ROS that accumulates within the diatom Phaeodactylum tricornutum . Laboratory experiments showed that individual cells of P . tricornutum respond differently to environmental stress , forming two distinct groups of either sensitive or resilient cells . Sensitive cells accumulated high levels of ROS within a cell compartment known as the chloroplast and eventually died . Whereas resilient cells were able to maintain low levels of ROS in the chloroplast and survived long after the other cells perished . Populations of genetically identical diatom cells also formed distinct groups of sensitive and resilient cells , demonstrating that these two opposing reactions to stress are not caused by genetic differences between cells . Lastly , Mizrachi et al . showed that how diatoms acclimate to stress depends on the amount of light they are exposed to . When in the dark , all cells became sensitive to oxidative stress , without forming distinct groups . But , when exposed to strong light that mimics the ocean surface , cells formed distinct groups within the population . This suggests that light regulates how susceptible these microscopic algae are to environmental stress . The different responses within a population may serve as a “bet-hedging” strategy , enabling at least some of the cells to survive unpredicted stressful conditions . The next challenge will be to find out whether algae growing in the oceans also use the same strategy and investigate what impact this has on diatom blooms .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology", "microbiology", "and", "infectious", "disease" ]	2019	Light-dependent single-cell heterogeneity in the chloroplast redox state regulates cell fate in a marine diatom
Prions are self-propagating protein aggregates that are characteristically transmissible . In mammals , the PrP protein can form a prion that causes the fatal transmissible spongiform encephalopathies . Prions have also been uncovered in fungi , where they act as heritable , protein-based genetic elements . We previously showed that the yeast prion protein Sup35 can access the prion conformation in Escherichia coli . Here , we demonstrate that E . coli can propagate the Sup35 prion under conditions that do not permit its de novo formation . Furthermore , we show that propagation requires the disaggregase activity of the ClpB chaperone . Prion propagation in yeast requires Hsp104 ( a ClpB ortholog ) , and prior studies have come to conflicting conclusions about ClpB's ability to participate in this process . Our demonstration of ClpB-dependent prion propagation in E . coli suggests that the cytoplasmic milieu in general and a molecular machine in particular are poised to support protein-based heredity in the bacterial domain of life . Prions are infectious , self-propagating protein aggregates first described in the context of scrapie ( Prusiner , 1982 ) , an example of a class of devastating neurodegenerative diseases known as the transmissible spongiform encephalopathies ( TSEs ) . Specifically , the prion form of a protein known as PrP is the causative agent of the TSEs , which afflicts humans and other mammals . Native PrP ( PrPC ) undergoes a dramatic change in conformation upon conversion to its prion form ( PrPSc ) , forming distinctive cross-β aggregates termed as amyloid ( Diaz-Espinoza and Soto , 2012 ) . Highly resistant to denaturation and proteolysis , PrPSc is infectious and templates the conformational conversion of PrPC molecules ( Caughey et al . , 2009 ) . Prion-like phenomena have also been described in budding yeast and other fungi . Since Wickner first invoked prions to account for two examples of non-Mendelian genetic elements in Saccharomyces cerevisiae ( Cox , 1965; Aigle and Lacroute , 1975; Wickner , 1994 ) , the study of fungal prion proteins has resulted in profound advances in the understanding of prion biology , including the first demonstration that purified prion protein aggregates are infectious ( Maddelein et al . , 2002; King and Diaz-Avalos , 2004; Tanaka et al . , 2004 ) . In general , such prion proteins exist in either a native , soluble form or a self-perpetuating , amyloid form with spontaneous conversion between forms representing a rare event ( Allen et al . , 2007; Lancaster et al . , 2010 ) . However , unlike PrPSc , yeast prions do not normally cause cell death . Instead , they can act as protein-based genetic elements that confer new phenotypes on those cells that harbor them ( True and Lindquist , 2000; Tuite and Serio , 2010; Newby and Lindquist , 2013 ) . Fungal prion proteins have been found to participate in diverse cellular processes ( Coustou et al . , 1997; True et al . , 2004; Suzuki et al . , 2012; Holmes et al . , 2013 ) . The conversion of these proteins to their prion forms typically results in a dominant loss-of-function phenotype ( Cox , 1965; Aigle and Lacroute , 1975 ) . A particularly well-characterized example involves the essential translation release factor Sup35 , which confers on cells a heritable nonsense suppression phenotype upon conversion to the prion form ( Cox , 1965; Ter-Avanesyan et al . , 1994; Patino et al . , 1996; Paushkin et al . , 1996 ) . Like other yeast prion proteins , Sup35 has a modular structure with a distinct prion domain ( PrD ) that mediates conversion to the prion form , [PSI+] . In the case of Sup35 , the essential prion determinants , which include a glutamine- and asparagine-rich segment and five complete copies of an imperfect oligopeptide repeat sequence , lie in the N-terminal domain ( N ) , whereas translation release activity resides in the C-terminal domain ( C ) ( Ter-Avanesyan et al . , 1993 ) . A highly charged middle region ( M ) increases the solubility of native Sup35 and enhances the mitotic stability of [PSI+] ( Liu et al . , 2002 ) . Together , Sup35 N and M function as a transferable prion-forming module ( NM ) that maintains its prionogenic potential when fused to heterologous proteins ( Li and Lindquist , 2000 ) . A distinctive property of the Sup35 conversion process in yeast is its dependence on the presence of a pre-existing prion , designated [PIN+] for [PSI+] inducibility factor ( Derkatch et al . , 1997 ) . Thus , yeast strains containing Sup35 in the non-prion form , [psi–] , support the spontaneous conversion to [PSI+] only if they contain [PIN+] , typically the prion form of the Rnq1 protein ( Derkatch et al . , 2000 ) . However , several other yeast prion proteins , including the New1 protein , have the capacity to function as [PIN+] in their prion forms ( Derkatch et al . , 2001; Osherovich and Weissman , 2001 ) . Importantly , the stable propagation of yeast prions—and thus , the heritability of their associated phenotypes—depends on the function of chaperone proteins ( Chernoff et al . , 1995 ) . Specifically , the AAA+ disaggregase Hsp104 is strictly required for the propagation of virtually all yeast prions characterized thus far , and several other chaperone proteins have been implicated in this process as well ( Liebman and Chernoff , 2012; Winkler et al . , 2012 ) . Various lines of evidence support the view that the essential role of Hsp104 with respect to prion propagation stems from its ability to fragment prion aggregates and thereby to generate smaller seed particles known as propagons that can be efficiently partitioned to daughter cells during cell division ( Paushkin et al . , 1996; Ness et al . , 2002; Cox et al . , 2003; Kryndushkin et al . , 2003; Satpute-Krishnan et al . , 2007; Higurashi et al . , 2008 ) . Accordingly , depletion or inhibition of Hsp104 in a prion-containing cell leads to prion loss in progeny cells . The molecular processes underlying prion biology constitute at least two distinct phases , namely , ( i ) the de novo conversion of a protein from its native to prion form , and ( ii ) the subsequent propagation of the self-perpetuating prion form over multiple generations . While studies have demonstrated that the bacterial cytoplasm can support the de novo formation of prion-like aggregates ( Sabaté et al . , 2009; Fernándes-Tresguerres et al . , 2010; Garrity et al . , 2010; Espargaró et al . , 2012; Gasset-Rosa et al . , 2014 ) , evidence for prion propagation—and thus , protein conformation-dependent heredity—in bacteria has remained elusive . We previously demonstrated that conversion of Sup35 NM to its prion form in Escherichia coli , as in S . cerevisiae , depends on [PIN+] , which is formed by providing the bacterial cells with the yeast New1 protein ( Garrity et al . , 2010 ) . This [PIN+] dependence provides an experimental framework for distinguishing between the initial conversion and subsequent propagation phases of the prion cycle . In particular , the formation of prion-like Sup35 NM aggregates can be induced in bacterial cells containing the New1 protein; subsequent depletion of the New1 protein from these cells reveals whether or not the bacterial cytoplasm can support the propagation of the Sup35 NM prion in the absence of [PIN+] . Here we show that bacteria can propagate the Sup35 prion in an infectious conformation over at least ∼100 generations under conditions that do not permit de novo prion formation . More specifically , we demonstrate maintenance of the Sup35 NM prion over multiple rounds of restreaking in E . coli cells no longer capable of synthesizing the New1 protein . Furthermore , we establish that propagation of the Sup35 NM prion in E . coli requires the disaggregase activity of ClpB , the bacterial ortholog of Hsp104 . The striking parallel between the requirements for both prion formation and prion propagation in yeast and bacteria , which are thought to have diverged more than 2 . 2 billion years ago , suggests that the paradigm of protein-based heredity may be more ancient than previously inferred ( DeSantis et al . , 2012 ) . Having previously shown that Sup35 NM can adopt an infectious amyloid conformation in the E . coli cytoplasm ( Garrity et al . , 2010 ) , we wished to determine whether or not E . coli cells could stably propagate Sup35 NM in its prion form . To address this question , we took advantage of the fact that conversion of Sup35 NM to its prion conformation in E . coli depends on the presence of New1 , mirroring features of the [PIN+] dependence of Sup35 prion formation in S . cerevisiae ( Figure 1A ) . Thus , our plan was to induce the formation of infectious Sup35 NM aggregates in E . coli cells containing the prionogenic module of New1 and then to monitor the fate of Sup35 NM over multiple generations after curing the cells of New1-encoding DNA . 10 . 7554/eLife . 02949 . 003Figure 1 . Conversion of Sup35 NM to its prion form in E . coli requires New1 . ( A ) Cartoon representation of how conversion of soluble Sup35 NM ( Sup35soluble ) to its amyloid conformation ( Sup35amyloid ) depends on the presence of New1 in its amyloid conformation ( New1amyloid ) . Sup35 NM and New1 ( black ) are depicted as fusions to mCherry ( red ) and mGFP ( green ) , respectively . ( B ) SDS-stable Sup35 NM aggregates are detected only in cells producing SDS-stable New1 aggregates as assessed by filter retention analysis . For each sample , undiluted lysate and three twofold dilutions are shown ( see ‘Materials and methods’ ) . Sup35 NM and New1 aggregates are no longer detected once boiled . The α-Sup35 antibody recognizes the Sup35 NM-mCherry-His6X fusion protein , and the α-GFP antibody detects the New1-mGFP fusion protein . ( C ) Intracellular full-length ( FL ) Sup35 NM fusion protein levels are comparable in the presence and absence of New1 as assessed by Western blot analysis . The α-RpoA antibody recognizes the α subunit of E . coli RNA polymerase . ( D ) Fluorescence images of representative cells containing Sup35 NM and New1 or Sup35 NM alone . For cells containing both fusion proteins , the mCherry channel , GFP channel , and merged images are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02949 . 00310 . 7554/eLife . 02949 . 004Figure 1—figure supplement 1 . α-Sup35 and α-His6X antibodies are interchangeable for detecting the Sup35 NM-mCherry-His6X fusion protein . ( A ) Probing nitrocellulose membranes with either α-Sup35 antibody ( Figure 1C ) or α-His6X antibody results in the detection of similar intracellular Sup35 NM fusion protein products in the presence or absence of New1 as assessed by Western blot analysis . Full-length ( FL ) Sup35 NM fusion proteins are indicated by arrows . ( B ) Probing cellulose acetate membranes with either α-His6X antibody ( Figure 3B ) or α-Sup35 antibody results in the detection of similar aggregate-positive and aggregate-negative samples in starter cultures ( ST ) and Round 1 ( R1 ) experimental clones as assessed by filter retention analysis . For each sample , undiluted lysate and three twofold dilutions are shown . Starter cultures of cells containing Sup35 NM and New1 and cells containing Sup35 NM alone serve as positive ( P ) and negative ( N ) controls , respectively . ( C ) Probing cellulose acetate membranes with either α-His6X antibody ( Figure 3D ) or α-Sup35 antibody results in the detection of similar aggregate-positive and aggregate-negative samples in starter cultures ( ST ) and R1 control clones as assessed by filter retention analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 02949 . 004 To facilitate these experiments , we fused Sup35 NM and New1 to two monomeric fluorescent proteins ( mCherry bearing a C-terminal hexahistidine tag and mGFP , respectively ) . The two fusion proteins were produced from compatible plasmids under the control of IPTG-inducible promoters . The plasmid encoding New1-mGFP ( pSC101TS-NEW1 ) bore a temperature-sensitive origin of replication , enabling us to cure cells of New1-encoding DNA and thereby deplete cells of the New1 fusion protein . As an initial test of our experimental system , we introduced the plasmid encoding Sup35 NM-mCherry-His6x ( pBR322-SUP35 NM ) together with either pSC101TS-NEW1 or an empty vector control ( pSC101TS ) into E . coli cells and induced the synthesis of the fusion proteins at the permissive temperature . After overnight growth , we detected SDS-stable Sup35 NM aggregates ( ‘Materials and methods’ ) only in cells producing the New1 fusion protein ( Figure 1B ) . As New1 can independently adopt an amyloid conformation in E . coli ( Garrity et al . , 2010 ) , we also detected SDS-stable New1 aggregates in cells containing both fusion proteins ( Figure 1B ) . Western blot analysis revealed that the intracellular levels of the Sup35 NM fusion protein were comparable in the presence and absence of the New1 fusion protein ( Figure 1C , Figure 1—figure supplement 1A ) . We also examined cells by fluorescence microscopy . In cells containing both fusion proteins , Sup35 NM formed twisted ring structures ( Garrity et al . , 2010 ) or large polar foci in 22 . 5% of cells ( N = 258 ) , whereas New1 formed punctate foci in 89 . 8% of cells ( N = 258 ) . ( Figure 1D ) . In contrast , Sup35 NM exhibited diffuse fluorescence in 100% of cells lacking New1 ( N = 532 ) ( Figure 1D ) . We then sought to determine whether or not E . coli cells could propagate SDS-stable Sup35 NM aggregates over multiple generations under conditions that do not permit the de novo formation of aggregates ( that is , in the absence of New1 ) . Our experimental protocol is illustrated in Figure 2 ( see also Figure 3A ) . We first induced fusion protein synthesis in cells transformed with pBR322-SUP35 NM and either pSC101TS-NEW1 ( experimental sample ) or pSC101TS ( control sample ) . These ‘starter cultures’ were grown overnight to allow for the formation of SDS-stable Sup35 NM aggregates in the experimental sample . The cells were then plated and grown at the non-permissive temperature to cure the cells of pSC101TS-NEW1 or pSC101TS , thereby generating a set of Round 1 ( R1 ) colonies . 20 R1 experimental colonies and 20 R1 control colonies were subsequently examined; each was ( a ) patched onto selective medium to test for loss of pSC101TS-NEW1 or pSC101TS , ( b ) restreaked to generate Round 2 ( R2 ) colonies , and ( c ) inoculated into liquid medium for overnight growth to test for the presence of SDS-stable Sup35 NM aggregates . Four separate experimental lineages ( L1E–L4E ) originating from ancestral R1 experimental colonies containing detectable Sup35 NM aggregates along with four separate control lineages ( L1C–L4C ) originating from ancestral R1 control colonies were then followed through Round 3 ( R3 ) and Round 4 ( R4 ) . For R2 and each subsequent round , 10 experimental colonies and 10 control colonies were analyzed . 10 . 7554/eLife . 02949 . 005Figure 2 . Experimental protocol for assessing the ability of E . coli cells to propagate SDS-stable Sup35 NM aggregates . Experiments are initiated with either a starter culture ( ST ) of cells containing Sup35 NM and New1 ( shown ) or a starter culture of cells containing Sup35 NM alone ( not shown ) . For each of the 4 lineages ( L1–L4 ) , the total number of generations over which Sup35 NM prion propagation is monitored corresponds to the number of cell divisions that occur in the absence of New1 during 4 rounds ( R1–R4 ) of growth on solid medium and an additional round of growth in liquid medium . Growth in the absence of New1 begins at the time the starter culture cells are plated at 37°C ( R1 ) . Single R1 colonies were found to contain ∼950 , 000 colony forming units ( CFUs ) , and the liquid cultures contain ∼108 CFUs per μl . Thus , prion propagation is monitored over 98 . 7 or ∼100 generations . We note that the presence of SDS-stable Sup35 NM aggregates in experimental starter culture cells does not represent prion propagation because the presence of New1 ( i . e . , [PIN+] ) enables the continuing de novo conversion of newly synthesized Sup35 NM to the prion form . DOI: http://dx . doi . org/10 . 7554/eLife . 02949 . 00510 . 7554/eLife . 02949 . 006Figure 3 . Converted Sup35 NM can remain in its prion conformation in E . coli cells lacking New1 . ( A ) Cartoon representation of how Sup35 NM can convert to its prion form in the presence of New1 and remain in the prion conformation after cells have been cured of New1-encoding DNA . Sup35 NM and New1 ( black ) are depicted as fusions to mCherry ( red ) and mGFP ( green ) , respectively . ( B ) SDS-stable Sup35 NM aggregates are detected in 8 of 20 Round 1 ( R1 ) experimental clones derived from a starter culture ( ST ) of cells containing Sup35 NM and New1 as assessed by filter retention analysis . Starter cultures of cells containing Sup35 NM and New1 and cells containing Sup35 NM alone serve as positive ( P ) and negative ( N ) controls , respectively . The four aggregate-positive clones selected to establish the four experimental lineages are indicated by asterisks . In all 20 R1 experimental samples , intracellular Sup35 NM fusion protein levels are comparable , and New1 fusion protein is not detectable as assessed by Western blot analysis . The α-His6X and α-Sup35 antibodies recognize the Sup35 NM-mCherry-His6X fusion protein ( see Figure 1—figure supplement 1B , C ) , the α-GFP antibody detects the New1-mGFP fusion protein , and the α-RpoA antibody recognizes the α subunit of E . coli RNA polymerase . ( C ) In all 20 R1 experimental samples , DNA encoding Sup35 NM is detectable whereas DNA encoding the prionogenic module of New1 is not detectable by PCR . ( D ) SDS-stable Sup35 NM aggregates are not detected in any of the 20 R1 control clones derived from a starter culture of cells containing Sup35 NM alone as assessed by filter retention analysis . The four aggregate-negative clones selected to establish the four control lineages are indicated by asterisks . Intracellular Sup35 NM fusion protein levels are comparable in all 20 R1 control samples . DOI: http://dx . doi . org/10 . 7554/eLife . 02949 . 006 All R1 experimental and control colonies ( 20 of each ) had lost pSC101TS-NEW1 or pSC101TS , respectively , as assessed by patching on selective medium ( data not shown ) . Moreover , the absence of NEW1 DNA was confirmed by PCR ( Figure 3C ) and the absence of New1 protein was confirmed by Western blot analysis ( Figure 3B ) . We detected SDS-stable Sup35 NM aggregates in 8 of 20 experimental samples ( Figure 3B ) and none of the control samples ( Figure 3D ) . We selected 4 of the 8 aggregate-positive clones ( Figure 3B , asterisks ) to establish the four experimental lineages and arbitrarily selected four aggregate-negative control clones ( Figure 3D , asterisks ) to establish the four control lineages . 2 of the 4 experimental lineages ( L1E and L3E ) retained SDS-stable Sup35 NM aggregates throughout the course of the experiment ( Figure 4A , Figure 4—figure supplement 1B ) . Of these two lineages , one maintained aggregates in 9 of 10 R4 clones ( Figure 4A ) and the other maintained aggregates in 7 of 10 R4 clones ( Figure 4—figure supplement 1B ) . We conclude that SDS-stable Sup35 NM aggregates can be propagated in E . coli for at least ∼100 generations in the absence of New1 ( Figure 5A ) . 10 . 7554/eLife . 02949 . 007Figure 4 . E . coli can propagate the Sup35 NM prion over ∼100 generations . ( A ) Experimental Lineage 1 ( L1E ) . An aggregate-positive Round 1 ( R1 ) experimental clone ( gray box ) derived from a starter culture ( ST ) of cells containing Sup35 NM and New1 is identified and restreaked to yield progeny Round 2 ( R2 ) clones ( gray bracket ) . All 10 R2 clones analyzed contain detectable SDS-stable Sup35 NM aggregates . An aggregate-positive R2 clone ( blue box ) is identified and restreaked to yield progeny Round 3 ( R3 ) clones ( blue bracket ) . Again , all 10 R3 clones analyzed contain SDS-stable Sup35 NM aggregates . An aggregate-positive R3 clone ( green box ) is identified and restreaked to yield progeny Round 4 ( R4 ) clones ( green bracket ) . 9 of 10 R4 clones analyzed contain SDS-stable Sup35 NM aggregates . The filter retention assay is used to detect SDS-stable Sup35 NM aggregates . Intracellular Sup35 NM fusion protein levels are comparable in all 40 samples as assessed by Western blot analysis . Starter cultures of cells containing Sup35 NM and New1 and cells containing Sup35 NM alone serve as positive ( P ) and negative ( N ) controls , respectively . The α-His6X and α-Sup35 antibodies recognize the Sup35 NM-mCherry-His6X fusion protein , and the α-RpoA antibody recognizes the α subunit of E . coli RNA polymerase . ( B ) Control Lineage 1 ( L1C ) . An aggregate-negative R1 control clone ( gray box ) derived from a starter culture of cells containing Sup35 NM alone is identified and restreaked to yield progeny R2 clones ( gray bracket ) , an aggregate-negative R2 clone ( blue box ) is identified and restreaked to yield progeny R3 clones ( blue bracket ) , and an aggregate-negative R3 clone ( green box ) is identified and restreaked to yield progeny R4 clones ( green bracket ) . No SDS-stable Sup35 NM aggregates are detectable in any sample . Intracellular Sup35 NM fusion protein levels are comparable in all 40 samples as assessed by Western blot analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 02949 . 00710 . 7554/eLife . 02949 . 008Figure 4—figure supplement 1 . The fate of Sup35 NM in experimental Lineages 2-4 . ( A ) Experimental Lineage 2 ( L2E ) . An aggregate-positive Round 1 ( R1 ) clone ( gray box ) derived from a starter culture ( ST ) of cells containing Sup35 NM and New1 is identified and restreaked to yield progeny Round 2 ( R2 ) clones ( gray bracket ) . 8 of 10 R2 clones analyzed contain detectable SDS-stable Sup35 NM aggregates as assessed by filter retention analysis . An aggregate-positive R2 clone ( blue box ) is restreaked to yield progeny Round 3 ( R3 ) clones ( blue bracket ) . 0 of 10 R3 clones analyzed contain detectable SDS-stable Sup35 NM aggregates . The apparent loss of aggregates coincides with the loss of detectable Sup35 NM fusion protein as assessed by Western blot analysis . Assaying the 10 Round 4 ( R4 ) progeny clones derived from an aggregate-negative R3 clone ( green box ) reveals that fusion protein levels can be restored without the recovery of SDS-stable Sup35 NM aggregates ( green bracket ) . Starter cultures of cells containing Sup35 NM and New1 and cells containing Sup35 NM alone serve as positive ( P ) and negative ( N ) controls , respectively . The α-His6X and α-Sup35 antibodies recognize the Sup35 NM-mCherry-His6X fusion protein , and the α-RpoA antibody recognizes the α subunit of E . coli RNA polymerase . ( B ) Experimental Lineage 3 ( L3E ) . 9 of 10 R2 clones analyzed contain detectable SDS-stable Sup35 NM aggregates , which are retained in 8 of 10 R3 progeny clones and 7 of 10 R4 progeny clones . ( C ) Experimental Lineage 4 ( L4E ) . 8 of 10 R2 clones analyzed contain detectable SDS-stable Sup35 NM aggregates , which are retained in 9 of 10 R3 progeny clones . 0 of 10 R4 clones analyzed contain detectable SDS-stable Sup35 NM aggregates . As in R3 of L2E ( A ) , the apparent loss of aggregates coincides with a dramatic drop in Sup35 NM fusion protein levels . DOI: http://dx . doi . org/10 . 7554/eLife . 02949 . 00810 . 7554/eLife . 02949 . 009Figure 4—figure supplement 2 . The fate of Sup35 NM in control Lineages 2–4 . ( A ) Control Lineage 2 ( L2C ) . An aggregate-negative Round 1 ( R1 ) clone ( gray box ) derived from a starter culture ( ST ) of cells containing Sup35 NM alone is identified and restreaked to yield progeny Round 2 ( R2 ) clones ( gray bracket ) . 0 of 10 R2 clones analyzed contain detectable SDS-stable Sup35 NM aggregates as assessed by filter retention analysis . An aggregate-negative R2 clone ( blue box ) is restreaked to yield progeny R3 clones ( blue bracket ) . 0 of 10 R3 clones analyzed contain detectable SDS-stable Sup35 NM aggregates . An aggregate-negative R3 clone ( green box ) is restreaked to yield progeny R4 clones ( green bracket ) . 0 of 10 R4 clones analyzed contain detectable SDS-stable Sup35 NM aggregates . Starter cultures of cells containing Sup35 NM and New1 and cells containing Sup35 NM alone serve as positive ( P ) and negative ( N ) controls , respectively . The α-His6X and α-Sup35 antibodies recognize the Sup35 NM-mCherry-His6X fusion protein , and the α-RpoA antibody recognizes the α subunit of E . coli RNA polymerase . ( B ) Control Lineage 3 ( L3C ) . As in L2C ( A ) , no SDS-stable Sup35 NM aggregates are detectable in progeny R2 , R3 , or R4 clones . ( C ) Control Lineage 4 ( L4C ) . As in L2C ( A ) and L3C ( B ) , no SDS-stable Sup35 NM aggregates are detectable in progeny R2 , R3 , or R4 clones . DOI: http://dx . doi . org/10 . 7554/eLife . 02949 . 00910 . 7554/eLife . 02949 . 010Figure 5 . Genealogy of E . coli cell lineages propagating Sup35 NM in an infectious prion conformation . ( A ) The fate of Sup35 NM in four experimental lineages ( L1E-L4E ) established from a starter culture of cells containing Sup35 NM and New1 is shown . Clones that maintain or lose the Sup35 NM prion are indicated by black or pink lines , respectively . Rounds 1–4 ( R1–R4 ) are depicted as gray arcs , with R1 situated at the center of the tree . Clones are designated as aggregate-positive if they contain SDS-stable Sup35 NM aggregates that are detectable in the undiluted sample and at least 1 of the 3 two-fold serial dilutions , as analyzed by filter retention . L1E and L3E retain SDS-stable Sup35 NM aggregates for the duration of the experiment ( Figure 4A , Figure 4—figure supplement 1B ) . L2E and L4E lose detectable SDS-stable Sup35 NM aggregates at R3 and R4 , respectively . In both cases , the loss of SDS-stable aggregates coincides with a dramatic yet apparently reversible drop in fusion protein levels ( Figure 4—figure supplement 1A , C; see ‘Discussion’ ) . Cells from four aggregate-positive L1E-R4 clones visualized by fluorescence microscopy are indicated by asterisks . ( B ) The fate of Sup35 NM in four control lineages ( L1C–L4C ) established from a starter culture of cells containing Sup35 NM alone is shown . None of the 120 clones analyzed contain SDS-stable Sup35 NM aggregates ( Figure 4B , Figure 4—figure supplement 2 ) . Cells from four aggregate-negative L1C-R4 clones visualized by fluorescence microscopy are indicated by asterisks . ( C ) Fluorescence images of representative cells corresponding to the four aggregate-positive R4 clones indicated by asterisks in ( A ) . ( D ) Fluorescence images of representative cells corresponding to the four aggregate-negative R4 clones indicated by asterisks in ( B ) . ( E ) E . coli cell extracts containing propagated , SDS-stable Sup35 NM aggregates are infectious when transformed into S . cerevisiae [psi−] cells . A starter culture ( ST ) of cells containing Sup35 NM and New1 contain infectious SDS-stable Sup35 NM aggregates capable of converting [psi−] yeast cells to [PSI+] . In contrast , a starter culture of cells containing Sup35 NM alone lacks detectable infectivity . Progeny cell extracts transformed into yeast are identified as RX-Y , where X corresponds to a round number and Y corresponds to a clone number assigned sequentially and clockwise according to ( A ) and ( B ) . Clones that gave rise to aggregate-negative progeny in the subsequent round are indicated by asterisks . Analysis of these data by Fisher's exact test indicates that the differences in the frequency of [PSI+] transformants observed with samples containing SDS-stable Sup35 NM aggregates compared with the sample containing soluble Sup35 NM are statistically significant ( p < 0 . 0001 ) . The percentages given refer to strong [PSI+] transformants; samples containing SDS-stable Sup35 NM aggregates ( but not samples containing soluble Sup35 NM ) also gave rise to weak [PSI+] transformants ( Figure 5—figure supplement 1 ) , but these were not quantified ( ‘Results’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02949 . 01010 . 7554/eLife . 02949 . 011Figure 5—figure supplement 1 . Bacterial cell extracts containing propagated , infectious Sup35 NM aggregates yield both strong and weak [PSI+] yeast transformants . The phenotypes of five representative strong [PSI+] ( A ) and weak [PSI+] ( B ) strains obtained by transforming S . cerevisiae [psi−] cells with E . coli cell extracts containing propagated , SDS-stable Sup35 NM-aggregates on 1/4 YPD agar before ( left ) and after ( right ) passage on YPD agar supplemented with 3 mM GuHCl . For the purposes of comparison , untransformed [psi−] , weak [PSI+] , and strong [PSI+] yeast strains are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02949 . 011 The remaining two experimental lineages ( L2E and L4E ) lost detectable SDS-stable Sup35 NM aggregates at R3 and R4 , respectively ( Figure 4—figure supplement 1A , C , Figure 5A ) . Moreover , the loss of aggregates manifested itself in all ten of the selected colonies at either R3 or R4 ( Figure 4—figure supplement 1A , C , Figure 5A ) . Curiously , the loss of SDS-stable Sup35 NM aggregates from a particular lineage coincided with a loss of detectable fusion protein , as assessed by Western blot analysis ( Figure 4—figure supplement 1A , C ) , suggesting that the loss of aggregates represented an indirect consequence of a radical drop in protein levels ( see ‘Discussion’ ) . We note that the observed drop in Sup35 NM-mCherry-His6x fusion protein levels is not irreversible , as exemplified by L2E , which exhibited a loss of detectable Sup35 NM aggregates coincident with an R3 drop in fusion protein levels and an apparent restoration of fusion protein levels by R4 in all 10 samples ( Figure 4—figure supplement 1A ) . Despite the presence of normal levels of fusion protein , SDS-stable Sup35 NM aggregates were not recovered in R4 of L2E , consistent with the expectation that prion propagation requires that prion protein synthesis be maintained above some threshold level ( Figure 4—figure supplement 1A , Figure 5A; Holmes et al . , 2013 ) . Critically , none of the samples ( 120 in total ) from any of the four control lineages contained detectable SDS-stable Sup35 NM aggregates ( Figure 4B , Figure 4—figure supplement 2 ) . Fluorescence microscopy revealed that cells containing propagated Sup35 NM aggregates exhibited smaller foci emanating from large aggregates typically localized at cell poles , a phenotype distinguished from experimental starter culture cells by the lack of twisted ring structures ( Figure 5C ) . However , we observed one instance of aggregate-positive R1 cells exhibiting twisted ring structures ( see Figure 6—figure supplement 2C ) . Whereas we cannot definitively assign the SDS-stable Sup35 NM aggregates detected by filter retention to those structures detected by fluorescence microscopy , we note that fluorescence microscopy of prion-containing yeast cells has also revealed structural diversity ( Derkatch et al . , 2001; Zhou et al . , 2001 ) . Furthermore , cells from aggregate-negative samples invariably exhibited diffuse fluorescence ( Figure 5D ) . 10 . 7554/eLife . 02949 . 012Figure 6 . Sup35 NM prion propagation in E . coli requires ClpB . ( A ) Cartoon representation of how Sup35 NM can convert to its prion form in the presence of New1 and ClpB but cannot propagate in the prion conformation after cells have been cured of New1- and ClpB-encoding DNA . Sup35 NM and New1 ( black ) are depicted as fusions to mCherry ( red ) and mGFP ( green ) , respectively . ClpB is depicted as a purple hexamer . ( B ) SDS-stable Sup35 NM aggregates are detected in 5 of 20 Round 1 ( R1 ) wild-type ( WT ) clones derived from a starter culture ( ST ) of wild-type cells containing Sup35 NM and New1 as assessed by filter retention analysis . In total , 17 of 60 R1 wild-type clones are aggregate-positive ( Figure 6—figure supplement 2A ) . Starter cultures of cells containing Sup35 NM and New1 and cells containing Sup35 NM alone serve as positive ( P ) and negative ( N ) controls , respectively . In all 20 R1 wild-type clones shown , full-length ( FL ) ClpB is detectable , Sup35 NM fusion protein levels are comparable , and New1 fusion protein is not detectable as assessed by Western blot analysis . The α-His6X and α-Sup35 antibodies recognize the Sup35 NM-mCherry-His6X fusion protein , the α-GFP antibody recognizes the New1–mGFP fusion protein , the α-ClpB antibody recognizes the E . coli ClpB chaperone , and the α-RpoA antibody recognizes the α subunit of E . coli RNA polymerase . Cells from four aggregate-positive R1 wild-type clones visualized by fluorescence microscopy ( Figure 6—figure supplement 2C ) are indicated by asterisks . ( C ) SDS-stable Sup35 NM aggregates are not detectable in R1 ΔclpB clones derived from a starter culture of ΔclpB cells containing Sup35 NM , New1 , and ectopically produced ClpB as assessed by filter retention analysis . In total , 0 of 60 R1 ΔclpB clones are aggregate-positive ( Figure 6—figure supplement 2B ) . Starter cultures of wild-type cells containing Sup35 NM and New1 and wild-type cells containing Sup35 NM alone serve as positive ( P ) and negative ( N ) controls , respectively . In all 20 R1 ΔclpB clones shown , Sup35 NM fusion protein levels are comparable , and neither ClpB nor New1 fusion protein is detectable as assessed by Western blot analysis . Cells from four aggregate-negative R1 wild-type clones visualized by fluorescence microscopy ( Figure 6—figure supplement 2D ) are indicated by asterisks . ( D ) Extract prepared from cells lacking ClpB is not infectious when transformed into S . cerevisiae [psi–] cells . Starter cultures of wild-type cells transformed with pBR322-SUP35 NM and pSC101TS-NEW1 as well as ΔclpB cells transformed with pBR322-SUP35 NM and pSC101TS-NEW1-clpB contain infectious SDS-stable Sup35 NM aggregates capable of converting [psi−] yeast cells to [PSI+] . Wild-type starter culture cells containing Sup35 NM alone lack detectable infectivity . An aggregate-positive R1 wild-type clone retains infectious Sup35 NM aggregates . In contrast , an aggregate-negative R1 ΔclpB clone lacks detectable infectivity , as does an aggregate-negative R1 wild-type clone . Analysis of these data by Fisher's exact test indicates that the differences in the frequency of [PSI+] transformants observed with samples containing SDS-stable Sup35 NM aggregates compared with the samples containing soluble Sup35 NM are statistically significant ( p < 0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02949 . 01210 . 7554/eLife . 02949 . 013Figure 6—figure supplement 1 . ΔclpB cells containing New1 and ectopically produced ClpB support the formation of SDS-stable Sup35 NM aggregates . ( A ) SDS-stable Sup35 NM aggregates are detected in wild-type ( WT ) cells producing SDS-stable New1 aggregates as assessed by filter retention analysis . SDS-stable Sup35 NM aggregates are also detected in ΔclpB cells containing SDS-stable New1 aggregates and ectopically produced ClpB . The α-His6X antibody detects the Sup35 NM-mCherry-His6X fusion protein , and the α-GFP antibody detects the New1-mGFP fusion protein . A lane cropped from the same immunoblot is indicated by a hash mark . Intracellular levels of full-length ( FL ) ClpB , Sup35 NM fusion protein , and New1 fusion protein are comparable in the presence and absence of New1 and ectopically produced ClpB as assessed by Western blot analysis . The α-ClpB antibody recognizes the E . coli ClpB chaperone , the α-Sup35 antibody recognizes the Sup35 NM fusion protein , and the α-RpoA antibody recognizes the α subunit of E . coli RNA polymerase . ( B ) Fluorescence images of representative wild-type cells containing Sup35 NM and New1 and ΔclpB cells containing Sup35 NM , New1 , and ectopically produced ClpB . The mCherry channel , GFP channel , and merged images are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02949 . 01310 . 7554/eLife . 02949 . 014Figure 6—figure supplement 2 . The fate of Sup35 NM in 40 wild-type R1 clones and 40 ΔclpB R1 clones . ( A ) SDS-stable Sup35 NM aggregates are detected in 12 of 40 Round 1 ( R1 ) wild-type ( WT ) clones derived from a starter culture ( ST ) of wild-type cells containing Sup35 NM and New1 as assessed by filter retention analysis . In total , 17 of 60 R1 wild-type clones are aggregate-positive ( Figure 6B ) . Starter cultures of cells containing Sup35 NM and New1 and cells containing Sup35 NM alone serve as positive ( P ) and negative ( N ) controls , respectively . The α-His6X antibody detects the Sup35 NM-mCherry-His6X fusion protein . ( B ) In contrast , 0 of 40 R1 ΔclpB clones derived from a starter culture of ΔclpB cells transformed with pBR322-SUP35 NM and pSC101TS-NEW1-clpB contain detectable SDS-stable Sup35 NM aggregates . In total , 0 of 60 R1 ΔclpB clones are aggregate-positive ( Figure 6C ) . The observed difference in the number of aggregate-positive clones of wild-type vs ΔclpB cells is statistically significant ( p < 0 . 0001 as determined by Fisher's Exact Test ) . ( C ) Fluorescence images of representative cells corresponding to the four aggregate-positive R1 wild-type clones indicated by asterisks in Figure 6B . Notably , wild-type clone R1-14 exhibits twisted ring structures . ( D ) Fluorescence images of representative cells corresponding to the four aggregate-negative R1 ΔclpB clones indicated by asterisks in Figure 6C . DOI: http://dx . doi . org/10 . 7554/eLife . 02949 . 014 We next sought to determine whether cells that maintained SDS-stable Sup35 NM aggregates in the absence of New1 contained infectious material capable of converting [psi−] yeast cells to [PSI+] . We prepared bacterial cell extracts from both experimental and control starter cultures , as well as aggregate-positive samples from each of the four experimental lineages . For each experimental lineage , we examined an arbitrarily chosen sample obtained from the last aggregate-positive round . In addition , for L2E and L4E , we examined , respectively , the R2 and R3 samples ( indicated by asterisks in Figure 5E ) that gave rise to aggregate-negative clones in the subsequent round . We used these bacterial extracts to transform S . cerevisiae spheroplasts prepared from a [pin−][psi−] strain . The use of a [pin−] recipient strain was critical as transient overproduction of Sup35 ( or Sup35 NM ) in [PIN+][psi−] strains significantly stimulates the conversion from [psi−] to [PSI+] ( Derkatch et al . , 1997 ) , whereas conversion in a [pin−][psi−] background requires the introduction of infectious seed material ( Tanaka and Weissman , 2006 ) . The experimental starter culture yielded [PSI+] yeast transformants at a frequency of ∼1% ( Figure 5E ) , consistent with our previous findings ( Garrity et al . , 2010 ) . Similarly , each of the aggregate-positive samples from the four experimental lineages yielded [PSI+] yeast transformants at a frequency of ∼1%; in contrast , the aggregate-negative samples yielded no [PSI+] yeast transformants ( Figure 5E ) . Among the [PSI+] transformants we obtained with the experimental samples , we observed both ‘strong’ and ‘weak’ strains ( Figure 5—figure supplement 1; Tanaka et al . , 2004; Frederick et al . , 2014 ) . We note that ∼1% corresponds only to the frequency of strong [PSI+] transformants and therefore represents a conservative estimate of E . coli cell extract infectivity; we did not attempt to quantify weak [PSI+] transformants because they are difficult to distinguish from [psi−] transformants on the medium utilized to isolate transformants . We conclude that E . coli cells can propagate Sup35 NM in an infectious prion conformation over at least ∼100 generations under conditions that do not permit de novo prion formation . The propagation of [PSI+] and other prions in yeast requires Hsp104 , an Hsp100-family ATP-dependent disaggregase that functions as a ring-shaped hexamer . Specifically , Hsp104 is thought to facilitate prion propagation by fragmenting large aggregates into smaller propagons that are subsequently disseminated during cell division ( Paushkin et al . , 1996; Ness et al . , 2002; Cox et al . , 2003; Kryndushkin et al . , 2003; Satpute-Krishnan et al . , 2007; Higurashi et al . , 2008 ) . We therefore investigated whether or not the propagation of Sup35 NM aggregates in E . coli requires ClpB , the bacterial ortholog of Hsp104 . To address this question , we sought to deplete cells of ClpB specifically during the propagation phase of our experiments . To accomplish this , we modified pSC101TS-NEW1 such that it also directed the expression of clpB under the control of its native promoter . When transformed into ΔclpB cells , pSC101TS-NEW1-clpB enabled us to grow starter cultures containing ClpB and subsequently to deplete both ClpB and New1 in cells plated at the non-permissive temperature ( Figure 6A ) . As expected , we detected SDS-stable Sup35 NM aggregates in ΔclpB starter culture cells transformed with pBR322-SUP35 NM and pSC101TS-NEW1-clpB ( Figure 6—figure supplement 1A ) . Furthermore , fluorescence microscopy revealed that these cells contained visible aggregates that were nearly indistinguishable from those in wild-type cells containing pBR322-SUP35 NM and pSC101TS-NEW1 ( Figure 6—figure supplement 1B ) . After plating ΔclpB starter culture cells containing pBR322-SUP35 NM and pSC101TS-NEW1-clpB at the non-permissive temperature to cure cells of ClpB- and New1-encoding DNA , we examined 60 R1 colonies for the presence of SDS-stable Sup35 NM aggregates . In parallel , we examined 60 R1 colonies derived from wild-type starter culture cells containing pBR322-SUP35 NM and pSC101TS-NEW1 . As before , every selected colony was patched onto selective medium to test for the loss of pSC101TS-NEW1-clpB or pSC101TS-NEW1 and inoculated into liquid medium for overnight growth to test for the presence of SDS-stable Sup35 NM aggregates . All selected colonies had lost the appropriate temperature-sensitive vector and the absence of New1 and/or ClpB was confirmed by Western blot analysis ( Figure 6B , Figure 6C ) . Whereas 17 of 60 ( 28% ) wild-type R1 samples tested aggregate-positive , all ΔclpB R1 samples tested aggregate-negative ( Figure 6B , Figure 6—figure supplement 2A , Figure 6C , Figure 6—figure supplement 2B ) . Western blot analysis revealed that the wild-type and ΔclpB R1 cells contained comparable amounts of Sup35 NM fusion protein ( Figure 6B , Figure 6C ) . Furthermore , yeast transformation assays confirmed the presence of infectious material capable of converting [psi−] yeast cells to [PSI+] in ΔclpB starter culture cells transformed with pBR322-SUP35 NM and pSC101TS-NEW1-clpB as well as in an aggregate-positive R1 clone derived from wild-type starter culture cells ( Figure 6D ) . In contrast , a ΔclpB R1 clone derived from ΔclpB starter culture cells containing pBR322-SUP35 NM and pSC101TS-NEW1-clpB as well as an aggregate-negative R1 clone derived from wild-type starter culture cells containing pBR322-SUP35 NM and pSC101TS lacked detectable infectivity ( Figure 6D ) . We conclude that cells lacking ClpB cannot propagate Sup35 NM in its infectious prion conformation . To investigate the mechanistic basis for the ClpB dependence of Sup35 NM prion propagation in E . coli , we devised a strategy that enabled us to test the abilities of specific ClpB mutants to support the propagation of SDS-stable Sup35 NM aggregates after their formation in the presence of wild-type ClpB . The disaggregase function of ClpB , which assembles as a two-tiered hexameric ring ( Lee et al . , 2003 ) , depends on its abilities to hydrolyze ATP , to translocate polypeptides through its central pore and to collaborate with DnaK ( the bacterial Hsp70 ) and its co-chaperones DnaJ and the nucleotide exchange factor GrpE ( reviewed in Doyle and Wickner , 2009 ) . Accordingly , we tested previously characterized ClpB mutants specifically defective for ( i ) ATP hydrolysis ( E279A/E678A ) ( Weibezahn et al . , 2003 ) ( ii ) substrate threading through the ClpB pore ( Y653A ) ( Weibezahn et al . , 2004 ) , and ( iii ) collaboration with DnaK ( E432A ) ( Oguchi et al . , 2012; Seyffer et al . , 2012; Carroni et al . , 2014 ) . We note that each of these ClpB mutants is fully proficient for oligomerization and only ClpB E279A/E678A is deficient in ATPase activity ( Mogk et al . , 2003; Weibezahn et al . , 2004; Oguchi et al . , 2012 ) . Our strategy required us to construct strains in which we could induce the production of a ClpB mutant specifically during the propagation phase of the experiment while providing wild-type ClpB during the formation phase of the experiment only . To accomplish this , we placed each of the mutant clpB alleles ( or the wild-type allele ) under the control of the anhydrotetracycline ( aTc ) -inducible promoter PLtetO-I ( Lutz and Bujard , 1997 ) , integrated these constructs onto the chromosome of our ΔclpB strain , and transformed the resulting strains with pBR322-SUP35 NM and pSC101TS-NEW1-clpB . As expected , we detected SDS-stable Sup35 NM aggregates in starter culture cells of all strains producing plasmid-encoded Sup35 NM-mCherry-His6X , New1-mGFP , and wild-type ClpB ( Figure 7A ) . To determine whether or not each of the mutants could support the propagation of these aggregates following the depletion of New1 and wild-type ClpB , we plated the starter culture cells at the nonpermissive temperature on solid medium lacking or containing increasing concentrations of aTc , generating sets of R1 colonies . We prepared cell extracts from scraped R1 colonies ( ‘Materials and methods’ ) and examined these extracts for the presence or absence of SDS-stable Sup35 NM aggregates . Whereas SDS-stable Sup35 NM aggregates were detected as a function of increasing aTc concentration in cells carrying the wild-type clpB allele , no aggregates were detected in cells harboring the clpB E279A/E678A , clpB Y653A , or clpB E432A allele at any concentration of aTc ( Figure 7A ) . Western blot analysis revealed that levels of chromosomally-encoded wild-type ClpB and each of the three disaggregase mutants were comparable in cell extracts prepared from colonies scraped off of plates containing 50 ng/ml aTc ( Figure 7B ) . Furthermore , replica plating confirmed that all colonies of R1 cells grown on medium supplemented with 50 ng/ml aTc had been cured of pSC101TS-NEW1-clpB ( Figure 7—figure supplement 1 ) . We conclude that ATP hydrolysis coupled to substrate translocation through the ClpB central pore and collaboration with DnaK are required for propagation of SDS-stable Sup35 NM aggregates in the absence of New1 . 10 . 7554/eLife . 02949 . 015Figure 7 . Propagation of SDS-stable Sup35 NM aggregates in E . coli requires ClpB disaggregase activity . ( A ) SDS-stable Sup35 NM aggregates are detected in starter cultures ( ST ) of ΔclpB cells containing pBR322-SUP35 NM , pSC101TS-NEW1-clpB , and one of four aTc-inducible chromosomal clpB alleles . Wild-type ( WT ) ClpB is depicted as a purple hexamer . ClpB E279A/E678A is unable to hydrolyze ATP , ClpB Y653A is pore-deficient , and ClpB E432A is unable to collaborate with DnaK . Propagated Sup35 NM aggregates are detected in scraped cell suspensions as a function of increasing aTc concentration only for Round 1 ( R1 ) clones producing wild-type ClpB . Sup35 NM aggregates are not detected at any aTc concentration in scraped cell suspensions of R1 clones producing ClpB disaggregase mutants or in R1 clones lacking ClpB . Lanes cropped from the same immunoblot are indicated by hash marks . The α-His6X antibody recognizes the Sup35 NM-mCherry-His6X fusion protein . ( B ) Wild-type and mutant ClpB levels along with Sup35 NM fusion protein levels are comparable in R1 clones grown on solid medium supplemented with 50 ng/ml aTc as assessed by Western blot analysis . The α-Sup35 antibody recognizes the Sup35 NM-mCherry-His6X fusion protein , the α-GFP antibody recognizes the New1-mGFP fusion protein , the α-ClpB antibody recognizes the E . coli ClpB chaperone , and the α-RpoA antibody recognizes the α subunit of E . coli RNA polymerase . DOI: http://dx . doi . org/10 . 7554/eLife . 02949 . 01510 . 7554/eLife . 02949 . 016Figure 7—figure supplement 1 . All Round 1 clones producing wild-type ClpB are cured of pSC101TS-NEW1-clpB . All Round 1 ( R1 ) clones derived from a starter culture of ΔclpB cells containing pBR322-NM SUP35 , pSC101TS-NEW1-clpB , and chromosomal aTc-inducible wild-type clpB lose pSC101TS-NEW1-clpB as assessed by replica plating from solid medium supplemented with carbenicillin ( Carb ) , chloramphenicol ( Cam ) , IPTG , and 50 ng/ml aTc to solid medium containing either Carb or Cam . pSC101TS-NEW1-clpB confers Cam resistance . DOI: http://dx . doi . org/10 . 7554/eLife . 02949 . 016 As only two of four lineages retained the prion for the full duration of our experiments , propagation of the Sup35 NM prion may be less stable in E . coli than in S . cerevisiae ( DiSalvo et al . , 2011 ) . However , as noted above , loss of the prion in two lineages was coincident with a dramatic drop in Sup35 NM fusion protein levels . Furthermore , this drop was evidently reversible as Sup35 NM fusion protein levels were restored in R4 of L2E without reappearance of the prion . We do not understand the mechanism underlying this reversible change in protein levels; however , we suggest that stochastic fluctuations in plasmid copy number may set the stage for such an event . We note that our experiments were performed in recA− cells , which should prevent plasmid rearrangements that might lead to a permanent loss of fusion protein coding capacity . The question of whether or not ClpB can substitute for Hsp104 in promoting Sup35 prion propagation has been addressed in a number of studies yielding conflicting indications . On one hand , several in vitro studies have provided evidence that Hsp104 ( Shorter and Lindquist , 2004 , 2006 , 2008; DeSantis et al . , 2012 ) , but not ClpB ( DeSantis et al . , 2012 ) , can fragment amyloid aggregates in the absence of auxiliary factors . Furthermore , whereas the presence of various combinations of S . cerevisiae Hsp70- and Hsp40-family proteins was found to modulate Hsp104 activity on amyloid substrates ( Shorter and Lindquist , 2008; DeSantis et al . , 2012 ) , ClpB appeared to remain inert even in the presence of bacterial Hsp70 ( DnaK ) , Hsp40 ( DnaJ ) , and nucleotide exchange factor GrpE despite exhibiting robust activity on various disordered protein aggregates in vitro ( DeSantis et al . , 2012 ) . On the other hand , the results of several in vivo studies suggest that ClpB , in the presence of appropriate co-chaperones , is competent to support Sup35 prion propagation in yeast ( Tipton et al . , 2008; Reidy et al . , 2012 ) . Based on an analysis of the in vivo activities of Hsp104/ClpB chimeras , Tipton et al . argue that prion replication in yeast requires that Hsp104 collaborate with its cognate Hsp70 chaperone system . A logical inference from their work is that the inability of ClpB to substitute for Hsp104 in supporting Sup35 prion propagation in S . cerevisiae is an indirect consequence of the inability of ClpB to cooperate with fungal co-chaperones . More recently , Reidy et al . provided direct support for this inference . In particular , Reidy et al . found that ClpB supported prion propagation in yeast provided that DnaK and GrpE were present . Interestingly , the activity of the bacterial disaggregase machinery in yeast was dependent on the fungal Hsp40-family Sis1 protein , consistent with prior work implicating Sis1 as a necessary component of the chaperone network required for prion propagation in yeast ( Higurashi et al . , 2008; Tipton et al . , 2008 ) . Our work demonstrates that no exogenous fungal accessory factors are required for prion propagation in bacteria . Taken together , these observations argue that the amyloid remodeling activity of Hsp104 is an evolutionarily conserved feature of the Hsp100-family chaperones , an inference that is strongly supported by our finding that propagation of the Sup35 NM prion in E . coli requires ClpB disaggregase activity . Despite the apparent prevalence of prions in the fungal kingdom , to date , no bacterial prion has been identified . Notably , the absence of cytoplasmic mixing during conjugation would preclude the discovery of prion-like phenomena by classic genetic approaches , which facilitated the discovery of prions in yeast based on the non-Mendelian inheritance of their associated phenotypes ( Cox , 1965; Aigle and Lacroute , 1975; Wickner , 1994 ) . Nevertheless , recent bioinformatic analyses of prokaryotic proteomes have revealed that bacterial and archaeal genomes encode many proteins containing glutamine- and asparagine-rich prion-like domains resembling those found in most confirmed and putative S . cerevisiae prions ( Alberti et al . , 2009; Espinosa Angarica et al . , 2013; Yuan et al . , unpublished data ) . Moreover , it is becoming increasingly clear that Q/N-richness at the level of primary amino acid sequence is neither a prerequisite for prion conversion ( Taneja et al . , 2007; Suzuki et al . , 2012 ) nor protein amyloidogenesis ( Goldschmidt et al . , 2010 ) . In fact , several bacteria utilize non-Q/N-rich amyloid-forming proteins to assemble extracellular appendages mediating surface attachment and biofilm formation ( Chapman et al . , 2002; Romero et al . , 2010 ) . These considerations—in conjunction with the work presented here—suggest that prions or prion-like proteins may exist as epigenetic reservoirs of phenotypic diversity in the bacterial domain of life . Bacteria experiments were performed with E . coli strain DH5αZ1 ( Lutz and Bujard , 1997 ) grown in LB ( Miller ) medium . To construct DH5αZ1 ΔclpB , a temperature-sensitive plasmid encoding the RecA protein ( pSC101TS-recA ) was constructed and transformed into DH5αZ1 cells . A ΔclpB::kan allele from strain JW2573 ( Keio collection ) was transferred to DH5αZ1 cells containing pSC101TS-recA via P1 transduction . Cells were subsequently cured of pSC101TS-recA by overnight growth and plating in the absence of antibiotic selection at the non-permissive temperature ( 37°C ) . To construct strains harboring chromosomal PLtetO-I-clpB alleles , plasmids pAY152 , pAY154 , pAY155 , pAY156 , and pAY157 were cloned in strain AY290 and integrated onto the chromosome of strain AY295 at attB ( HK022 ) . Single-copy integrants were selected on LB agar supplemented with kanamycin ( 10 μg/ml ) and verified by PCR as described ( Haldimann and Wanner , 2001 ) . Yeast experiments were performed with S . cerevisiae strain YJW187 [pin−][psi−] grown in yeast extract peptone dextrose ( YPD ) medium . For yeast infectivity assays , cell extracts were co-transformed with pRS316 into YJW187 spheroplasts; [PSI+] URA+ transformants were identified by plating the yeast cells in top agar containing synthetic defined medium lacking uracil and adenine ( SD-Ura-Ade ) and supplemented with 10 mg/ml adenine hemisulfate ( Sunrise Science , San Diego , CA ) . Further details concerning strains and plasmids are provided in Supplementary file 1 . Cells were transformed with pBR322-SUP35 NM and pSC101TS , pSC101TS-NEW1 , or pSC101TS-NEW1-clpB and grown at 30°C on LB agar supplemented with carbenicillin ( Carb , 100 μg/ml ) and chloramphenicol ( Cam , 12 . 5 μg/ml ) . Starter cultures were generated by growing transformants at 30°C in 6 ml of LB broth supplemented with Carb ( 100 μg/ml ) , Cam ( 12 . 5 μg/ml ) , and 10 μM IPTG to an OD600 of 2 . 0–2 . 5 . To cure cells of pSC101TS-derivatives and generate Round 1 ( R1 ) colonies , starter cultures were diluted ( 10−5 ) in pre-warmed ( 37°C ) LB broth supplemented with Carb ( 100 μg/ml ) and 10 μM IPTG . Diluted cells were grown at 37°C on pre-warmed ( 37°C ) LB agar supplemented with Carb ( 100 μg/ml ) and 10 μM IPTG . R1–R4 colonies were , ( a ) patched on LB agar supplemented with Cam ( 12 . 5 μg/ml ) , ( b ) restreaked and grown at 30°C on pre-warmed ( 30°C ) LB agar supplemented with Carb ( 100 μg/ml ) and 10 μM IPTG , and ( c ) inoculated and grown at 30°C in 6 ml LB broth supplemented with Carb ( 100 μg/ml ) and 10 μM IPTG . For analysis of ClpB disaggregase mutants , ∼1000 R1 colonies were gently scraped off LB agar plates containing Carb ( 100 μg/ml ) , 10 μM IPTG , and a range of aTc concentrations ( 0–500 ng/ml ) in 3 ml LB broth supplemented with Carb ( 100 μg/ml ) and 10 μM IPTG . Cell cultures and scraped cell suspensions were normalized to 8 ml of an OD600 of 1 . 0 and pelleted by centrifugation . Cell pellets were resuspended in 166 ml STC Buffer ( 1 M sorbitol , 10 mM Tris–HCl [pH 7 . 5] , 10 mM CaCl2 ) supplemented with 10 U of rLysozme ( Novagen , Germany ) and 0 . 1 U of OmniCleave endonuclease ( Epicentre , Wisconsin , MA ) , incubated at room-temperature for 30 min , and incubated on ice for an additional 30 min . Omnicleave endonuclease was omitted from samples destined for PCR analysis and yeast infectivity assays . Finally , samples were flash frozen and thawed on ice to yield unclarified lysates ( used in filter retention assays ) . To generate partially clarified lysates ( used in Western blot analysis and yeast transformation assays ) , unclarified lysates were subjected to two rounds of low-speed centrifugation , each at 500 RCF for 15 min at 4°C . 25 μl of unclarified lysates was added to 375 μl of BugBuster protein extraction reagent ( Novagen ) supplemented with 5 U of rLysozyme and 0 . 1 U of Omnicleave endonuclease and gently rocked at room-temperature for 30 min . Samples were challenged with 100 μl of 10% ( wt/vol ) SDS ( 2% SDS final concentration ) and gently rocked at room-temperature for an additional 30 min . For each sample , 100 μl of undiluted lysate and three twofold serial dilutions made in PBS containing 2% SDS were filtered through a 0 . 2-μm cellulose acetate membrane ( Advantec , Japan ) in a dot-blotting vacuum manifold . Samples on membranes were washed twice with 100 μl of PBS containing 2% SDS and twice with 100 μl of PBS . Cellulose acetate membranes ( used in filter retention assays ) and Hybond-C Extra nitrocellulose membranes ( used in Western blot analysis ) were blocked for 30 min in PBS containing 3% ( wt/vol ) milk . Membranes were probed with one of the following primary antibodies: anti-Sup35 ( yS-20 , Santa Cruz Biotechnology , Dallas , TX , 1:5000 ) , anti-His6X ( His-2; Roche , Indianapolis , IN , 1:10 , 000 ) , anti-GFP ( Roche , 1:10 , 000 ) , anti-RpoA ( NeoClone , Madison , WI , 1:10 , 000 ) , or anti-ClpB ( gift from S Wickner , 1:10 , 000 ) . Membranes were washed and probed with one of the following HRP-conjugated secondary antibodies: anti-goat IgG ( Santa Cruz Biotechnology , 1:10 , 000 ) , anti-mouse IgG ( Cell Signaling , Beverly , MA , 1:10 , 000 ) , or anti-rabbit IgG ( Cell Signaling , 1:10 , 000 ) . Proteins were detected with ECL Plus Western blot detection reagents ( GE Healthcare , Pittsburgh , PA ) and a ChemiDock XRS+ imaging system ( Bio-Rad , Hercules , CA ) . Protein concentrations of partially clarified E . coli cell extracts were determined by the bicinchoninic acid ( BCA ) assay ( ThermoFisher , Waltham , MA ) and normalized to ∼1 mg/ml . Protein transformations were performed as previously described ( Tanaka and Weissman , 2006; Garrity et al . , 2010 ) . Each and every putative [PSI+] transformant was ( a ) restreaked on 1/4 YPD agar to assess the [PSI+] phenotype , ( b ) restreaked on YPD containing 3 mM GuHCl to cure cells of [PSI+] , and ( c ) restreaked on 1/4 YPD to assess the [psi−] phenotype . Only those transformants exhibiting curability were scored as [PSI+] . Cells were spotted onto 1% ( wt/vol ) agarose pads consisting of Seakem LE Agarose ( Lonza , Walkersville , MD ) in PBS and visualized with an UplanFL N 100x/1 . 30 phase contrast objective mounted on an Olympus BX61 microscope . Images were captured with a CoolSnapHQ camera ( Photometrics , Tucson , AZ ) and the Metamorph software package ( Molecular Devices , Sunnyvale , CA ) . All fluorescence images were obtained from 10 ms exposures .	Unlike most infectious agents—such as viruses or bacteria—that contain genetic material in the form of DNA or RNA , a prion is simply an aggregate of misfolded proteins . Although they are not living organisms , these prion aggregates can self-propagate; when they enter a healthy organism , they cause existing , correctly folded proteins to adopt the prion fold . Within the aggregate , the prion proteins have a corrugated structure that allows them to stack together tightly , which in turn makes the aggregates very stable . As more prions are formed , they then trigger other protein molecules to misfold and join the aggregates , and the aggregates continue to grow and spread within the infected organism causing tissue damage and cell death . Prion diseases are well known in mammals , where the prion aggregates typically destroy tissue within the brain or nervous system . Bovine spongiform encephalopathy ( also commonly known as BSE or ‘mad cow disease’ ) is an example of a prion disease that affects cattle and can be transmitted to humans by eating infected meat . Prions also form in yeast and other fungi . These prions , however , do not cause disease or cell death; instead , yeast prions act as protein-based elements that can be inherited over multiple generations and which provide the yeast with new traits or characteristics . Although prions can form spontaneously in yeast cells , their stable propagation depends on so-called chaperone proteins that help to remodel the prion aggregates . Previous work has shown that bacterial cells can also support the formation of prion-like aggregates . The bacteria were engineered to produce two yeast prion proteins—one of which spontaneously formed aggregates that were needed to trigger the conversion of the other to its prion form . However , it was not known if bacterial cells could support the stable propagation of prions if the initial trigger for prion conversion was removed . Yuan et al . now reveal that the bacterium Escherichia coli can propagate a yeast prion for over a hundred generations , even when the cells can no longer make the protein that serves as the trigger for the initial conversion . This propagation depends on a bacterial chaperone protein called ClpB , which is related to another chaperone protein that is required for stable prion propagation in yeast . As such , the findings of Yuan et al . raise the possibility that , even though a prion specific to bacteria has yet to be identified , prions or prion-like proteins might also contribute to the diversity of traits found in bacteria . Furthermore , since both yeast and bacteria form and propagate prions in similar ways , such protein-based inheritance might have evolved in these organisms' common ancestor over two billion years ago .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2014	Prion propagation can occur in a prokaryote and requires the ClpB chaperone
In budding yeast , if the spindle becomes mispositioned , cells prevent exit from mitosis by inhibiting the mitotic exit network ( MEN ) . The MEN is a signaling cascade that localizes to spindle pole bodies ( SPBs ) and activates the phosphatase Cdc14 . There are two competing models that explain MEN regulation by spindle position . In the 'zone model' , exit from mitosis occurs when a MEN-bearing SPB enters the bud . The 'cMT-bud neck model' posits that cytoplasmic microtubule ( cMT ) -bud neck interactions prevent MEN activity . Here we find that 1 ) eliminating cMT– bud neck interactions does not trigger exit from mitosis and 2 ) loss of these interactions does not precede Cdc14 activation . Furthermore , using binucleate cells , we show that exit from mitosis occurs when one SPB enters the bud despite the presence of a mispositioned spindle . We conclude that exit from mitosis is triggered by a correctly positioned spindle rather than inhibited by improper spindle position . Asymmetric cell division is a common characteristic of development and is seen in diverse cell types ranging from Drosophila neuroblasts to mammalian oocytes . In order to produce viable progeny with distinct cell fates , asymmetrically dividing cells must coordinate nuclear position with the site of cytokinesis . When the spindle is mispositioned with respect to the cleavage plane , cell cycle progression is delayed until the spindle is correctly aligned . In the budding yeast Saccharomyces cerevisiae , coupling exit from mitosis with spindle position is particularly important because the site of cytokinesis forms independently of mitotic spindle position ( Pruyne et al . , 2004 ) . Thus budding yeast has evolved mechanisms that align the spindle and that ensure that cytokinesis only occurs after the nucleus has been partitioned between the mother cell and bud , the future daughter cell ( Bardin et al . , 2000; Miller et al . , 1999; Pereira et al . , 2000; Yeh et al . , 1995 ) . Budding yeast employs two redundant mechanisms to position the spindle along the mother – bud axis . The first positioning mechanism relies on the type V myosin motor Myo2 . Myo2 , together with its adaptor Kar9 and the plus-end microtubule binding protein Bim1 , positions the pre-anaphase spindle at the bud neck by pulling cytoplasmic microtubules ( cMTs ) along actin cables ( Beach et al . , 2000; Kopecká and Gabriel , 1998; Miller et al . , 1999; Miller and Rose , 1998; Palmer et al . , 1992 ) . The second pathway is active during anaphase and requires the minus-end microtubule motor protein dynein , together with its associated coactivating complex dynactin . Dynein , when anchored to the cell cortex by Num1 , aligns the spindle by essentially towing it along the cortex of the cell ( Farkasovsky and Küntzel , 2001; Heil-Chapdelaine et al . , 2000; Muhua et al . , 1994; Yeh et al . , 1995 ) . The consequences of eliminating either positioning pathway are minor but cells lacking both pathways are inviable ( Miller and Rose , 1998 ) . Cells that fail to segregate the nucleus into the bud will arrest in late anaphase with the spindle mispositioned in the mother cell compartment . If the cell manages to correct this positioning defect and threads the nucleus through the bud neck into the bud , it will then disassemble the anaphase spindle and exit from mitosis ( Yeh et al . , 1995 ) . Yeast cells link spindle position and exit from mitosis through the regulation of the essential phosphatase Cdc14 ( Bardin et al . , 2000; D'Aquino et al . , 2005; Pereira et al . , 2000; Pereira and Schiebel , 2005 ) . Cdc14 functions to reverse mitotic cyclin-CDK ( cyclin dependent kinase ) activity by dephosphorylating cyclin-CDK targets as well as by targeting cyclins for degradation ( Jaspersen et al . , 1998; Visintin et al . , 1998; Zachariae et al . , 1998 ) . These Cdc14 functions cause exit from mitosis , the final cell cycle transition that encompasses disassembly of the mitotic spindle and cytokinesis ( Stegmeier and Amon , 2004 ) . Cdc14’s essential role in exit from mitosis requires that its activity is tightly regulated . Cdc14 is kept inactive from G1 to metaphase by its inhibitor Cfi1/Net1 , which functions by sequestering the phosphatase in the nucleolus . It is only upon anaphase entry that Cdc14 is released from Cfi1/Net1 to spread throughout the cell where it antagonizes mitotic CDK activity and so returns the cell to G1 ( Shou et al . , 1999; Visintin et al . , 1999 ) . Two pathways control the activity of Cdc14: the Cdc14 early anaphase release network ( FEAR ) and the mitotic exit network ( MEN ) . The FEAR network serves to ensure anaphase spindle stability , spindle midzone assembly and proper rDNA segregation by transiently releasing Cdc14 from its inhibitor in the nucleolus during early anaphase ( reviewed in Rock and Amon [2009] ) . While not essential , this brief release of Cdc14 serves to ensure proper anaphase timing and primes the cell to efficiently exit from mitosis . In contrast , the MEN is responsible for sustained Cdc14 release during later stages of anaphase and is essential for cells to exit from mitosis ( Jaspersen et al . , 1998; Lee et al . , 2001; Stegmeier et al . , 2002; Visintin et al . , 1998; 1999 ) . The MEN is a GTPase signaling pathway whose constituents primarily localize to spindle pole bodies ( SPBs; yeast centrosomes ) . Regulation of the GTPase Tem1 is central to MEN control . When Tem1 is in its GTP-bound state , the MEN is active and cells will exit from mitosis ( Scarfone and Piatti , 2015 ) . Likewise when Tem1 is inactive , the MEN is off and cells will arrest in anaphase ( Geymonat et al . , 2002; Shirayama et al . , 1994 ) . Tem1 regulates a kinase cascade comprised of the PAK-like kinase Cdc15 and the protein kinase Dbf2 . Tem1 activates Cdc15 by recruiting it to the spindle pole body ( Rock and Amon , 2011; Visintin and Amon , 2001 ) . Cdc15 in turn recruits Dbf2 to spindle poles by creating a phospho-peptide binding domain on the SPB component and MEN scaffold Nud1 . In its phosphorylated state , Nud1 docks the Dbf2-activating subunit Mob1 ( Rock et al . , 2013 ) . Activated Dbf2-Mob1 together with Cdc5 then promote the sustained release of Cdc14 from the nucleolus through a largely uncharacterized mechanism ( Manzoni et al . , 2010; Mohl et al . , 2009 ) . Tem1 itself is controlled by two opposing factors , Bub2/Bfa1 and Lte1 . Bub2/Bfa1 functions as a GTPase-activating protein complex ( GAP ) for Tem1 and so inhibits the MEN ( Bloecher et al . , 2000; Geymonat et al . , 2002; Li , 1999; Shirayama et al . , 1994 ) . The GAP complex in turn is regulated by the protein kinase Kin4 . Kin4 localizes to the mother cell cortex as well as the mother cell-localized SPBs and functions to maintain GAP activity by preventing the inactivation of Bub2/Bfa1 by the Polo kinase Cdc5 ( D'Aquino et al . , 2005; Maekawa et al . , 2007; Pereira and Schiebel , 2005 ) . Lte1 localizes to the bud cell compartment and promotes exit from mitosis by preventing Kin4 localization to SPBs in the bud ( Bertazzi et al . , 2011; Falk et al . , 2011 ) . Lte1 displays homology with guanine nucleotide exchange factors ( GEFs ) ; however , whether Lte1 also functions as a GEF for Tem1 remains unknown . Spindle position regulates MEN activity . When the spindle is mispositioned , the MEN is inactive: Cdc14 is sequestered in the nucleolus and cells arrest in anaphase ( Bardin et al . , 2000 ) . This regulatory mechanism that prevents exit from mitosis in response to spindle misposition is called the spindle position checkpoint ( SPoC; [Muhua et al . , 1998] ) . Two models have been proposed to explain how spindle position regulates MEN activity . The 'zone model' proposes that the cell is partitioned into a MEN inhibitory zone in the mother cell compartment and a MEN activating zone in the bud ( Figure 1A ) ( Chan and Amon , 2010 ) . The MEN inhibitor Kin4 localizes to the mother cell , the MEN activator Lte1 to the bud ( Bardin et al . , 2000; D'Aquino et al . , 2005; Pereira et al . , 2000; Pereira and Schiebel , 2005 ) . In the event that anaphase spindle elongation occurs only in the mother cell , the spindle poles ( where Tem1 resides ) cannot escape the negative regulation of Kin4 and the MEN is kept inactive . Inhibition of Tem1 is only relieved once the spindle realigns along the mother-bud axis and one spindle pole exits the Kin4 inhibitory zone . The bud compartment promotes Tem1 activation through redundant mechanisms: 1 ) the bud is largely devoid of Kin4 and 2 ) the Kin4 inhibitor Lte1 prevents any small amount of Kin4 present in the bud from localizing to the daughter SPB . 10 . 7554/eLife . 14036 . 003Figure 1 . A system to induce spindle misposition . ( A ) Zone model of exit from mitosis . Yeast cells are partitioned into two zones: a MEN inhibitory zone in the mother cell compartment ( red ) and a MEN activating zone in the bud cell compartment ( green ) . If the spindle becomes misaligned in the inhibitory zone , MEN inhibitors such as Kin4 prevent Tem1 enrichment on SPBs thereby inhibiting exit from mitosis . It is only once one SPB escapes the MEN inhibitory zone and moves into the bud cell compartment that Tem1 can become enriched at the daughter-bound SPB and the cell can exit mitosis . Note that in this model , Tem1 is shown not to localize to SPBs in cells with mispositioned spindles . This is based on the observation that Tem1-13MYC does not localize to SPBs in cells with mispositioned spindles ( D’Aquino et al . , 2005 ) . ( B ) cMT - budneck model of exit from mitosis . If the spindle becomes misaligned in the mother compartment , cytoplasmic microtubules activate a checkpoint response through their interactions with factors at the bud neck . Once the spindle has realigned , the cytoplasmic microtubules are no longer in contact with the bud neck , the checkpoint signal is eliminated and cells exit from mitosis . ( C–D ) Wild type ( A33138 ) , kar9Δ ( A33729 ) and dyn1Δ ( A32922 ) cells harboring GFP-tagged α-tubulin were grown to mid-log in YEPD and arrested in G1 with 10 μg/mL of the α-factor pheromone at 25°C . The cultures were released into the cell cycle in YEPD and then loaded onto a Y04C CellASIC flow cell . Cells were imaged on the flow cell in synthetic complete pH 6 . 0 medium . ( C ) Quantification of the percent of anaphase cells which misposition their anaphase spindle . Anaphase was defined as any spindle measuring >2 μm . Aligned spindles were defined as those that entered anaphase with one spindle pole in the bud cell compartment . Mispositioned spindles were defined as those that entered anaphase with both spindle poles the mother cell compartment . ( D ) Time-lapse analysis of anaphase length . n =100 cells for each strain ( E–I ) osTIR1 ( A35699 ) and osTIR1 DYN-AID kar9Δ ( A35707 ) cells expressing GFP-tagged α-tubulin were grown in YEPD medium at 25°C and arrested in the G1 phase of the cell cycle with 10 μg/mL α-factor pheromone . Cells were released into the cell cycle in YEPD pH 6 . 0 medium and then monitored by live cell microscopy . Depletion of dyn1-AID was induced on a Cellasic flow cell with 100 μM auxin in SC pH 6 . 0 medium at 25°C . ( E ) Time-lapse analysis of anaphase length . Open squares indicate cells arrested in anaphase for more than 200 min . ( F ) Analysis of ploidy . Cells that were arrested and contained a misaligned spindle or cells that exited mitosis that contained an aligned spindle were categorized as 'euploid' . Cells that inappropriately exited mitosis and broke down the spindle in the mother cell compartment were categorized as 'multinucleate' . n=100 cells . ( G–I ) Montage of representative time-lapse images . The numbers at the top of the GFP images are time in minutes . DOI: http://dx . doi . org/10 . 7554/eLife . 14036 . 003 Support for the zone model comes from studies in which the localization of Kin4 and Lte1 have been switched . Targeting Lte1 to the mother cell leads to inappropriate mitotic exit in cells with misaligned spindles ( Bardin et al . , 2000; Bertazzi et al . , 2011; Castillon et al . , 2003; Geymonat et al . , 2009 ) . Targeting Kin4 to the bud and simultaneously inactivating its inhibitor , Lte1 , causes anaphase arrest even in cells with correctly positioned spindles ( Chan and Amon , 2010; Falk et al . , 2011 ) . A second model proposes that MEN activity is controlled by a microtubule-based checkpoint mechanism ( Figure 1B; henceforth the 'cMT - budneck model' ) ( Adames et al . , 2001; Moore et al . , 2009; Muhua et al . , 1998 ) . The model posits that stable contact between cytoplasmic microtubules and the bud-neck activates a checkpoint response that prevents cells from exiting mitosis . The hypothetical cMT checkpoint sensor would , according to this model , localize to the mother side of the septin ring ( Castillon et al . , 2003 ) . The model was proposed based on studies showing that cytoplasmic microtubule loss from the bud neck precedes anaphase spindle disassembly and exit from mitosis ( Adames et al . , 2001; Moore et al . , 2009 ) . Laser ablation of cytoplasmic microtubules interacting with the bud neck was further reported to trigger exit from mitosis ( Moore et al . , 2009 ) . Here we describe several experimental approaches aimed at distinguishing between the zone model and the cMT - budneck model . These analyses refute the cMT - budneck model and support the zone model . In the first approach we conducted live cell imaging to investigate the relationship between the presence of cMTs in the neck and exit from mitosis in cells with mispositioned spindles . As previously reported , we found that cMT loss from the bud neck does indeed precede exit from mitosis in cells that inappropriately breakdown their spindle in the mother cell compartment . However , our data show that loss of cMTs from the bud neck is not a cause but rather a consequence of exit from mitosis . We find , in cells wh exit from mitosis despite harboring a mispositioned spindle , that Cdc14 release from the nucleolus precedes rather than follows the disassembly of cytoplasmic microtubules and exit from mitosis . Second , we show that severing cytoplasmic microtubules does not lead to inappropriate exit from mitosis in cells with mispositioned spindles . Finally , we developed a method that allowed us to create cells containing two nuclei . We find that as long as one nucleus enters the bud during anaphase , cells will exit from mitosis , irrespective of whether the other nucleus is correctly or incorrectly positioned . Our data are inconsistent with a model where cMT-bud-neck interactions prevent exit from mitosis in cells with mispositioned spindles . Instead , they support the conclusion that spatial regulation of the MEN is controlled through the delivery of a MEN component bearing SPB into the bud . Inactivation of either Kar9 or Dyn1 causes a fraction of cells to transiently misposition their spindles ( Figure 1C ) . Such cells will then delay in anaphase until spindle position has been corrected ( Figure 1D ) ( Miller and Rose , 1998 ) . The relatively low penetrance and transient nature of the spindle positioning defect of kar9Δ and dyn1Δ cells has impeded the investigation of the consequences of spindle misposition on cell cycle progression . To overcome this limitation we developed a system to conditionally inactivate both the Kar9 and Dyn1 spindle positioning pathways . We depleted Dyn1 using the Indole-3-acetic acid ( IAA; auxin ) depletion system ( Nishimura et al . , 2009 ) . IAA is a naturally occurring plant hormone that promotes the degradation of proteins containing an AID degron sequence by targeting them for ubiquitinylation by the SCF-Tir1 ubiquitin ligase ( Dharmasiri et al . , 2005; Gray et al . , 2001; Kepinski and Leyser , 2005; Teale et al . , 2006 ) . We generated a strain carrying a DYN1-AID fusion and a deletion of KAR9 , henceforth the DYN1-AID kar9Δ strain . Live cell imaging showed that 92% of DYN1-AID kar9Δ cells initially mispositioned their spindles ( i . e . had spindles greater than 2 μm in length in the mother cell compartment ) upon IAA addition in contrast to 13% seen in the wild type controls . This finding indicated that the DYN1-AID kar9Δ system effectively inactivates spindle-positioning systems in the cell . To characterize the effects of spindle mispositioning on exit from mitosis we compared anaphase duration of cells with correctly aligned spindles to those with mispositioned spindles . Wild-type cells with correctly aligned spindles underwent anaphase within 19 . 2 ± 4 . 8 min ( Figure 1E and G ) . In contrast , DYN1-AID kar9Δ cells spent 65 . 9 ± 51 . 7 min in anaphase . This anaphase delay was highly variable . Most cells ( 85% ) eventually were able to pull the mispositioned spindle into the bud ( see Figure 1H for an example ) , which was followed by exit from mitosis . Only 6% of cells arrested with a mispositioned spindle for the duration of the movie analysis ( longer than 200 min , open squares in Figure 1E ) . The fact that the majority of DYN1-AID kar9Δ cells eventually managed to correctly align their spindles along the mother – bud axis suggests that cells harbor residual dynein activity , perhaps because the depletion is not complete . It is also possible that additional minor spindle positioning pathways exist in these cells ( Kirchenbauer and Liakopoulos , 2013; Segal et al . , 2002 ) . Although exit from mitosis was prevented in the majority of cells with mispositioned spindles , we observed inappropriate exit from mitosis in 9% of such cells , leading to the formation of anucleate and binucleate cells ( Figure 1F and I ) . This incomplete penetrance of the cell cycle arrest caused by spindle misposition has been observed previously ( Adames et al . , 2001; D'Aquino et al . , 2005; Pereira and Schiebel , 2005 ) . The reason why a small fraction of cells escapes the anaphase arrest caused by spindle misposition was , however , not understood . One hallmark of cells with mispositioned spindles is long cMTs that emanate from one or both SPBs through the bud neck and into the bud ( Figure 1H; 100 min time point ) . In the small fraction of cells that escape the anaphase arrest caused by spindle misposition , inappropriate mitotic exit is preceded by retraction of cytoplasmic microtubules from the bud neck ( Moore et al . , 2009 ) . Having established a tool to induce spindle misposition in many cells , we decided to reinvestigate this correlation . As reported previously , we found that cells with mispositioned spindles frequently display cMTs that contact the bud neck ( Figure 2 , cells 1–30; Video 1 ) . Also consistent with previous studies , we found that in cells that exit from mitosis despite harboring a mispositioned spindle , this cell cycle transition was preceded by the loss of cMT-bud neck interactions . Contact was lost approximately 2–8 min before spindle breakdown ( Figure 2 , cells 31–37; Video 2 ) . Additionally , we found a small number of cells that did not lose cMT-bud neck contact but entered the next cell cycle as assessed by budding ( Figure 2 , cells 38–40; Video 3 ) . These cells also did not completely breakdown their spindle and did not complete cytokinesis . Importantly , our analysis also revealed that inappropriate mitotic exit was not an obligatory consequence of loss of cMT – bud neck interactions . The majority of cells with mispositioned spindles lacked cMT-bud neck contacts for significant periods of time yet stayed arrested in anaphase ( Video 4; Figure 2 , cells 1–30 ) . We conclude that loss of cMT – bud neck interactions does not necessarily cause inappropriate mitotic exit in cells with mispositioned spindles . 10 . 7554/eLife . 14036 . 004Figure 2 . Analysis of cytoplasmic microtubules in the bud neck . Cells harboring osTIR1 DYN-AID kar9Δ and expressing GFP-labeled tubulin ( A35707 ) were grown and imaged as descried in Figure 1E–I . A table summarizing cMT-bud neck contact for cells that contained a mispositioned spindle for 60-min ( cells 1–30 ) or exited mitosis within that time frame ( cells 31–40 ) is shown . Each row shows the color-coded fate of one cell for the given time period , as well as whether it had a cMT in contact with the bud neck . cMT analysis was performed by assessing whether a cMT was present or absent from the bud neck . Cells in which the tip of a cMT interacted with the bud neck or where the cMT traversed the bud neck was categorized as 'cMT in bud neck' ( grey boxes ) Cells lacking any cMT in the bud neck are described as 'cMT absent from bud neck' ( black boxes ) . Movement of one spindle pole into the bud is described as 'spindle pole movement into bud' ( blue boxes ) . Inappropriate exit from anaphase was determined by the spindle morphology and is described as 'spindle breakdown' ( red boxes ) . A second category of inappropriate exit from mitosis was scored based on whether the cell rebudded without spindle collapse or cytokinesis ( yellow boxes ) . Due to the low frequency of inappropriate spindle breakdown in the mother compartment , this table shows all cells that inappropriately exit mitosis from 2 experiments ( cells 31–40 ) . The cells that remain euploid ( cells 1–30 ) are from experiment 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 14036 . 00410 . 7554/eLife . 14036 . 005Video 1 . A cell with a mispositioned spindle that has a cMT in the bud neck . osTIR1 DYN-AID kar9Δ cells expressing GFP-labeled α-tubulin ( A35707 ) were grown and imaged as described in Figure 1E-I . Depicted is a cell that mis-positions its spindle and displays cMTs that protrude into the bud . Time in minutes is show in the upper left-hand corner of the time-lapse image . A merged image of the DIC and GFP channels is shown on the left . The GFP channel alone is shown on the right . The DIC image was adjusted to maximize contrast . DOI: http://dx . doi . org/10 . 7554/eLife . 14036 . 00510 . 7554/eLife . 14036 . 006Video 2 . A cell with a mispositioned spindle that inappropriately exits mitosis in the mother cell compartment . osTIR1 DYN-AID kar9Δ cells expressing GFP-labeled α-tubulin ( A35707 ) were grown and imaged as described in Figure 1E-I . Depicted is a cell that mis-positions its spindle and then inappropriately exits mitosis in the mother cell . Time in minutes is show in the upper left-hand corner of the time-lapse image . A merged image of the DIC and GFP channels is shown on the left . The GFP channel alone is shown on the right . The DIC image was adjusted to maximize contrast . DOI: http://dx . doi . org/10 . 7554/eLife . 14036 . 00610 . 7554/eLife . 14036 . 007Video 3 . A cell with a mispositioned spindle that inappropriately exits mitosis but does not complete spindle disassembly or cytokinesis . osTIR1 DYN-AID kar9Δ cells expressing GFP-labeled α-tubulin ( A35707 ) were grown and imaged as described in Figure 1E-I . Depicted is a cell that mispositions its spindle and then inappropriately exits mitosis in the mother cell but does not complete cytokinesis or spindle disassembly . Time in minutes is show in the upper left-hand corner of the time-lapse image . A merged image of the DIC and GFP channels is shown on the left . The GFP channel alone is shown on the right . The DIC image was adjusted to maximize contrast . DOI: http://dx . doi . org/10 . 7554/eLife . 14036 . 00710 . 7554/eLife . 14036 . 008Video 4 . A cell with a mispositioned spindle that lacks cMT-bud neck interactions but arrests in anaphase . osTIR1 DYN-AID kar9Δ cells expressing GFP-labeled α-tubulin ( A35707 ) were grown and imaged as described in Figure 1E-I . Depicted is a cell that arrests in late anaphase with a mispositioned spindle . This cell lacks cMTs in the bud neck for a substantial time period in anaphase . Time in minutes is show in the upper left-hand corner of the time- lapse image . A merged image of the DIC and GFP channels is shown on the left . The GFP channel alone is shown on the right . The DIC image was adjusted to maximize contrast . DOI: http://dx . doi . org/10 . 7554/eLife . 14036 . 008 A previous study reported that eliminating cMT-bud neck interactions by ablating GFP-labeled microtubules causes exit from mitosis in cells with mispositioned spindles ( Moore et al . , 2009 ) . We conducted a similar analysis and found this not to be the case . We used a laser to ablate microtubules in cells either lacking both the KAR9 and DYN1 spindle positioning pathways or just the DYN1 positioning pathway . We ablated the cMT that interacted with the bud neck in 15 DYN1-AID kar9Δ cells , but exit from mitosis was not observed in the mother cell compartment within the time that we monitored cells post severing ( 69 min; Figure 3C ) . We also assessed exit from mitosis following laser ablation in these cells using a marker for cytokinesis . None of the 10 cells with mispositioned spindles in which we ablated the bud-neck interacting cMT exited mitosis as judged by loss of septin Cdc3 from the bud neck ( Figure 3A and C ) . To ensure that exposure of cells to the laser pulse did not cause cell cycle arrest , we targeted the cytoplasm of cells with correctly positioned spindles with the laser . In the 20 cells treated in this manner , exit from mitosis occurred within 2–24 min following application of the laser pulse ( Figure 3B; C , aligned category ) . Furthermore , when laser ablated cells with mispositioned spindles managed to realign their spindle , they also exited mitosis ( Figure 3C , spindle breakdown in the bud category ) . Lastly , we ablated cMTs in dyn1Δ cells . Exit from mitosis did not occur for the duration of the analysis ( 69 min ) in cells with mispositioned spindles in which cMTs were ablated . One cell succeeded in positioning its spindle correctly along the mother – bud axis and promptly exited mitosis thereafter . In 2 cells exit from mitosis followed ablation of cMTs . One cell exited mitosis 8 min post ablation and the other cell exited mitosis 69 min post ablation . The latter instance seems unlikely to be the consequence of loss of cMTs from the budneck because the cell initially lacked cMTs in the budneck for at least 7 min but exited mitosis much later ( 69 min post-ablation ) . The one cell that exited mitosis shortly after cMT ablation is well in line with the fraction of wild-type cells that exit from mitosis despite harboring a misaligned spindle ( Figure 1F ) . In summary , our results show that ablation of cMTs does not promote exit from mitosis in the vast majority ( 33/35 ) of cells analyzed ( Figure 3C , mispositioned spindles category ) . We conclude that although cMT retraction frequently precedes inappropriate exit from mitosis in cells with misaligned spindles , it is not the cause of exit from mitosis in these cells . 10 . 7554/eLife . 14036 . 009Figure 3 . cMT laser ablation does not promote exit from mitosis in cells with mispositioned spindles . ( A-B ) Cells harboring the osTIR1 DYN-AID kar9Δ constructs that also expressed GFP-tagged α-tubulin , mCherry-tagged Cdc3 and an NLS-mCherry ( A35143 ) were grown overnight to mid-log in YEPD at 25°C . Cells were then resuspended in synthetic complete medium supplemented with 100 μM IAA and incubated for 2–3 hr at 25°C . The cells were prepared on an agar pad for live cell microscopy . Two pre-ablation images were taken ( only one is shown ) before the cMT was cut . Post ablation cells were monitored for 9 min at 1-min intervals for cMTs and then 1 hr at 15-min intervals to follow cell cycle progression . The arrowheads indicate the approximate laser targeting site . ( A ) A montage of a cell with a mispositioned spindle where cMT bud neck interactions were disrupted due to microtubule severing . The DIC channel is optimized to enhance contrast . ( B ) A montage of a control cell with an aligned spindle where the laser was targeted to the cytoplasm . The DIC channel is optimized to enhance contrast . ( C ) Table summarizing cell cycle stage of aligned and misaligned spindles 69 min post ablation . The culturing conditions for cells in the first ( A35143 ) and second rows ( A34832: the same as A35143 but lacking Cdc3-mCherry ) of the table are the same as described above . In the third row , cells lacking DYN1 and expressing GFP-tagged α-tubulin and NLS-mCherry ( A34722 ) were grown overnight at 25°C to mid-log and then shifted to 16°C for 2–5 hr to enrich for cells with mispositioned spindles . The cells were then mounted on an agar pad for live cell microscopy . One set of pre-ablation images was taken before the cMT was cut . Post ablation cells were monitored as described above . DOI: http://dx . doi . org/10 . 7554/eLife . 14036 . 009 If loss of cMT-bud neck interactions does not induce inappropriate exit from mitosis in cells with misaligned spindles , what does ? To begin to address this question we asked whether inappropriate exit from mitosis in cells with mispositioned spindles relied on the same regulatory pathways that promote exit from mitosis in cells with correctly positioned spindles . Cdc14 is the key trigger of exit from mitosis ( reviewed in Stegmeier and Amon [2004] ) . Cdc14 release from the nucleolus during anaphase activates the phosphatase to trigger exit from mitosis . We used Cdc14 localization as the criterion to determine whether cMT retraction from the bud neck occurred before or after exit from mitosis . To examine Cdc14 localization we used a Cdc14-tdTomato fusion . This allele is slightly hypermorphic ( it causes inappropriate exit from mitosis in a small fraction of cells with mispositioned spindles after a substantial anaphase delay , Figure 4—figure supplement 1A ) but nevertheless accurately reflects the changes in Cdc14 localization during the cell cycle ( Figure 4—figure supplement 1B ) . Live cell imaging showed that Cdc14 release from the nucleolus preceded both , cMT retraction from the bud neck and mitotic spindle breakdown in cells that exited mitosis despite harboring a misaligned spindle . In 78 . 22 ± 3 . 0% of cells ( n ≥ 37 cells per biological replicate . 3 replicates were performed ) , Cdc14 release occurred approximately 5 min prior to loss of cMT – bud neck interactions and approximately 5–15 min prior to mitotic spindle breakdown ( Figure 4A and Figure 4—figure supplement 2 ) . The decrease in nucleolar Cdc14 signal intensity that occurred shortly prior to spindle breakdown was not caused by changes in nucleolar morphology that take place during anaphase . Signal intensity of Cdc14's nucleolar anchor Cfi1/Net1 did not change during anaphase ( Figure 4—figure supplements 3 and 4 ) . Instead , it appears that Cdc14 was released from the nucleolus . In the majority of cells ( 73 . 85 ± 4 . 4% ) Cdc14 release from the nucleolus occurred during anaphase ( Figure 4A and Figure 4—figure supplement 2A ) . However in a small fraction of these cells , 4 . 4 ± 4 . 1% , Cdc14 appeared to be fully released already at the metaphase to anaphase transition ( Figure 4C ) . 10 . 7554/eLife . 14036 . 010Figure 4 . Cdc14 Release from the nucleolus precedes cMT retraction in cells that inappropriately breakdown their anaphase spindles in the mother cell compartment . ( A–D ) A diploid strain ( A37463 ) homozygous for osTIR1 dyn1-AID kar9Δ and heterozygous for GFP-Tub1 and Cdc14-tdTomato was grown to midlog in Synthetic Complete medium . Cycling cells were imaged on a flow cell in Synthetic Complete medium supplemented with 100 μM IAA . ( A ) Representative images are shown for the anaphase release of Cdc14-tdTomato with respect to GFP-Tub1 cMT retraction . ( B ) The coefficient of variation ( the standard deviation divided by the mean ) was measured for Cdc14 pixel intensity for the cell pictured in the Figure 4A montage . Time ( in minutes ) is displayed on the X-axis and the zero time point reflects anaphase onset . ( C ) Representative images are shown for a cell with a mispositioned spindle in which complete Cdc14-td Tomato release from the nucleolus was observed at anaphase onset . cMTs are also shown for that cell during the same time interval . ( D ) The coefficient of variation ( the standard deviation divided by the mean ) was measured for Cdc14 pixel intensity for the cell pictured in the Figure 4C montage . Time ( in minutes ) is displayed on the X-axis . The zero time point reflects anaphase onset . See also: Figure 4—figure supplement 1–4 . DOI: http://dx . doi . org/10 . 7554/eLife . 14036 . 01010 . 7554/eLife . 14036 . 011Figure 4—figure supplement 1 . Analysis of the CDC14-tdTomato allele . ( A ) Cells homozygous for osTIR1 dyn1-AID kar9Δ and heterozygous for either GFP-Tub1 ( A38232 ) or GFP-Tub1 and Cdc14-td Tomato ( A37464 ) were grown overnight in YPD to mid-log and then loaded onto a flow cell and imaged in synthetic complete medium containing 100 μM auxin . Degree of ploidy was analyzed 3 hr after the addition of auxin . ( B ) Cells harboring Cdc14-3HA ( A1411 ) or Cdc14-tdTomato and GFP-Tub1 ( A38243 ) were grown overnight to mid-log in YEPD and then arrested in G1 with 10 μg/mL of the α-factor pheromone . Cells were then released into the cell cycle , fixed and Cdc14 localization and mitotic spindle length was analyzed . DOI: http://dx . doi . org/10 . 7554/eLife . 14036 . 01110 . 7554/eLife . 14036 . 012Figure 4—figure supplement 2 . Analysis of Cdc14 Release in cells with mis-positioned spindles . The diploid strain ( A37463 ) homozygous for osTIR1 dyn1-AID kar9Δ and heterozygous for GFP-Tub1 and Cdc14-tdTomato was grown to midlog in synthetic complete medium . Cycling cells were imaged as in Figure 4 . ( A ) Representative images are shown for Cdc14-tdTomato release and GFP-Tub1 cMT retraction in anaphase . ( B ) The coefficient of variation ( the standard deviation divided by the mean ) was measured for Cdc14 intensity for the cell pictured in ( A ) . Time ( in minutes ) is displayed on the X-axis and the zero time point reflects anaphase onset . DOI: http://dx . doi . org/10 . 7554/eLife . 14036 . 01210 . 7554/eLife . 14036 . 013Figure 4—figure supplement 3 . Montage of Cdc14 Release in cells with mis-positioned spindles containing the nucleolar marker Cfi1 . The diploid strain ( A38233 ) homozygous for osTIR1 dyn1-AID and kar9Δ and heterozygous for GFP-Tub1 , Cfi1-GFP and Cdc14-tdTomato was grown overnight in synthetic complete medium and then imaged as in Figure 4 . Depicted are time-lapse images showing GFP-Tub1 cMT retraction and Cdc14-tdTomato release with respect to Cfi1-GFP . The red-boxed time points highlight the time frame when Cdc14-tdTomato intensity decreases in the nucleolus yet the cell still harbors a cMT in contact with the bud neck ( arrowheads ) . Note that Cfi1-GFP intensity remains largely unchanged throughout the time course . Time ( in minutes ) is shown in the upper hand corner of each DIC image . DOI: http://dx . doi . org/10 . 7554/eLife . 14036 . 01310 . 7554/eLife . 14036 . 014Figure 4—figure supplement 4 . Montage of Cdc14 Release in cells with mis-positioned spindles containing the nucleolar marker Cfi1 . The diploid strain ( A38233 ) homozygous for osTIR1 dyn1-AID and kar9Δ and heterozygous for GFP-Tub1 , Cfi1-GFP and Cdc14-tdTomato was grown overnight in synthetic complete medium and then imaged as in Figure 4 . Depicted are time-lapse images showing GFP-Tub1 cMT retraction and Cdc14-tdTomato release with respect to Cfi1-GFP . The red-boxed time points highlight the time frame when Cdc14-tdTomato intensity decreases in the nucleolus yet the cell still harbors a cMT in contact with the bud neck ( arrowheads ) . Note that Cfi1-GFP intensity remains largely unchanged throughout the time course . Time ( in minutes ) is shown in the upper hand corner of each DIC image . DOI: http://dx . doi . org/10 . 7554/eLife . 14036 . 014 To more precisely analyze when Cdc14 was released from the nucleolus relative to cMT retraction , we calculated the coefficient of variation ( CV; standard deviation/mean ) of Cdc14-tdTomato pixel intensity in the whole cell over time ( Lu and Cross , 2010 ) . As Cdc14 is released from the nucleolus and spreads throughout the nucleus and later the cytoplasm , the standard deviation of pixel intensities will decrease as cells progress through anaphase . We should note that despite intense efforts , we were not able to normalize changes in Cdc14-tdTomato CV to that of a protein that localizes to the nucleolus in a constitutive manner . GFP-tagged nucleolar markers overlapped with the Tubulin-GFP signal , a construct that was necessary to determine when cMTs retract . Tags other than GFP , such as BFP , were too dim to detect by live cell microscopy . Despite this limitation , it was nevertheless clear that the coefficient of variation of Cdc14-td Tomato pixel intensity decreased before cMT retraction from the bud neck ( Figure 4B , D , Figure 4—figure supplement 2B ) . Not all cells showed release of Cdc14 from the nucleolus prior to cMT retraction from the bud neck . In 10 . 8 ± 4 . 9% of cells Cdc14 release from the nucleolus occurred concomitantly with cMT retraction . In the remaining cells ( 11 . 0 ± 2 . 2% ) that did not release Cdc14 prior to cMT retraction , Cdc14 release from the nucleolus was not detected prior to anaphase spindle breakdown . This is most likely because the fraction of Cdc14 that was released from the nucleolus was too small to detect by imaging . Importantly , we never observed that Cdc14 release occurred after cMT retraction from the bud neck . Consistent with the idea that Cdc14 triggers inappropriate exit from mitosis in cells with mispositioned spindles we find that depletion of Cdc14 suppressed exit from mitosis in the rare wild-type cells that escape the anaphase arrest when their spindles are mispositioned ( Figure 5A ) . We conclude that inappropriate mitotic exit in cells with mispositioned spindles is not due to loss of cMT – bud neck interactions . Rather , cMT retraction is a consequence of Cdc14 release from the nucleolus in these cells . We further propose that the reason why loss of cMT – bud neck interactions always precedes mitotic spindle breakdown in cells with mispositioned spindles that exit from mitosis is because disassembly of a single cMT occurs more quickly than disassembly of the mitotic spindle . 10 . 7554/eLife . 14036 . 015Figure 5 . Inhibition of Cdc14 , the MEN and the FEAR Network prevents Spindle Breakdown in the mother cell compartment . ( A ) osTIR1 dyn1-AID kar9Δ ( A35707 ) and osTIR1 dyn1-AID kar9Δ cdc14-1-AID ( A37895 ) cells harboring GFP-tagged α-tubulin were grown as described in Figure 1E–I . Cells were monitored by live cell microscopy and scored for inappropriate spindle breakdown in the mother cell compartment . n=100 for osTIR1 dyn1-AID kar9Δ and 225 for osTIR1 dyn1-AID kar9Δ cdc14-1-AID . ( B ) osTIR1 dyn1-AID kar9Δ ( A35707 ) , osTIR1 dyn1-AID kar9Δ cdc15-AS1 ( A36264 ) expressing GFP-tagged α-tubulin , were analyzed as in Figure 5A but with the addition of 20 μM NAPP1 to the medium . Cells were imaged in a Lab-Tek II chamber . n=100 for each strain . ( C ) osTIR1 dyn1-AID kar9Δ ( A35707 ) , osTIR1 dyn1-AID kar9Δ spo12Δ ( A35700 ) , osTIR1 dyn1-AID kar9Δ slk19Δ ( A36028 ) expressing GFP-tagged α-tubulin , were analyzed as in Figure 5A . n ≥ 284 for each strain . ( D ) Wild type ( A37753 ) or spo12Δ ( A37610 ) diploid strains homozygous for osTIR1 dyn1-AID kar9Δ and GFP-Tub1 and heterozygous for Cdc14-tdTomato were grown to mid log in synthetic complete medium . Cells with mispositioned spindles that had two distinct nucleoli were scored based on whether Cdc14-tdTomato was released from the nucleolus in late anaphase . Cells that did not exit mitosis in the mother cell compartment were monitored for 60 min . n≥22 cells . ( E ) cMT analysis was performed on osTIR1 dyn1-AID kar9Δ spo12Δ ( A35700 ) as described in Figure 2 . Each row shows the color-coded fate of one cell for the given time period , as well whether it had a cMT in contact with the bud neck . cMT analysis was performed by assessing whether a cMT was present or absent from the bud neck . Cells with a cMT end that was in the bud neck or in the bud was categorized as 'cMT in bud neck' ( grey boxes ) Cells lacking any cMT in the bud neck are described as 'cMT absent from bud neck' ( black boxes ) . Movement of one spindle pole into the bud is represented by the 'spindle pole movement into bud category ( blue boxes ) . Note: Cell #25 transiently moves a spindle pole into the bud cell compartment but does not exit from mitosis . This is most likely due the inefficient activation of the MEN in cells lacking the FEAR network . DOI: http://dx . doi . org/10 . 7554/eLife . 14036 . 015 Two signaling pathways control Cdc14 release from the nucleolus . The FEAR network promotes the transient release of the phosphatase from the nucleolus during early anaphase . The MEN is needed for the sustained release of the phosphatase during later stages of anaphase ( Stegmeier and Amon , 2004 ) . Our data showing that Cdc14 is inappropriately released in some cells with mispositioned spindles led us to investigate which pathway regulating Cdc14 was responsible for promoting inappropriate Cdc14 release in these cells . Not surprisingly , inhibition of the MEN , a pathway which is essential for exit from mitosis in all cells , suppressed the inappropriate mitotic exit that is observed in the 11% of cells that exit from mitosis when their spindle is mispositioned ( Figure 5B ) . We were , however , surprised to find that inactivation of the FEAR network , either by deleting SPO12 or SLK19 also prevented the inappropriate exit from mitosis in cells with mispositioned spindles ( Figure 5C ) ( Scarfone et al . , 2015 ) . Consistent with the idea that cMT – bud neck interactions are not regulating mitotic exit , we found that deleting SPO12 largely did not affect these interactions ( compare Figures 2 and 5E ) despite completely suppressing inappropriate mitotic exit in cells with mispositioned spindles . Finally , in cells with mispositioned spindles , we found that the release of Cdc14 from the nucleolus was prevented in cells lacking SPO12 ( Figure 5D ) . Together , our data lead to the following two conclusions . First , cMT – bud neck interactions are not responsible for preventing Cdc14 activation and exit from mitosis in response to mispositioned spindles . Instead , activation of Cdc14 causes cMT retraction from the bud neck and exit from mitosis in cells with mispositioned spindles that inappropriately exit from mitosis . Second , Cdc14 activation in the cells that exit from mitosis despite harboring a mispositioned spindle depends on the FEAR network . This observation raises the interesting possibility that it is high FEAR network activity that causes bypass of the anaphase arrest triggered by spindle misposition in some cells . A central tenet of the cMT - budneck model is the prediction that as long as a spindle is mispositioned , mitotic exit is inhibited . The 'zone model' or any other model that posits a mitotic exit-activating signal in the bud predicts the opposite . As long as a spindle is correctly positioned along the mother – bud axis exit from mitosis will occur , even if the cell also harbors a spindle that is mispositioned . To determine whether a MEN inhibitory signal caused by a misaligned spindle prevents MEN activation and hence exit from mitosis or whether a correctly aligned spindle activates the MEN and hence triggers mitotic exit , we generated cells with two nuclei ( henceforth , heterokaryons ) . When these cells undergo anaphase two main classes of cells are observed ( Figure 6A , B ) :10 . 7554/eLife . 14036 . 016Figure 6 . Analysis of Exit from Mitosis in prm3Δ heterokaryons . ( A ) Cartoon of the prm3Δ heterokaryon system showing the two main classes of heterokaryons obtained . ( B–F ) Heterokaryons were obtained by mating cells lacking PRM3 ( see Materials and methods for details ) . Briefly , G1 cells isolated by centrifugal , elutriation were mated at 30° for 2 hr and then loaded onto a Y04D CellASIC flow cell for imaging . Synthetic complete pH 6 . 0 medium was used in the flow cell during imaging and was supplemented with 100 μM IAA to induce spindle mispositioning . ( B ) Montages of binucleate zygotes created by mating cells lacking PRM3 . Binucleate diploids are homozygous for prm3Δ and osTIR1 and heterozygous for DYN-AID , kar9Δ , mCherry-labeled Cdc3 and GFP-labeled α-tubulin ( A37892 x A35570 ) . The DIC channel was adjusted to maximize contrast . ( C–D ) Analysis of anaphase kinetics of cells described in Figure 6B . Class 1: n= 16 . Class II: n= 21 ( E ) Binucleate diploids that were homozygous for prm3Δ , osTIR1 , DYN-AID , kar9Δ and heterozygous for mCherry-labeled Cdc3 and GFP-labeled α-tubulin ( A35570 x A35571 ) were analyzed for anaphase duration . Black diamonds indicated permanently arrested cells ( permanently arrested ≥320 min ) n=51 cells . ( F ) Montage of cells from Figure 6E . The DIC channel was adjusted to maximize contrast . DOI: http://dx . doi . org/10 . 7554/eLife . 14036 . 016 Class I cells: both spindles correctly align along the mother ( Figure 6A , B - top panel ) . Class II cells: one spindle aligns whereas the other one is misaligned ( Figure 6A , B - bottom panel ) . We generated heterokaryons using two different methods . In the first method , we created cells with two nuclei by mating two cells that lacked the nuclear fusion gene PRM3 ( Figure 6A ) ( Heiman and Walter , 2000; Shen et al . , 2009 ) . In cells in which both spindles were correctly aligned along the mother – bud axis ( Class I cells ) both spindles entered the bud in quick sequence during anaphase and exit from mitosis ( as judged by Cdc3 loss from the bud neck ) occurred promptly thereafter ( Figure 6C ) . Class II cells also exited mitosis even though only one spindle entered the bud during anaphase . Average anaphase duration for these cells was 15 . 7 ± 2 . 3 min , which was comparable to the average anaphase duration in cells in which both spindles were correctly positioned ( 16 . 6 ± 3 . 7 min; compare Figure 6C , D ) . Furthermore the time of entry of one spindle into the bud until exit from mitosis was the same in cells in which both spindles aligned correctly and cells in which one spindle was correctly aligned and the other was misaligned ( compare Figure 6C , D column 'bud entry to exit from mitosis' ) . Importantly , we were also able to obtain many heterokaryons where both anaphase spindles were mispositioned for prolonged periods of time ( henceforth 'Class II delayed' ) . These cells were severely delayed in anaphase ( Figure 6E , F ) . In the vast majority of these cells it was only once one spindle moved from the mother compartment into the bud that cells promptly exited mitosis ( 42/51 cells ) . Of the remaining cells , six cells ( 11 . 76% ) permanently arrested in anaphase with both spindles in the mother cell compartment and two inappropriately exited mitosis with both spindles in the mother cell ( 3 . 92% ) . Strikingly , we also noticed that in a very small fraction of cells ( 1/51 ) , exit from mitosis occurred even when the mispositioned spindle in the mother cell had not yet initiated anaphase ( Figure 7 ) . We conclude that movement of one spindle pole into the bud triggers exit from mitosis . 10 . 7554/eLife . 14036 . 017Figure 7 . Exit from Mitosis in a prm3Δ heterokaryon with one aligned anaphase spindle and one metaphase spindle . Montage of binucleate zygotes created by mating homozygous for prm3Δ , osTIR1 , DYN-AID , kar9Δ and heterozygous for mCherry-labeled Cdc3 and GFP-labeled α-tubulin ( A35570 x A35571 ) . The montage depicts a zygote with one aligned anaphase spindle and a second spindle in metaphase in the mother compartment . Both spindles exit mitosis at the same time with the metaphase spindle never going through anaphase . The DIC channel was adjusted to maximize contrast . DOI: http://dx . doi . org/10 . 7554/eLife . 14036 . 017 The second method to generate heterokaryons took advantage of the fact that cells undergoing meiosis can be returned to vegetative growth ( Simchen , 2009 ) . Diploid yeast cells sporulate in response to nutrient deprivation . Budding is suppressed and cells progress through premeiotic S-phase . When glucose-containing medium is supplied to cells that lack the CDK inhibitory kinase Swe1 following premeiotic S phase ( in pachytene of meiotic prophase I ) these cells will return to vegetative growth and undergo mitosis producing cells with two nuclei ( Figure 8A ) ( Tsuchiya and Lacefield , 2013 ) . We analyzed the mitotic cell cycle after the formation of a binucleate cell . In this method of generating heterokaryons , a third class of cells was observed: one spindle is pulled into the bud generating cells with a misaligned spindle in the mother and bud cell compartments ( Figure 8A ) . As in the heterokaryons generated by mating , cells in which both spindles were correctly aligned along the mother – bud axis ( Class I cells ) exited from mitosis as judged by the average anaphase spindle breakdown 15 . 8 ± 5 . 5 min after the two spindles had entered the bud ( Figure 8B ) . Class II cells also exited mitosis even though only one spindle entered the bud during anaphase . Anaphase duration was similar for both spindles and both spindles broke down soon after one spindle entered the bud ( Figure 8C ) . 10 . 7554/eLife . 14036 . 018Figure 8 . Analysis of Exit from Mitosis in swe1Δ heterokaryons . ( A ) Cartoon of swe1Δ return to growth heterokaryon system and depictions of cell type classes that were analyzed . Briefly , diploid cells lacking SWE1 , also harboring the meiotic prophase marker Zip1 , tagged with GFP as well as GFP-tagged Tub1 ( LY1043 ) , were induced to enter meiotic prophase through nutrient starvation . Upon entry into meiotic prophase ( as judged by the presence of Zip1-GFP positive cells ) , cells were returned to glucose-containing complete medium in microfluidic chambers and thus induced to grow mitotically . These cells were monitored by live cell microscopy . ( B ) Anaphase kinetics of Class I cells ( depicted in Figure 8A ) . Anaphase duration is classified as the time the cell spends with a spindle >2 μm to spindle breakdown . 'Bud entry to exit from mitosis' is defined as the time from when at least one spindle pole is in the bud to anaphase spindle disassembly . ( C ) Anaphase kinetics of Class II cells ( depicted in Figure 8A ) . Anaphase duration and bud entry to exit from mitosis are define as in Figure 8B . ( D ) Anaphase kinetics of Class III cells ( depicted in Figure 8A ) . Anaphase duration and bud entry to exit from mitosis are define as in Figure 8B . n=50 cells for each class . DOI: http://dx . doi . org/10 . 7554/eLife . 14036 . 018 Spindle disassembly of the nucleus in the mother cell also occurred concomitantly with spindle disassembly in the nucleus in the bud of class III cells ( Figure 8D ) . This result indicates that exit from mitosis is not triggered by a correctly positioned spindle but rather a spindle that is in the bud , as exit from mitosis occurred in class III cells even though the spindle in the bud was mispositioned . It is important to note however , that a spindle simply being in the bud is not sufficient to bring about exit from mitosis . Exit from mitosis only occurred after the bud-localized spindle had undergone anaphase . This observation is consistent with previous findings showing that anaphase entry is required for MEN activation and exit from mitosis ( Rock and Amon , 2011 ) ( data not shown ) . In summary , our heterokaryon analyses do not support the hypothesis that a dominant inhibitory signal originating from a mispositioned spindle prevents MEN activation and exit from mitosis . Instead , our data show that movement of the spindle into the bud as occurs during a cell cycle with a correctly positioned spindle activates the MEN and exit from mitosis . In budding yeast , the site of cytokinesis is determined long before cells undergo mitosis . Division by budding also means that the connection between mother cell and bud is small and the nucleus and other organelles must be squeezed through the bud neck to be accurately partitioned . Therefore division by budding not only requires sophisticated mechanisms to position the nucleus along the mother – bud axis , it also requires mechanisms to prevent cells from exiting mitosis and undergoing cytokinesis until the nuclei are partitioned between the mother and bud cell compartments . In 1995 , Yeh et al . demonstrated the existence of such a mechanism . They showed that cells with mispositioned spindles arrest in late anaphase and fail to exit from mitosis . Subsequently , Muhua et al . ( 1998 ) termed this regulatory mechanism the spindle position checkpoint ( SPoC ) . Ensuing studies showed that exit from mitosis is prevented in cells with misaligned spindles through the inhibition of the Mitotic Exit Network , the GTPase signaling cascade that promotes anaphase spindle disassembly , chromosome decondensation and cytokinesis by activating Cdc14 ( D'Aquino et al . , 2005; Pereira and Schiebel , 2005 ) . Two models have been proposed to explain how MEN activity is inhibited in response to spindle misposition: the cMT - budneck model and the zone model . The former posits that a MEN inhibitory activity is generated by a misaligned spindle , the latter that a MEN activating activity is produced by a correctly aligned spindle ( Adames et al . , 2001; Bardin et al . , 2000; Chan and Amon , 2010; Moore et al . , 2009 ) . In this paper we took advantage of a new inducible system to study mispositioned spindles to distinguish between these two models . Our data support the conclusion that a correctly aligned spindle promotes exit from mitosis . Our data , together with previous studies , further indicate that it is the movement of the MEN component-carrying SPB into the bud that signals exit from mitosis . Cytoplasmic microtubules continuously interact with the bud neck during spindle positioning prior to anaphase . However , once the nucleus traverses the bud neck during anaphase these interactions are lost . In cells that misposition their spindles and undergo anaphase in the mother cell , cMT – bud neck interactions persist . The proposal that it is these cMT – bud neck interactions that prevent exit from mitosis in cells with misaligned spindles stems from the analysis of cells that exit mitosis despite harboring a misaligned spindle ( Adames et al . , 2001; D'Aquino et al . , 2005; Pereira and Schiebel , 2005 ) . Previous work by Adames and Cooper demonstrated that cMT retraction from the bud neck precedes mitotic spindle breakdown in cells that exit mitosis with a mispositioned spindle ( Adames et al . , 2001 ) . This correlation led them to propose that cMT – bud neck interactions emit an inhibitory signal that prevents exit from mitosis . This hypothesis was supported by the observations that 1 ) elimination of cMT – bud neck interactions by microtubule ablation or 2 ) loss of cMTs brought about by the cold sensitive β-tubulin allele ( tub2-401 ) increased the frequency with which cells with mispositioned spindles inappropriately exit from mitosis ( Adames et al . , 2001; Moore et al . , 2009 ) . We also observed this striking correlation between cMT retraction from the bud neck and mitotic spindle breakdown , but several additional analyses demonstrate that this correlation does not indicate causality . First , retraction of cMTs occurs frequently and often for extended periods of time also in cells with-mispositioned spindles that do not inappropriately exit mitosis in the mother cell compartment . Second , in our cMT ablation studies we did not observe exit from mitosis following the loss of cMT – bud neck interactions . We cannot explain why our ablation results differ from those of Moore et al . ( 2009 ) but we note that mitotic exit that followed ablation of cMTs took a long time ( approximately 18 min ) in this previous study and regrowth of cMTs into the bud neck was also observed during the time it took until cells exited mitosis . Third , not all mutants that lack cytoplasmic microtubules exhibit an increased frequency in inappropriate exit from mitosis when the spindle is mispositioned . Gryaznova et al . ( accompanying paper ) found that cells lacking SPC72 arrest in anaphase when their spindles are mispositioned despite the absence of cMTs . The most conclusive demonstration that loss of cMT bud neck interactions does not trigger exit from mitosis was the analysis of Cdc14 localization . It clearly showed that cMT retraction from the bud neck did not precede Cdc14 release from the nucleolus but was a consequence thereof . We propose that the reason why cMTs invariably disassemble prior to the mitotic spindle in such cells is that disassembly of a single cMT upon mitotic CDK inactivation takes less time than the disassembly of a mitotic spindle that is composed of many microtubules . An inherently higher instability of cMTs compared to spindle microtubules could of course also explain this difference in disassembly timing . Together , our studies disfavor a mitotic exit inhibitory function of cMT – bud neck interactions . MEN signaling takes place at SPBs ( Maekawa et al . , 2007; Valerio-Santiago and Monje-Casas , 2011; Visintin and Amon , 2001 ) . The GTPase Tem1 and Polo kinase recruit the MEN kinase Cdc15 to SPBs where it is activated by an unknown mechanism ( Rock and Amon , 2011 ) . Regulators of the GTPase are strategically placed in the cell . Kin4 , the GTPase inhibitor localizes to the mother cell , the Kin4 inhibitor and hence MEN activator Lte1 localizes to the bud ( Bardin et al . , 2000; D'Aquino et al . , 2005; Pereira and Schiebel , 2005 ) . These localization patterns led us to propose that spindle position controls exit from mitosis through the establishment of a MEN activating compartment in the bud , a MEN inhibitory compartment in the mother cell and a sensor , the MEN component bearing SPB that shuttles between them . When both of the spindle pole bodies are in the MEN inhibitory mother cell compartment , the cell cannot exit from mitosis and arrests in anaphase . It is only once one MEN component-bearing spindle pole body moves into the mitotic exit-activating zone in the bud does the MEN become active and exit from mitosis occurs . It is important to emphasize that the zone model takes into account that not only are there MEN inhibiting factors in the mother cell compartment , but that there are also factors in the bud that promote exit from mitosis . Evidence that cells have both a negative zone in the mother cell compartment and a positive zone in the bud comes from the analysis of cells in which the MEN activating and inhibitory zones were switched . When the Kin4 inhibitor Lte1 is targeted to the mother cell , cells with mispositioned spindles inappropriately exit from mitosis ( Bardin et al . , 2000; Bertazzi et al . , 2011 ) . When Kin4 is targeted to the bud and its inhibitor Lte1 is inactivated , cells with correctly positioned spindles cannot exit from mitosis and arrest in anaphase ( Chan and Amon , 2010; Falk et al . , 2011 ) . The analysis of heterokaryons presented here also supports the zone model . Irrespective of whether or not a cell harbors a mispositioned spindle , exit from mitosis occurs once one anaphase spindle enters the bud . Our analysis of heterokaryons in which one of the two spindles gets pulled into the bud in its entirety further shows that it is the presence of a spindle pole in bud and not a correctly positioned spindle that leads to MEN activation . In cells that harbor one nucleus in the bud and one in the mother cell , both spindles are mispositioned , yet exit from mitosis occurs once the spindle in the bud has initiated anaphase . This finding further demonstrates that two signals are necessary for MEN activation in anaphase: ( 1 ) a spatial signal – the movement of a MEN bearing SPB into the bud and ( 2 ) a temporal signal that indicates that anaphase chromosome segregation has occurred ( Bardin et al . , 2000; Chan and Amon , 2010; Manzoni et al . , 2010; Rock and Amon , 2011 ) . Dissecting the molecular details of how these two signals interact to control Tem1 activity will be a critical next step in understanding how the MEN integrates spatial and temporal cues to control exit from mitosis . It has long been known that a small fraction of cells exit from mitosis despite the presence of a mispositioned spindle ( Adames et al . , 2001; D'Aquino et al . , 2005; Pereira and Schiebel , 2005 ) . We show here that this event is preceded by the release of Cdc14 from the nucleolus . Our data further indicate that this Cdc14 activation requires FEAR network function because inappropriate mitotic exit in cells with mispositioned spindles is completely prevented when FEAR network component encoding genes are deleted . FEAR network activity is under the control of the regulatory mechanisms governing the metaphase – anaphase transition ( Stegmeier et al . , 2002 ) . At this cell cycle transition , a checkpoint known as the spindle assembly checkpoint ( SAC ) inhibits entry into anaphase until all chromosomes have attached correctly to the mitotic spindle ( reviewed in Musacchio and Salmon ( 2007 ) ) . Once this has occurred , the SAC inhibition of a protease known as Separase is alleviated and the protease initiates chromosome segregation by cleaving cohesins , the protein complexes that hold sister chromatids together ( Nasmyth , 2002 ) . As Separase is also a component of the FEAR network ( Stegmeier et al . , 2002 ) , SAC activity also governs the release of Cdc14 from the nucleolus during early anaphase . It will be interesting to determine why there are cell-to-cell differences in Cdc14 release from the nucleolus when spindles are mispositioned . Metaphase duration could be a factor . Difficulties in mitotic spindle formation and correctly attaching chromosomes to the mitotic spindle , may lead to prolonged metaphase delays , during which FEAR network component levels could increase leading to a burst of FEAR network activity once the checkpoint is satisfied and cells enter anaphase . This could also explain why the cell cycle arrest following spindle misposition is especially leaky in the tub2-401 mutant , in which the SAC is activated . It is also possible that mitotic CDK activity , which inhibits FEAR network-mediated Cdc14 release from the nucleolus varies between individual cells . Especially high levels of activity could cause a more sustained FEAR-network-dependent release of Cdc14 from the nucleolus causing inappropriate exit from mitosis in some cells with mispositioned spindles . Irrespective of where this variability originates from , the fact that the cell cycle arrest is not absolute in cells with mispositioned spindles is interesting . One interpretation of this observation is that FEAR network activity exhibits cell-to-cell variability with biologically meaningful consequences . It is also possible that defects in the spindle positioning pathways also subtly affect spindle position control of the MEN in some cells but not others . However , we consider this latter possibility less likely because such a scenario predicts that the few cells with mispositioned spindles that exit mitosis inappropriately do so only after a very long arrest . This is not the case . The observation that spindle position control of the MEN and hence mitotic exit is not complete also raises the question of whether the leakiness of the arrest serves a biological function . Is it possible that under conditions where spindle misposition occurs at higher frequency ( i . e . in the cold ) that a not universally permanent cell cycle arrest provides an advantage ? Could binucleation and hence polyploidization provide a reservoir of cells with increased adaptability ? Further investigation is needed to better understand the molecular basis and importance of these cell-to-cell differences . Classically , checkpoint pathways are defined as surveillance mechanisms that monitor the proceedings of a ( cell cycle ) event and prevent the next one from occurring until the preceding event is completed or defects therein have been corrected ( Hartwell and Weinert , 1989 ) . That is , an inhibitory signal prevents cells from progressing to the next cell cycle stage if the preceding cell cycle stage is still ongoing or stalled ( Rao and Johnson , 1970 ) . Our data show that an activating signal in the bud can override any mitotic exit inhibiting signal that may emanate from a mispositioned spindle . These data argue against a checkpoint model in the classical sense to explain the anaphase arrest in response to spindle misposition . Instead they support a model where both spatially constrained positive and negative regulatory signals control the activity of a signal transduction pathway . Checkpoint regulation has been described for other asymmetric divisions ( Cheng et al . , 2008; O'Connell and Wang , 2000 ) . In Drosophila male germline stem cells with mispositioned centrosomes , cell cycle progression is delayed until the centrosomes properly align with respect to the mother-daughter axis of division ( Cheng et al . , 2008 ) . Additionally , the AMP-related kinase family member Par-1 ( of which Kin4 is also a member ) has been shown to be important in delaying cell cycle progression in response to mispositioned spindles in male germline stem cells ( Pereira and Yamashita , 2011; Yuan et al . , 2012 ) . Given these recent findings it is tempting to speculate that a mechanism similar to the one described for spatial control of the MEN by nuclear position , rather than a checkpoint mechanism , also operates in these stem cells . The analysis of cells with multiple centrosomes analogous to what has been described here could address this question . Yeast Strains are derivatives of W303 ( A2587 ) and are described in Supplementary file 1 . GFP-Tub1 is described in Straight et al . ( 1997 ) . The pCTS1-2xmCherry-SV40NLS plasmid was a gift from Drew Endy’s lab . The YIp211-CDC3-mCherry plasmid was a gift from the Erfei Bi’s lab and is described in Fang et al . ( 2010 ) . pFA6a-link-tdTomato-SpHis5 was a gift from Kurt Thorn ( Addgene plasmid # 44640 ) . Leon Chan , Thomas Eng , and Vinny Guacci constructed the pGPD1-OsTIR1-LEU2 and pFA6-3V5-IAA17-KanMx6 plasmids and these were received as gifts from the D . Koshland and K . Weis labs . All gene deletions and C-terminal tags were constructed by the standard PCR-based procedures ( Longtine et al . , 1998; Sheff and Thorn , 2004 ) . All reported statistical error calculations are standard deviations . A biological replicate refers a replicate that was performed using the same experimental conditions but distinct yeast samples . Indirect immunofluorescence microscopy to detect Cdc14-3HA and Tub1 was performed as described in Kilmartin and Adams ( 1984 ) and ( Visintin et al . , 1999 ) . Fixed cell microscopy of GFP-Tub1 and Cdc14-tdTomato was performed by fixing cells in a 4% paraformaldehyde and 3 . 4% sucrose solution for 3 min . Fixed cells were washed in potassium phosphate buffer ( 0 . 1 M KPO4 , pH 7 . 5 and 1 . 2 M sorbitol ) and then treated with 1% Triton for 5 min . These cells were then resuspended potassium phosphate buffer and imaged . Imaging was performed on a Zeiss Axio Observer . Z1 inverted microscope ( Zeiss . Thornwood , NY ) with an ORCA-ER C4742-80 CCD camera ( Hamamatsu Corporation . Middlesex , NJ ) and an X-Cite Series 120 arc lamp ( Life Sciences & Industrial Division . Ontario , Canada ) . Image acquisition and analysis was performed with Molecular Devices Metamorph Software ( Molecular Devices . Sunnyvale , CA ) . Growth conditions for live cell imaging are described in the figure legends with the exception of Figures 6 and 7 ( see prm3Δ heterokaryon protocol below ) . All imaging was done at 25°C . Imaging for the return to growth experiment ( Figure 8 ) was performed with a Nikon Ti-E inverted microscope ( Nikon Instruments Inc . Melville , NY ) equipped with a 60X Plan APO 1 . 4NA objective , a GFP filter , and a CoolSNAPHQ2 CCD camera ( Photometrics , Tucson , AZ ) , controlled by Nikon Elements software . Z stacks of four to eight sections were acquired in 10 min intervals for 12 hrs with a 12 . 5% ND filter and exposure times of 30-500ms . Imaging described in Figure 5B was performed using Nunc Lab-Tek II Chambered Coverglass incubation chambers ( Thermo Fisher Scientific . Cambridge , Mass ) , on a DeltaVision Elite microscope platform ( GE Healthcare Bio-Sciences , Pittsburgh , PA ) . This microscope platform consisted of an InsightSSI solid state light source , an UltimateFocus hardware autofocus system and a model IX-71 , Olympus microscope controlled by SoftWoRx software . Time-lapse images were acquired with a 60X Plan APO 1 . 42NA objective and a CoolSNAP HQ2 camera . All other live cell imaging experiments , with the exception of the microtubule ablation experiment ( see below ) , were performed on a Zeiss Axio Observer . Z1 inverted microscope ( Zeiss . Thornwood , NY ) with a Heliophor Pumped Phosphor Light Engine ( Chroma Technology Corp ( 89 North ) . Bellows Falls , VT ) . Imaging data were collected using a Hamamatsu ORCA-ER C4742-80 CCD camera ( Hamamatsu Corporation . Middlesex , NJ ) run by Molecular Devices Metamorph Software ( Molecular Devices . Sunnyvale , CA ) . Cells were imaged in a CellASIC Y04C or Y04D flow cell chambers ( EMD Millipore Billerica , Massachusetts ) . Microtubules were cut using a Coherent OBIS 405LX laser ( Coherent Inc . Santa Clara , CA ) . Two pre-ablation images were acquired to confirm that a cytoplasmic microtubule was present in the bud neck . The microtubule was cut with one 250 ms , 405 nm laser pulse and severing was confirmed by acquiring a Z-series of nine images spaced at 0 . 6 μm . Microtubule contact with the bud neck was followed at 1 min intervals for 8 min post ablation . Images of the ablated cells were acquired for up to one hour at 15 ± 1 min time intervals to determine cell cycle stage . Imaging was performed on a Nikon Eclipse Ti microscope with a Clara CCD camera ( Andor Technology . South Windsor , Connecticut ) and a Nikon Intensilight arc lamp . See figure legend for culture methods to generate the swe1Δ heterokaryons . To generate the prm3Δ heterokaryons , both MATa and MATα cells were grown overnight to mid-log phase in YEPD medium at room temperature . Cells were centrifuged and resuspended in YEP and then briefly sonicated using a Branson 250 Sonifier ( Branson Ultrasonics Corporation , Dansbury CT ) . These cells were then loaded into a Beckman elutriation rotor JE 5 . 0 ( Beckman Coulter , Brea , CA ) which was cooled to 4°C and equilibrated with YEP at 2400 rpm . Cells were loaded into the elutriation chamber at a speed to 20 mL/min and then equilibrated in the elutriator for 20–30 min at a pump speed of ~10 mL/min . G1 cells were collected at a pump speed of ~23 mL/min . Harvested G1 cells were concentrated using a Konte filtration system and then resuspended to a final OD600nm of 5 . 0 in YEPD . 200 μL of these cells were then plated on a YEPD agar plate and incubated at 30°C for ~2 hr . The resultant population that contained zygotes was washed off of the agar plate using YEPD medium and loaded into a CellASIC Y04D flow chamber for time-lapse analysis .	Most cells duplicate their genetic material and then separate the two copies before they divide . This is true for budding yeast cells , which divide in an unusual manner . New daughter cells grow as a bud on the side of a larger mother cell and are eventually pinched off . To make healthy daughter cells , yeast must share their chromosomes between the mother cell and the bud . This involves threading the chromosomes through a small opening called the bud neck , which connects the mother cell and the bud . A surveillance mechanism in budding yeast monitors the placement of the molecular machine ( called the spindle ) that separates the chromosomes before a cell divides . This mechanism stops the cell from dividing if the spindle is not positioned correctly . Two models could explain how an incorrectly positioned spindle prevents budding yeast from dividing . The first model proposes that yeast cells do not divide if protein filaments ( called microtubules ) touch the bud neck . This only occurs if the spindle is not properly threaded into the bud through the small opening of the bud neck . The second model proposes that specific proteins required for cell division ( which are found at the ends of the spindle ) are inhibited while they are inside the mother cell . This means that the cell cannot divide until one end of its spindle moves out of the mother cell and into the bud . Now , Falk et al . report the results of experiments that aimed to distinguish between these two models . First , a laser was used to cut the spindle filaments in live yeast cells . This stopped the filaments from touching the neck between the mother cell and the bud , but did not cause the cell to divide . Therefore , these results refute the first model . Next , Falk et al . generated yeast cells that had essentially been tricked into forming two separate spindles before they started to divide . As would be predicted by the second model , these cells could divide as long as an end from at least one of the spindles entered the bud . These findings strongly suggest that the second model provides the best explanation for how yeast cells sense spindle position to control cell division . The findings also lend further support to previous work that showed that activators of cell division are found in the bud , while inhibitors of cell division are found in the mother cell . Finally , in a related study , Gryaznova , Caydasi et al . identify a protein at the ends of the spindle that acts like a regulatory hub to coordinate cell division with spindle position . Their findings also suggest that the surveillance mechanism is switched off in the bud to allow the cell to divide .	[ "Abstract", "Introduction", "Results", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2016	Spatial signals link exit from mitosis to spindle position
Small fly eyes should not see fine image details . Because flies exhibit saccadic visual behaviors and their compound eyes have relatively few ommatidia ( sampling points ) , their photoreceptors would be expected to generate blurry and coarse retinal images of the world . Here we demonstrate that Drosophila see the world far better than predicted from the classic theories . By using electrophysiological , optical and behavioral assays , we found that R1-R6 photoreceptors’ encoding capacity in time is maximized to fast high-contrast bursts , which resemble their light input during saccadic behaviors . Whilst over space , R1-R6s resolve moving objects at saccadic speeds beyond the predicted motion-blur-limit . Our results show how refractory phototransduction and rapid photomechanical photoreceptor contractions jointly sharpen retinal images of moving objects in space-time , enabling hyperacute vision , and explain how such microsaccadic information sampling exceeds the compound eyes’ optical limits . These discoveries elucidate how acuity depends upon photoreceptor function and eye movements . The acuity of an eye is limited by its photoreceptor spacing , which provides the grain of the retinal image . To resolve two stationary objects , at least three photoreceptors are needed for detecting the intensity difference in between . To resolve two moving objects is harder , as vision becomes further limited by each photoreceptor’s finite integration time and receptive field size ( Srinivasan and Bernard , 1975; Juusola and French , 1997; Land , 1997 ) . Nevertheless , animals - from insects to man - view the world by using saccades , fast movements , which direct their eyes to the surroundings , and fixation intervals between the saccades , during which gaze is held near stationary ( Land , 1999 ) . Because of photoreceptors’ slow integration-time , saccades should blur image details and these are thought to be sampled when gaze is stabilized . Thus , information would be captured during fixations whilst during saccades animals would be effectively blind . This viewpoint , however , ignores fast photoreceptor adaptation , which causes perceptual fading during fixation ( Ditchburn and Ginsborg , 1952; Riggs and Ratliff , 1952 ) , reducing visual information and possibly rendering perception to mean light only . Therefore , to maximize information and acuity , it is plausible that evolution has optimized photoreceptor function in respect to visual behaviors and needs . We have now devised a suite of new experimental and theoretical methods to study this question both in time and over space in Drosophila R1-R6 photoreceptors . The Drosophila compound eyes are composed of ~750 seemingly regular lens-capped modules called ommatidia , which should provide the fly a panoramic visual field of low optical resolution ( Barlow , 1952; Land , 1997 ) . Each ommatidium contains eight photoreceptor cells ( R1-R8 ) , pointing to seven different directions . The ultraviolet and blue-green-sensitive outer photoreceptors , R1-R6 , initiate the motion vision pathway , whilst the central R7 and R8 , which lie on top of each other , detect different colors from one direction ( Wardill et al . , 2012 ) . Owing to the eye’s neural superposition principle , R1 , R2 , R3 , R4 , R5 and R6 , each from a separate neighboring ommatidium , also point to the same direction . By pooling their output for synaptic transmission , the photoreceptor spacing ( spatial resolution ) effectively matches the ommatidium spacing ( average interommatidial angle , Δφ = 4 . 5o ( Götz , 1964; Land , 1997; Gonzalez-Bellido et al . , 2011 ) but the signal-to-noise ratio of the transmitted image could improve by √6 ( de Ruyter van Steveninck and Laughlin , 1996; Zheng et al . , 2006 ) . Here we show how evolution has improved Drosophila vision beyond these classic ideas , suggesting that light information sampling in R1-R6 photoreceptors is tuned to saccadic behavior . Our intracellular recordings reveal that R1-R6s capture 2-to-4-times more information in time than previous maximum estimates ( Juusola and Hardie , 2001a; Song et al . , 2012; Song and Juusola , 2014 ) when responding to high-contrast bursts ( periods of rapid light changes followed by quiescent periods ) that resemble light input from natural scenes generated by saccadic viewing . Biophysically-realistic model simulations suggest that this improvement largely results from interspersed ‘fixation’ intervals , which allow photoreceptors to sample more information from phasic light changes by relieving them from refractoriness ( Song et al . , 2012; Song and Juusola , 2014; Juusola et al . , 2015 ) . Remarkably , over space , our intracellular recordings , high-speed microscopy and modeling further reveal how photomechanical photoreceptor contractions ( Hardie and Franze , 2012 ) work together with refractory sampling to improve spatial acuity . We discover that by actively modulating light input and photoreceptor output , these processes reduce motion blur during saccades and adaptation during gaze fixation , which otherwise could fade vision ( Ditchburn and Ginsborg , 1952; Riggs and Ratliff , 1952; Land , 1997 ) . The resulting phasic responses sharpen retinal images by highlighting the times when visual objects cross a photoreceptor’s receptive field , thereby encoding space in time ( see also: Ahissar and Arieli , 2001; Donner and Hemilä , 2007; Rucci et al . , 2007; Kuang et al . , 2012a; Kuang et al . , 2012; Franceschini et al . , 2014; Viollet , 2014 ) . Thus , neither saccades nor fixations blind the flies , but together improve vision . Incorporation of this novel opto-mechano-electric mechanism into our ‘microsaccadic sampling’-model predicts that Drosophila can see >4 fold finer details than their eyes’ spatial sampling limit – a prediction directly confirmed by optomotor behavior experiments . By demonstrating how fly photoreceptors’ fast microsaccadic information sampling provides hyperacute vision of moving images , these results change our understanding of insect vision , whilst showing an important relationship between eye movements and visual acuity . To work out how well a Drosophila R1-R6 photoreceptor can see the world , we compared intracellular recordings with realistic theoretical predictions from extensive quantal light information sampling simulations ( Appendixes 1–3 ) , having the following physical limits and properties ( Song et al . , 2012; Song and Juusola , 2014; Juusola et al . , 2015; Song et al . , 2016 ) : As previously described for a variety of other stimuli ( Song et al . , 2012; Song and Juusola , 2014; Juusola et al . , 2015 ) , we found a close correspondence between the recordings and simulations ( waveforms , noise , adaptation dynamics and information transfer ) for all the tested stimuli , establishing how refractory quantal sampling is tuned by light changes . Conversely , control models without refractoriness or based on the Volterra black-box method ( Juusola and French , 1997 ) failed to predict R1-R6s’ information sampling and adaptation dynamics . Nevertheless , these limitations and differences gave us vital clues into the hidden/combined mechanisms that underpin photoreceptor function ( Appendixes 2–9 ) . We now analyze and explain the key results step-by-step . A well-known trade-off of fast adaptation is that it causes perceptual fading during fixation ( Ditchburn and Ginsborg , 1952; Riggs and Ratliff , 1952 ) , and to see the world requires motion or self-motion: body , head and eye movements ( Hengstenberg , 1971; Land , 1973; Franceschini and Chagneux , 1997; Schilstra and van Hateren , 1998; Blaj and van Hateren , 2004; Martinez-Conde et al . , 2013 ) , which remove adaptation . However , it remains unclear whether or how the fly photoreceptors’ information sampling dynamics is tuned to visual behaviors to see the world better . To start unravelling these questions , we first surveyed what kind of stimuli drove their information transfer maximally ( Figure 1 ) , ranging from high-contrast bursts , in which transient intensity fluctuations were briefer than Drosophila’s normal head/body-saccades ( Fry et al . , 2003; Geurten et al . , 2014 ) , to Gaussian white-noise ( GWN ) . These stimuli tested systematically different contrast and bandwidth patterns over R1-R6s’ diurnal encoding gamut . Intracellular recordings ( Figure 1A ) revealed that photoreceptors responded most vigorously to high-contrast bursts , which contained fast transient events with darker intervals . Figure 1B shows the averages ( signals; thick ) and individual responses ( thin ) of a typical R1-R6 , grouped by the stimulus bandwidth and mean contrast . For all the bandwidths ( columns ) , the responses increased with contrast , while for all the contrasts ( rows ) , the responses decreased with the increasing bandwidth ( left ) . Therefore , the slowest high-contrast bursts ( red; top-left ) with the longest darker intervals , which theoretically ( Song et al . , 2012; Song and Juusola , 2014; Juusola et al . , 2015 ) should relieve most refractory microvilli ( Appendixes 1–3 ) , evoked the largest peak-to-peak responses ( 43 . 4 ± 5 . 6 mV; mean ± SD , n = 16 cells; Figure 1—figure supplement 1 ) . Whereas the fastest low-contrast GWN ( blue; bottom-right ) , which would keep more microvilli refractory , evoked the smallest responses ( 3 . 7 ± 1 . 1 mV; n = 4 ) . Notably , whilst all the stimuli were very bright , the largest responses ( to bursts ) were induced at the dimmest background ( BG0 , darkness ) and the smallest responses ( to GWN ) at the brightest background ( BG1 . 5 ) ( Figure 1B ) . Thus , the mean emitted photon rate and light information at the source was lower for the bursts and higher for the GWNs ( the signal-to-noise ratio of the observable world increases with brightening illumination; e . g . Appendix 2—figures 5D and H ) . However , in very bright stimulation , the global mean light intensity ( over the experiment ) becomes less critical for good vision as the eye self-regulates its own input more . Photons galore are lost to intracellular pupil ( Howard et al . , 1987; Song and Juusola , 2014 ) and refractory microvilli ( Song et al . , 2012 ) , which reduce quantum efficiency . Although a R1-R6’s receptive field could be bombarded by 106–109 photons/s ( in daylight ) , due to the dramatic drop in quantum efficiency , the photoreceptor could only count up ∼80 , 000–800 , 000 quantum bumps/s ( Appendix 2 ) . Therefore , the stimulus contrast and bandwidth , which drive the dynamic quantum bump rate changes , summing up the photoreceptor output , are confounded with changes in mean intensity . And , as such , this stimulus design , by containing four different BGs , makes it difficult to see the exact contributions of contrast , bandwidth and mean in controlling the responses . Information theoretical analysis ( Figure 2 and Figure 2—figure supplement 1 ) indicated that the response differences largely reflected differences in their quantum bump counts . The maximum signal power spectra to bursty stimuli could be up to ~6 , 000 times larger than that of the noise , which was effectively stimulus-invariant ( Figure 2—figure supplement 2A ) . Because the noise power spectrum largely represents the average quantum bump’s frequency composition ( Wong et al . , 1982; Juusola and Hardie , 2001a; Song and Juusola , 2014 ) , the bumps adapted to a similar size . Here , given the brightness of the stimuli , the bumps had light-adapted close to their minimum ( Juusola and Hardie , 2001a ) . Thereby , the larger responses simply comprised more bumps . Moreover , with Poisson light statistics , the response precision - how well it estimated photon flux changes - should increase with the square root of bump count until saturation; when more microvilli remained refractory ( Song and Juusola , 2014 ) . Accordingly , signaling performance ( Figure 2A , C ) increased both with the stimulus bandwidth ( left ) and contrast ( right ) , until these became too fast to follow . Information transfer peaked at 100 Hz bursts , which allocated the R1-R6’s limited bandwidth and amplitude range near-optimally , generating the broadest frequency ( Figure 2A and Figure 2—figure supplement 1A ) and ( Gaussian ) voltage distributions ( Figure 2B and Figure 2—figure supplement 1B ) . Thus , with the right mixture of bright ‘saccadic’ bursts ( to maximally activate microvilli ) and darker ‘fixation’ intervals ( to recover from refractoriness ) forming the high-contrast input , a photoreceptor’s information transfer approached the capacity ( Shannon , 1948 ) , the theoretical maximum , where every symbol ( voltage value ) of a message ( macroscopic voltage response ) is transmitted equally often ( Figure 2C and Figure 2—figure supplement 1C ) . Remarkably , this performance ( 610–850 bits/s ) was 2-to-4-times of that for GWN ( 200–350 bits/s ) , which has often been used for characterizing maximal encoding ( Juusola and Hardie , 2001a ) , and twice of that for rich naturalistic stimuli ( 380–510 bits/s ) ( Song and Juusola , 2014 ) ( Figure 2—figure supplement 3 ) . GWN , especially , lacks longer darker events , which should make microvilli refractory ( Song and Juusola , 2014 ) with fewer sampled photons limiting information transfer ( Appendixes 2–3 ) . There are two reasons why these information rate estimates , which were calculated from equal-sized data chunks by the Shannon formula ( Equation 1 , Material and methods ) , should be robust and largely bias-free . First , apart from the responses to 20 Hz high-contrast bursts ( Figure 2B , red trace ) , the responses to all the other stimuli had broadly Gaussian signal and noise distributions , obeying the Shannon formula’s major assumptions ( Shannon , 1948 ) . Second , our previous tests in comparing the Shannon formula to triple extrapolation method ( Juusola and de Polavieja , 2003 ) , which is directly derived from Shannon’s information theory , have shown that for sufficiently large sets of data both these methods provide similar estimates even for this type of highly non-Gaussian responses ( ~5–20% maximal differences ) ( Song and Juusola , 2014; Dau et al . , 2016 ) . And , indeed , new tests using additional recordings to longer stimulus repetitions ( Figure 2—figure supplement 4 ) indicated the same . Thus here , the Shannon formula should provide a sufficiently accurate information estimate also for the 20 Hz high-contrast burst responses , making this evaluation fair ( see Appendix 2 ) . These findings were largely replicated by stochastic simulations ( Figures 3–4 ) . A biophysically realistic photoreceptor model , which contains 30 , 000 microvilli ( Song et al . , 2012 ) , sampled light information much like a real R1-R6 , generating authentic responses to all the test stimuli ( Figure 3A–B ) . Yet , markedly , the model lacked the intracellular pupil ( or any structural adaptation ) , which protects microvilli from saturation ( Howard et al . , 1987; Song and Juusola , 2014 ) , and network connections ( Zheng et al . , 2006; Rivera-Alba et al . , 2011; Wardill et al . , 2012 ) . In real photoreceptors , the pupil screens off excess light to maximize information transfer ( Howard et al . , 1987; Song and Juusola , 2014 ) . Similarly , in the simulations , the mean light intensity of each stimulus was optimized ( Appendix 2 ) for maximum information ( Figure 4A–C ) , establishing the photon absorption rate for a R1-R6 photoreceptor’s best signaling performance ( bits/s ) . At its peak , the model transferred 633 ± 20 bits/s ( mean ± SD; Figure 4C ) for 100 Hz bursts of 8 × 105 photons/s , with further brightening reducing information as more microvilli became refractory . This performance matches that of many real R1-R6s ( Figure 2—figure supplement 1C ) , but is ~200 bits/s less than in some recordings ( Figure 2C ) . The real R1-R6s , on balance , receive extra information from their neighbors ( Rivera-Alba et al . , 2011; Wardill et al . , 2012 ) , which through superposition ( Zheng et al . , 2006 ) sample information from overlapping receptive fields . In other words , since our stimuli ( from a white LED ) were spatially homogenous , these synaptic feedbacks should be able to enhance the system’s signal-to-noise by averaging the photoreceptors’ independent photon count estimates from the same visual area , reducing noise ( Zheng et al . , 2006; Juusola and Song , 2017 ) . Moreover , as their rhabdomere sizes ( Figure 5A–B ) and connectivity vary systematically ( Rivera-Alba et al . , 2011 ) , each R1-R6 receives different amounts of information ( Figure 5C–D ) ( see also: Wardill et al . , 2012 ) . Here , R6s , with large rhabdomeres ( Figure 5B ) and gap-junctions to R8 ( Figure 5C ) , should receive the most ( Wardill et al . , 2012 ) , suggesting that the best performing cells ( e . g . Figures 1 and 2 ) might be of the R6-type ( Figure 5E ) . And yet whilst R7s also share gap-junctions with R6s ( Shaw et al . , 1989 ) , our stimuli contained little UV component to drive them . Encoding efficiency for the different stimuli ( Figure 2D and Figure 2—figure supplement 1D ) was determined as the ratio between the related photoreceptor and light information rates ( Routput/Rinput ) ; with Rinput estimated from the simulated Poisson stimulus repeats , which maximized information in R1-R6 model output ( Figures 3B and 4C; Appendix 2 ) . Thus , as Rinput included the photon loss by the intracellular pupil and other structural adaptations ( Howard et al . , 1987; Song and Juusola , 2014 ) , it was less than at the light source . Moreover , in vivo , the combined stimulus information captured simultaneously by other photoreceptors in the retina network must be more than that by a single R1-R6 ( Zheng et al . , 2006 ) . E . g . as summation reduces noise , the signal-to-noise of a postsynaptic interneuron , LMC , which receives similar inputs from six R1-R6s , can be √6-times higher than that of a R1-R6 ( de Ruyter van Steveninck and Laughlin , 1996; Zheng et al . , 2006 ) , but lower than what is broadcasted from the source ( Song and Juusola , 2014 ) . Thus , information is lost during sampling and processing , with the analysis obeying data processing theorem ( Shannon , 1948; Cover and Thomas , 1991 ) . Finally , as the LED light source’s photon emission statistics were untested ( if sub-Poisson , Rinput would be higher ) , the efficiency estimates represented the theoretical upper bounds . We found that encoding efficiency for both the recordings ( Figure 2D and Figure 2—figure supplement 1D ) and simulations ( Figure 4D ) weakened with the increasing bandwidth ( left ) but less so with contrast ( right ) . This was because Rinput estimates ( Appendix 2 ) increased monotonically with bandwidth ( Song and Juusola , 2014 ) and contrast , while Routput for bandwidth did not ( Figure 2C ) . However , as predicted , some recordings showed >100% efficiency for 20 Hz bursts , presumably due to their extra network information ( Figure 5 and Figure 5—figure supplement 1 ) ( Zheng et al . , 2006; Wardill et al . , 2012; Dau et al . , 2016 ) . A locomoting Drosophila generates ~1–5 head/body-saccades/s , which direct its gaze in high velocities to the surroundings ( Fry et al . , 2003; Geurten et al . , 2014 ) . Here , our recordings and simulations suggested that the refractoriness in R1-R6s’ phototransduction , together with network inputs , might be tuned for capturing information during such fast light changes in time . To test this idea more directly , we used published body yaw velocities ( Geurten et al . , 2014 ) of a walking Drosophila ( Figure 6A ) to sample light intensity information from natural images ( of characteristic 1/fn-statistics [van Hateren , 1997a] ) ( Figure 6B ) . This resulted in time series of contrasts ( Figure 6C , blue ) that ( i ) mimicked light input to a R1-R6 photoreceptor during normal visual behavior , containing fixations , translational movements and saccadic turns . As controls , we further used light inputs resulting from corresponding ( ii ) linear median ( red ) and ( iii ) shuffled ( gray ) velocity walks across the same images ( Video 1 ) . These stimuli were then played back to R1-R6s in intracellular experiments and stochastic refractory model simulations . We found that saccadic viewing of natural images ( Figure 6C , i ) , even without visual selection ( i . e . without the fly choosing what it gazes ) , transformed the resulting light input to resemble the bursty high-contrast stimulation ( Video 1 ) , which maximized photoreceptor information ( Figures 1–2 ) . Such inputs had increasingly sparse light intensity difference ( first derivative ) distributions in respect to those of the linear walks or GWN stimulation ( Figure 6D–E; Appendix 3 ) . Specifically , the saccadic walks contained fixation periods that retained the same light input values for longer durations than the linear walks , which lacked these periods , causing the ~63% higher peak in the saccadic histogram ( Figure 6D ) . Saccadic walking also enhanced the proportion of large intensity differences between two consecutive moments , seen as ~18% higher histogram flanks than those for linear walking ( p=3 . 65 × 10−9 , pair-wise t-test for the combined 0 . 5–1 . 0 and -0 . 5--1 ranges ) . These dynamics drove refractory sampling efficiently ( Song and Juusola , 2014 ) , enabling a R1-R6 to better utilize its output range , and thus capture more information than through the median or shuffled velocity viewing ( Figure 6F; Figure 6—figure supplement 1; cf . Figure 2—figure supplement 3 ) . Altogether , these results ( Figures 1–6 ) imply that saccades and fixations improve a R1-R6’s neural representation of the world in time . Furthermore , as behaviors modulate visual inputs in a sensorimotor-loop , bursty spike trains from the brain ( Franceschini et al . , 1991; Franceschini and Chagneux , 1994; Tang and Juusola , 2010 ) , which direct the gaze through self-motion , may have evolved with photoreceptors’ information sampling dynamics to better detect changes in the world . So when a freely-moving fly directs its gaze to visual features that are relevant for its behavior , its R1-R6’s information capture may become optimized for the imminent task . However , visual behaviors should also affect spatial acuity ( Srinivasan and Bernard , 1975; Juusola and French , 1997; Land , 1997; Geurten et al . , 2014 ) . Hence , we next asked how R1-R6s see saccadic light changes over space . A Drosophila’s head/body-saccades generate fast phasic photoreceptor movements , which ought to blur retinal images ( Srinivasan and Bernard , 1975; Juusola and French , 1997; Land , 1997 ) . Moreover , saccades – when dominated by axial rotation - provide little distance information ( Land , 1999 ) because objects , near and far , would move across the retina with the same speed . Therefore , it has been long thought that visual information is mostly captured during translational motion and gaze fixation , and less during saccades . To test this hypothesis , we reasoned that object motion and self-motion shape a photoreceptor’s light input the same way . Thus , the influence of eye movements ( and motion blur ) on a R1-R6’s ability to resolve objects could be measured in experiments , where , instead of moving the eye , the objects were moved over its stationary receptive field ( Figure 7A; Appendixes 4–6 ) . Using this approach , we recorded individual R1-R6s' voltage responses ( Figure 7B; black traces ) to a pair of bright dots ( each 1 . 7° in size and 6 . 8o apart , as seen by the fly ) , moving at constant speed across their receptive field in front-to-back direction . The movements were either fast ( 205 o/s ) or double-fast ( 409 o/s ) , both within the head/body saccadic velocity range of a walking Drosophila ( Figure 6A–B: 200–800 o/s ) ( Geurten et al . , 2014 ) , and were presented against a dark or lit background ( note: during a free flight ( Fry et al . , 2003 ) , saccadic velocities may reach 2 , 000 o/s ) . Importantly , the dots’ angular separation was less than the half-width of a R1-R6’s receptive field ( Figure 7C ) at the two backgrounds ( ∆ρdark = 9 . 47 ± 1 . 57° , n = 19 cells; ∆ρlight = 7 . 70 ± 1 . 27° , n = 6; mean ± SD; Figure 7—figure supplements 1 and 2 ) and 1 . 5-times the average interommatidial angle ( Δφ ~ 4 . 5° ) , which should determine Drosophila’s visual acuity ( Gonzalez-Bellido et al . , 2011 ) . Thus , these fast-moving point objects tested the theoretical limit of what a R1-R6 should be able to resolve . We further estimated each cell’s respective impulse response ( Appendix 6 ) . Then following the classic theory of compound eyes’ resolving power ( Srinivasan and Bernard , 1975; Juusola and French , 1997; Land , 1997 ) , we calculated each R1-R6’s expected voltage output to the moving dots by convolving its impulse response with its measured dark- or light-adapted receptive field . These Volterra-model ( Juusola and French , 1997 ) predictions ( Figure 7B–C; blue ) were then compared to the actual recordings ( black ) . Remarkably in all these tests , the recordings showed distinctive responses to the two dots ( Figure 7B ) , as two peaks separated by a trough . The relative magnitude of this amplitude separation was quantified as resolvability , using the Raleigh criterion ( Juusola and French , 1997 ) ( Figure 7C ) . However , in marked contrast , the model predictions failed to resolve the double-fast dots , instead blurring into one broad response in both adapting states ( Figure 7D; blue vs . black bars , respectively ) . The predictions for the fast dots were also poorer than the measured responses . Thus , a photoreceptor’s real resolving power was significantly better and less affected by motion blur than predicted by classic theory ( Appendix 6 ) . We next asked whether this better-than-expected resolving power resulted from synaptic interactions ( Zheng et al . , 2006; Freifeld et al . , 2013 ) by using hdcJK910 mutants ( Figure 7E , red traces ) , in which photoreceptors lacked their neurotransmitter , histamine ( Burg et al . , 1993 ) ( Appendixes 4–6 ) . Because hdcJK910 R1-R6s cannot transmit information to their post-synaptic targets ( Dau et al . , 2016 ) ( LMCs , which initiate the motion detection pathways ( Joesch et al . , 2010 ) , and the amacrine cells ) , neither could these photoreceptors receive any light-driven interneuron feedback modulation ( Dau et al . , 2016 ) . Therefore , if the synaptic interactions improved the wild-type output to the moving dots , then hdcJK910 R1-R6s , which lacked these interactions , should show diminished resolvability . But this was never observed . Instead , we found that hdcJK910 R1-R6s resolved the dots at least equally well as the wild-type ( Figure 7D , red ) . Thus , high acuity did not result from synaptic inputs but was intrinsic to photoreceptors . We also calculated Δρ needed to explain the spatial acuity of the recordings . The example ( Figure 7F ) is from a R1-R6 , which had the narrowest light-adapted receptive field ( Δρ = 5 . 73o ) ( Figure 7—figure supplement 2 ) . Its response resolved the two fast-moving dots with a 40 . 5% dip . However , the Volterra model prediction , using its receptive field , only resolved the dots with a 12 . 5% dip ( cf . Figure 7D ) . In fact , for 41 . 0% resolvability , its Δρ would need to narrow to 3 . 70o ( from 5 . 73o ) . Thus , for the prediction to match the recording , the receptive field would have to narrow at least by one-third . Because the required ( predicted ) acceptance angles of R1-R6s were always much narrower ( ≤ 4o ) than the actual measurements ( ∆ρdark = 9 . 47 and ∆ρlight = 7 . 70; see above ) , measurement bias cannot explain this disparity . We further discovered that R1-R6 recordings often showed phasic directional selectivity ( Figure 7G ) , with the responses rising and decaying faster to back-to-front than to front-to-back moving dots . We asked whether these lag-time differences originated from asymmetric photomechanical photoreceptor contractions . Namely , atomic-force microscopy has revealed minute ( <275 nm ) vertical movements on the surface of dissected Drosophila eyes , generated by contraction of individual microvilli as PIP2 is hydrolyzed from the inner leaflet of the lipid bilayer ( Hardie and Franze , 2012 ) . Here , we reasoned that if the ommatidium lenses were effectively fixed and R1-R8s levered to the retinal matrix , the contractions ( Video 2 ) might be larger in situ , moving and shaping the photoreceptors’ receptive fields along some preferred direction . Such mechanical feedback could then reduce light input to R1-R8s , making it more transient and directional . To probe this idea , we recorded in vivo high-speed videos of photoreceptor rhabdomeres ( viewed by optical neutralization of the cornea ) inside the eyes reacting to blue-green light flashes ( 470 + 560 nm ) ( Figure 8A ) . The recordings were performed under far-red ( >720 nm ) illumination , which is nearly invisible to Drosophila ( Appendix 7 ) . We found that 8–20 ms after a flash the rhabdomeres , which directly faced the light source at the image center , shifted rapidly towards the anterior side of their ommatidia ( Figure 8B ) . These local movements were faster and larger the brighter the flash ( Figure 8C ) , and reached their intensity-dependent maxima ( 0 . 2–1 . 2 µm; Figure 8D ) in 70–200 ms , before returning more slowly to their original positions ( Appendix 7 analyses hdcJK910-rhabdomere responses ) . Because the mean R1-R6 rhabdomere tip diameter is ~1 . 7 µm ( Figure 5B ) , a bright flash could shift it more than its half-width sideways . Consequently , the fast rhabdomere movements , whilst still ~3 times slower than their voltage responses ( Figure 8C , wine ) , adapted photoreceptors photomechanically by shifting their receptive fields by 0 . 5–4 . 0o , away from directly pointing to the light source . Video footage at different eye locations indicated that light-activated rhabdomeres moved in back-to-front direction along the eye’s equatorial ( anterior-posterior ) plane ( Figure 8E , red; Video 3 ) , with little up-down components ( black ) . Therefore , as each ommatidium lens inverts projected images , the photoreceptors’ receptive fields should follow front-to-back image motion . This global motion direction , which corresponds to a forward locomoting fly’s dominant horizontal optic flow field , most probably explains the phasic directional selectivity we found to opposing image motions ( Figure 7F; Appendix 8 ) . Thus , the responses to back-to-front moving dots were faster because the dots entered and exited each contracting photoreceptor's front-to-back moving receptive field earlier; whereas the dots moving in the opposite direction stayed slightly longer inside each receptive field . Video analyses further revealed that the first rhabdomere movement was the largest ( Figure 8E ) , but 1 s dark intervals , as used in Figure 7 , could resensitize the photoreceptors for the next ( ~0 . 5 µm ) movements . Even <100 ms dark periods rescued noticeable motility ( Figure 2—figure supplement 2E ) . To inspect how rhabdomere contractions affected the cornea lens system’s image projection , we scanned ommatidia by z-axis piezo steps , with the imaged focal plane travelling down from the lens surface into rhabdomeres ( Figure 8F; Video 4 ) , delivering flashes at predetermined depths . Crucially , we found that the ommatidium lens stayed nearly still , while specific pigment and cone cells , which are connected to the rhabdomere tips by adherens junctions ( Tepass and Harris , 2007 ) , formed a narrow aperture that moved with the rhabdomeres but only half as much . Thus , as the lens system was immobile but the aperture and sensors ( rhabdomeres ) underneath swung differentially , the light input to the moving rhabdomeres was shaped dynamically . This implied that , during saccadic image motion , R1-R6s’ receptive fields might not only move but also narrow ( Appendixes 7–8; Video 2 ) . Essentially , light input to a R1-R6 was modulated by the photoreceptor itself ( Figure 8F ) . To estimate how these photomechanics influenced encoding , we implemented them in stochastic model simulations . We then compared how the predicted light inputs of the classic theory ( Figure 8G ) and the new ‘microsaccadic sampling’-hypothesis ( Figure 8H ) would drive R1-R6 output during the saccadic dot stimulation . In the classic theory , the rhabdomere is immobile ( ii ) . Therefore , light input of two moving dots was a convolution of two broad Gaussians ( i ) that fused together ( iii ) , making them irresolvable to phototransduction ( iv ) ; this also flawed the Volterra-models ( Figure 7 ) . In the new hypothesis , instead , as microvilli became light-activated ( ii ) , the rhabdomere contracted away from the focal point , and then returned back more slowly , recovering from refractoriness . And because its receptive field moved and narrowed concurrently ( its acceptance angle , ∆ρ , halved to 4 . 0o ) , the light input of two moving dots transformed into two intensity peaks ( iii ) , in which time-separation was enhanced by the rhabdomere’s asymmetric motion . Crucially , with such input driving the refractory photon sampling model , its output ( iv ) closely predicted the responses to the two moving dots ( Figure 8I and Figure 8—figure supplement 1 ) . Interestingly , early behavioral experiments in bright illumination ( Götz , 1964 ) suggested similarly narrow R1-R6 acceptance angles ( ~3 . 5o ) . Because of the close correspondence between R1-R6 recordings and the new hypothesis ( Appendixes 6–9 ) , we used it further to predict whether Drosophila possessed hyperacute vision ( Figure 9 ) . We asked whether ‘saccade-fixation-saccade’-like behaviors , when linked to refractory photon sampling and photomechanical photoreceptor contractions , allowed encoding in time finer spatial details than the compound eye’s optical limit ( Δφ ~4 . 5o ) . R1-R6 output was simulated to two bright dots 1-4o apart , crossing its receptive field at different speeds at 25°C . We found that if the dots , or a Drosophila , moved at suitable speed , a photoreceptor should resolve them well ( Figure 9A ) , with this performance depending upon the inter-dot-distance . When the dots/eye moved at 10 o/s , a R1-R6 may capture image details at 1o resolution . But with slower movement ( ≤2 . 5 o/s ) , adaptation should fuse the dots together , making them neurally unresolvable . Conversely , 3o-apart-dots should be seen at 5–100 o/s speeds and 4o-apart-dots even during fast saccades ( 200–300 o/s ) . Thus , the ‘microsaccadic sampling’-hypothesis implied that Drosophila had hyperacute vision over a broad speed range ( Figure 9B ) , and through its own self-motion , could adjust the resolution of its neural images . Further comparisons of model outputs with and without refractoriness indicated that it extends the speed range of hyperacute vision ( Appendix 8 ) . Again , intracellular recordings corroborated these predictions ( Figure 9C and Figure 9—figure supplement 1 ) , demonstrating how acuity could be enhanced by encoding space in time . These results meant that the unexpectedly fine temporal responses of R1-R6s ( Figures 7–9 ) could be used by downstream neurons ( Zheng et al . , 2006; Joesch et al . , 2010; Rivera-Alba et al . , 2011; Wardill et al . , 2012; Behnia et al . , 2014 ) , which can have even faster dynamics ( Juusola et al . , 1995b; Uusitalo et al . , 1995; Zheng et al . , 2006 ) , for spatial discrimination between a single passing object from two passing objects , even if these objects were less than an interommatidial angle apart . The fly brain could then integrate information from hyperacute moving objects and use it for directing behaviors . To test this prediction , we investigated the spatial resolution of Drosophila vision through their optomotor behavior in a conventional flight simulator system , which used brightly-lit high-resolution prints for panoramic scenes ( Figure 10; Appendix 10 ) . We asked whether tethered Drosophila possessed motion vision hyperacuity by recording their yaw torque ( optomotor response ) to vertical black-and-white bar panoramas with <4 . 5o wavelengths , which slowly rotated ( 45 o/s ) to clockwise and counterclockwise . We found that every tested fly responded down to ~1o panoramic bar resolution ( Figure 10A and Figure 10—figure supplement 1 ) with their responses becoming smaller the finer its bars ( Figure 10A–C ) . Importantly , because these responses consistently followed the rotation direction changes , they were not caused by aliasing . Thus , optomotor behavior verified that Drosophila see the world at least in 4-fold finer detail than what was previously thought . Moreover , when a fine-grained ( 3 . 9o ) panoramic image was rotated faster ( Figure 10D ) , the response declined as predicted ( cf . two dots 4o apart in Figure 9A ) . This result is consistent with photoreceptor output setting the perceptual limit for vision and demonstrates that Drosophila see hyperacute details even at saccadic speeds ( Figure 10D–F ) . We have provided deep new insight into spatiotemporal information processing in Drosophila R1-R6 photoreceptors and animal perception in general . Our results indicate that the dynamic interplay between saccades and gaze fixation is important for both the maintenance and enhancement of vision already at the photoreceptor level . This interplay , which is commonly observed in locomoting Drosophila ( Geurten et al . , 2014 ) , makes light input to photoreceptors bursty . We showed that high-contrast bursts , which resemble light input during a fly’s saccadic behaviors , maximize photoreceptors’ information capture in time , and provided evidence that such encoding involves four interlinked mechanisms . Light input is first regulated by two processes inside photoreceptors: slower screening pigment migration ( intracellular pupil , 1–10 s ) and much faster photomechanical rhabdomere contractions ( 0 . 01–1 s ) . These modulations have low noise ( Figure 2—figure supplement 2 ) , enabling refractory photon sampling by microvilli to enhance information intake in phasic stimulus components . Finally , asymmetric synaptic inputs from the network differentiate individual R1-R6 outputs . Remarkably , over space , these mechanisms further sharpen neural resolvability by ~4 fold below the theoretical limit of the compound eye optics , providing hyperacute vision . Further analyses imply that these mechanisms with systematic rhabdomere size variations combat aliasing ( Appendixes 2 and 5 ) . Thus , with microsaccadic sampling , a fly’s behavioral decisions govern its visual information/acuity trade-off . To see the finest image details it should scan the world slowly , which probably happens during normal gaze fixation . But gaze fixation cannot be too slow; otherwise , adaptation would fade vision . Conversely , by locomoting faster , in a saccadic or bursty fashion , visual information capture in time is increased ( see also: Juusola and de Polavieja , 2003 ) , while surprisingly little spatial details about its surroundings would be lost . This viewing strategy corresponds well with the recent human psychophysics results and modeling of ganglion cell firing ( Rucci and Victor , 2015 ) , which indicate that microsaccades and ocular drift in the foveal region of the retina actively enhance perception of spatial details ( Rucci et al . , 2007; Poletti et al . , 2013; Rucci and Victor , 2015 ) . Interestingly , here our findings further imply that , in Drosophila , the extraction of phasic stimulus features , which characterize object boundaries and line elements in visual scenes , already starts during sampling and integration of visual information in the microvilli , at the first processing stage ( rather than later on in the retinal network or in the brain ) . Our results make a general prediction about the optimal viewing strategy for maximizing information capture from the world . Animals should fixate gaze on darker features , as this resensitizes photoreceptors by relieving their refractory sampling units ( e . g . microvilli ) . And then , rapidly move gaze across to brighter image areas , as saccadic crossings over high-contrast boundaries enhance information intake by increasing photoreceptors’ sample ( quantum bump ) rate changes/time ( Appendix 9 ) . Given the high occurrence of eye/head-saccades in animals with good vision ( Land , 1999 ) , it seems plausible that their photoreceptors could also have adapted encoding dynamics to quicken response modulation , reducing motion blur . Therefore , if information sampling biophysics in rods and cones were matched to microsaccadic eye movements , this could provide a mechanistic explanation to the old paradox: how saccadic gaze fixation provides stable perception of the world , while curtailing motion blur effects . 2–10 day old wild-type red-eyed ( Canton-S and Berlin ) fruit flies ( Drosophila melanogaster ) and hdcJK910-mutants were used in the experiments . Other transgenic and mutant Drosophila tests and controls are explained in specific Appendixes . Drosophila were raised at 18°C in a 12 hr/12 hr dark/light cycle and fed on standard medium in our laboratory culture . Sharp microelectrode recordings from Drosophila R1-R6 photoreceptors were detailed before ( Juusola and Hardie , 2001a; Juusola et al . , 2016 ) , and we only list the key steps here . Flies were immobilized to a conical holder by beeswax ( Juusola and Hardie , 2001a ) ( Figure 1A ) . A small hole , the size of a few ommatidia , was cut in the dorsal cornea for the recording electrode and sealed with Vaseline to prevent tissue from drying . R1-R6s’ intracellular voltage responses were recorded to different spatiotemporal light patterns ( see below ) using sharp filamented quartz or borosilicate microelectrodes ( 120–220 MΩ ) , filled with 3 M KCl . A blunt reference electrode , filled with fly ringer , was inserted in the head capsule . The flies’ temperature was kept either at 19 ± 1 or 25 ± 1°C by a feedback-controlled Peltier device , as indicated in the figures . The recordings were performed after 1–2 min of dark adaptation , using the discontinuous clamp method with a switching frequency 20–40 kHz . The electrode capacitance was compensated using the head-stage output voltage . To minimize effects of damage and external noise on the analysis , only stable recordings of low-noise and high sensitivity were chosen for this study ( sometimes lasting several hours ) . Such photoreceptors typically had resting potentials <-60 mV in darkness and >45 mV responses to saturating test light pulses ( Juusola and Hardie , 2001a ) . Because of short-term adaptive trends , we removed the first 3–10 responses to repeated stimulation from the analysis and used the most stable continuous segment of the recordings . Information theoretical methods for quantifying responses of approximately steady-state-adapted fly photoreceptors to different stimuli were described in detail before ( Juusola and Hardie , 2001b; Juusola and de Polavieja , 2003; Song et al . , 2012; Song and Juusola , 2014 ) . Below we list the key approaches used here . Cornea-neutralization method with antidromic far-red ( >720 nm ) illumination was used to observe deep pseudopupils ( Franceschini and Kirschfeld , 1971b ) ( photoreceptor rhabdomeres ) in the Drosophila eye at 21°C . A high-speed camera ( Andor Zyla , UK; 500 frames/s ) , connected to a purpose-built microscope system , recorded fast rhabdomere movements in vivo to blue-green light stimuli ( 470 + 535 nm peaks ) , which were delivered orthodromically into the eye . The method details , mutant and transgenic Drosophila used and the related image analyses are explained in Appendix 7 . Open-loop configuration was used to test hyperacute motion vision . Wild-type flies were tethered in a classic torque meter ( Tang and Guo , 2001 ) with heads fixed , and lowered by a manipulator into the center of a black and white cylinder ( spectral full-width: 380–900 nm ) . A flying fly saw a continuous panoramic scene ( 360° ) , which in the tests contained multiple vertical stripes ( black and white bars of equal width ) . The control was a diffuse white background . After viewing the still scene for 1 s , it was spun counterclockwise by a linear stepping motor for 2 s , stopped for 2 s before rotating clockwise for 2 s , and stopped again for 1 s . This 8 s stimulus was repeated 10 times and each trial , together with the fly's yaw torque responses , was sampled at 1 kHz ( Wardill et al . , 2012 ) . Flies followed the stripe scene rotations , generating yaw torque responses ( optomotor responses to right and left ) , the strength of which reflected the strength of their motion perception . The flies did not follow the white control scene rotations . The panoramic scenes had ±360° azimuth and ±45° elevation , as seen by the fly . The stripe scenes had 1 . 0 contrast and their full-wavelength resolutions were either hyperacute ( 1 . 16° or 2 . 88o ) or coarse ( 14 . 40o ) , giving the inter-bar-distances of 0 . 58o , 1 . 44o and 7 . 20o , respectively . The white scene has zero contrast . The tested scene rotation velocities were 45 , 50 , 200 and 300°/s . The fly eye dissection , fixation embedding , sectioning and imaging protocols for EM ( Figure 5A ) are described in Appendix 5 . Test responses were compared with their controls by performing two-tailed t-tests to evaluate the difference in the compared data . Welch’s t-test was used to accommodate groups with different variances for the unpaired comparisons . In the figures , asterisks are used to mark the statistical significance: ns indicates p>0 . 05 , ∗ indicates p≤0 . 05 , ∗∗ indicates p≤0 . 01 , and ∗∗∗ indicates p≤0 . 001 . Custom written simulation and analyses software used in this study can be downloaded under GNU General Public License v3 . 0 from: https://github . com/JuusolaLab/Microsaccadic_Sampling_Paper . A copy is archived at https://github . com/elifesciences-publications/Microsaccadic_Sampling_Paper .	Fruit flies have five eyes: two large compound eyes which support vision , plus three smaller single lens eyes which are used for navigation . Each compound eye monitors 180° of space and consists of roughly 750 units , each containing eight light-sensitive cells called photoreceptors . This relatively wide spacing of photoreceptors is thought to limit the sharpness , or acuity , of vision in fruit flies . The area of the human retina ( the light-sensitive surface at back of our eyes ) that generates our sharpest vision contains photoreceptors that are 500 times more densely packed . Despite their differing designs , human and fruit fly eyes work via the same general principles . If we , or a fruit fly , were to hold our gaze completely steady , the world would gradually fade from view as the eye adapted to the unchanging visual stimulus . To ensure this does not happen , animals continuously make rapid , automatic eye movements called microsaccades . These refresh the image on the retina and prevent it from fading . Yet it is not known why do they not also cause blurred vision . Standard accounts of vision assume that the retina and the brain perform most of the information processing required , with photoreceptors simply detecting how much light enters the eye . However , Juusola , Dau , Song et al . now challenge this idea by showing that photoreceptors are specially adapted to detect the fluctuating patterns of light that enter the eye as a result of microsaccades . Moreover , fruit fly eyes resolve small moving objects far better than would be predicted based on the spacing of their photoreceptors . The discovery that photoreceptors are well adapted to deal with eye movements changes our understanding of insect vision . The findings also disprove the 100-year-old dogma that the spacing of photoreceptors limits the sharpness of vision in compound eyes . Further studies are required to determine whether photoreceptors in the retinas of other animals , including humans , have similar properties .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "neuroscience" ]	2017	Microsaccadic sampling of moving image information provides Drosophila hyperacute vision
In vivo two-photon calcium imaging currently allows us to observe the activity of multiple neurons up to ~900 µm below the cortical surface without cortical invasion . However , many important brain areas are located deeper than this . Here , we used an 1100 nm laser that underfilled the back aperture of the objective together with red genetically encoded calcium indicators to establish two-photon calcium imaging of the intact mouse brain and detect neural activity up to 1200 μm from the cortical surface . This imaging was obtained from the medial prefrontal cortex ( the prelimbic area ) and the hippocampal CA1 region . We found that neural activity before water delivery repeated at a constant interval was higher in the prelimbic area than in layer 2/3 of the secondary motor area . Reducing the invasiveness of imaging is an important strategy to reveal the intact brain processes active in cognition and memory . Two-photon calcium imaging reveals the in vivo activity of multiple neurons at cellular and subcellular resolution ( Jia et al . , 2010; Ohki et al . , 2005 ) . Recent work demonstrates that by exciting red-fluorescent calcium indicators with a laser at wavelengths of 1000–1100 nm through a cranial window , it is possible to image neural activity in the mouse sensory cortex at depths of 800–900 µm from the cortical surface ( corresponding to layers 5 and 6 ) ( Dana et al . , 2016; Tischbirek et al . , 2015 ) . However , for functional imaging of deeper regions such as the medial prefrontal cortex , hippocampus , and basal ganglia , invasive penetration is unavoidable; it is necessary to insert a microlens or a microprism into the cortical tissue , or to remove the cortical tissue above the target region ( Attardo et al . , 2015; Dombeck et al . , 2010; Low et al . , 2014; Pilz et al . , 2016 ) . The difficulty of deep imaging is mainly caused by refractive index mismatch and light scattering within the tissue ( Helmchen and Denk , 2005; Tung et al . , 2004 ) . These can be weakened when objectives with low numerical aperture ( NA ) are used because the angle of light emitted from the objective is smaller and the light-path length within the tissue is shorter than when high NA objectives are used . However , when low NA objectives are used , the spatial resolution and collection efficiency of emitted fluorescent signals are worse than when high NA objectives are used . If a high NA objective is used in conjunction with underfilling of the back aperture by the excitation laser , the collection efficiency of the fluorescent signal remains high . If the effective NA for the excitation light is small but sufficient to resolve single neurons ( 10–15 μm along the Z axis; Lecoq et al . , 2014 ) , this technique may increase the maximal depth for two-photon calcium imaging of neuronal somata . Although this technique has been theoretically predicted and partially demonstrated in the skin ( Helmchen and Denk , 2005; Tung et al . , 2004 ) , it has not been applied to deep imaging of neural activity in behaving animals . In addition , long-wavelength excitation light and red-fluorescent genetically encoded calcium indicators ( red GECIs; Dana et al . , 2016; Inoue et al . , 2015; Ohkura et al . , 2012 ) are suitable for deep imaging because light scattering is weaker at longer wavelengths . Here , we demonstrate that by underfilling high NA objectives to reduce the effective NA for excitation to approximately 0 . 5 and exciting red GECIs with an 1100-nm laser , we could detect the activity of multiple neurons in the medial prefrontal cortex ( the prelimbic [PL] area ) and the hippocampal CA1 region at depths of 1 . 0–1 . 2 mm in behaving mice without the need for invasive penetration or removal of cortical tissue . To reduce the effective NA , the back aperture of a high NA ( 1 . 00 ) objective was underfilled with a diameter-narrowed 1100 nm laser beam ( 1/e2-width was 7 . 2 mm , compared with the back aperture of 14 . 4 mm ) ( Figure 1A; Matsuzaki et al . , 2008 ) . The effective NA was calculated to be roughly 0 . 5 ( see Materials and methods ) . To examine the effect of underfilling the objective on spatial resolution , 2-μm-diameter fluorescent beads embedded in 2% agarose were imaged through the same glass window used for in vivo GECI imaging ( approximately 0 . 77 mm total thickness ) . When the objective was underfilled , the full-widths at half-maximum ( FWHMs ) were 2 . 28 ± 0 . 05 μm ( mean ± s . d . , n = 5 beads ) laterally and 6 . 95 ± 0 . 13 μm ( n = 5 beads ) axially ( Figure 1Bi ) . These values were greater than the FWHMs of 2 . 15 ± 0 . 05 μm ( n = 4 beads ) laterally and 4 . 37 ± 0 . 06 μm ( n = 4 beads ) axially when the objective was overfilled ( Figure 1Bii ) , but comparable with those used for two-photon calcium imaging of multiple neurons with cellular resolution ( Lecoq et al . , 2014; Sadakane et al . , 2015; Stirman et al . , 2016 ) . Next , we examined whether underfilling the objective was effective for deep imaging in the mouse cerebral cortex . Adeno-associated virus ( AAV ) carrying the tdTomato gene was injected into the intact medial frontal cortex ( mFrC ) of 2- to 3-month-old mice . Two to three weeks post-injection , we imaged tdTomato-expressing neurons in the mFrC at depths of 100–1200 µm from the cortical surface in anesthetized and head-restrained mice through a cranial window in underfilled and overfilled configurations ( Figure 1C and Video 1 ) . We adjusted the laser power at the front aperture of the objective such that it was equal in both configurations . For all depths , the mean bright fluorescent signal ( Kobat et al . , 2009 ) , which was assumed to reflect the fluorescence from tdTomato-expressing neurons , was higher in the underfilled than in the overfilled configuration ( Figure 1D , E and Figure 1—source data 1 ) . The number of circles detected by the Hough transform method ( see Materials and methods ) , which is assumed to reflect the number of fluorescent neuronal somata , was similar between the two configurations at depths of 200–600 µm from the cortical surface ( Figure 1F , G and Figure 3—source data 1 ) . However , the number of circles detected at depths > 600 µm was larger in the underfilled than in the overfilled configuration ( Figure 1E ) . Thus , underfilling the objective was more effective at detecting neuronal morphology at depths > 600 µm . Next , we determined whether underfilling the objective allowed us to detect neural activity at depths > 900 µm from the cortical surface . Three to four weeks after an injection of AAV carrying the R-CaMP1 . 07 gene ( Ohkura et al . , 2012 ) into the intact mFrC of 2- to 3-month-old mice , we observed R-CaMP1 . 07-expressing neurons in the mFrC at depths of 100–1200 µm in awake and head-restrained mice ( Figure 2A–C and Video 2 ) . Using laser power of 170–180 mW at the front aperture of the objective , we detected calcium transients at depths of 1 . 0–1 . 2 mm ( Figure 2D , E ) . This laser power was higher than that used for two-photon imaging of cortical layers 5 and/or 6 ( the maximum power used was: 150 mW in Dana et al . , 2016; 114 mW in Masamizu et al . , 2014; 170 mW in Tischbirek et al . , 2015 ) . Therefore , we examined whether imaging deep mFrC with a 180 mW , 1100 nm laser caused inflammation and heating-induced responses . We used anti-Glial Fibrillary Acidic Protein ( GFAP ) as a marker for activated astrocytes , anti-Ionized calcium binding adapter molecule 1 ( Iba1 ) as a marker for activated microglia , and anti-Heat Shock Protein 70/72 ( HSP70/72 ) as a marker for heat-induced responses in glial cells and neurons ( Figure 3; Podgorski and Ranganathan , 2016 ) . Immunostaining intensity was quantified under five conditions: without imaging ( as a negative control ) , after 15-min imaging at 800–900 µm depths , after 30-min imaging at 900–1100 µm depths , after 30-min imaging at 300–400 µm depths ( as a positive control ) , and after 30-min imaging at 200–300 µm depths , using an objective overfilled with a 200 mW , 920 nm laser and a slow scanning mode ( as a strong positive control ) . The immunoreactivity after 15- or 30-min imaging at 800–1100 µm depths was not different from that in negative control mice ( Figure 3 , Figure 3—figure supplement 1 , and Figure 3—source data 1 ) . By contrast , imaging at 300–400 µm depths with an 1100 nm laser caused significantly higher anti-GFAP and anti-Iba1 immunoreactivity than deep imaging ( Figure 3D–F and Figure 3—source data 1 ) . Imaging at 200–300 µm depths with a 920 nm laser in the slow scanning mode caused significantly higher immunoreactivity of all antibodies than deep imaging ( Figure 3D–F , Figure 3—figure supplement 2 , and Figure 3—source data 1 ) . These results indicate that our immunostaining assays are sensitive enough to detect laser-induced tissue damage and are consistent with a study reporting that heat-induced cell responses occur when two-photon imaging is performed at a 250 µm depth with ≥~300 mW laser power at 920 nm ( Podgorski and Ranganathan , 2016 ) . However , 15–30 min two-photon imaging of the deep area with an 1100 nm laser at 180 mW power did not cause any apparent histological injury . We also examined whether neural activity was affected by deep imaging . In a 25-min continuous imaging session at 900–1100 µm depths in awake mice , we compared the activity of imaged neurons in the first 5-min period with that in the last 5-min period . There was no difference in the mean inferred activities ( by a constrained non-negative matrix factorization algorithm; Pnevmatikakis et al . , 2016 ) between the two periods ( Figure 3—figure supplement 3 ) . This indicates that two-photon imaging of the deep area with ~180 mW laser power did not appear to alter neural excitability . To demonstrate the utility of this method for identifying neural functions in deep areas in the intact brain , we examined neural activity in the mFrC over ~1 mm depth during simple conditioning . Head-restrained mice were conditioned to the delivery of a drop of water with an inter-delivery interval of 20 s ( Figure 4A ) . As each session progressed ( one session per day ) , the licking response rate to water delivery increased and licking became faster ( Figure 4A–D ) . From the fourth–fifth sessions onwards , we performed two-photon calcium imaging of the mFrC at cortical depths of 100–1200 µm ( Video 3 ) . The imaging fields were classified into three areas according to depth ( Paxinos and Franklin , 2007 ) : the superficial area ( 100–300 µm , corresponding to layer 2/3 in the secondary motor area , M2 ) , the middle area ( 300–800 µm , corresponding to layer 5 in M2 ) , and the deep area ( 800–1200 µm , roughly corresponding to layer 6 in M2 and the PL area ) . In all three areas , approximately 50% of neurons showed a peak in the mean ( trial-averaged ) activity during the 5 s after water delivery ( Figure 4E , F and Figure 4—figure supplement 1 ) , which was presumably related to licking and water acquisition ( Figure 4B ) . Additionally , approximately 30% of neurons in all three areas showed a peak in the mean activity during the 10 s before water delivery ( pre-reward period; Figure 4E , F and Figure 4—figure supplement 1 ) . The sequential distribution of the times of peak activity was not an artifact of ordering the neurons according to the time of peak activity , as the ratio of the mean activity around the peak activity to the baseline activity ( ridge-to-background ratio; Harvey et al . , 2012 ) was significantly higher than that of shuffled data ( Figure 4G , H ) . Additionally , the sequential distribution of neurons with peak activity during the pre-reward period was not an artifact ( Figure 4—figure supplement 2 ) . As the mFrC demonstrates strong activity before movement starts ( Friedman et al . , 2015; Kim et al . , 2016a; Pinto and Dan , 2015; Sul et al . , 2011 ) , we focused on the activity during the pre-reward period . When 5 s windows were chosen from the pre-reward period , the ridge-to-background ratios of deep area neurons with peak activity during each 5 s window were frequently higher than those in the shuffled data ( Figure 4—figure supplement 3 ) . To determine whether the activity pattern across trials was stable for individual neurons with peak activity during the pre-reward period , we calculated the correlation coefficient between the times of peak activity of two randomly separated groups of trials ( Figure 4—figure supplement 4A , B; see details in Materials and methods section ) and found that it was higher in the deep area than in the superficial area ( Figure 4—figure supplement 4C ) . This indicates that the PL neurons reliably code the neural activity during the pre-reward period . In addition to the mPrC , we examined whether neural activity in the hippocampus can be imaged without removal of the neocortical tissue lying above it ( Figure 5A ) . CA1 GFP-expressing neurons can be detected by two-photon microscopy in 4-week-old mice , but not in 6- to 9-week-old mice ( Kawakami et al . , 2013 ) . Therefore , we injected AAV-jRGECO1a ( Dana et al . , 2016 ) into the hippocampus of mice aged between 12 and 14 days , and then performed imaging after another 2 weeks . For imaging , the 15 . 1 mm back aperture of the objective ( NA 1 . 05 ) was underfilled with a 7 . 2 mm laser beam ( Figure 5A ) . When we deepened the focal plane below the white matter to depths of 900–1000 µm , we observed densely distributed fluorescent neurons typically located in the CA1 pyramidal layer ( Figure 5B , C and Video 4 ) , as described previously ( Dombeck et al . , 2010 ) , and clearly detected spontaneous calcium transients from these neurons ( Figure 5C ) . No cell death or strong damage was apparent after 15 min of imaging ( Figure 5D , E ) . By contrast , we could not detect any neural morphology or activity in the CA1 region of the mice when they were 3 months old . In the present surgery schedule , the period used for reward delivery conditioning and imaging was shorter , and the young mice were conditioned more slowly than the adult mice used for imaging of the mFrC ( compare the middle and right panels in Figure 6A and Figure 4B , C ) . In the fourth conditioning session , we conducted two-photon calcium imaging of hippocampal CA1 neurons ( Video 5 ) . The majority of active neurons showed peak activity 0–5 s after water delivery ( Figure 6B ) . The time of peak activity of neurons showing peak activity during the pre-reward period was not stable across trials ( Figure 6C , D ) . Here , we demonstrated that underfilling the objective was effective for deep ( 600–1200 µm from the cortical surface ) imaging of neural morphology and activity in the mouse brain over the course of several days . This may be due not only to reduced light scattering but also to the high fluorescent signal in the underfilled configuration; fluorescence is integrated over a larger focal volume in the underfilled configuration than in the overfilled configuration . Deep imaging of neural activity required a relatively high-power laser . However , we confirmed that 15- and 30-min imaging did not produce any apparent morphological or functional damage to the brain tissue . This might be because the excitation ( or density of photons ) at the focal center is lower in the underfilled configuration than in the overfilled configuration ( Helmchen and Denk , 2005 ) and the heat derived from the 1100 nm photon absorbance is relatively low ( Hale and Querry , 1973 ) . We could not image the infralimbic area at >1200 µm depths or the hippocampus in adult mice . Adult hippocampus is difficult to image because , as the mouse becomes older , the myelination of the white matter increases ( Bockhorst et al . , 2008 ) . A recent study demonstrated that three-photon calcium imaging with a 1300 nm laser with a high power per pulse ( ~60 nJ ) can access adult hippocampal CA1 neurons ( Ouzounov et al . , 2017 ) . This suggests that light at around 1100 nm may also penetrate the highly myelinated white matter if a laser with higher average power or higher power per pulse ( Kawakami et al . , 2015 ) is used ( the power per pulse in this study was 2 . 3 nJ ) . In addition , reduction of the effective NA to ~0 . 35 ( corresponding to an axial resolution of ~10 µm; Lecoq et al . , 2014; Stirman et al . , 2016 ) , introduction of adaptive optics to compensate for light scattering ( Ji et al . , 2010 ) , and further improvement of the signal-to-noise ratio of red GECIs ( Dana et al . , 2016; Inoue et al . , 2015 ) will certainly be helpful for imaging multicellular activity in the infralimbic area and the hippocampus in adult intact mice . The major disadvantage of reducing the effective NA is a decrease in spatial resolution . In this study , the axial resolution was approximately 7 µm , which is sufficient to resolve single neurons and is unlikely to cause cross-talk between neurons closely located along the Z axis ( Lecoq et al . , 2014 ) . However , if dendritic branches and spines , and axonal branches and boutons , are the imaging target , then the spatial resolution may not be sufficient . When adaptive optics are used , YFP-expressing dendritic spines can be resolved at a depth of 600 µm ( Wang et al . , 2015 ) . Thus , for deep imaging of subcellular activity , a slightly underfilled objective ( with an effective NA of ~0 . 7 ) combined with adaptive optics may be useful . We found that neural activity during the pre-reward period was more robust in the deep area ( the PL area ) than in the superficial area of the mFrC ( M2 ) . The PL area is strongly related to the processing of motivation- , attention- , and reward-related information ( Friedman et al . , 2015; Kim et al . , 2016b; Otis et al . , 2017; Pinto and Dan , 2015 ) , whereas M2 is strongly related to action selection and motor planning ( Li et al . , 2015; Sul et al . , 2011 ) . Thus , PL activity during the pre-reward period might reflect the expectation of reward delivery , including motivation , reward prediction , or attention to the timing of the water delivery . However , the conditioning in this study did not require any change in mouse brain state before water delivery . Conducting deep imaging during decision-making tasks will help us to understand the hierarchical and/or parallel processing occurring across the PL and M2 areas during decision-making and action . In the intact brain , it is easy to change the field of view parallel to the cortical surface . An 8-mm-wide glass window can be used for long-term imaging of the whole dorsal neocortex in the mouse ( Kim et al . , 2016a ) . Objectives with wide fields of view ( >3 mm ) developed for two-photon imaging ( Sofroniew et al . , 2016; Stirman et al . , 2016; Tsai et al . , 2015 ) can cover the mFrC and the neocortex lying above the hippocampus . The hippocampus connects the mFrC through the thalamus and the entorhinal cortex ( Jin and Maren , 2015; Varela et al . , 2014 ) and is thought to associate spatial , temporal , and reward information , which are required for goal-directed decision-making ( Wikenheiser and Schoenbaum , 2016 ) . Here , the pre-reward activity in hippocampal CA1 neurons was not stable , likely because the conditioning period in the young mice was not sufficient to form such activity . If the adult hippocampus can be imaged through a wide-field cranial window and objective , the neural activity in both areas could be imaged simultaneously . Deep and wide-field two-photon calcium imaging of the intact brain will substantially aid our understanding of the brain circuits that integrate multimodal information in decision-making . All animal experiments were approved by the Institutional Animal Care and Use Committee of The University of Tokyo , Japan ( Medicine-P16-012 ) . All mice were provided with food and water ad libitum and housed in a 12:12 hr light–dark cycle . The mice were not used for other experiments before this study . Male C57BL/6 mice ( aged 2–3 months , SLC , Shizuoka , Japan ) were utilized for mFrC imaging . Male and female C57BL/6 mice ( aged 12–40 days in the young mice , and 2–3 months in the adult mice; Japan SLC , Shizuoka , Japan ) were utilized for the imaging experiments in the hippocampus . For experiments using young mice , pups were weaned at P30 , and then group-housed until the imaging window was implanted . In this study , two red-fluorescent genetically encoded calcium indicators , R-CaMP1 . 07 ( Ohkura et al . , 2012 ) , jRGECO1a ( Dana et al . , 2016 ) and a calcium-insensitive red-fluorescent protein , tdTomato , were used . For imaging of R-CaMP1 . 07 , the GCaMP3 DNA of pAAV-human synapsin I promoter ( hSyn ) -GCaMP3-WPRE-hGH polyA ( Masamizu et al . , 2014 ) was replaced with R-CaMP1 . 07 DNA from a pN1-R-CaMP1 . 07 vector construct ( Ohkura et al . , 2012 ) . rAAV2/1-hSyn-R-CaMP1 . 07 ( 1 . 3 × 1013 vector genomes/ml ) was produced with pAAV2-1 and purified as described previously ( Kaneda et al . , 2011; Kobayashi et al . , 2016 ) . tdTomato was expressed via a 1:1 cocktail of viral solutions including rAAV2/1-CaMKII-Cre ( 3 . 16 × 1010 vector genomes/ml ) and rAAV2/1-CAG-FLEX-tdTomato ( 5 . 1 × 1012 vector genomes/ml ) . rAAV2/1-CaMKII-Cre , rAAV2/1-CAG-FLEX-tdTomato and rAAV2/1-hSyn-NES-jRGECO1a ( 2 . 95 × 1013 vector genomes/ml ) were obtained from the University of Pennsylvania Gene Therapy Program Vector Core . The mice were water-deprived in their home cages and maintained at 80–85% of their normal weight throughout the experiments . During the behavioral conditioning , mice were set within a body chamber and head-fixed with custom-designed apparatus ( O’Hara , Tokyo , Japan; Hira et al . , 2013 ) . A spout was set in front of their mouth , and a 4 μl drop of water was delivered from the spout at a time interval of 20 s . The mice were allowed to lick at any time , and licking behavior was monitored by an infrared LED sensor . The rate of water delivery that incurred at least one lick during 2 s after the delivery was defined as the responsive rate . The duration of the daily conditioning sessions was 40–60 min . At the end of each session , the mice were allowed to freely gain water drops ( total water consumption was ~1 ml per session ) . On rest days ( typically weekends ) , the mice had free access to a 3% agarose block ( 1 . 2 g per day ) in the cage . Two-photon imaging was conducted using an FVMPE-RS system ( Olympus , Tokyo , Japan ) equipped with a 25 × water immersion objective ( for imaging of the mFrC: XLPLN25XSVMP , numerical aperture: 1 . 00 , working distance: 4 mm , Olympus; for imaging of the hippocampus: XLPLN25XWMP2 , numerical aperture: 1 . 05 , working distance: 2 mm , Olympus ) and a broadly tunable laser with a pulse width of 120 fs and a repetition rate of 80 MHz ( Insight DS +Dual , Spectra Physics , CA , USA ) , set at a wavelength of 1100 nm . Fluorescence emissions were collected using a GaAsP photomultiplier tube ( Hamamatsu Photonics , Shizuoka , Japan ) . To shorten the light-path length within the tissue , the back aperture of the objective was underfilled with the diameter-shortened ( 7 . 2 mm , in comparison with that of the back aperture of 14 . 4 mm or 15 . 1 mm ) laser beam . When the objective ( XLPLN25XSVMP ) was underfilled , the effective NA was calculated to be roughly 0 . 5 ( i . e . , 1 . 00 × 7 . 2/14 . 4 ) . When scanning the center of the glass window at a depth of 1 . 2 mm from the cortical surface , the laser was assumed to be not clipped by the glass window ( 0 . 5 < 1 . 33 sin ( tan−1 [0 . 75/1 . 2] ) =0 . 70 ) . During the imaging experiments , the mouse head was fixed and the body was constrained within a body chamber under the microscope ( OPR-GST , O’Hara; Masamizu et al . , 2014 ) . Before the first imaging session began for each mouse , the angle of the stage on which the mouse chamber was placed was finely adjusted to set the glass window perpendicular to the optical axis . This was accomplished by the imaging of microbeads on the surface of the glass window ( Kawakami et al . , 2015 ) . The frame acquisition rate was 30 frames/s , with a resonant scanning mirror for the X axis and a galvanometric scanning mirror for the Y axis , the pixel dwell time was 0 . 067 µs , and the size of the imaging fields was generally 512 × 512 pixels ( 0 . 994 µm/pixel ) or 512 × 160 pixels , with three-frame averaging to increase the signal-to-noise ratio . For the strong control condition in the immunostaining experiment , the laser wavelength was tuned to 920 nm , galvanometric scanning mirrors were used for the horizontal and vertical axes , and the pixel dwell time was 200 µs . The collection collar of the objective was adjusted so that the imaging plane was well resolved . For XYZ imaging , the collar was adjusted so that deep planes were well resolved . The laser power was gradually increased from the cortical surface to the deep imaging plane . XYZ image stacks were acquired with a resonant scanner and 16–frame averaging per XY-plane . The step size was 2 . 5 µm unless otherwise noted . In the comparative experiments using underfilled and overfilled objectives , the laser power at the front aperture of the objective was measured in both objective configurations , and adjusted such that it was equal at each depth of imaging in the same anesthetized mice . The transmission ratio of the overfilled objective to the underfilled objective was approximately 1 . 5 . The depth of the functional imaging plane was up to 1200 µm from the cortical surface ( n = 62 planes in the mFrC from 11 mice expressing R-CaMP1 . 07 , n = 6 in the hippocampus from three mice expressing jRGECO1a ) . The duration of one imaging session was 15–20 min unless otherwise noted , and 1–4 imaging sessions from different depths were performed in a daily experiment . For each mouse , imaging was conducted for 1–5 days . Analyses were performed using MATLAB ( R2016a , version 9 . 0 . 0 . 341360; MathWorks , MA , USA , RRID:SCR_001622 ) and Fiji software ( Schindelin et al . , 2012 , RRID:SCR_002285 , http://imagej . net/Fiji ) . Raw image sequences acquired on the FVMPE-RS system were loaded into MATLAB using custom-written scripts ( http://github . com/YR-T/oir2stdData; copy archived at https://github . com/elifesciences-publications/oir2stdData ) . To estimate the difference between images obtained with underfilled and overfilled objectives in tdTomato-expressing animals , we calculated the bright fluorescent signal and extraction of geometric characteristics from the raw image at each depth . The bright fluorescent signal in each imaging plane was defined as the average value of the brightest 0 . 1% pixels ( Kobat et al . , 2009 ) . To estimate the number of fluorescent circular structures , Hough transform-based detection of circles using a built-in MATLAB function ( imfindcircles provided in the image processing toolbox ) with a radius range of 6–12 pixels ( approximately 6–12 µm ) was applied to each XY plane . As described above , virus was injected into at 800–1200 µm from the cortical surface . This might explain why the number of detected circles increased as the imaging depth increased to 1000 µm ( Figure 1F ) . Motion correction for calcium imaging was performed by phase-correlation using the Suite2P package ( Pachitariu et al . , 2016 , http://github . com/cortex-lab/Suite2P ) . After the motion correction , images were three frame-averaged before being analyzed . A constrained non-negative matrix factorization ( cNMF ) algorithm was employed to extract neural activities from a time series of images ( Pnevmatikakis et al . , 2016 , http://github . com/epnev/ca_source_extraction ) . Then , extracted active components with soma-like contours were selected and those with dendrite- or axon-like contours were removed via visual inspection . The number of extracted active components during the conditioning experiment was as follows: 80 . 50 ± 26 . 29 ( mean ± s . d . , n = 12 fields ) in the superficial area of the mFrC , 74 . 66 ± 14 . 89 ( n = 35 fields ) in the middle area of the mFrC , 65 . 53 ± 27 . 64 ( n = 15 fields ) in the deep area of the mFrC , and 22 . 17 ± 16 . 32 ( n = 6 fields ) in the hippocampal CA1 region . The noise variances in the power spectrum density at high frequency estimated by the cNMF algorithm were as follows ( mean ± s . d . ) : 14 . 42 ± 5 . 11 ( n = 12 fields ) in the superficial area of the mFrC , 22 . 00 ± 10 . 27 ( n = 35 fields ) in the middle area of the mFrC , 21 . 69 ± 12 . 63 ( n = 15 fields ) in the deep area of the mFrC , and 19 . 45 ± 3 . 68 ( n = 6 fields ) in the hippocampal CA1 region . The detrended relative fluorescence changes ( ∆F/F ) were calculated with eight percentile values over an interval of ±30 s around each sample time point ( Dombeck et al . , 2007 ) . Traces of ∆F/F from 10 s before to 10 s after the water delivery in those deliveries with at least one lick during 2 s after the delivery were used for the analyses . The ridge-to-background ratio was used for the estimation of the distribution of the time of peak activity ( Harvey et al . , 2012 ) . To create a shuffled ∆F/F trace of each neuron , the time point of the actual ∆F/F trace was circularly shifted by a random amount for each trial and then trial-averaged . For each neuron , the ridge ∆F/F was defined as the mean ∆F/F over 12 frames ( 100 ms/frame ) surrounding the time of peak activity , and the background ∆F/F was defined as the mean ΔF/F in the other data points . The ridge ∆F/F was then divided by the background ∆F/F . The trial-by-trial stability of the time of peak activity of the neurons that had their peak activity during the pre-reward period ( −10 s to 0 s ) was evaluated as follows: in each session , all trials were randomly divided into two groups , and the trial-averaged activity in each group was calculated for each neuron . To remove the effects of different sample sizes across the three mFrC areas and the hippocampus , 50 neurons were randomly chosen from all imaging fields in each area . The time of peak activity in one group was plotted against that in the other , and the Pearson’s correlation coefficient was determined . Thus , if the timing of the peak activity of each neuron was constant across trials , the correlation coefficient should be 1 . This procedure was repeated 1000 times , and the 95% confidence interval was determined for each of the areas . When the lower bound of the 95% confidence interval was above zero , it was concluded that the time of peak activity was not random across trials . To estimate the difference in the trial-by-trial stability of the time of peak activity between pairs of the three areas in the mFrC ( Figure 4—figure supplement 4C ) , the mean correlation coefficients were compared using a permutation test . For each pair from the superficial , middle , and deep areas , all neurons with peak activity during the pre-reward period were randomly reassigned to one of the two areas . For each area with reassigned neurons , the correlation coefficient between the times of peak activity of the two randomly separated groups of trials was calculated , and the absolute difference of the correlation coefficients between the two areas was estimated . This procedure was repeated 10 , 000 times , and the distribution of the absolute differences between the two areas was determined . Following this , the statistical significance was determined according to whether or not the absolute difference in the mean correlation coefficients between the two areas with original neurons assigned ( Figure 4—figure supplement 4B ) was above the 95th percentile of the resampled distribution corrected using the Bonferroni method . The difference in the distribution of the correlation coefficients was not due to differences in animal behavior because the mean response rate and reaction time during imaging experiments were not different among the three areas ( response rate , 98 . 98 ± 0 . 72% in the superficial area , 98 . 63 ± 0 . 39% in the middle area , and 98 . 88 ± 0 . 58% in the deep area ( mean ± s . e . m . ) , p=0 . 88 , one-way ANOVA; reaction time , 224 . 08 ± 40 . 66 ms in the superficial area , 235 . 46 ± 31 . 44 ms in the middle area , and 272 . 89 ± 49 . 28 ms in the deep area ( mean ± s . e . m . ) , p=0 . 77 , one-way ANOVA ) . The mice were deeply anesthetized with ketamine ( 74 mg/kg ) and xylazine ( 10 mg/kg ) and transcardially perfused with 40 ml of phosphate buffered saline ( PBS ) and 40 ml of 4% paraformaldehyde in PBS ( Wako , Osaka , Japan ) 16–24 hr after the last in vivo imaging session . In the experiment to assess deep-imaging-induced damage to brain tissue , mice were perfused approximately 1 month ( for imaging with 920 nm laser ) or 5–10 days ( under other conditions ) after cranial window implantation . The brains were removed and postfixed with the same fixative at 4°C for longer than 12 hr . For immunostaining , the brains were cut into coronal sections with a thickness of 50–100 µm . Slices were washed in PBS-X ( 0 . 5% triton-X in PBS ) containing 10% normal goat serum , and then incubated with one of the primary antibodies ( 1:500 dilution of rabbit anti-GFAP , G9269 , Sigma-Aldrich , MO , USA , RRID:AB_477035; 1:500 dilution of rabbit anti-Iba1 , 019–19741 , Wako , RRID:AB_839504; 1:400 dilution of mouse anti-HSP70/72 , ADI-SPA-810-F , Enzo Life Sciences , NY , USA , RRID:AB_10616513 ) overnight at 4°C . Afterwards , slices were washed in PBS-X and incubated with species-appropriate Alexa Fluoro-488 conjugated secondary antibody ( 1:500 dilution of anti-rabbit IgG for GFAP and Iba1 antibodies; 1:500 dilution of anti-mouse IgG for HSP70/72 antibody ) . After staining the cell nuclei with fluorescent Nissl stain ( 1:200 NeuroTrace 435/455 or NeuroTrace 640/660 , N21479 or N21483 , Thermo Fisher Scientific , MA , USA ) , the slices were mounted on glass slides with Fluoromount/Plus mounting medium ( Diagnostic BioSystems , CA , USA ) . Fluorescence images were acquired with an upright fluorescence microscope ( BX53 , Olympus ) and a CCD camera ( Retiga 2000R , Q Imaging , BC , Canada ) or all-in-one fluorescence microscope ( BZ-X700 , Keyence , Osaka , Japan ) , and analyzed with Fiji software ( Schindelin et al . , 2012 ) , RRID:SCR_002285 ) . To calculate the immunoreactivity of glial and heat-shock protein activation , regions of interest with approximately 1 × 1 mm covering the cortical surface and imaged depth were selected at the center of the imaging site and the mirror position in the contralateral hemisphere . Three to five slices per immunolabel were selected for the calculation in each animal . To compensate for signal dispersion in each slice , the mean fluorescence intensity on the treated side was normalized according to the mean intensity on the contralateral side for each immunolabel . The ratios of immunoreactivity in the AAV-injected hemisphere to that in the contralateral hemisphere in mice without imaging were similar to those reported in Podgorski and Ranganathan ( 2016 ) but were greater than one . This may be because the cranial window implantation took place only 5–10 days before the tissue fixation and because of remaining damage from AAV injection on the ipsilateral side . Data are presented as mean ± s . d . , and the Wilcoxon rank-sum tests , paired t-test , one-way ANOVA and post-hoc multiple comparison with Tukey-Kramer method , Spearman’s correlation tests , Pearson’s correlation tests , and permutation tests described above were used for statistical comparisons . Pairwise comparisons were two-tailed unless otherwise noted . Error bars in graphs represent the s . e . m . No statistical tests were run to predetermine the sample size , and blinding and randomization were not performed .	Microscopes can now reveal what individual cells are doing inside a living brain . In a technique called two-photon microscopy , light-sensitive proteins are introduced into the brain cells . A laser then shines light of a specific wavelength into the brain . Whenever one of the proteins in an active brain cell absorbs some light from the laser , it gives off light that a sensor can detect . Yet , a two-photon microscope could only "see" up to about 900 micrometers from the brain’s surface . This is because light scatters as it travels through brain tissue . Shorter wavelengths scatter the most; so two-photon microscopes use infrared lasers , which have a longer wavelength than visible light . Even so , structures deeper within the brain like the hippocampus and medial prefrontal cortex remained out of range . The only way to see these structures – which are involved in memory and planning – was to damage the brain by inserting a lens or by removing the overlying tissue . But such damage may also change brain activity . Kondo et al . have now found a way to image brain cells up to 1 , 200 micrometers below the surface of an intact mouse brain . The new approach uses an optimized microscope and a laser that generates even longer wavelength light . It also makes use of proteins that give off red light , rather than yellow or green . These changes made it possible to view activity in the medial prefrontal cortex and hippocampus . The brain cells showed no signs of damage after about 30 minutes of viewing . This suggests that the approach does not cause overheating or kill cells . Many questions remain about what happens deep within an active brain . By allowing neuroscientists to follow the activity of brain cells over months , for example as an animal learns a task , these improvements to two-photon microscopy could lead to new insights into the processes of learning and decision-making . Kondo et al . hope that other researchers will find more ways to use the refined technique in their own experiments .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "tools", "and", "resources", "neuroscience" ]	2017	Two-photon calcium imaging of the medial prefrontal cortex and hippocampus without cortical invasion
Magneto- and electro-encephalography ( MEG/EEG ) non-invasively record human brain activity with millisecond resolution providing reliable markers of healthy and disease states . Relating these macroscopic signals to underlying cellular- and circuit-level generators is a limitation that constrains using MEG/EEG to reveal novel principles of information processing or to translate findings into new therapies for neuropathology . To address this problem , we built Human Neocortical Neurosolver ( HNN , https://hnn . brown . edu ) software . HNN has a graphical user interface designed to help researchers and clinicians interpret the neural origins of MEG/EEG . HNN’s core is a neocortical circuit model that accounts for biophysical origins of electrical currents generating MEG/EEG . Data can be directly compared to simulated signals and parameters easily manipulated to develop/test hypotheses on a signal’s origin . Tutorials teach users to simulate commonly measured signals , including event related potentials and brain rhythms . HNN’s ability to associate signals across scales makes it a unique tool for translational neuroscience research . Modern neuroscience is in the midst of a revolution in understanding the cellular and genetic substrates of healthy brain dynamics and disease due to advances in cellular- and circuit-level approaches in animal models , for example two-photon imaging and optogenetics . However , the translation of new discoveries to human neuroscience is significantly lacking ( Badre et al . , 2015; Sahin et al . , 2018 ) . To understand human disease , and more generally the human condition , we must study humans . To date , EEG and MEG are the only noninvasive methods to study electrical neural activity in humans with fine temporal resolution . Despite the fact that EEG/MEG provide biomarkers of almost all healthy and abnormal brain dynamics , these so called ‘macro-scale’ techniques suffer from difficulty in interpretability in terms of the underlying cellular- and circuit-level events . As such , there is a need for a translator that can bridge the ‘micro-scale’ animal data with the ‘macro-scale’ human recordings in a principled way . This is the ideal problem for computational neural modeling , where the model can have specificity at different scales . To address this need , we developed the Human Neocortical Neurosolver ( HNN ) , a modeling tool designed to provide researchers and clinicians an easy-to-use software platform to develop and test hypotheses regarding the neural origin of their data . The foundation of the HNN software is a neocortical model that accounts for the biophysical origin of macroscale extracranial EEG/MEG recordings with enough detail to translate to the underlying cellular- and network-level activity . HNN’s graphical user interface ( GUI ) provides users with an interactive tool to interpret the neural underpinnings of EEG/MEG data and changes in these signals with behavior or neuropathology . HNN’s underlying model represents a canonical neocortical circuit based on generalizable features of cortical circuitry , with individual pyramidal neurons and interneurons arranged across the cortical layers , and layer-specific input pathways that relay spiking information from other parts of the brain , which are not explicitly modeled . Based on known electromagnetic biophysics underlying macroscale EEG/MEG signals ( Jones , 2015 ) , the elementary current generators of EEG/MEG ( current dipoles ) are simulated from the intracellular current flow in the long and spatially-aligned pyramidal neuron dendrites ( Hämäläinen et al . , 1993; Ikeda et al . , 2005; Jones , 2015; Murakami et al . , 2003; Murakami and Okada , 2006; Okada et al . , 1997 ) . This unique construction produces equal units between the model output and source-localized data ( ampere-meters , Am ) allowing one-to-one comparison between model and data to guide interpretation . The extracranial macroscale nature of EEG/MEG limits the space of signals that are typically observed and studied . The majority of studies focus on quantification of event related potentials ( ERPs ) and low-frequency brain rhythms ( <100 Hz ) , and there are commonalities in these signals across tasks and species ( Buzsáki et al . , 2013; Shin et al . , 2017 ) . HNN’s underlying mathematical model has been successfully applied to interpret the mechanisms and meaning of these common signals , including sensory evoked responses and oscillations in the alpha ( 7–14 Hz ) , beta ( 15–29 Hz ) and gamma bands ( 30–80 Hz ) ( Jones et al . , 2009; Jones et al . , 2007; Lee and Jones , 2013; Sherman et al . , 2016; Ziegler et al . , 2010 ) , and changes with perception ( Jones et al . , 2007 ) and aging ( Ziegler et al . , 2010 ) . The model has also been used to study the impact of non-invasive brain stimulation on circuit dynamics measured with EEG ( Sliva et al . , 2018 ) , and to constrain more reduced ‘neural mass models’ of laminar activity ( Pinotsis et al . , 2017 ) . In the clinical domain , HNN’s model has also been applied to study MEG-measured circuit deficits in autism ( Khan et al . , 2015 ) . Despite these examples of use , the complexity of the original model and code hindered use by the general community . The innovation in the new HNN software is the construction of an intuitive graphical user interface to interact with the model without any coding . We offer several free and publicly-available resources to assist the broad EEG/MEG community in using the software and applying the model to their studies . These resources include an example workflow , several tutorials ( based on the prior studies cited above ) to study ERPs and oscillations , and community-sharing resources . HNN’s GUI is designed so that researchers can simultaneously view the model's net current dipole output and microscale features ( including layer-specific responses , individual cell spiking activity , and somatic voltages ) in both the time and frequency domains . HNN is constructed to be a hypothesis development and testing tool to produce circuit-level predictions that can then be directly tested and informed by invasive recordings and/or other imaging modalities . This level of scalability provides a unique tool for translational neuroscience research . In this paper , we outline biophysiological and physiological background information that is the basis of the development of HNN , give an overview of tutorials and available data and parameter sets to simulate ERPs and low-frequency oscillations in the alpha , beta , and gamma range , and describe current distribution and online resources ( https://hnn . brown . edu ) . We discuss the differences between HNN and other EEG/MEG modeling software packages , as well as limitations and future directions . The HNN GUI is designed to allow researchers to link macro-scale EEG/MEG recordings to the underlying cellular- and network-level generators . Currently available visualizations include a direct comparison of simulated electrical sources to recorded data with calculated goodness of fit estimates , layer-specific current dipole activity , individual cell spiking activity , and individual cell somatic voltages ( Figure 4B–D ) . Results can be visualized in both the time and frequency domain . Based on its biophysically detailed design , the output of HNN’s model and recorded source-localized data have the same units of measure ( Am ) . By closely matching the output of the model to recorded data in an interactive manner , users can test and develop hypotheses on the cell and network origin of their signals . The process for simulating evoked responses or brain rhythms from a single region of interest is to first define the network structure and then to ‘activate’ the network with exogenous driving input based on your hypotheses and simulation experiment . HNN’s template model provides the initial network structure . The choice of ‘activation’ to the network depends on the simulation experiment . The GUI design is motivated by our prior published studies and was built specifically to simulate sensory evoked responses , spontaneous rhythms , or a combination of the two ( Jones et al . , 2009; Jones et al . , 2007; Khan et al . , 2015; Lee and Jones , 2013; Sherman et al . , 2016; Sliva et al . , 2018; Ziegler et al . , 2010 ) . The tutorials described in the Results section below detail examples of how to ‘activate’ the network to simulate sensory evoked responses and spontaneous rhythms . Here , we outline a typical simulation experiment workflow . In practice , users apply the following interactive workflow , as in Figure 4 and detailed further in the tutorials with an example tactile evoked response from somatosensory cortex ( data from Jones et al . , 2007 ) . ( Step 1 ) Load EEG/MEG data ( blue ) . ( Step one is optional . ) ( Step 2 ) Define the cortical column network structure . The default template network is automatically loaded when HNN starts . Default parameters describing the local network can be adjusted by clicking the Set Parameters button on the GUI and then Local Network Parameters , or directly from the Local Network Parameters button on the GUI ( Figure 4A ) . ( Step 3 ) ‘Activate’ the local network by defining layer-specific , exogenous driving inputs ( Figure 3B , C ) . The drive represents input to the local circuit from thalamus and/or other cortical areas and can be in the form of ( i ) spike trains ( single spikes or bursts of rhythmic input ) that activate post-synaptic targets in the local network , ( ii ) current clamps ( tonic drive ) , or ( iii ) noisy ( Poisson ) synaptic drive . The choice of input parameters depends on your hypotheses and ‘simulation experiment’ . In the example simulation , predefined evoked response parameters were loaded in via the Set Parameters From File button and choosing the file ‘ERPYes100Trials . param’; this is also the default evoked response parameter set loaded when starting HNN ( Figure 4B ) . The Evoked Input parameters are then viewed in the Set Parameters dialog box under Evoked Inputs ( Figure 4C ) . The Evoked Inputs parameters are described further in the tutorials below . ( Step 4 ) Run simulation and directly compare model output ( black ) and data ( purple ) with goodness of fit calculations ( root mean squared error , RMSE , between data and averaged simulation ) ( Figure 4D ) . ( Step 5 ) Visualize microcircuit details , including layer-specific responses , cell membrane voltages , and spiking profiles by choosing from the View pull down menu ( Figure 4D , E , F ) . ( Step 6 ) Adjust parameters through the Set Parameters dialog box to develop and test predictions on the circuit mechanisms that provide the best fit to the data . With any parameter adjustment , the change in the dipole signal can be viewed and compared with the prior simulation to infer how specific parameters impact the current dipole waveform . Prior simulations can be maintained in the GUI or removed . For ERPs , automatic parameter optimization can be iteratively applied to tune the parameters of the exogenous driving inputs to find those that provide the best initial fit between the simulated dipole waveform and the EEG/MEG data ( see further details below ) . ( Step 7 ) To infer circuit differences across experimental conditions , once a fit to one condition is found , adjustments to relevant cell and network parameters can be made ( guided by user-defined hypotheses ) , and the simulation can be re-run to see if predicted changes account for the observed differences in the data A list of the GUI-adjustable parameters in the model can be found in the ‘Tour of the GUI’ section of the tutorials on our website . HNN’s GUI was designed so that users could easily find the adjustable parameters from buttons and pull down menus on the main GUI leading to dialogue boxes with explanatory labels . As a specific example on how to use HNN as a hypothesis testing tool , we have used HNN to evaluate hypothesized changes in EEG-measured neural circuit dynamics with non-invasive brain stimulation ( Figure 5 ) . We measured somatosensory evoked responses from brief threshold-level taps to the middle finger tip before and after 10 min of ~10 Hz transcranial alternating current stimulation ( tACS ) over contralateral somatosensory cortex ( see Sliva et al . , 2018 for details ) . The magnitude of an early peak near ~70 ms in the tactile evoked response increased after the tACS session ( Figure 5 , top left ) . Based on prior literature , we hypothesized that the observed difference was due to changes in synaptic efficacy in the local network induced by the tACS ( Kronberg et al . , 2017; Rahman et al . , 2017 ) . To test this hypothesis , we first used HNN to simulate the pre-tACS evoked responses , following the evoked response tutorial in our software ( see Tutorial below ) . Once the pre-tACS condition was accounted for , we then adjusted the synaptic gain between the excitatory and inhibitory cells in the network using the HNN GUI and re-simulated the tactile evoked responses . We tested several possible gain changes between the populations . HNN showed that a two-fold increase in synaptic strength of the inhibitory connections , as opposed to an increase in the excitatory connections or in total synaptic efficacy , could best account for the observed differences in the data ( compare blue in red curves in Figure 5 ) . By viewing the cell spiking profiles in each condition ( Figure 5 , bottom right ) , HNN further predicted that the increase in the magnitude of the ~70 ms peak coincided with increased firing in the inhibitory neuron population and decreased firing in the excitatory pyramidal neurons in the post-tACS compared to the pre-tACS window . These detailed predictions can guide further experiments and follow-up testing in animal models or with other human imaging experiments . Follow up testing of model derived predictions is described further in the alpha/beta tutorial below . HNN’s tutorials are designed to teach users how to simulate the most commonly studied EEG/MEG signals , including sensory evoked responses and low-frequency oscillations ( alpha , beta , and gamma rhythms ) by walking users through the workflow we applied in our prior studies of these signals . The data and parameter sets used in these studies are distributed with the software , and the interactive GUI design was motivated by this workflow . In completing each tutorial , users will have a sense of the basic structure of the GUI and the process for manipulating relevant parameters and viewing results . From there , users can begin to develop and test hypotheses on the origin of their own data . Below we give a basic overview of each tutorial . The HNN website ( https://hnn . brown . edu ) provides additional information and example exercises for further exploration . To ease the process of narrowing in on parameter values representing a user’s hypothesized model , we have added a model optimization tool in HNN . Currently , this tool automatically estimates parameter values that minimize the error between model output and features of ERP waveforms from experiments . Parameter estimation is a computationally demanding task for any large-scale model . To reduce this complexity , we have leveraged insight of key parameters essential to ERP generation , along with a parameter sensitivity analysis , to create an optimization procedure that reduces the computational demand to a level that can be satisfied by a common multi-core laptop . Two primary insights guided development of the optimization tool . First , exogenous proximal and distal driving inputs are the essential parameters to first tune to get an initial accurate representation of an ERP waveform . Thus , the model optimization is currently designed to estimate the parameters of these driving inputs defined by their synaptic connection strengths , and the Gaussian distribution of their timing ( see dialog box in Figure 11B ) . In optimizing the parameters of the evoked response simulations to reproduce ERP data distributed with HNN ( e . g . see ERP tutorial ) , we performed sensitivity analyses that estimated the relative contributions of each parameter to model uncertainty , where a low contribution indicated that a parameter could be fixed in the model and excluded from the estimation process to decrease compute time ( see Supplementary Materials ) . Second , an intuitive insight that was confirmed by parameter sensitivity analysis is that the influence of each exogenous input on the simulated dipole varies over time , with the highest influence during and just after the time of the input ( see Supplementary Material ) . We used this knowledge to create a stepwise optimization process , only estimating parameter values for one input at a time , where the objective of each optimization is to minimize a weighted root mean squared error ( RMSE ) measure between simulated and experimental data only during the relevant time window ( see Materials and methods ) . This stepwise estimation reduces the complexity of the optimization problem and saves time . Each step in the process searches for parameter estimates using the COBYLA optimization algorithm ( Powell , 1994 ) ( see Materials and methods for detailed explanation of the stepwise optimization procedure ) . In this example , we describe an application of the model optimization tool for estimating parameters to simulate data representing the SI evoked response to a brief suprathreshold level tactile stimulation -- which is 100% detected ( Figure 11A ) . This evoked response is similar to that shown in Figure 4 , where the signal was elicited from a perceptual threshold level stimulation - at 50% detection . We start from the parameter file fitted to the 50% detection scenario , and use HNN’s model optimization feature to find parameter estimates that provide a better fit the suprathreshold-level experimental data . The data from this study is also included in the HNN distribution ( ‘SI_SupraT . txt’ ) . Many models of neocortical circuitry , with varying levels of complexity , have been developed to simulate LFP , EEG/MEG and/or ECoG ( e . g . , Barrès et al . , 2013; Kiebel et al . , 2008; Reimann et al . , 2013; Sanz Leon et al . , 2013 ) . Several modeling tools and associated documentation are also available to build user defined neocortical models for general use that are not domain specific , such as NEURON ( https://neuron . yale . edu/neuron/ ) , NetPyNE ( Dura-Bernal et al . , 2019 ) , the Brain Modeling Toolkit/Bionet ( Gratiy et al . , 2018 ) , and the Brain Simulation Platform from the European Union Human Brain Project . Among the current modeling software designed specifically for study of EEG/MEG signals ( e . g . , The Virtual Brain [TVB] https://thevirtualbrain . org , Dynamic Causal Modeling [DCM] of E/MEG within the Statistical Parametric Mapping [SPM] software https://www . fil . ion . ucl . ac . uk/spm/ , and LFPy https://lfpy . readthedocs . io/en/latest/; Hagen et al . , 2018; Kiebel et al . , 2008; Sanz Leon et al . , 2013 ) , HNN’s model , goals , and capabilities are unique . The goal of HNN is to provide a user-friendly graphical interface to a validated biophysically detailed model of neocortical circuitry , and to teach the community , regardless of neural modeling or coding experience , how to interact with the model to study the neural origin of commonly measured macroscale EEG/MEG signals . This includes studying ERPs , and low frequency alpha , beta and gamma rhythms . HNN’s construction and tutorials are based on knowledge and workflows developed in prior published studies . As with other open-source software , continued application of HNN to new use cases means that software users can add to and improve upon the examples distributed with HNN . The level of biophysical detail included in HNN’s model and the calculation of the primary electrical currents from the intracellular dendritic current flow in multi-compartment pyramidal neurons enables one-to-one comparison between model output and source localized data in units of Am . HNN was specifically designed for interpreting microscale cellular- and circuit-level activity from single regions of interest . The cell and network level details provided can guide targeted testing and make connections to studies in animal models . Below we describe ways in which HNN’s goals and construction are distinct from other current domain specific EEG/MEG modeling software , namely LFPy , TVB , DCM . LFPy is a Python package that provides a set of Python libraries and associated documentation on how to apply these scripts to simulate multi-scale signals , including current dipole , LFP , ECoG , M/EEG sensor signals , in user defined multi-compartment neuron models and networks built in NEURON or NeuroML ( Hagen et al . , 2018 ) . LFPy does not contain a GUI and is designed for users who have experience in neural modeling and Python . Users define their own workflows to simulate signals of interest that can be compared to data . The LFPy Python classes are likely to provide a useful framework for expanding the utility of HNN to include multi-area simulations , and simulations of LFP and EEG/MEG sensor level signals , as described in Limitations and future directions below . DCM applied to EEG/MEG data is also a non-GUI based scripting tool , using Matlab . Users assume an active set of distributed sites , that is nodes , in the brain that contribute to a recorded signal . The neural activity of a node is simulated using ‘neural mass’ representations in which the activity ( e . g . firing rate ) of a population of neurons is simulated with a reduced number of variables ( Kiebel et al . , 2008 ) . The recorded data is fit to the assumed nodes and directionality of interactions between nodes statistically inferred . TVB is designed to simulate large-scale network interactions also using reduced neural mass representations . Active nodes across the whole brain are assumed to contribute to the recorded signal and connectivity between nodes is informed by individualized tractography data ( Sanz Leon et al . , 2013 ) . Multi-scale EEG/MEG and/or fMRI data can be fit to the model . One advantage of this approach is that propagation of activity across the brain can be studied ( e . g . spread of seizure ) , unlike HNN which is currently restricted to interpreting detailed activity in a single region of interest . Indeed , many prior models of EEG/MEG rely on reduced representations of neural activity , including neural mass and/or mean field approximations ( Breakspear et al . , 2004; Jansen and Rit , 1995; Jirsa and Haken , 1996; Kiebel et al . , 2008; Sanz Leon et al . , 2013; Woolrich and Stephan , 2013 ) . Such simplifications may be necessary to ensure mathematical or computational tractability of models that address interactions between multiple areas or whole brain activity ( Breakspear , 2017 ) . However , that tractability comes at the cost of suppressing or eliminating the ability to evaluate cellular-level details of individual spiking units and dendritic currents , or to perform one-to-one comparisons between model and data; explicit goals of HNN . One of the greatest challenges in computational neural modeling is deciding the appropriate scale of model to use to answer the question at hand . There is always a tradeoff between model complexity and computational efficiency , ease of use , and interpretability . As discussed above , this tradeoff underlies different scales of modeling in various EEG/MEG modeling software . HNN’s model was chosen to be minimally sufficient to accurately account for the biophysical origin of the primary currents that underlie EEG/MEG signals in a single brain area; namely , the net intracellular current flow in the apical dendrites of pyramidal neurons that span across the cortical layers and receive layer-specific synaptic input from other brain areas . HNN’s model was also constructed to maintain known canonical features of neocortical circuitry including , excitatory/inhibitory ratios , layer specific synaptic interactions , and cell spiking behaviors ( see Parameter Tuning above , and Materials and methods ) . While HNN’s model is a reduction of the full complexity of neocortical circuits , it has been successful in interpreting the origin of extracranially measured macro-scale EEG/MEG signals that likely rely on canonical macroscale features of neocortical circuitry and not on finer details of the underlying structure . A future direction discussed below is to expand HNN to simulate extracellular local field potential signals ( LFPs ) , and sensor level signals , whose accuracy may require additional model detail and whose implementation can be aided by other existing tools such as LFPy and NetPyNe ( discussed further below ) . Any conclusions made with HNN are based on the underlying model assumptions that are important for users to understand . These assumptions are outlined in detail in the Materials and methods section , in our prior publications , and on our website . Parameter optimization is a computationally challenging problem in any large-scale model . The process for parameter tuning to study ERPs and oscillations in HNN’s underlying model is detailed above . Based on our prior studies and sensitivity analyses ( see Supplementary Materials ) , we have identified that the timing and strength of the layer specific exogenous drive to the local network is critical in defining the timing and peaks of sensory evoked responses . As such , HNN currently includes a tool to optimize these parameters based on reducing the error between simulated evoked response waveforms and recorded data . Due to the non-stationary nature of spontaneous brain rhythms ( e . g . Figure 6 and Figure 10 ) error reduction based on matching waveform features is not as straightforward , and other signal features may be necessary to consider for optimization ( e . g . PSD peak amplitudes , see Figure 8 and Jones et al . , 2009 ) . Future expansions of HNN will include the ability to optimize over other user defined parameters , and to minimize errors between model output and various features of recorded data , with an estimate of the sensitivity of various parameters to these features . Given enough compute power , large parameter sweeps could be implemented in HNN to generate families of models for template matching to given waveforms via machine learning algorithms . This would serve as an alternative means for circuit interpretation without interactive hypothesis development and testing . At present , HNN can be run on high performance computers through the Neuroscience Gateway Portal ( www . nsgportal . org ) and Amazon Web Services ( https://aws . amazon . com ) , see also Dissemination in Materials and methods . Currently , all conclusions made in HNN are derived from the template neocortical column model provided . Another important step in expanding HNN’s utility will be to enable users to define their own cells and circuits to use within the HNN framework . While the HNN code is open source and adaptable for advanced users , it is difficult for those without expertise in computational neural modeling in Neuron/Python to expand . Therefore , work is in progress to convert HNN’s underlying neural model to the NetPyNe simulation language ( www . netpyne . org ) ( Dura-Bernal et al . , 2019 ) . NetPyNe is a neural modeling platform enabling flexible cell and network development . This conversion will facilitate the ability to expand HNN to the study of activity from and between multiple cortical areas and the thalamus . NetPyNe is designed with both a GUI and command line ( CLI ) interface facilitating the construction of code that is readily accessible and human-readable . In expanding HNN to the NetPyNe language , HNN will also embrace the dual GUI and CLI capabilities , enabling the specification of architectures and parameters to be scriptable so that simulations and analyses can scale-up beyond manual operations . HNN is designed to simulate source-localized current dipole signals produced by neurons . Source localization is currently viewed as an independent process . The output from any source localization algorithm can be compared to HNN’s simulated output . In future expansions of HNN , we plan to integrate HNN’s ‘bottom up’ simulations , with ‘top down’ source localization estimates using minimum-norm-estimate ( MNE ) software ( www . martinos . org/mne ) ( Gramfort et al . , 2013; Gramfort et al . , 2014 ) , providing an all-in-one software tool for source localization and circuit-based interpretation . In doing so , parameter estimation in each software package may benefit from direct knowledge and constraints from the other . Additionally , HNN’s utility will be expanded to include estimation of forward fields through the brain to simulate and visualize LFPs , current-source density , and sensor-level EEG/MEG signals , facilitating comparison to these recording modalities . We have shown that HNN can be a useful tool to interpret the impact of noninvasive brain stimulation ( NIBS ) on EEG-measured circuit dynamics ( Figure 5 , Sliva et al . , 2018 ) . HNN was used to test specific hypotheses on tACS-induced modulation of synaptic dynamics by accounting for EEG signal differences in pre-tACS compared to post-tACS periods . A useful expansion of HNN will be to include simulations of the fields induced in the brain by NIBS ( e . g . , with finite-element-estimates Windhoff et al . , 2013 ) and to directly couple these fields to the modeled neurons . This integration would facilitate studying the effects of NIBS on real-time EEG signals and could lead to improved NIBS paradigms . In total , HNN’s present distribution and planned expansions are aimed at providing a one-of-a-kind , user-friendly software tool for translational neuroscience research that is accessible to a wide scientific and clinical community . Membrane voltages in each simulated compartment were calculated using the standard Hodgkin-Huxley parallel conductance equations , and current flow between compartments follows from cable theory as accounted for in NEURON . Extending the prior work of Bush and Sejnowski ( 1993 ) , active ionic currents were included in both the somatic and dendritic compartment of the cells of the pyramidal neurons , and in the single compartment of the inhibitory neurons . For the pyramidal neurons , the membrane resistance was increased and membrane capacitance was decreased from the Bush and Sejnowski’s values by the same 1 . 3 scaling factor as the compartment sizes described above ( Rm23 , 474 cm2 for L5 and L2/3; Cm 0 . 85 and Cm0 . 6195 F/cm2 for L5 and L2/3 , respectively ) to maintain the input resistances in the cells of 45 M for the L5 and 110 M for L2/3 ( Douglas et al . , 1991 ) . The axial resistance for each cell was Ra200 cm ( Segev et al . , 1992 ) . The parameters regulating the active currents were tuned to replicate known in vitro firing patterns in response to somatic current injection . The kinetic equations and NEURON code used for each of these currents were as used by Mainen and Sejnowski ( 1996 ) and downloaded from http://senselab . med . yale . edu/senselab/modeldb/ . The maximal conductances of each current were constant throughout the soma and dendrite ( Bekkers , 2000; Korngreen and Sakmann , 2000; Migliore and Shepherd , 2002; Stuart and Sakmann , 1994 ) and were chosen to produce adapting spikes in the L2/3 PNs and bursting in the L5 PNs to current injected in the soma ( 1 nA for 100 ms ) representative of neurons classified as regular spiking and intrinsically bursting , respectively ( Moore and Nelson , 1998; Silva et al . , 1991; Zhu and Connors , 1999 ) . The inhibitory neurons were tuned to represent basket cells and produced regular fast spiking dynamics to injected current , as in other cortical network models ( Garabedian et al . , 2003; Jones et al . , 2000; Pinto et al . , 2003 ) . The following table displays the ion channels and mechanisms in each cell type in the model ( X ) indicates the presence of the channel/mechanism in the cell type , see online code for full equations . In the table above , Na ( fast ) /K ( fast ) are the fast sodium and potassium channels responsible for generating action potentials . Km is the muscarine sensitive potassium channel , with a relatively slow time-constant and KCa is the calcium-dependent potassium channel , which contributes to hyperpolarization after calcium influx into the cell . The L- and T-type calcium ( Ca ) channels represent the high-threshold and low-threshold activated calcium channels which together with the hyperpolarization-activated cyclic nucleotide gated channel ( HCN ) contribute to bursting . Ca decay represents the calcium extrusion pump , which causes intracellular calcium to decay towards a baseline level . Leak represents the passive channel , with constant conductance . Dipole represents the mechanism that takes into account the primary axial current flow within pyramidal neuron dendrites , responsible for the generation of simulated signals comparable to MEG/EEG recordings . For more details see Jones et al . ( 2009 ) . HNN’s default template neocortical model includes neurons arranged in three dimensions . The XY plane is used to array cells on a regular grid while the Z-axis specifies cortical layer . HNN’s default model contains a regular 10 × 10 grid ( arbitrary units ) of pyramidal neurons in layer 2/3 and layer five for a total of 200 pyramidal neurons , with interneurons interleaved regularly in a 3–1 ratio ( see Figure 3D ) . The local synaptic architecture in Figure 3A was based on an abundance of animal studies and , in particular , studies of the mouse/rat somatosensory cortex ( Bernardo et al . , 1990a; Bernardo et al . , 1990b; for review , see Thomson et al . , 2002; Thomson and Bannister , 2003 , and Bannister , 2005 ) . Inhibitory synaptic connections onto PNs were located on the soma ( Freund et al . , 1986; Kisvárday et al . , 1985; Somogyi et al . , 1983 ) , and excitatory synapses contacted the basal and apical oblique dendrites ( Deuchars et al . , 1994; Feldmeyer et al . , 2002; Lübke et al . , 1996; Thomson and Bannister , 1998 ) . Synaptic dynamics were modeled with bi-exponential functions . The rise and decay time constants and reversal potentials were based on experiments and the original neocortical model in Jones et al . ( 2009 ) , and are generally as follows: AMPA ( 0 . 5 ms , 1 . 0 ms , 0 mV ) ; NMDA ( 1 . 0 ms , 20 . 0 ms , 0 mV ) ; GABAA ( 0 . 5 ms , 5 . 0 ms , −80 mV ) , GABAB ( 1 . 0 ms , 20 . 0 ms , −80 mV ) . Within a cortical layer there is recurrent connectivity between neurons of a given type ( PN to PN , interneuron to interneuron ) , PN to interneuron connectivity , and synaptic inhibition from interneurons onto PNs . The following synaptic connections are present across cortical layers: layer 2/3 PNs to layer 5 PNs , layer 2/3 interneurons to layer 5 PNs , layer 2/3 PNs to layer five interneurons . The conductance of the synaptic connections within the local network grid were defined with a symmetric 2D Gaussian spatial profile , with a delay incorporate into the synaptic connection between two cells defined by and inverse Gaussian ( Jones , 1986; Kaas and Garraghty , 1991 ) . There is all-to-all connectivity between any two populations of synaptically-coupled neurons . Synaptic weights between the neurons are scaled inversely by the distance in the XY plane ( arbitrary units ) between the neurons ( d ) using exponential fall-off following e-d2/λ2 , and space constant λ , which depends on pre- and post-synaptic type ( Table 2 below ) . The synaptic delays are scaled in proportion to the XY plane distance ( d ) between the neurons following 1/e-d2/λ2 , to account for the larger propagation distance ( note that the λ value is determined using values in Table 2 ) . With increasing d between neurons , the synaptic weights decay , while the synaptic delays increase . The connectivity details are based on known neocortical anatomy and local circuit wiring patterns , as derived from the literature . Further details on connectivity are available on HNN’s website and prior publications . At rest , the default model does not generate activity . HNN provides several ways to activate the local cortical column with layer specific excitatory synaptic input representing thalamo-cortical , and/or cortical-cortical and noisy/tonic drive . The user defines the choice of driving input to the network , based on their simulation experiment , as described in Results . Exogenous driving networks are not explicitly modeled , rather the user defines trains or bursts of action potentials representing these inputs that excite the local network via AMPA or NMDA synaptic connections to distinct layers and cellular compartments . These inputs are referred to as proximal and distal drive based on the PN dendritic contact location . Proximal inputs contact basal and oblique dendrites of PN and somas of the inhibitory neurons in L2/3 and L5 , and distal inputs contact distal dendrites of the PN in L2/3 and L5 and somas of the inhibitory neurons in L2/3 only , as shown in Figure 3 . The trains of action potentials , or tonic/noisy input , that the user defines are created in specific dialog boxes in the GUI and represent either Evoked , Rhythmic , Tonic , or Poisson Inputs , as motivated by our prior studies and tutorials described in Results . Axial current flow between any two neighboring model compartments i , j is defined as iaxial = ( vi - vj ) /raxial , where vi , vj , and raxial are the voltages in compartment i , j , and the resistance between the compartments , respectively . In order to convert this axial current into a dipole signal , we apply a length scaling where the axial current is scaled by the inter-compartment distance along the vertical axis . The length scaling means that for the longer apical dendrites of layer five pyramidal neurons , the contribution will be larger than from the shorter layer 2/3 pyramidal neuron apical dendrites . Note that the orientation of the dendrites relative to the vertical axis also influences the contribution to the dipole signal . For example , the horizontally-oriented oblique dendrites which do not have any vertical length component , do not contribute to the dipole signal , whereas for basal dendrites oriented at 45 degrees from the vertical axis , the scaling is -2/2 ( note the negative sign is because these dendrites are pointing downward ) . The contribution from all neighboring compartments within a neuron is integrated and then added to a value across the set of all pyramidal neurons . As a result of the multiplication between axial current and length , the model dipole output signal has the same units of measure as the experimental data ( Am ) ( Okada et al . , 1997; Murakami et al . , 2003; Murakami and Okada , 2006; Jones et al . , 2007; Hagen et al . , 2018 ) . HNN includes a method to optimize ERP simulations . The optimization procedure was uniquely designed to minimize the RMSE between model output and ERP waveforms in a stepwise manner that decreases parameter exploration and saves compute time . This procedure takes advantage of the assumption that the exogenous proximal and distal driving inputs are essential parameters to tune to get an accurate representation of an ERP waveform . Additionally , it applies the knowledge that , with probabilistic certainty , features of the dipole waveform at a particular point in time cannot be influenced by an exogenous driving input that begins after that point in time . Since exogenous inputs are modeled as Gaussian processes , the likelihood of occurrence can be modeled by a probability distribution function ( PDF ) normally distributed with a given mean and standard deviation . Figure 11—figure supplement 1A shows the PDFs of the inputs for the suprathreshold example described in the results Figure 11 . An input’s contribution to the ERP will begin when there is a non-zero probability of occurrence and persist for a duration commensurate with the input’s cumulative distribution function ( CDF ) , shown in Figure 11—figure supplement 1B . This clearly illustrates that from 20 to 50 ms , the input labeled ‘Proximal 1’ is the unique contributor to the waveform . After 50 ms , effects from Distal one begin , thus adding new parameters that contribute to the waveform fit and reduce the relative contribution of Proximal 1 ( from full to partial ) . It follows that each successive driving input will have a time window where it is most likely to have a unique and dominant effect . As such , our approach to model optimization is to divide the process into smaller steps where only a single input’s parameters are estimated before proceeding to optimize the next input . To implement this procedure , we developed a new goodness of fit measure that amplifies the importance of maximizing the fit at points of unique contribution ( e . g . 20–50 ms for Proximal 1 , Figure 11—figure supplement 1C ) and diminished the importance of fitting to later points where other inputs contribute more to the fit . We began with standard root mean squared error ( RMSE ) RMSE=∑t=0T ( x1 , t−x2 , t ) 2Twhere t is the current simulation time , from 0 to simulation completion ( T ) , and x1 , t is the simulated dipole at t , and x2 , t is the experimental data point . Then we adapted RMSE to include weight functions specific for input k at time t , wRMSEk=∑t=0Twk ( t ) ( x1 , t−x2 , t ) 2∑t=0Twk ( t ) where an assignment of wk ( t ) =1 for all t would be equivalent to RMSE . For each input k , we first defined a weight distribution function , wk ( t ) , as the Unique Contribution Index ( UCI ) , which starts from the CDF of input k and simply subtracts the CDF of subsequent inputs , with a lower bound of 0 ( Figure 11—figure supplement 1C ) . Equivalently , UCIk ( t ) =CDFk ( t ) −∑i=k+1NCDFi ( t ) , where N is the number of exogenous driving inputs in the simulation . Figure 11—figure supplement 1C shows that Proximal 1’s influence is unique up to 50 ms , Distal one has a dominant , but not unique contribution near 70 ms , and Proximal two is dominant after ~100 ms . When the UCI is applied as a weighting function in the wRMSE equation above , we observed that some optimization steps would negatively impact the fit in regions after the peak in UCI , where the errors had been down-weighted , requiring subsequent optimization steps to attempt to ‘correct’ the fit . Our solution was to instead define the weight function using the Extended Contribution Index ( ECI ) , which includes a term that delays the weight function’s return to 0 , extending the window of data points that have an impact on wRMSE further into the simulation . This achieves a balance between optimal parameter estimates for the current step and providing a good starting point for following optimization steps . ECI is defined byECIk ( t ) =CDFk ( t ) −∑i=k+1NCDFi ( t ) A ( μi−μkT ) , where μi and μk are the mean start times of the next input and the current input , respectively . Simulation length is represented by T and A is an empirically derived constant . We arrived at a value of 1 . 6 for A as a factor that appropriately minimized the contribution of inputs proportional to the delay between their onset and the kth input currently being optimized . The effect of the ECI’s decay term can be seen in Figure 11—figure supplement 1D where the ECI for Proximal one extends further than the corresponding UCI , and the ECI of Distal one remains significant through the end of the simulation . Since points where ECIk , t approximately equal 0 will have a negligible impact on wRMSEk , we define a threshold of 0 . 01 where wRMSEk is calculated for the window starting when ECIk , t rises above 0 . 01 and ending when ECIk , t drops below 0 . 01 . For the first exogenous driving input , it is likely that the window will end before the completion of the simulation . In that first step , simulations can be stopped early , reducing the time required for simulating each candidate parameter set in that step . The final step in our model optimization process is to vary all free parameters from all inputs using regular RMSE to measure goodness of fit . Like each previous step , the number of simulations run is limited . So this primary purpose of this final step is to make small corrections , not perform all-at-once optimization ( which would likely require thousands of simulations ) . It also provides an opportunity to rebalance the contributions from multiple inputs in regions where there is a high degree of parameter inter-dependence . However , if the user is certain that they want to perform all-at-once optimization ( which would likely require many more simulations ) , they could set the number of simulations for all steps except the last one to 0 , and specify a very large number of simulations for the final step . For each optimization step , HNN uses the COBYLA optimization algorithm ( Powell , 1994 ) , which supports bound constraints as defined by the user for each parameter . We have found COBYLA converges at a local minimum faster than the PRAXIS algorithm ( Brent , 1973 ) as implemented in NEURON’s multiple run fitter .	Neurons carry information in the form of electrical signals . Each of these signals is too weak to detect on its own . But the combined signals from large groups of neurons can be detected using techniques called EEG and MEG . Sensors on or near the scalp detect changes in the electrical activity of groups of neurons from one millisecond to the next . These recordings can also reveal changes in brain activity due to disease . But how do EEG/MEG signals relate to the activity of neural circuits ? While neuroscientists can rarely record electrical activity from inside the human brain , it is much easier to do so in other animals . Computer models can then compare these recordings from animals to the signals in human EEG/MEG to infer how the activity of neural circuits is changing . But building and interpreting these models requires advanced skills in mathematics and programming , which not all researchers possess . Neymotin et al . have therefore developed a user-friendly software platform that can help translate human EEG/MEG recordings into circuit-level activity . Known as the Human Neocortical Neurosolver , or HNN for short , the open-source tool enables users to develop and test hypotheses on the neural origin of EEG/MEG signals . The model simulates the electrical activity of cells in the outer layers of the human brain , the neocortex . By feeding human EEG/MEG data into the model , researchers can predict patterns of circuit-level activity that might have given rise to the EEG/MEG data . The HNN software includes tutorials and example datasets for commonly measured signals , including brain rhythms . It is free to use and can be installed on all major computer platforms or run online . HNN will help researchers and clinicians who wish to identify the neural origins of EEG/MEG signals in the healthy or diseased brain . Likewise , it will be useful to researchers studying brain activity in animals , who want to know how their findings might relate to human EEG/MEG signals . As HNN is suitable for users without training in computational neuroscience , it offers an accessible tool for discoveries in translational neuroscience .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "tools", "and", "resources", "neuroscience" ]	2020	Human Neocortical Neurosolver (HNN), a new software tool for interpreting the cellular and network origin of human MEG/EEG data
Maintenance of energy homeostasis depends on the highly regulated storage and release of triacylglycerol primarily in adipose tissue , and excessive storage is a feature of common metabolic disorders . CIDEA is a lipid droplet ( LD ) -protein enriched in brown adipocytes promoting the enlargement of LDs , which are dynamic , ubiquitous organelles specialized for storing neutral lipids . We demonstrate an essential role in this process for an amphipathic helix in CIDEA , which facilitates embedding in the LD phospholipid monolayer and binds phosphatidic acid ( PA ) . LD pairs are docked by CIDEA trans-complexes through contributions of the N-terminal domain and a C-terminal dimerization region . These complexes , enriched at the LD–LD contact site , interact with the cone-shaped phospholipid PA and likely increase phospholipid barrier permeability , promoting LD fusion by transference of lipids . This physiological process is essential in adipocyte differentiation as well as serving to facilitate the tight coupling of lipolysis and lipogenesis in activated brown fat . Evolutionary pressures for survival in fluctuating environments that expose organisms to times of both feast and famine have selected for the ability to efficiently store and release energy in the form of triacylglycerol ( TAG ) . However , excessive or defective lipid storage is a key feature of common diseases such as diabetes , atherosclerosis , and the metabolic syndrome ( Greenberg et al . , 2011 ) . The organelles that are essential for storing and mobilizing intracellular fat are lipid droplets ( LDs ) ( Walther and Farese , 2012 ) . They constitute a unique cellular structure where a core of neutral lipids is stabilized in the hydrophilic cytosol by a phospholipid monolayer embedding LD proteins . While most mammalian cells present small LDs ( <1 μm ) ( Suzuki et al . , 2011 ) , white ( unilocular ) adipocytes contain a single giant LD that occupies most of their cell volume . In contrast , brown ( multilocular ) adipocytes hold multiple LDs of smaller size that increase the LD surface/volume ratio , which facilitates the rapid consumption of lipids for adaptive thermogenesis ( Cinti , 2012 ) . The exploration of new approaches for the treatment of metabolic disorders has been stimulated by the rediscovery of active brown adipose tissue ( BAT ) in adult humans ( Virtanen et al . , 2009; Cypess and Kahn , 2010 ) and by the induction of multilocular brown-like cells in white adipose tissue ( WAT ) ( Harms and Seale , 2013 ) . The multilocular morphology of brown adipocytes is a defining characteristic of these cells along with expression of genes such as Ucp1 . The acquisition of a unilocular or multilocular phenotype is likely to be controlled by the regulation of LD growth . Two related proteins , CIDEA and CIDEC , promote LD enlargement in adipocytes ( Wu et al . , 2014; Puri et al . , 2007; 2008 ) , with CIDEA being specifically found in BAT . Together with CIDEB , they form the CIDE ( cell death-inducing DFF45-like effector ) family of LD proteins , which have emerged as important metabolic regulators ( Xu et al . , 2012 ) . Different mechanisms have been proposed for LD enlargement , including in situ neutral lipid synthesis , lipid uptake , and LD–LD coalescence ( Kuerschner et al . , 2008; Wilfling et al . , 2013; Boström et al . , 2007 ) . The study of CIDE proteins has revealed a critical role in the LD fusion process in which a donor LD progressively transfers its content to an acceptor LD until it is completely absorbed ( Gong et al . , 2011 ) . However , the underlying mechanism by which CIDEC and CIDEA facilitate the interchange of TAG molecules between LDs is not understood . In the present study , we have obtained a detailed picture of the different steps driving this LD enlargement process , which involves the stabilization of LD pairs , phospholipid binding , and the permeabilization of the LD monolayer to allow the transference of lipids . To examine the processes controlling LD enlargement in brown adipocytes , we followed LD dynamics using time-lapse microscopy . During differentiation of immortalized brown pre-adipocytes , large LDs were formed by the fusion of pre-existing LDs ( Video 1 ) . This fusion process was characterized by a slow and progressive reduction in the volume of a donor LD until it was completely absorbed by an acceptor LD ( Figure 1A ) , which is characteristic of CIDE activity . As CIDEA is selectively expressed in brown adipocytes and could have a prominent role in the acquisition of their multilocular morphology , we explored the effects of its expression in undifferentiated pre-adipocytes . After inducing CIDEA , LDs in pre-adipocytes showed an equivalent dynamic pattern to that observed in differentiating brown cells , with the progressive fusion of the initial LDs until a few large LDs remained in the cell ( Video 2 ) . LD fusion was achieved by the slow transference of lipids between LDs , and was preceded by the formation of small clusters of interacting LDs ( Figure 1B ) . Given the importance of this process in adipocyte dynamics , we decided to undertake a comprehensive molecular analysis . 10 . 7554/eLife . 07485 . 003Video 1 . Lipid droplet ( LD ) enlargement in differentiating immortalized brown adipose tissue ( imBAT ) cells . Immortalized brown pre-adipocytes were induced to differentiate by incubation for 48 hr + 6 hr with the described differentiation cocktails . The cell displays the characteristic LD enlargement pattern triggered by CIDE proteins , defined by the progressive fusion by lipid transference of the pre-existing LDs . DOI: http://dx . doi . org/10 . 7554/eLife . 07485 . 00310 . 7554/eLife . 07485 . 004Figure 1 . CIDEA promotes lipid droplet ( LD ) enlargement by transference of lipids . ( A ) Live imaging of the LD dynamics during the differentiation of a brown pre-adipocyte , showing the characteristic CIDE-triggered LD enlargement , characterized by the progressive transference of lipids from a donor to an acceptor LD until it is completely absorbed . ( B ) Live imaging of the LD dynamics in an undifferentiated 3T3-L1 cell 6 hr after infection with adenoviral particles carrying the Cidea gene . Red arrows highlight the transient formation of irregularly shaped LD clusters , while yellow arrows mark the fusion of two droplets by transference of lipids . ( C ) CIDEA-v5 expression in Hela cells induces LD enlargement . An enrichment in CIDEA-v5 ( red ) can be observed in the contact site between two LDs ( green ) . ( D ) Detail of LD fusion by slow transference of lipids in a 3T3-L1 cell stably expressing CIDEA-v5 . DOI: http://dx . doi . org/10 . 7554/eLife . 07485 . 00410 . 7554/eLife . 07485 . 005Video 2 . Lipid droplet ( LD ) enlargement induced by CIDEA . LD dynamics in undifferentiated 3T3-L1 cells 6 hr after infection with adenoviral particles carrying the mouse Cidea gene . After CIDEA induction , the initial individual LDs form stable contacts reflected by small irregularly shaped clusters of LDs . These interacting LDs undergo an enlargement process by lipid transference , characterized by the progressive enlargement of the acceptor LD and shrinkage of the donor LD until only a few large LDs remain in the cell . DOI: http://dx . doi . org/10 . 7554/eLife . 07485 . 005 The ectopic expression of full-length CIDEA induced the formation of large LDs through LD fusion by lipid transfer ( Figure 1C , D ) . Control cells , which lacked expression of CIDEA , did not show LD enlargement ( Figure 1C ) . As many proteins are constructed of domains , which are conserved across families and serve as their main structural and functional units , we assessed the conserved regions within the CIDE proteins . By comparing the 217 amino acid ( aa ) sequence of CIDEA with that of CIDEB and CIDEC , four highly conserved regions could be identified ( Figure 2A and Figure 2—figure supplement 1 ) . The N-terminal ( N-term ) domain of CIDEA is composed of a basic region ( 2–72 aa ) followed by an acidic sequence ( 73–110 aa ) . These distinctly differently charged regions are indicated by protein crystallography studies to be important for the dimerization of CIDE domain proteins ( Lugovskoy et al . , 1999; Wang et al . , 2012; Sun et al . , 2013; Lee et al . , 2013 ) . The CIDEA C-terminal ( C-term ) is rich in basic aas and contains a highly conserved region ( 126–155 aa ) and a basic and hydrophobic sequence ( 162–197 aa ) . Based on this sequence analysis , we created an extensive collection of v5-tagged CIDE point and deletion mutants to test their effects on LDs ( Figure 2 ) . Interestingly , certain mutations such as R171E/R175E promoted the aggregation of the cellular LDs in a few 'bunch of grapes'-like LD clusters , but were unable to induce the transference of lipids between them ( Figure 2B ) . In other cases , as with the expression of CIDEA- ( 116–217 ) -v5 , the LDs remained small and dispersed throughout the cytoplasm despite the protein being normally localized at their surface . Finally , some versions of CIDEA , such as CIDEA- ( 1–118 ) -v5 , showed no LD localization and did not affect their size , number , or distribution . Together with the time-lapse results , this indicates that the molecular mechanism of CIDEA is composed of three discrete phases: LD targeting , LD–LD docking , and LD growth . 10 . 7554/eLife . 07485 . 006Figure 2 . Mapping the functional domains of CIDEA . ( A ) Amino acid ( aa ) sequence of murine CIDEA highlighting the residues conserved in either CIDEB or CIDEC ( grey underline ) , or in both proteins ( black underline ) . The substituted aa in mutant constructs appear in red , and a positively charged sequence necessary for the TAG transfer step is encircled in orange . Four highly conserved regions are defined and symbolized by colour boxes in a linear representation of CIDEA-v5 . The theoretical isoelectric point of each fragment is indicated inside the boxes . ( B ) Representative images of the different phenotypes observed in Hela cells overexpressing mutated forms of CIDEA-v5 24 hr after transfection cells were treated with oleic acid and incubated for further 18 hr prior to fixation . Cells were classified into six major phenotypes . Cells expressing fully active forms of CIDEA had few and large LDs ( Type I ) . In some mutants , the large LDs remained attached to many small LDs , indicating that lipid transfer was inefficient or inactive for some LDs ( Type II ) . When CIDEA alterations blocked the lipid transfer process , the LDs remained small and grouped in a few large clusters ( Type III ) . If this was accompanied by inefficient LD–LD docking , the cells contained a number of small LD clusters combined with isolated LDs ( Type IV ) . The CIDEA forms that could not stabilize LD–LD interactions displayed a phenotype similar to the mock transfected cells , with most of the LDs dispersed through the cytoplasm ( Type V ) . Finally , some CIDEA constructs were unable to target the LDs , indicating an alteration of the LD-binding domain ( Type VI ) . ( C ) Morphologic distribution of cells expressing each of the studied CIDE constructs . The phenotypic distribution was performed in a minimum of three independent experiments for every construct ( n>50 cells ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07485 . 00610 . 7554/eLife . 07485 . 007Figure 2—figure supplement 1 . Alignment of amino acid sequences of CIDEA , CIDEB , and CIDEC . Clustal format alignment of murine CIDEA , CIDEB , and CIDEC , performed using T-COFFEE ( www . tcoffee . org ) . Amino acid sequences were obtained from http://ensembl . org . "" means residues are identical in all sequences in the alignment; ":" means conserved substitutions have been observed , " . " means that semi-conserved substitutions are observed . Four highly conserved regions are defined by colour boxes . DOI: http://dx . doi . org/10 . 7554/eLife . 07485 . 007 All the CIDEA constructs that showed impaired LD localization contained deletions or mutations in the C-term hydrophobic and basic region ( 162–197 aa ) ( Figure 2C ) . In fact , the last 66 aas of CIDEA were sufficient for LD localization , as shown with the expression of CIDEA- ( 152–217 ) -v5 , whereas it lacked the ability to facilitate the docking of LDs ( Figures 2C and 4A ) . Although it is known that the C-term domain of CIDE proteins is essential for LD localization and enlargement ( Liu et al . , 2009; Christianson et al . , 2010 ) , only the structure of the N-term domain has been solved ( Lugovskoy et al . , 1999; Wang et al . , 2012; Sun et al . , 2013; Lee et al . , 2013 ) . The CIDE-N domain ( Pfam reference PF02017 ) has been determined in members of the CIDE family ( PBD Codes: 2eel ( hCIDEA ) , 1D4B ( hCIDEB ) , 4MAC ( mCIDEC ) , 4ikg ( mCIDEC ) ) to aas 40–117 , 34–100 , and 41–118 in hCIDEA , hCIDEB , and mCIDEC , respectively . Thus , the sequence 163–180 that we found essential for LD targeting ( Figure 2C ) lacks direct structural information to date . We therefore predicted its structure using in silico approaches . The region displayed high probability of a helical conformation with a strongly amphipathic character ( Figure 3A , B , and Figure 3—figure supplement 1 ) . To experimentally confirm the presence of an amphipathic helix in the LD-targeting domain of CIDEA , circular dichroism ( CD ) spectroscopy was used to estimate the secondary structure of a synthetic peptide corresponding to residues 158–185 in CIDEA . The CD spectra in the presence of 0 . 1% n-dodecyl-β-D-maltopyranoside confirmed the presence of α-helical structure ( Figure 3C ) . 10 . 7554/eLife . 07485 . 008Figure 3 . CIDEA targets the LD monolayer through a cationic amphipathic helix . ( A ) Secondary structure of CIDEA predicted by SWISS-MODEL server . ( B ) Helical wheel representation of the putative amphipathic α-helix ( 163–180 ) generated at http://heliquest . ipmc . cnrs . fr/ . ( C ) Circular dichroism ( CD ) spectra of a 28-aa peptide corresponding to the 158–185 sequence of CIDEA ( 41 μM ) solubilized in 50 mM potassium phosphate , pH 6 . 2 plus 0 . 1% n-dodecyl-β-D-maltopyranoside . ( D ) A Hela cell expressing HA-CIDEA- ( 1–120 ) -v5 ( red ) or HA-CIDEA- ( 1–117 ) - ( 163–180 ) ( red ) , showing the inclusion of aas 163–180 enhances LD localization and the ability to promote LD docking . The phenotypic distribution was performed in a minimum of three independent experiments for every construct ( n>50 cells ) . HA signal in LDs was only detected in a proportion of the cells where HA-CIDEA constructs had induced LD enlargement or clustering , possibly due to the formation of CIDEA complexes reducing antibody accessibility to the HA epitope at the N-term . ( E ) A Hela cell expressing CIDEA- ( F166R/V169R/L170R ) -v5 ( red ) showing aa substitutions to compromise amphipathicity of the helix disrupt LD targeting , and a Hela cell expressing CIDEA- ( K167E/R171E/R175/E ) -v5 ( red ) showing amino acid ( aa ) substitutions to invert the charge of the helix but maintaining amphipathicity retains predominantly LD localization . DOI: http://dx . doi . org/10 . 7554/eLife . 07485 . 00810 . 7554/eLife . 07485 . 009Figure 3—figure supplement 1 . Conservation of amino acids ( aas ) for CIDEA amphipathic helix across vertebrate species . ( A ) Alignment of aa sequences for CIDEA over a range of vertebrate species corresponding to the amphipathic α-helix ( murine aas 163–180 ) . Amino acid sequences were obtained from http://ensembl . org and initially aligned using WebPRANK at http://ebi . ac . uk . Sequences were then grouped based on species phylogeny . ( B ) Helical wheel representation of the CIDEA putative amphipathic α-helix for indicated species ( corresponding to murine aas 163–180 ) generated at http://heliquest . ipmc . cnrs . fr/ . DOI: http://dx . doi . org/10 . 7554/eLife . 07485 . 00910 . 7554/eLife . 07485 . 010Figure 4 . Liquid droplet ( LD ) –LD docking and CIDEA interactions . ( A ) A Hela cell expressing CIDEA- ( 152–217 ) -v5 showing normal recruitment to LDs , but no LD docking . A Hela cell expressing CIDEA-△ ( 126–155 ) -v5 showing normal LD–LD docking but inefficient LD enlargement as revealed by the presence of clusters of small and large LDs . Representative images are shown of experiments performed in a minimum of three independent experiments for every construct ( n>50 cells ) . ( B ) Co-immunoprecipitation ( co-IP ) assays between CIDEA-HA and different CIDEA-v5 constructs . The observed CIDEA–CIDEA interaction was driven by the C-term domain and required the presence of the 126–155 aa sequence . ( C , D ) Co-IP assays showing CIDEA interactions with CIDEB , CIDEC , and PLIN5 . Each co-IP assay was performed at least in triplicate , producing similar results in each experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 07485 . 010 As some LD proteins are known to be bound to the LD membrane through amphipathic helices ( Hinson and Cresswell , 2009; Krahmer et al . , 2011 ) , we tested if this short sequence was sufficient for LD targeting . While HA- ( 1–120 ) -CIDEA showed no LD localization and had no effect on LD distribution or size , HA-CIDEA- ( 1–117 ) - ( 163–180 ) was partially localized on the LD surface and promoted LD clustering ( Figure 3D ) . Furthermore , the deletion of this C-term sequence in CIDEA-△ ( 163–179 ) -v5 completely eliminated LD localization in most of the cells ( Figure 2B , C ) , confirming its role in LD targeting . However , partial LD localization could be observed in a small percentage of cells , together with the presence of LD clusters . This was also observed in CIDEA-△ ( 162–197 ) -v5 and was particularly frequent in CIDEA- ( N172X ) -v5 , which contains a deletion in the middle of the helix . In contrast , no LD targeting could be observed for the N-term fragment alone ( 1–118 aa ) ( Figure 2C ) . This may indicate that other regions in C-term may contribute to LD localization either by directly binding the LD membrane or by interacting with other LD proteins . Similarly , LD targeting was compromised when the amphipathic character of the helix was disrupted in CIDEA- ( F166R/V169R/L170R ) -v5 by introducing cationic aas in its hydrophobic face . Although LD localization was only lost in a small percentage of cells , in the remaining cells the LD staining was accompanied by a predominantly cytosolic localization ( Figures 2C and 3E ) . In contrast , the predominant LD localization of wild type ( wt ) CIDEA was maintained in CIDEA- ( K167E/R171E/R175E ) -v5 , which presents a charge inversion of the helix but maintains its amphipathic properties ( Figures 2C and 3E ) . In addition to its role in LD targeting , our data indicate that the cationic amphipathic helix in the C-term participates in the TAG transference step of CIDEA activity , as the charge inversion ( K167E/R171E/R175E ) did not affect LD targeting but completely blocked LD enlargement ( Figure 2C ) . Despite not being essential for LD targeting , the cationic aas in the helix are highly conserved in vertebrates Figure 3—figure supplement 1A . K167 is 100% conserved across all vertebrate species examined . R171 was conserved across vertebrates including birds , snakes , lizards , crocodiles , turtles , marsupials , placental mammals , and monotremes , although not in fish . R175 is also highly conserved , with only birds , dolphins , and the Nile Tilapia ( a fish ) lacking this residue . Remarkably , an amphipathic helix is predicted in CIDEA of all the vertebrate species examined ( Figure 3—figure supplement 1B ) . The absence of negative charges in the helix appeared to be an essential condition to permit TAG transference , as a single inverted charge mutation such as R171E or R175E was sufficient to block LD enlargement ( Figure 2C ) . In contrast , conservative substitutions such as R171K or K167R did not affect CIDEA activity , and even the substitution of the three basic aas with histidine in ( K167H/R171H/R175H ) -CIDEA-v5 was compatible with the formation of large LDs , strongly supporting the conclusion from sequence comparison that positive charges are required at these positions . As histidine has a lower pKa value than arginine and lysine , it can carry a positive charge depending on the pH and local environment , which could explain the activity retained by this protein . Deletions in the N-term domain of CIDEA impaired LD–LD docking , as shown by the increase in cells displaying isolated LDs ( Figure 2C ) . Furthermore , LD clustering could be induced by forcing the LD localization of the N-term fragment through conjugation with the 18-aa amphipathic helix ( HA-CIDEA- ( 1–117 ) - ( 162–180 ) ) ( Figure 3C ) . As the N-term of CIDEA forms a highly polarized structure that is prone to dimerize ( Lugovskoy et al . , 1999 ) , we hypothesized that LD–LD docking was induced by the N-term–N-term interaction of CIDEA molecules in adjacent LDs ( trans complexes ) . However , the C-term fragment ( 116–217 ) retained some degree of LD–LD docking activity ( Figure 2C ) , indicating that an additional interaction site could be present in this region . In fact , a complete blocking of LD clustering was only observed with the shorter fragment 152–217 , which lacks the N-term and a section of the C-term ( Figure 4A ) . The formation of CIDEA–CIDEA complexes was confirmed by co-immunoprecipitation ( co-IP ) of CIDEA-v5 with CIDEA-HA . Surprisingly , co-IP was observed with CIDEA- ( 116–217 ) -v5 but not CIDEA- ( 1–118 ) -v5 , indicating that the C-term was responsible for that interaction ( Figure 4B ) . Interestingly , a similar percentage of the input was co-immunoprecipitated for constructs producing highly clustered LDs ( CIDEA- ( R171E/R175E ) -v5 ) and constructs showing few LD–LD contacts ( CIDEA-v5 or CIDEA- ( 116–217 ) -v5 ) . Hence , this C-term interaction is largely independent of the presence of LD–LD contacts , indicating that it may also occur in cis . Within the C-term region , the deletion of the 162–197 sequence did not affect the co-IP whereas the signal was largely reduced in CIDEA-△ ( 126–155 ) -v5 ( Figure 4B ) , indicating that this conserved region was involved in the C-term interaction . However , the residual interaction still detectable by co-IP could sustain the LD-docking activity , as cells expressing this construct displayed normal LD clustering ( Figure 4A , B ) . CIDEA- ( 152–217 ) -v5 ( Figures 2 and 4A ) , which showed no LD clustering and lacked both the 126–155 interaction site and the N-term domain , displayed a further reduction on the co-IP signal ( Figure 4B ) . Therefore , trans complexes through N-term dimerization would be responsible for the LD clusters and weak co-IP signal observed in CIDEA- ( 126–155 ) -v5 . The lack of co-IP between the N-term fragment and the full-length CIDEA could be due to conformational and positional factors favouring the interaction between the HA-tagged full-length proteins in the LD or between the cytosolic v5-tagged N-term fragments . In fact , co-IP between N-term fragments of CIDEC was previously reported ( Sun et al . , 2013 ) . This interaction could be disrupted with the point mutations E87Q/D88N or R55E as predicted by the crystal structure of the N-term fragment , which reveals the formation of homodimers in which the positively charged R46 , R55 , and R56 in one molecule interact with negative residues in the other ( E87 and D88 ) ( Sun et al . , 2013 ) . Interestingly , we found that the equivalent mutations in CIDEA ( E79Q/D80N and R47E ) impaired LD docking , while R47Q and R47A , which would not create repulsions between the interacting domains , did not affect CIDEA activity ( Figure 2C ) . Taken together , these results suggest that both the C-term dimerization site ( 126–155 ) and the N-term domain of CIDEA can contribute to LD–LD docking by forming complexes with its counterparts on the adjacent LD . The differential expression of CIDEA and CIDEC in BAT and WAT could be related to the acquisition of multilocular or unilocular morphologies in brown and white adipocytes ( Barneda et al . , 2013 ) . While the ectopic expression of both CIDEA and CIDEC produce LD enlargement in a similar manner , specific differences in their activity and regulation could achieve discrete outcomes . In fact , whereas deletion of the N-term domain of CIDEA blocks LD enlargement ( Figure 2C ) , it has been described that the C-term fragment of CIDEC retains its activity ( Gong et al . , 2011; Jambunathan et al . , 2011 ) . Here we show that similar to CIDEA , the N-term of CIDEC is involved in LD–LD docking , as its deletion increases the fraction of cells displaying isolated LDs ( Figure 2C ) . However , in the cells where the C-term of CIDEC could effectively induce LD–LD docking , large LDs were observed instead of LD clusters , showing that although its docking efficiency is reduced , this region of CIDEC is sufficient for docking and enlarging the LDs . This differs from CIDEA , in which the C-term fragment cannot induce LD enlargement despite retaining a partial LD docking activity . In addition to the intrinsic differences between CIDEC and CIDEA , their activity could be affected by the interaction with additional proteins . While PLIN1 interacts with CIDEC , but not CIDEA , and potentiates its activity ( Sun et al . , 2013; Grahn et al . , 2013 ) , we have observed that CIDEA interacts with PLIN5 ( Figure 4D ) , which is rich in BAT ( Harms and Seale , 2013; Zhou et al . , 2003 ) . In addition to PLIN5 , CIDEA showed high affinity for both CIDEB and CIDEC , while it did not co-IP with DFF40 or DFF45 , which share homology with the N-term domain of CIDE proteins ( Figure 4C ) . As BAT cells express high levels of both CIDEA and CIDEC , the formation of CIDE heterocomplexes could be involved in the regulation of LD enlargement to retain the multilocular state . To further characterize the interaction of CIDEA with the LD membrane , we utilized lipid strips to investigate the affinity of CIDEA for different lipids present in mammalian cell membranes and found that it selectively bound a set of anionic phospholipids ( Figure 5A ) . The interaction with phosphatidic acid ( PA ) was of particular interest , as increased levels of this phospholipid have been linked with LD fusion ( Fei et al . , 2011 ) and the identification of enzymes such as AGPAT3 and LIPIN-1γ in LDs supports the existence of in situ generation and consumption of PA ( Wilfling et al . , 2013; Wang et al . , 2011 ) . PA binding was confirmed by the strong affinity of CIDEA-v5 for PA beads ( Figure 5B ) , which was greatly reduced by pre-incubation of the lysate with soluble PA , but not phosphatidylcholine ( PC ) . Although the N-term fragment showed some residual affinity , the main PA-binding site of CIDEA was in the C-term region containing the amphipathic helix ( 163–180 ) ( Figure 5C ) . The charge inversion of its three cationic amino acids resulted in the loss of affinity for PA beads in the inactive mutant ( K167E/R171E/R175E ) -CIDEA-v5 without affecting its LD localization ( Figure 5C ) , linking PA binding with the TAG-transference step ( Figure 5D ) . 10 . 7554/eLife . 07485 . 011Figure 5 . CIDEA is a phosphatidic acid ( PA ) -binding protein . ( A ) Lipid strip assay showing the affinity of CIDEA-v5 for certain anionic phospholipids . ( B ) Interaction of CIDEA-v5 with PA beads . Binding was reduced by pre-incubation of the lysate with soluble PA , but not phosphatidylcholine ( PC ) . ( C ) The affinity for PA beads was highly reduced in CIDEA-v5 constructs with alterations in its C-term hydrophobic and basic region ( 162–197 ) . ( D ) CIDEA- ( K167E/R171E/R175E ) -v5 localizes to LDs and induces their clustering but cannot promote their enlargement by lipid transfer . Representative images are shown of experiments performed in a minimum of three independent experiments for every construct ( n>50 cells ) . ( E ) Circular dichroism spectra of the synthetic wild type ( wt ) or mutant ( F166R/V169R/L170R ) CIDEA peptides encompassing aas 158–185 solubilized in 25 mM sodium phosphate ( pH 7 . 2 ) at concentrations of 70 μM ( wt ) and 47 μM ( mutant ) . Peptide samples were prepared in the absence and presence of increasing amounts of DPLC or DLPC:DLPA ( 9:1 molar ratio ) . ( F–H ) Coarse-grained molecular dynamics ( CG-MD ) simulations of peptide interactions with LDs ( PC: hydrophobic chains , transparent blue , polar heads , opaque blue; TAG: hydrophobic chains , dark brown , glycerol chain , light brown; PA: hydrophobic chains , transparent red , polar heads , opaque red; peptides: yellow , with cationic aa in blue and anionic in red ) . ( F ) Selected time points of the wt helix simulation with PC-LDs . At 124 ns the helix initiates the contact through its hydrophobic face , being rapidly embedded in the phospholipid monolayer . TAG molecules can abandon the neutral lipid core and are integrated in the hydrophobic region of the phospholipid monolayer . ( G ) Distance between the peptide and LD centre of mass ( COM ) versus time for the different helices with a PC-LD and a PC:PA-LD . The dashed line represents the approximate location of LD phospholipid head groups . ( H ) Different views of the configuration of the LD helix at the end of the simulations . Interaction between the polar head of PA and the helix can be observed for the wt and F166R/V169R/L170R but not K167E/R171E/R175E . ( I–K ) Comparison of full-length hCIDEC-v5 and the lipodystrophy-associated truncation hCIDEC- ( E186X ) -v5 , including LD localization and morphology ( I ) , co-IP with CIDEA-HA ( J ) , and affinity for PA beads ( K ) . Each co-IP , PA-binding assay , and lipid strip assay was performed at least in triplicate , producing similar results in each experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 07485 . 01110 . 7554/eLife . 07485 . 012Figure 5—figure supplement 1 . Secondary structure determination of CIDEA amino acids 158–185 ( wild type and F166R/V169R/L170R ) by CDPro DATABASE 4 ( 43 soluble proteins ) using the CONTINLL program . DOI: http://dx . doi . org/10 . 7554/eLife . 07485 . 01210 . 7554/eLife . 07485 . 013Figure 5—figure supplement 2 . Computational prediction of the amphipathic helix and LD interactions . CG-MD simulations of peptide interactions with LDs ( PC: hydrophobic chains , transparent blue , polar heads , opaque blue; TAG: hydrophobic chains , dark brown , glycerol chain , light brown; PA: hydrophobic chains , transparent red , polar heads , opaque red; peptides: yellow , with cationic aa in blue and anionic in red ) . Selected time points of the wt helix and K167E/R171E/R175E helix simulation with PC-LDs or PC:PA-LDs are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 07485 . 01310 . 7554/eLife . 07485 . 014Figure 5—figure supplement 3 . TAG infiltration into the phospholipid monolayer . ( A ) 2D number density maps of TAG molecules along the z direction from the CG-MD simulations . TAG frequency was measured over 76 ns before and after the docking of the wt helix to the membrane . The differential distribution before and after docking reveals an increase in the presence of TAG molecules in the membrane , which was revealed by the red spots in the outer arch and blue in the inner . ( B ) CG-MD simulations of peptide interactions with LDs ( PC: hydrophobic chains , transparent blue , polar heads , opaque blue; TAG: hydrophobic chains , dark brown , glycerol chain , light brown; PA: hydrophobic chains , transparent red , polar heads , opaque red; peptides: yellow , with cationic aa in blue and anionic in red ) . Selected time points of the wt helix with PC-LD show TAG molecules in the LD membrane were increased after CIDEA peptide docking ( 170 vs 24 ns ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07485 . 01410 . 7554/eLife . 07485 . 015Figure 5—figure supplement 4 . Computational prediction of PA docking to the amphipathic helix structure of CIDEA . ( A ) Charge-smoothed potential of the predicted putative amphipathic helix of murine CIDEA ( aas 163–180 ) . Stick representation has been employed to highlight the positively charged residues K167 , R171 , and R175 . ( B ) The nine top-ranked models from molecular docking of PA and the amphipathic helix using Autodock Vina . Interaction of the PA phosphate group with R171 and R175 is observed in eight of them , while the hydrophobic chains of the phospholipid can adopt multiple configurations . DOI: http://dx . doi . org/10 . 7554/eLife . 07485 . 015 To investigate if PA affects the structure of the amphipathic helix , we repeated the CD analyses in phosphate buffer in the presence and absence of DLPC lipid vesicles with and without DLPA . Fitting of the CD data suggested a low ( ~5% ) helical content for the wt peptide when analysed in phosphate buffer alone , and indicated a predominantly sheet/coil structure in the absence of detergent or liposomes . The presence of DLPC liposomes stabilized a sharp increase in α-helical structure ( up to 40% ) and an equivalent reduction of sheet content ( Figure 5E and Figure 5—figure supplement 1 ) , yielding higher helical content than that observed in n-dodecyl-β-D-maltopyranoside micelles ( ~25% , Figure 3C ) . In contrast , the induction of helix formation by DLPC was not observed in a mutant peptide carrying the substitutions impairing LD targeting in CIDEA- ( F166R/V169R/L170R ) -v5 ( Figure 5E ) . This mutant peptide remained predominantly random coil in the absence and presence of DLPC liposomes . Interestingly , fitting of the CD data indicated significant helical content for both the wt and mutant peptides in the presence of DLPC:DLPA ( 9:1 ) vesicles ( Figure 5E and Figure 5—figure supplement 1 ) . This indicates that the interaction with the negatively charged phospholipid PA can compensate for the excess of positive charges in CIDEA- ( F166R/V169R/L170R ) -v5 . To obtain more detailed insight into the interaction of the amphipathic helix with phospholipids and the role of PA in this process , the interaction of the amphipathic helix with LDs was modelled using coarse-grained molecular dynamics ( CG-MD ) simulations ( Figure 5F–H ) . CG-MD simulations are well established for lipid-containing systems ( Marrink et al . , 2007 ) , including LDs ( Mohammadyani et al . , 2014 ) , and have the advantage over full atomistic simulations in that the time scales required are much smaller allowing us to compare different LD compositions and helix mutants in the large multimolecular LD system . The wt ( 163–180 ) helix ( CTSFKAVLRNLLRFMSYA ) diffused towards the LD containing 400 palmitoyl-oleoyl-glycero-phosphocholine ( POPC ) and 200 TAG molecules where it interacted at its full length with the LD surface and penetrated into the hydrophobic region of the phospholipid monolayer covering the TAG core ( Figure 5F , G ) . A similar behaviour was observed by the charge-inverted mutant K167E/R171E/R175E ( Figure 5G and Figure 5—figure supplement 2 ) , supporting the experimental result that these mutations do not affect LD localization in CIDEA ( Figures 2C and 3E ) . In contrast , no interaction with the LD was observed with the non-amphipathic F166R/L169R/V169R ( Figure 5G ) , which also impairs LD binding in CIDEA ( Figures 2C and 3E ) and which was unable to attain stable secondary structure as evidenced by CD ( Figure 5E ) . Similarly , a non-amphipathic α-helix in N-term ( SSLQELISKTLDVLVITT ) also showed no interaction with the LD ( Figure 5G ) . To study the effect of PA on LD structure and interaction with the CIDEA helix , we replaced 10% of the PC molecules with PA . The equilibration of the system resulted in a slight deformation of the spherical shape of the LD ( Figure 5—figure supplement 2 ) . The wt helix made a stable complex with this LD at even earlier simulation times than with the PC-only containing LDs ( Figure 5G and Figure 5—figure supplement 2 ) . The triple-E replacement mutant was also able to bind this LD , and even the F166R/L169R/V169R mutant was now able to interact with the membrane ( Figure 5G ) . This result fits well with the CD results where the addition of PA also rescued the helix induction in this mutant peptide through interaction with the liposomes ( Figure 5E ) . Interestingly , while the presence of PA permitted the accommodation of the F166R/L169R/V169R helix in the LD monolayer , it could not penetrate as deep towards the TAG core as the wt or helix . The average distance of the peptide to the centre of the LD ( ± SEM ) was 5 . 6 ± 0 . 04 nm and 5 . 7 ± 0 . 03 nm for wt and K167E/R171E/R175E , respectively . In the presence of PA , the distance was 5 . 6 ± 0 . 02 nm , 5 . 7 ± 0 . 02 nm , and 6 . 1 ± 0 . 02 for wt , K167E/R171E/R175E , and F166R/L169R/V169R , respectively ( also see Figure 5G , H ) . This result confirms that the presence of the hydrophobic face was necessary for proper helix insertion in the LD monolayer . The CG-MD simulations not only shed light on the interaction between the helix and the LD , but also provided an indication of the mechanism by which this process could lead to LD enlargement by TAG transference . We observed that TAG molecules were able to escape the LD core and were integrated in the hydrophobic section of the membrane ( Figure 5H ) . This TAG infiltration was increased after the docking of the wt helix in the membrane ( Figure 5—figure supplement 3 ) , suggesting that CIDEA could promote the migration of TAG into the membrane as an intermediate state prior to the transference to the acceptor LD . To complete the transference , the hydrophobic TAG molecules should overcome the energy barrier constituted by the phospholipid polar heads and water molecules in the LD–LD interface . Interestingly , we observed that the wt helix could attract PA molecules in its vicinity by the interaction of its cationic residues with the negatively charged polar head of PA ( Figure 5H ) . A direct interaction of the amphipathic helix with PA was also indicated by molecular docking using Autodock Vina , which supported the role of the K167 , R171 , and R175 residues in the interaction ( Figure 5—figure supplement 4 ) . Remarkably , CG-MD simulations showed that whereas the non-amphipathic cationic helix F166R/L169R/V169R also interacted with PA molecules from its superficial docking position in the LD membrane , the anionic amphipathic mutant K167E/R171E/R175E was docked in a PA-depleted area and avoided the PA molecules ( Figure 5H ) . Although TAG infiltration was also observed in this simulation and the helix was well embedded in the membrane , its inability to attract PA molecules could be responsible for the lack of TAG transference activity in CIDEA- ( K167E/R171E/R175 ) -v5 ( Figure 5D ) . Taken together , these results indicate that CIDEA binds the LD by embedding a cationic amphipathic helix into the LD monolayer and that once there , it can interact with PA molecules , which could facilitate TAG transference . We found that PA binding was a feature common to all three members of the CIDE protein family ( Figure 5C ) . Intriguingly , we determined that an inactive CIDEC identified in a patient with lipodystrophy ( Rubio-Cabezas et al . , 2009 ) contained a truncation ( E186X ) in the predicted PA-binding site . Although hCIDEC- ( E186X ) -v5 and the equivalent mCIDEA- ( N172X ) -v5 were localized in LDs in a high percentage of cells , they were completely unable to induce LD enlargement ( Figures 2C and 5I ) . LD clustering activity and its ability to interact with CIDE proteins was not altered in hCIDEC- ( E186X ) -v5 ( Figure 5I , J ) , but it showed no affinity for PA ( Figure 5K ) . Thus , PA binding could be involved in the lipid transfer phase of CIDE activity . To confirm the requirement of PA binding , we examined the effect of PA depletion on CIDEA activity . While substantial alterations in the phospholipid composition of mammalian cells often compromise their viability , yeast cells offer a wide range of genetically modified strains with well-characterized alterations in phospholipid metabolism ( Figure 6A ) ( Henry et al . , 2012 ) . Thus , despite the absence of CIDE homologues in yeast ( Wu et al . , 2008 ) , we explored the functionality of CIDEA in wt and genetically modified strains of Saccharomyces cerevisiae ( Figure 6—source data 1 ) . 10 . 7554/eLife . 07485 . 016Figure 6 . CIDEA is functional in yeast and requires PA . ( A ) Pathway showing the reactions catalyzed by the enzymes altered in the studied yeast strains . ( B ) Stable expression of mCIDEA-v5 in three transformed yeast clones . ( C ) Frequency distribution of the diameter of the largest LD per cell . ( D ) LD staining in the studied yeast strains transformed with pRS316-CYC1p-Cidea or the empty vector . ( E , F ) Quantification of LD size and number per cell in randomly acquired images ( 100–200 cells/condition ) . CIDEA activity in yeast was measured by its ability to increase the percentage of cells with supersized LDs ( E ) and reduce the total number of LDs per cell ( F ) . ( G , H ) Effect of CIDEA and PAH1-7A expression in the cellular levels of PA ( G ) and its synthesis rate ( H ) . Three different yeast clones per condition were analysed , and results are shown as the mean ± SEM . One-way ANOVA with Bonferroni post-test was performed to determine significant differences due to the presence of CIDEA ( p<0 . 05; ***p<0 . 001 ) . ( I–L ) Coexpression of hLIPIN-1γ-v5 and CIDEA-HA in Hela cells . ( I ) Representative immunofluorescence images showing LD staining ( blue ) in Hela cells expressing CIDEA-HA ( green ) in the presence or absence of hLIPIN-1γ-v5 ( red ) . Twenty-four hours after overexpression of hLIPIN-1γ-v5 , cells were transfected with pcDNA3 . 1/Cidea-HA and incubated for a further 24 hr . ( J ) Phenotypic distribution in randomly selected cells ( n>50 ) showing the average values for three independent experiments . ( K ) Co-IP assay in lysates of transfected Hela cells . ( L ) PA beads binding assay for hLIPIN-1γ-v5 . Each co-IP and PA-binding assay was performed at least in triplicate , producing similar results in each experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 07485 . 01610 . 7554/eLife . 07485 . 017Figure 6—source data 1 . List of yeast strains used in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 07485 . 017 Murine CIDEA could be stably expressed in yeast cells ( Figure 6B ) , producing an increase in the size of their LDs ( Figure 6C ) . Yeast cells expressing wt CIDEA , but not the inactive R171E/R175E mutant , contained fewer and larger LDs than the control ( Figure 6D–F ) , indicating that murine CIDEA was functional in these cells . By measuring the frequency of supersized LDs ( diameter above 0 . 5 μm ) and the total number of LDs in strains with altered lipid metabolism , we could determine the yeast strains in which CIDEA was able to induce LD enlargement ( Figure 6E , F ) . CIDEA was inactive in cells defective in phospholipase D ( pld1△ ) ( Rose et al . , 1995 ) , which catalyzes the production of PA from PC . CIDEA activity was also abrogated in cells expressing a hyperactive form of the PA phosphohydrolase ( PAH1-7A ) ( Choi et al . , 2010; Choi et al . , 2012 ) . These results indicate that PA is necessary for CIDEA activity . In addition , we observed that total cellular PA levels were increased by CIDEA , an effect that was prevented by the expression of PAH1-7A ( Figure 6G ) . As the PA synthesis rate was not affected ( Figure 6H ) , CIDEA could be protecting a pool of PA from degradation . The deletion of diacylglycerol kinase ( dgk1△ ) ( Han et al . , 2008 ) did not affect CIDEA activity . DGK1 is important for the generation of phospholipids from TAG as cells exit from stasis ( Fakas et al . , 2010 ) , but its deletion has not been shown to have a great effect on PA levels under normal growth conditions . Regarding PAH1 , its deletion produces dramatic cellular effects ( Santos-Rosa et al . , 2005 ) , including defective LD formation ( Adeyo et al . , 2011; Fakas et al . , 2011 ) . As this alteration in LDs can be compensated by the deletion of DGK1 , we chose to use the dgk1△pah1△ strain , observing normal LD enlargement by CIDEA ( Figure 6E , F ) . CIDEA was also able to further increase the LD size in the cho2△ strain , which lacks the phosphatidylethanolamine ( PE ) methylation pathway for PC synthesis , and has been shown to present supersized LDs and high levels of PA and PE ( Fei et al . , 2011 ) . The CIDEA-induced LD enlargement in yeast was not due to a mere coating effect protecting LDs against lipases , as it was functional in the tgl3△tgl4△tgl5△ strain , which lacks lipase activity . As expected , CIDEA could not induce the appearance of LDs in dga1△lro1△are1△are2△ cells ( Figure 6D ) , which are deficient in the enzymes required for TAG and steryl ester synthesis and contain no LDs ( Sandager et al . , 2002 ) . To study the role of PA-dependent CIDEA action in mammalian cells without compromising other PA-dependent cellular processes , we specifically degraded this phospholipid in LDs by overexpressing a LD-localized isoform of PA phosphohydrolase ( LIPIN-1γ ) ( Wang et al . , 2011; Han and Carman , 2010 ) . While CIDEA-HA displayed normal activity in cells co-transfected with an empty vector , it was unable to promote LD enlargement in cells expressing LIPIN-1γ-v5 ( Figure 6I , J ) . LIPIN-1γ-v5 showed affinity for PA but it did not co-IP with CIDEA , indicating that its inhibitory effects were not due to a direct interaction between these proteins ( Figure 6K , L ) . Taken together , these results reveal that the mechanism of action of CIDEA involves direct interaction with PA molecules in the LD monolayer . The Cidea gene is highly expressed in BAT , induced in WAT following cold exposure ( Rosell et al . , 2014 ) , and is widely used by researchers as a defining marker to discriminate brown or brite adipocytes from white adipocytes ( Harms and Seale , 2013; Zhou et al . , 2003 ) . As evidence indicated a key role in the LD biology ( Hallberg et al . , 2008 ) , we have characterized the mechanism by which CIDEA promotes LD enlargement , which involves the targeting of LDs , the docking of LD pairs , and the transference of lipids between them . The lipid transfer step requires the interaction of CIDEA and PA through a cationic amphipathic helix . Independently of PA binding , this helix is also responsible for anchoring CIDEA in the LD membrane . Finally , we demonstrate that the docking of LD pairs is driven by the formation of CIDEA complexes involving the N-term domain and a C-term interaction site . CIDE proteins appeared during vertebrate evolution by the combination of an ancestor N-term domain and a LD-binding C-term domain ( Wu et al . , 2008 ) . In spite of this , the full process of LD enlargement can be induced in yeast by the sole exogenous expression of CIDEA , indicating that in contrast to SNARE-triggered vesicle fusion , LD fusion by lipid transference does not require the coordination of multiple specific proteins ( Risselada and Grubmuller , 2012 ) . While vesicle fusion implies an intricate restructuring of the phospholipid bilayers , LD fusion is a spontaneous process that the cell has to prevent by tightly controlling their phospholipid composition ( Krahmer et al . , 2011 ) . However , although phospholipid-modifying enzymes have been linked with the biogenesis of LDs ( Gubern et al . , 2008; Andersson et al . , 2006 ) , the implication of phospholipids in physiologic LD fusion processes has not been previously described . Complete LD fusion by lipid transfer can last several hours , during which the participating LDs remain in contact . Our results indicate that both the N-term domain and a C-term dimerization site ( aa 126–155 ) independently participate in the docking of LD pairs by forming trans interactions ( Figure 7 ) . Certain mutations in the dimerization sites that do not eliminate the interaction result in a decrease of the TAG transference efficiency , reflected by the presence of small LDs docked to enlarged LDs . This suggests that in addition to stabilizing the LD–LD interaction , the correct conformation of the CIDEA complexes is necessary for optimal TAG transfer . Furthermore , the formation of stable LD pairs is not sufficient to trigger LD fusion by lipid transfer . In fact , although LDs can be tightly packed in cultured adipocytes , no TAG transference across neighbour LDs is observed in the absence of CIDE proteins ( Gong et al . , 2011 ) , showing that the phospholipid monolayer acts as a barrier impermeable to TAG . Our CG-MD simulations indicate that certain TAG molecules can escape the neutral lipid core of the LD and be integrated within the aliphatic chains of the phospholipid monolayer . This could be a transition state prior to the TAG transference , and our data indicate that the docking of the amphipathic helix in the LD membrane could facilitate this process . However , the infiltrated TAGs in LD membranes in the presence of mutant helices , or even in the absence of docking , suggests that this is not enough to complete the TAG transference . 10 . 7554/eLife . 07485 . 018Figure 7 . Proposed molecular mechanism . ( A ) CIDEA targets the LD through its C-term amphipathic helix and once diffused to the LD surface , it forms cis CIDEA complexes by interacting through the C-term ( 126–155 ) region . When two CIDEA-containing LDs make contact , trans interactions between CIDEA molecules in each droplet can be established , which will facilitate the docking of the LDs . Both the N-term and the C-term would contribute by dimerizing with their counterparts of the neighbour LD . This trans interaction will anchor the CIDEA complexes in the LD–LD contact site , promoting a local enrichment of CIDEA . The monolayers of the two LDs will be maintained at short distance by the CIDEA complex . The amphipathic helices , embedded in the hydrophobic region of the membrane , will interact with the cone-shaped PA , creating a local perturbation in the phospholipid barrier that will increase its permeability to TAG . ( B ) The docking of the amphipathic helix to the membrane could facilitate the integration of TAG molecules within the phospholipid hydrophobic tails . Although the helix will be stabilized with its cationic residues pointing outwards , it will interact with PA molecules in its vicinity , which could be pulled out of the monolayer by the helix molecular dynamics . This could create a transitory discontinuity in the polar barrier that will reduce the energy required to transfer the TAG molecules present in the membrane . This alteration , together with the microenvironment created by the CIDEA complex , will reduce the energy barrier necessary to transfer TAG molecules between LDs , allowing the LD growth by lipid transference . DOI: http://dx . doi . org/10 . 7554/eLife . 07485 . 018 To be transferred to the adjacent LD , the TAGs integrated in the hydrophobic region of the LD membrane should cross the energy barrier defined by the phospholipid polar heads , and the interaction of CIDEA with PA could play a role in this process , as suggested by the disruption of LD enlargement by the mutations preventing PA binding ( K167E/R171E/R175E ) and the inhibition of CIDEA after PA depletion . The minor effects observed with more conservative substitutions in the helix suggest that the presence of positive charges is sufficient to induce TAG transference by attracting anionic phospholipids present in the LD membrane . PA , whose requirement is indicated by our PA-depletion experiments , is a cone-shaped anionic phospholipid that could locally destabilize the LD monolayer by favouring a negative membrane curvature that is incompatible with the spherical LD morphology ( Kooijman et al . , 2005 ) . Interestingly , while the zwitterion PC , the main component of the monolayer , stabilizes the LD structure ( Krahmer et al . , 2011 ) , the negatively charged PA promotes their coalescence ( Fei et al . , 2011 ) . This is supported by our CG-MD results which resulted in a deformation of the LD shape by the addition of PA . We propose a model in which the C-term amphipathic helix positions itself in the LD monolayer and interacts with PA molecules in its vicinity , which might include trans interactions with PA in the adjacent LD . The interaction with PA disturbs the integrity of the phospholipid barrier at the LD–LD interface , allowing the LD to LD transference of TAG molecules integrated in the LD membrane ( Figure 7 ) . Additional alterations in the LD composition could facilitate TAG transference , as differentiating adipocytes experience a reduction in saturated fatty acids in the LD phospholipids ( Arisawa et al . , 2013 ) , and in their PC/PE ratio ( Hörl et al . , 2011 ) , which could increase the permeability of the LD membranes; we previously observed that a change in the molecular structures of TAG results in an altered migration pattern to the LD surface ( Mohammadyani et al . , 2014 ) . During LD fusion by lipid transfer , the pressure gradient experienced by LDs favours TAG flux from small to large LDs ( Gong et al . , 2011 ) . However , the implication of PA , a minor component of the LD membrane , could represent a control mechanism , as it is plausible that the cell could actively influence the TAG flux direction by differently regulating the levels of PA in large and small LDs , which could be controlled by the activity of enzymes such as AGPAT3 and LIPIN-1γ ( Wilfling et al . , 2013; Wang et al . , 2011 ) . This is a remarkable possibility , as a switch in the favoured TAG flux direction could promote the acquisition of a multilocular phenotype and facilitate the browning of WAT ( Barneda et al . , 2013 ) . Interestingly , Cidea mRNA is the LD protein-encoding transcript that experiences the greatest increase during the cold-induced process by which multilocular BAT-like cells appear in WAT ( Barneda et al . , 2013 ) . Furthermore , in BAT , cold exposure instigates a profound increase in CIDEA protein levels that is independent of transcriptional regulation ( Yu et al . , 2015 ) . The profound increase in CIDEA is coincident with elevated lipolysis and de novo lipogenesis that occurs in both brown and white adipose tissues after β-adrenergic receptor activation ( Mottillo et al . , 2014 ) . It is likely that CIDEA has a central role in coupling these processes to package newly synthesized TAG in LDs for subsequent lipolysis and fatty acid oxidation . Importantly , BAT displays high levels of glycerol kinase activity ( Bertin , 1976; Bertin et al . , 1984 ) that facilitates glycerol recycling rather than release into the blood stream , following induction of lipolysis ( Portet et al . , 1974 ) , which occurs in WAT . Hence , the reported elevated glycerol released from cells depleted of CIDEA ( Zhou et al . , 2003 ) is likely to be a result of decoupling lipolysis from the ability to efficiently store the products of lipogenesis in LDs , therefore producing a net increase in detected extracellular glycerol . This important role of CIDEA is supported by the marked depletion of TAG in the BAT of Cidea-null mice following overnight exposure to a temperature of 4°C ( Zhou et al . , 2003 ) and by our finding that CIDEA-dependent LD enlargement is maintained in a lipase-negative yeast strain . Cidea and the genes that are required to facilitate high rates of lipolysis and lipogenesis are associated with the 'browning' of white fat either following cold exposure ( Rosell et al . , 2014 ) or in genetic models such as RIP140 knockout WAT ( Kiskinis et al . , 2014 ) . The induction of a brown-like phenotype in WAT has potential benefits in the treatment and prevention of metabolic disorders ( Whittle et al . , 2013 ) . Differences in the activity and regulation of CIDEC and CIDEA could also be responsible for the adoption of unilocular or multilocular phenotypes . In addition to their differential interaction with PLIN1 and 5 , we have observed that CIDEC is more resilient to the deletion of the N-term than CIDEA , indicating that it may be less sensitive to regulatory post-translational modifications of this domain . This robustness of CIDEC activity together with its potentiation by PLIN1 could facilitate the continuity of the LD enlargement in white adipocytes until the unilocular phenotype is achieved . In contrast , in brown adipocytes expressing CIDEA the process would be stopped at the multilocular stage , for example , due to post-translational modifications that modulate the function or stability of the protein or alteration of the PA levels in LDs . Further work will be required to characterize the physiological differences between CIDEC and CIDEA , and determine the influence of their interacting partners and the role of proteins that are able to alter the LD PA levels , such as Lipin-1γ . Abnormal accumulation of large LDs have also been observed in non-adipocyte cells under other pathological conditions such as liver steatosis and atherosclerosis ( Krahmer and Walther , 2013 ) . As enhanced expression of CIDE proteins have been linked to these conditions ( Li et al . , 2010; Zhou et al . , 2012; Matsusue et al . , 2008 ) , the modulation of CIDE-triggered LD enlargement represents a potential therapeutic strategy that requires the elucidation of its molecular mechanism . In summary , we found that during LD fusion by lipid transference , CIDEA ensures the close proximity of the LD membranes by forming trans complexes through its N-term and C-term dimerization sites . This protein complex will be anchored in the LD–LD interface , forming the molecular environment necessary for TAG transport across the membrane . Finally , the amphipathic helix embedded in the LD membrane interacts with the cone-shaped phospholipid PA , generating a local perturbation of the monolayer integrity that would increase its permeability to TAG and enable its exportation to the acceptor LD . The new mechanistic insight into the molecular events underpinning LD dynamics revealed by this study highlights CIDEA and PA production as targets for therapeutic modulation of LD accumulation . The coding region of murine Cidea , Cideb , Cidec , Dff40 , and Dff45 were cloned into the vector pcDNA3 . 1D/V5-His-TOPO ( Invitrogen , Paisley , UK ) to obtain the v5-tagged versions of the proteins ( Hallberg et al . , 2008 ) . The human full-length and truncated forms of Cidec were subcloned into pcDNA3 . 1D/V5-His-TOPO from their GFP constructs ( Rubio-Cabezas et al . , 2009 ) and Lipin-1γ-v5 was constructed from pGH321 ( Han and Carman , 2010 ) . Mutations and deletions were generated with the QuikChange Lightning Kit ( Agilent ) . Tagged proteins were detected by using antibodies against v5 ( R96025; Invitrogen ) , HA ( H6908; Sigma ) , or GFP ( ab1218; Abcam ) . 3T3-L1 cells were maintained in Dulbecco’s modified Eagle’s medium ( DMEM ) containing 4 . 5 g/l glucose and L-glutamine supplemented with 10% newborn calf serum ( NCS; Invitrogen ) and penicillin/streptomycin at 37°C and 5% CO2 . Hela cells were cultured in similar conditions but with 10% FBS ( Invitrogen ) . Transfections were performed using Lipofectamine 2000 ( Invitrogen ) . Stable cell lines expressing CIDEA-v5 were generated by transfection of 3T3-L1 cells with pcDNA3 . 1/Cidea-v5 , followed by selection with G418 ( Invitrogen ) . The imBAT cell line was generated by the retroviral transduction of primary brown adipocytes with SV40 large-T antigen tsA58 mutant and differentiated as previously described ( Hallberg et al . , 2008 ) . Cells in gelatin-coated glass bottom dishes were stained with 0 . 1–0 . 5 μg/ml BODIPY 493/503 in the appropriate culture medium with 20 mM HEPES in the absence of serum . After 10 min at 37°C , 10% FBS was restored and the dish was equilibrated at 37°C in a Leica SP5 confocal microscope . Time-lapse Z-stacks were acquired every 2 min and represented as their maximum projection . 3T3-L1 cells were analysed 6 hr after infection with an adenovirus vector expressing CIDEA ( Hallberg et al . , 2008 ) . For the imBAT differentiation experiments , pre-adipocytes were incubated for 48 hr with differentiation cocktail ( Hallberg et al . , 2008 ) , and medium was changed to DMEM:F12 with 10% FBS , 1nM T3 , and 170 nM insulin for 6 hr before staining . Cells on glass coverslips were fixed in 4% paraformaldehyde and permeabilized with blocking solution ( BS: 0 . 5% BSA , 0 . 05% Saponin , 50 mM NH4Cl in PBS ) . Cells were incubated overnight at 4°C with primary antibodies diluted in BS , and for 1 hr at room temperature with secondary antibodies ( conjugated to Alexa488 and Alexa555 , Invitrogen ) . Cells were stained in PBS with 2 μg/ml BODIPY 493/503 or 1:200 dilution of LipidTox Deep Red for 15 min and mounted in ProLong Gold antifade reagent ( all from Invitrogen ) . Images were acquired in a Leica TCS SP5 microscope . For the phenotypic distribution of Hela cells expressing modified CIDEA-v5 constructs , cells were treated with oleic acid 24 hr after transfection and incubated for a further 18 hr prior to fixation . Phenotype classification was performed by visual analysis of randomized samples in a minimum of three independent experiments for each construct ( n>50 cells ) Liposomes were prepared by dissolving lipid ( 1 , 2-dilauroyl-sn-glycero-3-phosphocholine ( DLPC ) , 12:0 PC ) or a mixture of DLPC and DLPA ( 1 , 2-dilauroyl-sn-glycero-3-phosphate , 12:0 PA; at a 9:1 DLPC:DLPA molar ratio ) ( Echelon Biosciences , USA ) in 3:1 chloroform:MeOH and drying under vacuum using rotary evaporation . The resulting thin films were left to dry under vacuum overnight to remove all residual solvent , reconstituted in 25 mM sodium phosphate buffer ( pH 7 . 2 ) to a final lipid concentration of 3 . 3 mg/mL , and subjected to four times freeze-thaw-sonicate cycles . The vesicles were incubated at 37°C for 20 min prior to CD measurements . CD experiments were undertaken with a synthetic wt ( SYDIRCTSFKAVLRNLLRFMSYAAQMTG ) CIDEA peptide ( Pepmic , Suzhou , China ) encompassing aas 158–185 solubilized at a concentration of 41 μM ( based on absorbance at 280 nm ) in 50 mM potassium phosphate , pH 6 . 2 plus 0 . 1% n-dodecyl-β-D-maltopyranoside and analysed by CD in a Jasco J-815 spectrometer ( Jasco UK , Great Dunmow , UK ) . Additional CD experiments with the same wt peptide and a mutant ( F166R/V169R/L170R ) ( SYDIRCTSRKARRRNLLRFMSYAAQMTG ) were carried out using a Jasco J-1500 spectropolarimeter ( Jasco UK ) equipped with a Peltier thermally controlled cuvette holder and 1 mm path-length quartz cuvettes ( Starna; Optiglass , Hainault , UK ) . Spectra were recorded between 190 and 300 nm with a data pitch of 0 . 2 nm , a bandwidth of 2 nm , a scanning speed of 100 nm min–1 and a response time of 1 second . Peptides were solubilized in 25 mM sodium phosphate ( pH 7 . 2 ) at concentrations of 70 μM ( wt ) and 47 μM ( mutant ) . Peptide samples were prepared in the absence and presence of DPLC and DLPC:DLPA ( 9:1 molar ratio ) vesicles and CD spectra were acquired at 37°C . Data shown were averaged from four individual spectra after subtraction of the appropriate buffer/vesicle CD spectrum . All CD data were analysed using the CDPro suite of programs . The output of the individual programs CDSSTR and CONTINLL provided the estimated percentages of α-helix , β-sheet , turn , and unstructured regions , using the IB = 4 database of 43 soluble proteins with CD data from 190–250 nm . Secondary structure propensity of full-length CIDEA was predicted using DSSP ( Arnold et al . , 2006 ) . The amphipathic helix sequence CTSFKAVLRNLLRFMSYA ( 163–180 aa ) was submitted to the PEP-FOLD online de novo peptide structure prediction server using default settings ( Maupetit et al . , 2009 ) . PA was docked to the PEP-FOLD predicted structure using default settings in a single simulation by AutoDock Vina53 ( http://vina . scripps . edu ) ( Trott et al . , 2009 ) . Lipid and protein structures were converted from pdb into pdbqt format using MGL Tools54 . A grid box was centred at coordinates 35 . 651 , 35 . 471 , 35 . 569 with 34 Å units in x , y , and z directions to cover the entire helix . AutoDock Vina reports the nine lowest energy conformations , which were inspected using PyMOL software ( www . pymol . org ) . According to binding affinity and visual inspection , without RMSD clustering , the best-fit model has been selected . CG-MD simulations were used to predict the structure of a LD and its putative interaction with the amphipathic helix using a 4 to 1 atom mapping for both , lipids and protein ( Marrink et al . , 2007; Monticelli et al . , 2008 ) . A LD composed of a hydrophobic core containing 200 glyceryl trioleate or TAG molecules surrounded by a phospholipid monolayer containing 400 POPC molecules previously reported was used as the starting configuration ( Mohammadyani et al . , 2014 ) . A second LD containing PA consisting of a hydrophobic core of 200 TAG molecules , and a phospholipid monolayer with 364 POPC molecules and 36 palmitoyl-oleoyl-glycero-phosphatidic acid ( POPA ) was compiled using the same procedure . A rectangular simulation box including LD , amphipathic helix , water , and ions was energy minimized and pre-equilibrated . All MD runs were carried out for 200 ns under NPT conditions . The CG-MD simulation of the LD–helix interaction was carried out using the MARTINI CG force field developed by Marrink et al . ( version 2 . 0 ) ( Marrink et al . , 2007 ) . All simulations were performed using the GROMACS simulation package version 4 . 6 . 5 ( http://www . gromacs . org/ ) . The system was weakly coupled to an external temperature bath at 310 K ( Berendsen et al . , 1984 ) . The pressure was weakly coupled to an external bath at 1 bar using an isotropic pressure scheme ( Berendsen et al . , 1984 ) . Visualization and analysis was performed using the VMD v . 1 . 9 visualization software ( Humphrey et al . , 1996 ) . Distances and density maps were computed using analysis tools ( g_dist and g_densmap ) in the GROMACS package ( http://www . gromacs . org ) ( Van Der Spoel et al . , 2005 ) . Cells were lysed in 50 mM Tris ( pH 8 . 0 ) , 150 mM NaCl , 1% TRITON X-100 with protease inhibitor cocktail ( Roche ) . Anti-HA antibody ( H6908; Sigma ) or anti-V5 antibody ( R96025; Invitrogen ) was bound to Dynabeads Protein G ( Invitrogen ) and incubated with the lysate to immunoprecipitate the tagged proteins following manufacturer’s instructions . Cell lysates or IP fractions in Laemmli buffer were analysed by Western blot . Each co-IP experiment was performed at least in triplicate , producing similar results in each experiment with a representative image presented . In vitro translated CIDEA-v5 was synthesized from pcDNA3 . 1/Cidea-v5 using the TnT Coupled Wheat Germ Extract System ( Promega ) and verified by Western blot . The cell-free preparation of CIDEA-v5 was probed with Membrane Lipid Strips ( Echelon Biosciences ) following the manufacturer’s instructions . Protein affinity for PA was examined in pull-down assays using PA covalently linked to agarose beads ( PA beads ) ( Manifava et al . , 2001 ) . Cells were lysed in 50 mM Tris-HCl pH 8 . 0 , 50 mM KCl , 10 mM EDTA , 0 . 5% Nonidet P-40 , and protease inhibitors . Lysates were sonicated and centrifuged at 14000g prior to incubation with the PA beads as previously described ( Manifava et al . , 2001 ) . Competition experiments with soluble phospholipids were performed by supplementing the cell lysate with 1 , 2-dilauroyl-sn-glycero-3-phosphate 12:0 PC ( DLPA ) or 1 , 2-dilauroyl-sn-glycero-3-phosphocholine ( DLPC ) ( Echelon Biosciences ) . Each PA-binding experiment was performed at least in triplicate , producing similar results in each experiment with a representative image presented . The S . cerevisiae strains used in this study are listed in Supplementary file 1 . To express CIDEA in yeast , a codon-optimized version of the mouse Cidea gene was generated by artificial gene synthesis ( GeneOracle ) , and subcloned into pRS316-CYC1p . Wt BY4742 ( Brachmann et al . , 1998 ) and genetically modified yeast strains were transformed with pRS316-CYC1p-Cidea and stable transformants were selected in synthetic media minus uracil . Leucine selection was used for the expression of PAH1-7A with pHC204 ( Choi et al . , 2010 ) . Yeast cells in synthetic media cultured overnight at 30°C were diluted to OD600 = 0 . 1 and allowed to grow until mid-logarithmic phase ( OD600 = 0 . 5 ) before fixation with 4% formaldehyde and LD staining with 2 μg/ml BODIPY 493/503 . For the automatic quantification of LDs , random microscopy images were acquired using a Delta Vision RT system ( Applied Precision ) . Maximum intensity and integrated intensity projections were created from the deconvolved image stacks using ImageJ . A custom written CellProfiler pipeline ( Carpenter et al . , 2006 ) automatically identified individual yeast cells and measured their number and size of circle shaped LDs . Supersized LDs were defined as the LDs with a diameter above 0 . 5 μm . To measure the total levels of phospholipids in yeast , cells were grown overnight in synthetic medium at 30°C in the presence of 20 μCi/mL [32P]-orthophosphate . Cultures were then diluted to OD600 = 0 . 1 maintaining the label and were allowed to grow until OD600 = 0 . 5 . To analyse de novo synthesis of glycerophospholipids , cells were grown to OD600 = 0 . 5 in synthetic medium and incubated with 100 μCi/mL [32P]-for 20 min . Lipids were extracted and quantified by two-dimensional chromatography , as described by ( Gaspar et al . , 2006 ) .	If other energy sources become unavailable , cells fall back on stores of fatty molecules called lipids . These are held in membrane-enclosed compartments in the cell called lipid droplets , which in mammals are particularly abundant in fat cells called adipocytes . There are two main types of adipocytes: white adipocytes have a single giant lipid droplet , whereas brown adipocytes contain many smaller droplets . Proteins embedded in the membrane that surrounds a lipid droplet help to control the droplet’s growth and when it releases lipids . For example , a protein called CIDEA , which is only found in brown adipocytes , helps lipid droplets to grow by enabling one droplet to transfer its contents to another droplet . However , little is known about how this occurs . By combining cell biology , biophysical and computer modelling approaches , Barneda et al . investigated how normal and mutant forms of CIDEA affect the growth of lipid droplets . These experiments identified a helix in the structure of CIDEA that embeds it in the membrane , from where it can then interact with CIDEA proteins on other lipid droplets to hold the droplets together . In addition , the helix interacts with a molecule in the lipid droplet membrane called phosphatidic acid . Barneda et al . suggest that this interaction helps to transfer the contents of one droplet to another by making it easier for lipids to move through the droplets’ membranes . The next challenge is to characterize the mechanisms that control CIDEA activity to influence the formation of the multiple lipid droplets that distinguish brown and BRITE ( brown-in-white ) adipocytes from white adipocytes . The lipid droplets in brown adipocytes are an important target for research to combat obesity , due to the 'burning' rather than storing of lipids that occurs in these cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2015	The brown adipocyte protein CIDEA promotes lipid droplet fusion via a phosphatidic acid-binding amphipathic helix
Pheromones , chemical signals that convey social information , mediate many insect social behaviors , including navigation and aggregation . Several studies have suggested that behavior during the immature larval stages of Drosophila development is influenced by pheromones , but none of these compounds or the pheromone-receptor neurons that sense them have been identified . Here we report a larval pheromone-signaling pathway . We found that larvae produce two novel long-chain fatty acids that are attractive to other larvae . We identified a single larval chemosensory neuron that detects these molecules . Two members of the pickpocket family of DEG/ENaC channel subunits ( ppk23 and ppk29 ) are required to respond to these pheromones . This pheromone system is evolving quickly , since the larval exudates of D . simulans , the sister species of D . melanogaster , are not attractive to other larvae . Our results define a new pheromone signaling system in Drosophila that shares characteristics with pheromone systems in a wide diversity of insects . Different insect species use a diversity of pheromones to establish and organize their social encounters ( Karlson and Lüscher , 1959; Shorey , 1973 ) . For example , pheromones mediate courtship , aggression , alarm signaling , parental care , navigation , aggregation and many other behaviors ( Shorey , 1973; de Bruyne and Baker , 2008 ) . While the use of chemical cues is widespread amongst insects ( Symonds and Elgar , 2008 ) and other taxa , including mammals ( Dulac and Torello , 2003 ) , the majority of our understanding of the molecular and neural mechanisms that mediate insect pheromone signaling has come from studies of a single species , Drosophila melanogaster . This is because many powerful genetic tools have been developed to study gene and neural circuit function in Drosophila . Most studies of pheromones in Drosophila have focused on pheromones related to sex ( Dahanukar and Ray , 2011; Fernández and Kravitz , 2013 ) . These studies have provided a rich mechanistic understanding of the receptor genes and sensory neurons that detect these cues , and of the circuit architectures that mediate these behaviors , linking the detection of specific molecules to appropriate behavioral outputs . In contrast , pheromones that organize spatial behaviors like navigation or aggregation have been less intensively studied in Drosophila ( Bartelt et al . , 1985 ) , despite the fact that such pheromones are critical for the survival of many insects . For example , ants employ substrate-born pheromones to establish and maintain foraging trails ( Steck , 2012 ) and caterpillars use pheromones to organize mass migrations ( Fitzgerald , 1976 , 2003 ) . To date , it has not been possible to study such problems in Drosophila because no pheromones were known that exclusively mediate aggregation or trail following . Nonetheless , several previous studies have provided suggestive evidence that Drosophila larval behavior is influenced by social cues . First , Drosophila larvae are robustly attracted to odors produced by other larvae in food ( Durisko and Dukas , 2013 ) . Second , when in close proximity on a food source , larvae both aggregate ( Durisko et al . , 2014 ) and engage in a form of co-operative digging , which may effectively increase their rate of feeding ( Wu et al . , 2003; Xu et al . , 2008 ) . Finally , in natural conditions , where different species of drosophilids can be found sharing the same food resource , larvae preferentially pupariate with conspecific larvae and avoid pupariating with larvae of other species ( Beltramí et al . , 2012 ) . Each of these three phenomena , it has been proposed , may be mediated by chemical cues produced by larvae or by larval activity . We have discovered an attractive pheromone-signaling pathway in Drosophila larvae . We discovered two novel pheromones— ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid—that larvae deposit on substrate and that act as larval attractants . We identified a larval chemosensory neuron that is responsive to these compounds and that is required for behavioral attraction to these pheromones . We found that , in addition to their previously described roles in sex-pheromone detection , two members of the pickpocket family of DEG/ENaC channel subunits , pickpocket23 and pickpocket29 , are required for detecting ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid . Finally , we have found evolved differences in the production of these pheromones in the genus Drosophila , which , along with a repellent cue , cause differences in larval social behavior between these species . Our work provides the first window into social signaling during the larval stage of Drosophila and demonstrates that these mechanisms , and this social behavior , have evolved rapidly . Many insects employ fatty acid derived pheromones in the context of aggregation and navigation , and our work in Drosophila establishes a model system with which to further explore the sensory mechanisms and neural circuits that process this class of chemical cues . In preliminary assays of Drosophila larval behavior in dense populations , we noticed that larvae seemed to be attracted to regions of assay plates that had been occupied previously by other larvae . We developed an assay to quantify this effect ( Figure 1A ) . We allowed several hundred early third instar larvae to crawl on agarose plates for 20 min . Then , we cut away half the agarose surface , replaced it with fresh agarose , and placed a single new larva on the plate . We found that test larvae spent more time on the larval-treated substrate than on the untreated control substrate ( Figure 1B ) , suggesting that the population of larvae had deposited an attractive pheromone on the substrate . 10 . 7554/eLife . 04205 . 003Figure 1 . A single class of chemosensory receptors is required for larvae to respond to attractive larval-derived cues . ( A ) An assay to quantify larval attraction to substrate-born attractive pheromone . A high density of wild-type early third instar larvae were allowed to crawl over the surface of an agarose plate for 20 min . The larvae and half of the treated agarose ( green ) were removed and replaced with fresh control agarose ( white ) . The fraction of time spent on each surface by single larvae was used to calculate a preference index score ( Preference Index = ( time on test substrate—time on control substrate ) /total time ) . ( B ) A behavioral neuronal-silencing screen to identify the receptor neurons required for response to larval attractive pheromones . Data points represent the preference index scores of individual larvae in response to the larval cue in this assay . In this and subsequent figures , horizontal small lines represent means and vertical lines represent ± one standard deviation . Genotypes displayed significant heterogeneity ( ANOVA , F9 , 255 = 9 . 75 , p < 0 . 0001 ) . Control larvae carrying the empty attP2 landing site with no GAL4 insertion and a copy of UAS-tetanus toxin ( UAS-TNT ) were attracted to the larval cue , but not to untreated substrate . Silencing neurons expressing ppk23 with TNT blocked the response to the cue . Driving an inactive variant of tetanus toxin , IMPTV , with ppk23-GAL4 did not block attraction . Mutations affecting ppk23 or ppk29 also blocked attraction . Silencing neurons with Janelia enhancer fragment R58F10-GAL4 blocked the response to the cue . Driving IMPTV with R58F10-GAL4 did not block attraction . In this and subsequent figures , statistical comparisons between treatments were performed using the Tukey honest significant difference test and are reported for comparisons significant at p < 0 . 05 or below . ( C ) A schematic of neurons comprising the main external sensory organs in the larval chemosensory system . Receptor neurons in the terminal organ ganglion ( TOG; dark blue ) send sensory terminals into the terminal organ ( TO ) , and synaptic terminals into the subesophagael zone ( SEZ ) . Receptor neurons in the dorsal organ ganglion ( DOG; light blue ) send sensory terminals mainly into the dorsal organ ( DO ) , and synaptic terminals into both the antennal lobe and the SEZ . Three neurons ( green ) in the DOG send sensory terminals into the dorsolateral papillum of the TO , and project to the SEZ . ( D ) ppk23 expressing neurons innervating the terminal organ ( arrow ) , marked by membrane-targeted green fluorescent protein , mCD8:GFP ( green , anti-GFP ) . Neuron cell bodies in the dorsal organ ganglion and terminal organ ganglion are labeled with anti-elaV ( magenta ) . The TOG and DOG are marked by an arrowhead and asterisk , respectively . ( E ) The axon terminals of ppk23 expressing neurons in the larval brain ( green , anti-GFP ) . Neuropil is marked by antibodies against the perisynaptic marker bruchpilot ( nc82 , magenta ) . ( F ) R58F10 sensory neurons innervate the terminal organ ( arrow ) . R58F10 marked by membrane-targeted green fluorescent protein , mCD8:GFP ( anti-GFP , green; elaV , magenta ) . The dorsal organ is indicated with an arrowhead . ( G ) R58F10 neuron synaptic terminals project into the suboesophageal zone ( arrow ) . Secondary lineage neurons in the ventral nerve cord are also marked by R58F10 ( arrowhead ) ( anti-GFP , green; nc82 , magenta ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04205 . 00310 . 7554/eLife . 04205 . 004Figure 1—figure supplement 1 . Sex-pheromone receptor genes with known larval expression are not required to respond to attractive larval pheromones . Data points are the preference index scores of individual larvae in response to larval residue in the assay shown in Figure 1A . Lines represent means plus or minus one standard deviation for each dataset . Control larvae carrying the empty attP2 landing site with no GAL4 insertion and a copy of UAS-TNT were attracted to the larval cue , but not to untreated substrate . Blocking all olfactory neurons using Or83b-GAL4 , or gustatory receptors implicated in pheromone responses with Gr32a-GAL4 and Gr33a-GAL4 , or mutations affecting these genes , did not block larval attraction to the cue . DOI: http://dx . doi . org/10 . 7554/eLife . 04205 . 00410 . 7554/eLife . 04205 . 005Figure 1—figure supplement 2 . The expression patterns of ppk23-GAL4 and ppk29-GAL4 in larval sensory organs and the central nervous system . ( A ) ppk23-GAL4 driving the expression of mCD8:GFP ( green ) in the anterior sensory organs of third instar larvae stained for neuronal nuclei ( anti-elaV; magenta ) and DAPI ( white ) . ppk23-GAL4 is expressed in cells innervating the terminal organ ( arrowhead ) , and an internal sensory neuron in the esophagus ( arrow ) . ( B and C ) ppk29-GAL4 driving the expression of mCD8:GFP ( green ) in the third instar larval brain stained for neuropil ( anti-nc82 , magenta ) . ppk29-GAL4 is expressed widely throughout the brain ( Liu et al . , 2012; Thistle et al . , 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04205 . 00510 . 7554/eLife . 04205 . 006Figure 1—figure supplement 3 . A synaptic silencing screen to identify sensory neurons required for attractive pheromone-driven behavior . We silenced subsets of candidate sensory neurons with tetanus toxin , using a collection of driver lines from the Janelia enhancer GAL4 collection that drive expression in anterior sensory neurons . Data points represent the preference index scores of individual larvae for the larval cue in the assay shown in Figure 1A . Means ( horizontal red lines ) ± one standard deviation ( vertical black lines ) are plotted beside the data . DOI: http://dx . doi . org/10 . 7554/eLife . 04205 . 006 We hypothesized that larvae detect the substrate-born cue through receptor neurons in the external larval chemosensory system . This system is composed primarily of the dorsal organ and the terminal organ , which are both located on the larval head and in total contain 67–72 sensory neurons ( Stocker , 2008 ) ( Figure 1C ) . We sought to address two questions . First , which sensory neurons in these organs , if any , mediate attraction to the larval-derived cue ? Second , are any of the genes required for detecting adult sex pheromones required also to detect larval pheromones ? First , we tested if the substrate-born attractant is detected by larval neurons expressing specific odorant receptor or gustatory receptor genes that mediate adult social behaviors like courtship and aggression ( Kurtovic et al . , 2007; Miyamoto and Amrein , 2008; Moon et al . , 2009; Wang and Anderson , 2010; Liu et al . , 2011; Wang et al . , 2011 ) . Silencing synaptic transmission by expressing tetanus toxin ( Sweeney et al . , 1995 ) in Orco-GAL4 , Gr32a-GAL4 , or Gr33a-GAL4 expressing cells failed to block attraction to the larval cue ( Figure 1—figure supplement 1 ) . Similarly , mutations in these receptor genes had no effect on this behavior ( Figure 1—figure supplement 1 ) . Thus , larval neurons expressing odorant receptors and the gustatory receptors Gr32a and Gr33a are unlikely to be required to mediate attraction to the larval pheromone . Several members of the pickpocket family of degenerin/DEG/ENaC channel subunit genes are expressed in gustatory neurons on the leg segments and labellum of adult flies and are required for detecting sex pheromones ( Liu et al . , 2012; Lu et al . , 2012; Thistle et al . , 2012; Toda et al . , 2012; Yuan et al . , 2013 ) . We tested if two members of this family , pickpocket23 ( ppk23 ) and pickpocket29 ( ppk29 ) , are required also for the detection of the attractive larval cue . First , we tested ppk23 . Silencing ppk23-GAL4 expressing neurons with tetanus toxin reduced the attraction of larvae to the cue , as did a mutation deleting the ppk23 locus , Δppk23 ( Figure 1B ) . ppk23-GAL4 is expressed in 10 neurons that innervate the terminal organ , the main larval taste organ ( Figure 1C , D ) , and in one internal sensory neuron located along the larval pharynx ( Figure 1—figure supplement 2A ) . A subset of sensory neurons that innervate the terminal organ have their cell bodies located in the dorsal organ ganglion ( TO dorsolateral neurons ) , while the remaining terminal organ cells are housed in the terminal organ ganglion ( Colomb et al . , 2007 ) . ppk23 labels all three TO dorsolateral neurons . ppk23 neurons project widely throughout the subesophageal zone ( Figure 1E ) , the brain region that integrates sensory information related to taste and feeding ( Vosshall and Stocker , 2007; Stocker , 2008 ) . Next , we examined ppk29 . Larvae carrying a deletion of the ppk29 gene displayed reduced attraction to larval residue compared with controls ( Figure 1B ) . We examined ppk29-GAL4 expression and found that it is expressed ubiquitously in larvae ( Figure 1—figure supplement 2B , C ) and therefore is not useful for defining the requirement for ppk29 specific sensory neurons . Together , these results indicate that both ppk23 and ppk29 are required for attraction of larvae to larval residue . To refine the identity of candidate pheromone receptor neurons further , we screened a subset of the Janelia enhancer fragment GAL4 lines ( Jenett et al . , 2012 ) that were characterized as driving sparse patterns of expression in anterior sensory neurons ( Li et al . , 2013 ) . We drove tetanus toxin with each line and found that neuronal inactivation in several lines blocked the larval attraction to larval residue ( Figure 1B and Figure 1—figure supplement 3 ) . We focused further experiments on one of these lines that failed to respond to the larval residue , R58F10 , and that drives expression in only a single bilateral TO dorsolateral neuron ( Figure 1C , F ) . This neuron is therefore a subset of the three TO dorsolateral neurons marked by ppk23 . R58F10 neurons send straight , mostly unbranched projections into the subesophagael zone ( Figure 1G ) . No molecules that mediate social interactions between Drosophila larvae are known . To identify the molecules that larvae may use as attractive pheromones , we took advantage of the fact that the attractive activity was deposited on substrate by crawling larvae . We found that the activity could be extracted off glass upon which larvae had crawled using either hexane or acetone as solvent ( Figure 2A , B ) . Gas chromatography-mass spectrometry of these extracts revealed that they contained seven common saturated fatty acids and fatty acid derivatives and two rare fatty acid monoenes , ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid ( Figure 2C , Figure 5—source data 1 ) . We did not detect the adult sex pheromones ( Z ) -11-vaccenyl acetate ( cVA ) , 7-tricosene , 7 , 11-heptocosadiene , or CH503 in these extracts . While we observed ( Z ) -9-tetradecenoic acid in extracts of whole larval bodies , we did not observe significant levels of this compound in the residue deposited by larvae , suggesting that ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid are the major tetradecanoic derivatives deposited by larvae . ( Z ) -5-tetradecenoic acid is found also as a component of a defensive , ant-repulsive secretion released by some thrips ( Suzuki et al . , 2004 ) . To our knowledge , ( Z ) -7-tetradecenoic acid has not been found in any other insect . 10 . 7554/eLife . 04205 . 007Figure 2 . ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid are attractive larval pheromones . ( A ) A method for extracting the attractive cue deposited on substrate . Larvae were allowed to crawl over the inner surface of a glass vial for 30 min . Larvae were removed and replaced by solvent . Treated solvent was deposited on one side of an agarose plate , and untreated solvent on the other . ( B ) The attraction of wild-type D . melanogaster larvae to extracts from larval-treated glass using hexane , acetone , or ethanol as solvent compared to control solvent . Data points are the preference index scores of individual wild-type larvae . Hexane and acetone extracts elicited similar levels of attraction as plates treated directly with larvae . Ethanol extracts did not . ( C ) Chromatograph from GC analysis of D . melanogaster larval hexane extracts . The major components are saturated fatty acids , fatty acid monoenes and dienes . Each peak is labeled as follows: ( 1 ) dodecanoic acid , ( 2 ) ( Z ) -5-tetradecenoic acid , ( 3 ) ( Z ) -7-tetradecenoic acid , ( 4 ) tetradecanoic acid , ( 5 ) ( Z ) -9-hexadecenoic acid , ( 6 ) hexadecanoic acid , ( 7 ) ( Z , Z ) -9 , 12-octadecadienoic acid , ( 8 ) ( Z ) -9-octadecenoic acid , and ( 9 ) octadecanoic acid , methyl ester . Small amounts of ( Z ) -9-tetradecenoic acid ( * ) were also detected between peaks ( 3 ) and ( 4 ) . ( D ) Schematic of a multiple-larva assay to measure the attractive activity of individual compounds identified from larval residue . Test surfaces were coated in a checkerboard pattern with candidate molecules ( green ) or vehicle solvent ( white ) . ( E ) A behavioral screen measuring the attraction of control ( attP2;UAS-TNT , black ) , R58F10-GAL4;UAS-TNT ( red ) , and R58F10-GAL4; UAS-IMPTV ( blue ) larvae to individual compounds found in the D . melanogaster larval extracts . Data points are the preference index scores of small groups ( 10-15 individuals ) of larvae in response to 50 fmol/cm2 of candidate molecule . Compound treatments ( ANOVA; F9 , 113 = 22 . 3 , p < 0 . 0001 ) and genotype by compound interactions ( 2-way ANOVA; F1 , 232 = 17 . 7 , p < 0 . 0001 ) displayed significant heterogeneity . Larvae were attracted to ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid . Silencing the synaptic activity of the sensory neurons expressing R58F10-GAL4 strongly inhibited the response to these molecules . Attraction was partially restored when driving the inactive TNT gene product , IMPTV . ( F ) The response of attP2;UAS-TNT ( black ) , and R58F10-GAL4;UAS-TNT ( red ) larvae to 50 fmol/cm2 ( Z ) -5-tetradecenoic acid over time . ( G ) The timecourse of response to 50 fmol/cm2 ( Z ) -7-tetradecenoic acid . Solid lines represent the mean preference index score at each timepoint . Shaded areas represent ± one standard deviation . ( H ) Dose response curves of larval preferences to ( Z ) -5-tetradecenoic acid ( orange ) , ( Z ) -7-tetradecenoic acid ( green ) , and saturated tetradecanoic acid ( grey ) . Shaded areas represent ± one standard deviation . ( I , J ) The preference index scores ( I ) and preference time course ( J ) of attP2;UAS-TNT ( black ) , R58F10-GAL4;UAS-TNT ( red ) , and R58F10-GAL4; UAS-IMPTV ( blue ) larvae in response to an equimolar mixture ( ( 25 fmol:25 fmol ) /cm2 ) of ( Z ) -5-tetradecenoic and ( Z ) -7-tetradecenoic acids . DOI: http://dx . doi . org/10 . 7554/eLife . 04205 . 00710 . 7554/eLife . 04205 . 008Figure 2—figure supplement 1 . Time courses of the attraction of larvae to individual compounds identified from larval extracts at 50 fmol/cm2 . Time courses of larval preference to ( A ) vehicle , or to 50 fmol/cm2 ( B ) dodecanoic acid , ( C ) tetradecanoic acid , ( D ) ( Z ) -9-hexadecanoic acid , ( E ) hexadecanoic acid , ( F ) ( Z , Z ) -9 , 12-octadecadienoic acid , ( G ) ( Z ) -9-octadecanoic acid , ( H ) octadecanoic acid , methyl ester over a 300 s behavioral assay . Solid lines represent the mean preference index score and shaded areas represent one standard deviation at each time point of the assay period . Control larvae ( attP2;UAS-TNT , black ) were not attracted to any of the common compounds identified from in larval extracts at this concentration compared to vehicle controls . Silencing the synaptic activity of the sensory neurons expressing R58F10-GAL4 ( red ) did not reduce the response to these compounds . DOI: http://dx . doi . org/10 . 7554/eLife . 04205 . 00810 . 7554/eLife . 04205 . 009Figure 2—figure supplement 2 . Chemical analysis of synthetic and natural pheromones . ( A ) Tandem gas chromatography–mass spectroscopy ( GC–MS ) traces for synthetic ( Z ) -5-tetradecenoic acid , ( Z ) -7-tetradecenoic acid , and ( Z ) -9-tetradecenoic acid after fatty acid methyl ester ( FAME ) preparation . ( B ) Traces from the same samples presented in panel A run under different conditions , along with FAME preparation of natural pheromone extract . The retention times of synthetic ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid match to peaks present in the natural extract . ( C , D ) 1H NMR spectra of ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid , respectively . ( E , F ) Tandem liquid chromatography–mass spectroscopy ( LC-MS ) of purified ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid , as detected by TIC ( total ion current , mass spectrometry ) , respectively . ( G , H ) m/z composition of the dominant peak in chromatograms ( E ) and ( F ) respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 04205 . 00910 . 7554/eLife . 04205 . 010Figure 2—figure supplement 3 . A behavioral screen measuring the attraction of larvae to individual compounds at 0 . 5 nmol/cm2 . ( A ) Data points are the preference index scores of small groups ( 10–15 individuals ) of larvae in response to 0 . 5 nmol/cm2 of candidate molecule . Lines represent means and one standard deviation for each dataset . Treatments displayed significant heterogeneity ( ANOVA; F8 , 120 = 2 . 9 , p < 0 . 0001 ) . Control larvae ( attP2;UAS-TNT , black ) were attracted to all compounds at this concentration compared to vehicle controls ( p < 0 . 05 , post hoc Tukey test for multiple comparisons ) . Silencing the synaptic activity of the sensory neurons expressing R58F10-GAL4 ( red ) strongly inhibited the response to ( Z ) -5-tetradececnoic acid and ( Z ) -7-tetradecenoic acid and weakly inhibited the response to ( Z , Z ) -9 , 12-octadecadienoic acid . ( B–J ) Time courses for the experiments summarized in panel ( A ) Solid lines represent the mean preference index scores and shaded areas represent plus or minus one standard deviation at each time point of the assay period . DOI: http://dx . doi . org/10 . 7554/eLife . 04205 . 010 To determine which , if any , of the nine compounds deposited by larvae acts as a larval attractant , we measured the attractive activity of synthetic versions of each molecule using a checkerboard-layout behavioral assay ( Figure 2D ) , which allowed us to assay larvae at a density similar to the single plate assay ( Figure 1A ) , but at a higher throughput . Control larvae were attracted to low concentrations ( 50 fm/cm2 ) of ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid , but not to any other compound ( Figure 2E and Figure 2—figure supplement 1 ) . Moreover , larvae showed a sustained response to these two molecules over the entire duration of a 5-min assay ( Figure 2F , G ) . Dose response curves revealed that ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid are strongly attractive to larvae across a broad range of concentrations ( Figure 2H ) . Larvae responded most strongly to pheromone concentrations of 50 fm/cm2 and this effect decreased at higher concentrations . A similar phenomenon has been observed for pheromones in multiple other species where higher doses induce a less robust behavioral response , or even begin to elicit alarm or escape behaviors , compared with lower doses ( Leal et al . , 1989; Shimizu et al . , 2003 ) . Finally , we assayed the larval response to an equimolar mixture ( ( 25 fmol:25 fmol ) /cm2 ) of the two compounds . Larvae were attracted to a blend of these compounds , but were less responsive to this mixture than to a presentation of each molecule alone , which suggests an interaction between the two pheromones ( Figure 2I , J ) . These results indicate that larvae deposit the two compounds ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid on substrate and that low concentrations of these compounds provide an attractive cue to other larvae . Many of the nine compounds identified in larval residue when tested alone at high concentrations ( 0 . 5 nmol/cm2 ) were at least mildly attractive to larvae ( Figure 2—figure supplement 3 ) . This concentration corresponds to the concentration of compounds in extract of larval residue generated by hundreds of individuals . Thus , in principle , many or all of these larval compounds might contribute to larval social behavior . To determine which , if any , of these compounds are detected by R58F10 neurons , we tested the behavioral response of larvae to synthetic compounds with and without silenced R58F10 neurons . Silencing R58F10 neurons with tetanus toxin blocked larval attraction to ( Z ) -5-tetradecenoic acid , ( Z ) -7-tetradecenoic acid , and a mixture of the two pheromones , but had no significant effect on the response to the other compounds at 50 fmol/cm2 ( Figure 2E , I ) . Driving a mutant form of tetanus toxin carrying two point mutations that block its cleavage of synaptobrevin ( IMPTV ) caused an intermediate response to these pheromones . This suggests that some of the effect of driving TNT in R58F10 may arise from indirect toxicity . Other groups have also found that expressing IMPTV can sometimes produce intermediate phenotypes ( Kain et al . , 2013 ) . These results demonstrate that larval attraction to previously occupied substrate can be mediated by the sensitivity of R58F10 neurons to ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid . We next assayed whether R58F10 sensory neurons respond directly to ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic using a genetically encoded activity reporter , GCaMP6 ( Chen et al . , 2013 ) . We tested a range of pheromone concentrations estimated to encompass those encountered by larvae in the previously described behavioral assays . R58F10 neurons were excited in response to ( Z ) -5-tetradecenoic acid at 0 . 01 nM , 1 nM , and 100 nM . R58F10 neurons were not significantly excited by ( Z ) -7-tetradecenoic acid at 0 . 01 nM , but responded to higher concentrations ( Figure 3A , B ) . Moreover , at 1 nM , which corresponds to the concentration that evoked the peak behavioral preference ( Figure 2H ) , R58F10 neurons exposed to ( Z ) -5-tetradecenoic acid or ( Z ) -7-tetradecenoic acid were strongly elevated in their integrated fluorescence over the length of the trial compared to vehicle controls ( Figure 3C ) . We did not observe a significant increase in the integrated fluorescence when the compounds were tested at 0 . 01 nM and 100 nM ( Figure 3C ) , suggesting that the dose-response relationship of R58F10 to these pheromones is complicated . 10 . 7554/eLife . 04205 . 011Figure 3 . R58F10 sensory neurons are excited by ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid . ( A ) Example traces of the excitation of R58F10 sensory neurons in response to control vehicle , 1 nM ( Z ) -5-tetradecenoic acid , ( Z ) -7-tetradecenoic acid , tetradecanoic acid , or ( Z ) -9-octadecanoic acid measured using the calcium–sensitive activity reporter , GCaMP6-S , over time . Excitation is reported as the percent change in GCaMP fluorescence over baseline ( %ΔF/F ) . The solid black bar along the x-axis signifies the onset and duration of stimulus . ( B and C ) Stimulation with either ( Z ) -5-tetradecenoic acid or ( Z ) -7-tetradecenoic acid across a broad range of concentrations evoked a significant increase in the maximum percent change in GCaMP fluorescence ( max %ΔF/F ) ( B ) and in the integrated fluorescence ( arbitrary units ) ( C ) of individual trials over 150 s compared to control buffer . Compound treatments displayed significant heterogeneity ( ANOVA; F8 , 125 = 3 . 26 , p < 0 . 01; ANOVA; F8 , 123 = 3 . 02 , p < 0 . 01 ) . Data points represent recordings from individual cells . ( D and E ) The stimulation of two other ppk23-expressing sensory neurons innervating the dorsolateral terminal organ sensillum with R58F10 , marked by Gr33a-GAL4 with 1 nM ( Z ) -5-tetradecenoic acid or ( Z ) -7-tetradecenoic acid evoked a significant increase in the maximum percent change in GCaMP fluorescence ( max %ΔF/F ) ( D ) and in the integrated fluorescence ( arbitrary units ) ( E ) , compared to control . Compound treatments displayed significant heterogeneity ( ANOVA; F2 , 80 = 6 . 15 , p < 0 . 01; ANOVA; F2 , 80 = 4 . 23 , p < 0 . 05 ) . Data points represent recordings from individual cells . Data are pooled for both cells . DOI: http://dx . doi . org/10 . 7554/eLife . 04205 . 011 We explored the specificity of R58F10 neurons for ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid by testing if other compounds with similar structure that were present in the larval extracts could evoke activity in these cells . R58F10 neurons were not excited by tetradecanoic acid , which lacks a double bond compared to these pheromones , by ( Z ) -9-octadecanoic acid , a structurally similar , but longer , fatty acid monoene , nor by vehicle controls ( Figure 3A–C ) . These data demonstrate that R58F10 sensory neurons are directly excited by , and exhibit a high degree of specificity for , the two larval pheromones ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid . We wondered whether ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid are detected only by R58F10 , or if other sensory neurons—for example , other neurons that express ppk23—are also responsive to these compounds . Specifically , we tested if the two neurons that innervate the dorsolateral terminal organ sensillum along with R58F10 are sensitive to larval pheromone using GCaMP6 . Both of these cells are marked by Gr33a-GAL4 expression , and as noted previously , also express ppk23 . We were unable to morphologically discriminate between these two cells , so we pooled the individual fluorescence data from both cells . Both ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid elicited activity in at least one of these neurons ( Figure 3D , E ) . While neither cell is necessary for the attraction to larval residue ( Figure 1—figure supplement 1 ) , these cells may control other aspects of larval behavior driven by these compounds . Since we observed that both ppk23 and ppk29 were required for larvae to respond to a full bouquet of larval residue ( Figure 1 ) , we next tested if these genes were required for detecting ( Z ) -5-tetradecenoic and ( Z ) -7-tetradecenoic acid specifically . The ppk23-GAL4 expression pattern includes the larval pheromone sensing neuron R58F10 . As expected , silencing ppk23-GAL4 expressing neurons with tetanus toxin reduced larval attraction to both ( Z ) -5-tetradecenoic acid ( Figure 4A , B ) and ( Z ) -7-tetradecenoic acid ( Figure 4E , F ) compared to control larvae . Removing ppk23 gene function also reduced the attraction of larvae to both of these compounds compared to heterozygous controls ( Figure 4A , C , E , G ) . We next tested if ppk23 function is directly required in the sensory neuron R58F10 by rescuing ppk23 expression specifically with R58F10-GAL4 in a Δppk23 mutant background . The expression of ppk23 in R58F10 cells partially restored the attraction of larvae to both larval pheromones ( Figure 4A , D , E , H ) . These results indicate that ppk23 gene function is required specifically in R58F10 neurons for response to the larval pheromones ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid . 10 . 7554/eLife . 04205 . 012Figure 4 . ppk23 and ppk29 are required for detecting larval pheromones . ( A and E ) The mean index scores from preference assays measuring the attraction of control , ppk23 silenced , Δppk23 mutant , and Δppk23 mutant larvae rescued with ppk23 expression specifically in R58F10 neurons to synthetic ( Z ) -5-tetradecenoic acid ( A ) or ( Z ) -7-tetradecenoic acid ( E ) . Data points are the mean preference index scores of small groups ( 10-15 individuals ) of larvae in response to 50 fmol/cm2 of candidate molecule over the course of the 300 s assay . Genotypes displayed significant heterogeneity ( ANOVA; F5 , 68 = 18 . 03 , p < 0 . 0001 , F5 , 78 = 9 . 03 , p < 0 . 0001 , respectively ) . The mean preference index over the time-course of the larval response to either ( Z ) -5-tetradecenoic acid ( B–D ) or ( Z ) -7-tetradecenoic acid ( F–H ) . Silencing the synaptic activity of ppk23 expressing neurons , as well as a mutation deleting ppk23 blocked the attraction of larvae to ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid . The mean preference index scores ( I , L ) and timecourse ( J , K , M , N ) from preference assays measuring the attraction of control and Δppk29 mutant larvae to synthetic ( Z ) -5-tetradecenoic acid ( I , J , K ) or ( Z ) -7-tetradecenoic acid ( L , M , N ) . A mutation deleting ppk29 blocked the attraction of larvae to both ( Z ) -5-tetradecenoic acid ( p < 0 . 001 ) and ( Z ) -7-tetradecenoic acid ( p < 0 . 001 ) . Stimulation of R58F10 neurons mutant for ppk23 or ppk29 with either ( Z ) -5-tetradecenoic acid or ( Z ) -7-tetradecenoic acid evoked no significant increase in the maximum percent change in GCaMP fluorescence ( max %ΔF/F ) ( ANOVA; F5 , 79 = 0 . 45 , p < 0 . 83 ) ( O ) or in the integrated fluorescence ( arbitrary units ) ( ANOVA; F5 , 79 = 0 . 90 , p < 0 . 81 ) ( P ) of individual trials over 150 s compared to control buffer . DOI: http://dx . doi . org/10 . 7554/eLife . 04205 . 012 We tested next if ppk29 is required for detecting these larval pheromones . Larvae homozygous for a deletion of ppk29 were not attracted to ( Z ) -5-tetradecenoic acid ( Figure 4I , J ) and displayed weaker attraction to ( Z ) -7-tetradecenoic acid ( Figure 4L , M ) than did heterozygous control larvae . The expression of ppk29 in R58F10 cells did not rescue the attraction of larvae to ( Z ) -5-tetradecenoic acid ( Figure 4I , K ) , but partially restored the attraction of larvae to ( Z ) -7-tetradecenoic acid ( Figure 4L , N ) . These results indicate that ppk29 gene function is required specifically in R58F10 neurons for response to the larvae pheromone ( Z ) -7-tetradecenoic acid . That ppk29 expression in R58F10 failed to rescue larval attraction to ( Z ) -5-tetradecenoic acid indicates that either ppk29 is required in another cell , or that the expression of ppk29 by R58F10-GAL4 is inadequate to rescue this behavior . Finally , we performed calcium imaging to test if ppk23 and ppk29 are required for R58F10 sensory neurons to be excited by these pheromones . R58F10 neurons mutant for either ppk23 or ppk29 were not excited by 1 nM ( Z ) -5-tetradecenoic acid nor by ( Z ) -7-tetradecenoic acid ( Figure 4O , P ) . Due to technical limitations we were unable to identify ppk23 heterozygotes that would serve as an appropriate positive control for this experiment . All together , these results demonstrate that both ppk23 and ppk29 are required for robust attraction to ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid . To our knowledge , this is the first identification of genes required for pheromone detection at the larval stage in any insect . Pheromone signals often evolve quickly between species ( Symonds and Elgar , 2008 ) , including between species of the genus Drosophila ( Jallon and David , 1987; Ferveur , 2005; Shirangi et al . , 2009 ) . To determine if larval pheromone signaling has evolved within the D . melanogaster species group , we investigated social behavior in two sister species of D . melanogaster , Drosophila simulans and Drosophila sechellia . These species diverged from D . melanogaster approximately 3–5 MYA and diverged from each other approximately 0 . 5 MYA ( Kliman et al . , 2000; Lachaise and Silvain , 2004 ) . First , we tested if D . sechellia and D . simulans larvae deposit residue that is attractive to other larvae . We found that third instar D . sechellia larvae are attracted , and D . simulans larvae are not attracted , to substrate treated with conspecific larvae ( Figure 5 ) . These data are consistent with recent independent observations that D . simulans larvae are less likely to form aggregations than are D . melanogaster larvae ( Durisko et al . , 2014 ) . We reasoned that this difference could be caused by evolved changes in D . simulans leading to either ( 1 ) loss of sensitivity to an attractant , ( 2 ) loss of production of an attractant , ( 3 ) production of a repellent molecule , or a combination of these factors . 10 . 7554/eLife . 04205 . 013Figure 5 . A reduction of ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid production underlies an evolved difference in attractive pheromone signaling . ( A ) The attraction of D . sechellia , D . simulans , or D . melanogaster larvae to larval residue from each species , to synthetic ( Z ) -7-tetradecenoic acid , or to extracts of larval residue . Data points for residue and extract experiments represent preference index scores of individual larvae using the assays described in Figure 1A , and Figure 2A , respectively . Datapoints measuring the response to synthetic ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid represent the preference of 8–12 larvae using the checkerboard assay ( Figure 3A ) . Two-way ANOVA of residue experiments reveals a significant effect of residue , but not of test larvae ( 2-way ANOVA; residue F3 , 199 = 40 . 0 , p < 0 . 0001 ) . The residue from D . simulans is not attractive to any of the three species . Both D . sechellia and D . simulans are attracted to synthetic ( Z ) -7-tetradecenoic acid . D . melanogaster larvae are attracted to hexane and acetone extracts from D . sechellia larval residue , but not D . simulans larval residue . ( B ) Chromatographs obtained from D . sechellia ( magenta ) and D . simulans ( green , inverted ) larval hexane extracts . ( 1 ) dodecanoic acid , ( 2 ) ( Z ) -5-tetradecenoic acid , ( 3 ) ( Z ) -7-tetradecenoic acid , ( 4 ) tetradecanoic acid , ( 5 ) ( Z ) -9-hexadecanoic acid , ( 6 ) hexadecanoic acid , ( 7 ) ( Z , Z ) -9 , 12-octadecadienoic acid , ( 8 ) ( Z ) -9-octadecanoic acid , ( 9 ) octadecanoic acid , methyl ester . Inset , a portion of the D . sechellia ( magenta ) , D . simulans ( green ) , and D . melanogaster ( black ) chromatograph traces showing the relative quantities of tetradecenoic acid isomers ( Z ) -5-tetradecenoic acid ( 2 ) , ( Z ) -7-tetradecenoic acid ( 3 ) , and ( Z ) -9-tetradecenoic acid . ( C ) Quantification of the amount of ( Z ) -5-tetradecenoic and ( Z ) -7-tetradecenoic acid relative to saturated tetradecanoic acid in D . simulans , D . sechellia , and D . melanogaster extracts . Both ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid are reduced in D . simulans relative to D . melanogaster and D . sechellia . DOI: http://dx . doi . org/10 . 7554/eLife . 04205 . 01310 . 7554/eLife . 04205 . 014Figure 5—source data 1 . Summary of gas chromatographs of extract from larval residue . Compound ID , Retention times , and the area under each peak from D . melanogaster , D . sechellia , and D . simulans larval residue extracts . DOI: http://dx . doi . org/10 . 7554/eLife . 04205 . 01410 . 7554/eLife . 04205 . 015Figure 5—figure supplement 1 . D . sechellia and D . simulans larvae are attracted to ( Z ) -5-tetradecenoic acid . Both D . sechellia and D . simulans larvae are attracted to ( Z ) -5-tetradecenoic acid , but not to vehicle controls . Data points are the preference index scores of small groups ( 10–12 individuals ) of larvae in response to 0 . 5 nmol/cm2 of candidate molecule . Bars represent means ± one standard deviation for each treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 04205 . 015 To determine if D . simulans larvae have lost sensitivity to an attractive pheromone , we tested whether D . simulans larvae were attracted to residue produced by D . melanogaster and D . sechellia . D . simulans larvae were as attracted to these residues , as were D . melanogaster and D . sechellia ( Figure 5A ) . Furthermore , D . simulans larvae were attracted to synthetic ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid , and their preference was indistinguishable from the attraction of D . melanogaster and D . sechellia to these molecules ( Figure 2E , Figure 5A; Figure 5—figure supplement 1 ) . Together , these data demonstrate that sensitivity to both ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid is unimpaired in D . simulans . Next , to determine if D . simulans does not produce attractive pheromones , we tested first if the other species were attracted to D . simulans residue . We found that both D . melanogaster and D . sechellia larvae were indifferent to D . simulans residue , but were attracted to D . sechellia residue ( Figure 5A ) . To determine if ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid production was specifically lost or attenuated in D . simulans , we tested if an attractive cue could be extracted from glass treated with larvae of these species using either hexane or acetone as a solvent . Extracts from glass treated with D . simulans larvae were not attractive to wild-type D . melanogaster larvae . In contrast , both D . sechellia hexane and acetone extracts were attractive to wild-type D . melanogaster larvae ( Figure 5A ) . We found that levels of both ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid are reduced in D . simulans relative to D . sechellia and D . melanogaster ( Figure 5B , C , Figure 5—source data 1 ) . These results suggest that larvae of D . simulans produce less of the attractive cues ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid than do D . melanogaster and D . sechellia larvae . Finally , we explored whether D . simulans produces a repellent cue . This hypothesis was supported , initially , by the observation that acetone extract of D . simulans larval residue was mildly repulsive to D . simulans larvae ( Figure 6B , C ) . We reasoned that if D . simulans produced a repellent , then this repellent may be able to block the attractiveness of ( Z ) -5-tetradecenoic acid and/or ( Z ) -7-tetradecenoic acid . We tested this hypothesis by supplementing D . simulans larval extract with the attractive pheromones ( Figure 6A ) . While ( Z ) -5-tetradecenoic acid alone was attractive to D . simulans , this activity was blocked completely by the addition of D . simulans extract ( Figure 6B , D ) . These data are consistent with the hypothesis that an unidentified repellent is present in the D . simulans extract . In contrast , the addition of ( Z ) -7-tetradecenoic acid to D . simulans extract partially rescued the attraction of D . simulans to the extract , but not to the level observed for ( Z ) -7-tetradecenoic acid tested alone ( Figure 6B , E ) . Last , while D . simulans larvae were attracted to a mixture of the two pheromones alone , the addition of both pheromones to D . simulans extract conferred no attractive activity ( Figure 6B , F ) . Taken together , these results suggest that the reduced levels of ( Z ) -7-tetradecenoic acid production , combined with an unidentified repellent cue released by D . simulans , generate the evolved differences in larval social behavior between these closely related species . Furthermore , the unidentified repellent cue may interact with ( Z ) -5-tetradecenoic acid to overcome the residual attractive activity of ( Z ) -7-tetradecenoic acid . 10 . 7554/eLife . 04205 . 016Figure 6 . The evolved change in attractiveness of D . simulans is caused by the reduced production of ( Z ) -7-tetradecenoic acid and production of an unidentified repellent cue . ( A ) A schematic of an assay to measure the effect of larval pheromones in the context of D . simulans extract . D . simulans cues were extracted off glass treated with larvae and then supplemented with ( Z ) -5-tetradecenoic acid , ( Z ) -7-tetradecenoic acids or an equimolar mixture of the two compounds , and then presented to larvae using a checkerboard assay . ( B ) The attraction of D . simulans larvae to acetone , D . simulans acetone extract , D . simulans acetone extract supplemented with 0 . 5 nmol/cm2 synthetic ( Z ) -5-tetradecenoic acid , 0 . 5 nmol/cm2 synthetic ( Z ) -7-tetradecenoic acid , D . simulans acetone extract supplemented with synthetic ( Z ) -7-tetradecenoic acid , a mixture of synthetic ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid , and D . simulans extract supplemented with the pheromone mixture . Treatments displayed significant heterogeneity ( ANOVA , F7 , 103 = 28 . 1 , p < 0 . 0001 ) . Data points are the preference index scores of small groups ( 10-15 individuals ) of larvae in response to each condition . ( C–F ) The mean preference index score over time for the experiments summarized in ( B ) . Dark lines represent the mean preference index score for each timepoint . Shaded areas represent one standard deviation . ( B , C ) D . simulans larvae are weakly repulsed by D . simulans residue ( one-sample t-test; p < 0 . 01 ) . ( B , D ) D . simulans extract blocks the attractive effect of synthetic ( Z ) -5-tetradecenoic acid resulting in weak repulsion ( one-sample t-test; p < 0 . 001 ) ( B , E ) The supplementation of synthetic ( Z ) -7-tetradecenoic acid partially rescues the activity of D . simulans extract . ( B , F ) D . simulans extract blocks the activity of a mixture of ( Z ) -5-tetradecenoic and ( Z ) -7-tetradecenoic acid resulting in weak repulsion ( one-sample t-test; p < 0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04205 . 016 We found that two members of the pickpocket family of DEG/ENaC channel subunit genes , ppk23 and ppk29 , are required for detecting ( Z ) -5-tetradecenoic and ( Z ) -7-tetradececoic acid . These genes play broad , non-redundant roles in sex pheromone-mediated behaviors , courtship and aggression , and are required for detecting the male pheromones , cVA , ( Z ) -5-pentacosene , and ( Z ) -7-tricosene , and the female pheromones , 7 , 11-heptacosadiene and 7 , 11-nonacosadiene ( Lu et al . , 2012; Thistle et al . , 2012; Toda et al . , 2012; Yuan et al . , 2013 ) . Each of these molecules are derived from long chain fatty acids and have relatively simple double bond structures , which raises the possibility that ppk23 and ppk29 detect features that are common to these pheromones . How members of the ppk family mechanistically contribute to pheromone detection has not been resolved . Members of this family form heteromeric protein complexes ( Eskandari et al . , 1999; Benson and et al . , 2002 ) . It has been proposed by several research groups that the responses to specific pheromones could be conferred by other channel subunits or by an accessory protein that forms a complex with ppk23 and/or ppk29 . Alternatively , ppk23 and ppk29 may act more generally in pheromone transduction or to maintain the excitability of neurons . Recent work has demonstrated that ppk29 is expressed more broadly than initially characterized , and that ppk29 mRNA regulates a second channel gene ( seizure ) to control membrane excitability of motor neurons through a protein independent mechanism ( Zheng et al . , 2014 ) . In either case , our work broadens the roles that ppk23 and ppk29 play in detecting pheromones . It is likely that other receptor genes are also expressed in R58F10 and confer its specificity for larval pheromones . Until our work , only one Drosophila pheromone signaling system where both the pheromone and a single discrete class of sensory neurons that mediate a pheromone-driven behavior had been described . This is the adult male-derived sex pheromone cis-vaccenyl acetate ( cVA ) and the Or65a/Or67d-expressing olfactory receptor neurons that signal both female attractiveness and elicit male–male aggression in adult flies . cVA is a volatile pheromone , and its detection is processed via the DA1 glomerulus in the antennal lobe ( Fernández and Kravitz , 2013 ) . The fact that cVA-driven behaviors are elicited by a single class of sensory neurons has allowed further anatomical and functional dissection of the downstream circuit , including identification of sexually dimorphic aspects of cVA processing ( Datta et al . , 2008; Ruta et al . , 2010; Kohl et al . , 2013 ) . In contrast , there is currently a much poorer understanding of how contact-mediated pheromones are detected and decoded by downstream circuits . These pheromones , unlike cVA , are detected by gustatory neurons and the signals are processed in the subesophagael zone and ventral nerve cord ( Thorne et al . , 2004 ) . Some of the genes and sensory neurons required to detect these pheromones have been described ( Dahanukar and Ray , 2011 ) and in a few cases , the sensitivity of some pheromone-sensing neurons to contact-mediated signals has been characterized using physiology and/or calcium imaging in the periphery ( Lacaille et al . , 2007; Thistle et al . , 2012; Toda et al . , 2012 ) . However , the genes and other genetic tools used to manipulate these sensory neurons are expressed broadly in multiple cell types and body regions ( Thorne et al . , 2004; Miyamoto and Amrein , 2008; Moon et al . , 2009; Weiss at al . , 2011; Liu et al . , 2012; Lu et al . , 2012; Starostina et al . , 2012; Thistle et al . , 2012; Toda et al . , 2012; Fan et al . , 2013 ) , and some of these genes are required also for detecting aversive compounds other than pheromones ( Moon et al . , 2009; Lee et al . , 2010; Weiss et al . , 2011; Ling et al . , 2014 ) . In addition , the complicated pattern of central innervation from multiple cells types in the subesophagael zone and ventral nerve cord has made it difficult to identify the subset of these cells that are pheromone responsive , and to characterize the downstream circuits . Our study directly links the detection of two novel contact-mediated pheromones— ( Z ) -5-tetradecenoic and ( Z ) -7-tetradecenoic acid—by the receptor neuron R58F10 to larval attraction . Furthermore , because R58F10 labels only a single pair of bilateral sensory neurons , our work provides the first clean entry point to determine how pheromones are processed in the subesophagael zone . We found that compounds deposited by larvae of related species in the D . melanogaster subgroup differ in their attractiveness to other larvae . D . simulans has evolved in at least two ways . First , the reduced attractiveness of D . simulans is caused , at least in part , by a reduction in ( Z ) -7-tetradecenoic acid production . Second , an unidentified repellent cue present in D . simulans residue blocks the attractive activity of the reduced levels of ( Z ) -5-tetradecenoic acid , and may act through a ( Z ) -5-tetradecenoic acid-dependent mechanism to block larval attraction to ( Z ) -7-tetradecenoic acid . Interactions between pheromones that cause behaviorally antagonistic effects , such as we have observed , have been studied most intensively in the context of plume following during moth courtship ( Baker , 2008 ) . In many species , female moths release blends of volatile attractive pheromones to attract courting males . The major components of these blends are often shared in closely related sympatric species of moths , while these blends differ in minor components . These species-specific minor components can block the attraction of males to the blend ( Vickers and Baker , 1992; Quero and Baker , 1999; Lelito et al . , 2008 ) , a mechanism that likely reduces interspecific courtship . An alternative hypothesis that can explain the evolved difference in larval attraction that we observed is that larvae are exquisitely sensitive to the ratio of ( Z ) -5-tetradecenoic acid to ( Z ) -7-tetradecenoic acid . We believe that this explanation is unlikely , because , although the levels of both pheromones are reduced in D . simulans , the ratio of these two pheromones is not significantly different from the ratios observed in D . sechellia and D . melanogaster ( Figure 6C ) . These results illustrate how a relatively simple change in pheromone profiles can have a profound effect on the social behavior of a species . It is not clear why D . simulans has evolved to be less attractive to other larvae . It is possible that D . simulans has evolved to avoid costs associated with generating aggregation cues ( Wertheim et al . , 2002 , 2003 ) . For example , aggregation could lead to an increase in intra-specific competition over food sources , which may outweigh the density-dependent advantage larvae experience while clustered . Alternatively , the costs associated with providing positional information by releasing pheromones , such as the increased risks of predation , parasitism , or even cannibalism by conspecific larvae ( Hassell , 2000; Vijendravarma et al . , 2013 ) might in some situations outweigh the benefits of aggregation . Finally , the geographic range of D . simulans is largely overlapping with other species in this subgroup , and larvae of different species can often be found exploiting the same resource . While D . simulans produces little ( Z ) -5-tetradecenoic acid or ( Z ) -7-tetradecenoic acid , it is still attracted to these molecules . When sharing habitat with heterospecific larvae , it is possible that D . simulans larvae could benefit from eavesdropping on aggregation signals produced by other species while avoiding the direct and indirect costs of producing these compounds themselves . Organisms , such as nematodes , or life stages , such as Drosophila larvae , that possess relatively simple behavioral repertoires may be thought to have relatively simple social lives . However , recent studies have revealed rich detail about their social environments . For example , nematodes produce a complex set of pheromone signals mediated by a diverse family of molecules called ascarisides ( Braendle , 2012 ) . These molecules act to coordinate aspects of development and serve as both aggregation and dispersal cues ( Macosko et al . , 2009; Yamada et al . , 2010; Srinivasan et al . , 2012 ) . Both the synthesis and behavioral responses to these pheromones are evolving rapidly , and may be adaptations related to ecological specialization of these nematodes ( Braendle , 2012; Choe et al . , 2012 ) . Our identification of Drosophila larval aggregation pheromones represents a similarly unexpected level of complexity in invertebrate behavior , and may be only the first example of multiple pheromones that mediate larval social experience in Drosophila . Drosophila stocks were maintained under standard laboratory conditions . Stocks used were Or83b-GAL4 ( provided by T . Lee ) , orco1 ( Larsson et al . , 2004 ) , Gr32a-VP16 and Gr32a1 ( Miyamoto and Amrein , 2008 ) , Gr33a-GAL4 and Gr33a1 ( Moon et al . , 2009 ) , ppk23-GAL4 , Δppk23 , UAS-ppk23 , ppk29-GAL4 , Δppk29 , and UAS-ppk29 ( Thistle et al . , 2012 ) , attP2 ( UAS_unc84-2XGFP ) ( provided by GL Henry ) , UAS-TNT-E , UAS-IMPTV1A ( Sweeney et al . , 1995 ) , yw; P{w[+mC]=UAS-mCD8::GFP . L}LL5 , and the Janelia enhancer fragment collection ( Jenett et al . , 2012 ) . Canton S was used as wild-type D . melanogaster . 20XUAS-GCAMP6-S ( attP40 ) was obtained from the Janelia GENIE project . D . sechellia 14021-0248 . 28 was obtained from the Drosophila Species Stock Center at the University of California , San Diego . D . simulans 5 is an isofemale stock collected in Princeton , New Jersey by M . Womack . Data were plotted using the Matlab script errorbarjitter , available at http://www . mathworks . com/matlabcentral/fileexchange/33658-errorbarjitter . For all comparisons , we performed ANOVAs with significance values estimated using a permutation approach ( Anderson , 2011 ) using custom MATLAB scripts: http://www . mathworks . com/matlabcentral/fileexchange/44307-randanova1 and http://www . mathworks . com/matlabcentral/fileexchange/44308-randanova2 . P-values for comparisons of individual treatments were estimated using a Tukey honest significant difference test for multiple comparisons . All behavioral responses were recorded from early third instar larvae at 25–26°C and 50% humidity . Larvae were isolated using standard procedures ( Louis et al . , 2008 ) . All video recordings were performed with SONY DCR-HC52 MiniDV Handycam Camcorders , backlit with a white lightboard . To measure the activity of larval residue , 100 mm × 15 mm polystyrene petri dishes ( Fisher Scientific International , Inc . , Hampton , New Hampshire ) filled with 2% agarose were treated with a high density of larvae , nearly coating the entire surface for 30 min . The larvae were removed and discarded , and half of the larval-treated agarose was cut away and replaced with untreated 2% agarose . Single test larvae were deposited in the center of each prepared assay plate and their locations were recorded for 10 min . Larval positions were analyzed afterward by hand in 10 s intervals . To isolate and assay the activity larval extract , we treated pre-washed WHEATON Glass 20 ml Scintillation Vials with isolated larvae standardized by volume for 30 min . We removed the larvae and extracted the larval residue off the glass surface for 20 min with 2 ml ( for behavioral assays ) or 200 µl ( for chemical analysis ) solvent . To measure the activity of extracts , we coated half of an agarose petri dish with extract and half with solvent . We deposited a single larva in the center of each prepared assay plate and recorded their locations for 10 min . Larva positions were determined afterward by hand in 30 s intervals . To test the attraction of larvae to individual compounds , each synthetic compound was solubilized in acetone and deposited onto a Corning 245 mm Square BioAssay Dish filled with 2% agarose . We used a 16 square stainless steel stencil adhered to the agarose to print a checkerboard pattern of control and treated squares . The dimensions of each square was 25 mm × 25 mm . Small groups ( 8–12 ) of larvae were place in the center of the assay area and monitored for 5 . 5 min . Larvae were tracked and preference indexes were calculated using custom MATLAB software ( Supplementary file 1 ) and using the MATLAB Image Processing Toolbox . Dodecanoic acid ( L4250-100G ) , tetradecanoic acid ( M3128-10G ) , ( Z ) -9-hexadecenoic acid ( P9417-100MG ) , hexadecanoic acid ( P0500-10G ) , ( Z ) -9-octadecenoic acid ( O1008-5G ) , and ( Z , Z ) -9 , 12-octadecanoic acid ( L1376-1G ) were obtained from Sigma-Aldrich , Inc . St . Louis , MO ( Z ) -5-tetradecenoic acid and ( Z ) -7-tetradecenoic acid were synthesized by Shanghai Medicilon , Inc . , Shanghai , China . Larval anterior sensory organs and brains were dissected and stained using standard protocols ( Colomb et al . , 2007 ) with 7E8A10 anti-elav ( Developmental Studies Hybridoma Bank , Iowa City , Iowa ) , nc82 anti-bruchpilot ( Developmental Studies Hybridoma Bank ) , anti-GFP ( Millipore , Billerica , MA ) , and Alexa Fluor dyes ( 1:200 , Invitrogen ) . Samples were imaged on a Leica DM5500 Q Microscope using Leica Microsystems LAS AP software , and analyzed with Fiji ( Schindelin et al . , 2012 ) . Fatty acid methyl ester ( FAME ) preparation: A reagent solution was prepared by adding 20 µl of a 2M TMS diazomethane solution ( Aldrich ) to 1 ml of analytical grade dichloromethane ( B&J ) . After vortexing , 200 µl of dry methanol was added and the solution was vortexed again . Although the reagent solution could be used for several days without any detectable degradation , it was prepared fresh for each set of samples . Each sample was dried with a gentle stream of N2 , and 50 µl of the reagent solution was then added . After vortexing , samples were placed in an 80°C oven for 30 min . After cooling to room temperature the samples were again dried down with a gentle stream of N2 and re-dissolved in 50 µl of dichloromethane for EI-GC/MS analyses . Known amounts of fatty acid standards were prepared and analyzed in the same way . Double bond locations were established by GC/MS after conversion of the methylesters to pyrrolidideds following the procedures of Andersson ( Andersson , 1978 ) . The method was modified for small sample analyses as follows: A reagent stock solution was prepared by adding 10 µl of glacial acetic acid ( Sigma/Aldrich ) to 1 ml of dry pyrrolidine ( Sigma/Aldrich ) . The dichloromethane was removed from each FAME sample ( including standards ) and dried by a stream of N2; 50–100 µl of the pyrrolidide reagent was then added . The capped samples were heated in an 80°C oven for 1 hr and afterward allowed to cool to room temperature . Then 500 µl of dichloromethane was added and the solution was extracted 2× with 500 µl of water . The organic phase was dried by a stream of N2 , and 50–100 µl of dichloromethane was then added . After vortexing , the samples were analyzed by EI-GC/MS . Known amounts of fatty acid standards were prepared and analyzed in the same way as FAME and pyrrolidies . Tentatively indentified fatty acids were confirmed by synthesis followed by GC/MS analyses as FAME and pyrrolidide derivatives . GC/MS FAME analyses: All samples were injected as 1-µl aliquots of dichloromethane extracts onto a gas chromatograph ( HP 6890 ) equipped with 30 m length , 0 . 25-mm internal diameter , 0 . 25-µm film thickness DB-1 capillary column ( Agilent , Palo Alto , CA , USA ) , interfaced to a 5973 Mass Selective Detector , operating in electron impact mode . Samples were introduced by splitless injection at 240°C . The Oven was held at 30°C for 1 min after injection and then increased 10°C/min to 260°C and held at 260°C for 6 min . The carrier gas was helium at an average velocity of 30 cm/s . An EI spectra library search was performed using the NIST11 library . When available , mass spectra and retention times were compared to those of known standards . To better facilitate separation and thus MS identification of specific unsaturated FAMEs , the GC temperature program was changed to 30°C for 1 min after injection and then programmed for a temperature increase of 15°C/min to 150°C followed by 5°C/min to 260°C and held at 260°C for 4 min . To analyze the heavier pyrrolidide derivatives the injector temperature was increased to 260°C . The GC oven was held at 30°C for 1 min after injection and then increased 20°C/min to 200°C then by 5°C/min to 280°C and held at 280°C for 4 . 5 min . The EI spectra were interpreted using the NIST library and standard procedures for pyrrolidide interpretation ( Andersson and Holman , 1974; Andersson , 1978 ) . The cuticle of third instar larvae was too thick to observe GCAMP fluorescence in DOG neurons , but signal could be observed through second instar larva cuticle . The activity of pheromone-sensing neurons was performed on second instar larvae carrying two copies of the Janelia enhancer R58F10-GAL4 and two copies of 20XUAS-GCaMP6-S ( the slow , sensitive variant ) . Larvae were isolated and washed as described above . Hydrophobic test compounds were mobilized in adult hemolymph saline ( AHS ) ( Wang et al . , 2003 ) using detergent , CYMAL-7 ( 1 . 5 critical micelle count ( CMC ) ) ( Affymetrix; C327 , Santa Clara , CA ) . Larvae were partially dissected in AHS and mounted on a custom-built , acrylic , 75 mm × 25 mm slide , with a 2 mm slot cut down the center for test compound delivery ending in a small reservoir . Larvae were positioned perpendicular to and facing the slot and loosely fixed into position with a 22 mm coverslip ( Fisher Scientific; 12-542-B . ) and the slide was flushed with AHS . GCaMP fluorescence was measured by scanning at 1 Hz on a Leica DM5500 Q Microscope using Leica Microsystems LAS AP software . Neurons were imaged for 30 s to establish baseline fluorescence , after which 200 µl AHS with 1 . 5 CMC CYMAL-7 ( control ) or AHS with 1 . 5 CMC CYMAL-7 and 1 nM test compound were gently injected into the central slot . Recordings continued for 150 s . Changes in fluorescence were calculated using MATLAB Image Processing Toolbox and Fiji . To calculate the appropriate concentrations of pheromone to test , we posited that the larva might detect molecules within 1 mm2 in behavioral assays , and within 1 mm3 around its terminal organ in calcium imaging experiments . We therefore estimate that 50 fm/cm2 and 1 nM amount to 3 . 0 × 108 molecules and 6 . 0 × 108 molecules , respectively .	The release of chemical signals called pheromones is a common tactic used by animals in many social situations , such as to attract potential mates or to follow trails left by other members of their colony . Larvae of the fruit fly Drosophila melanogaster—a species commonly studied in the laboratory—gather together when sharing a food source and then cooperate in a way that may increase how efficiently they feed . It has been proposed that pheromones coordinate this behavior , but no larval pheromones had been identified . Mast et al . noticed that Drosophila larvae crawling on a surface tended to occupy areas where other larvae had crawled before . This suggested that larvae had left attractive chemicals on the surface . Mast et al . identified two such substances by analyzing the chemicals left on the surface and then by testing the response of larvae to each compound . Ultimately , Mast et al . found that a single sensory neuron in the larva is responsible for detecting these attractive chemical signals . Furthermore , two genes called pickpocket23 and pickpocket29 control this response . These genes were previously known for their roles in detecting sex pheromones , and they are members of a diverse family of calcium channel subunits that are involved in detecting multiple ‘sensory modalities’ such as touch and taste . When either pickpocket23 or pickpocket29 are inactivated , larvae ignore the social cues left by their neighbors . Mast et al . also looked for an evolutionary role for these pheromones . Larvae of a closely related fly species called Drosophila simulans produce a subtly different blend of compounds to D . melanogaster , and this blend is not attractive to any of the species tested . While Drosophila simulans larvae were not attracted to the cues left by their own species , they were attracted to the pheromones produced by Drosophila melanogaster , indicating that they retain the sensory mechanisms to detect and respond to these pheromones . These results suggest that larvae experience a rapidly evolving , complex , pheromone-rich environment that may help them tailor their behavior to survive .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "neuroscience" ]	2014	Evolved differences in larval social behavior mediated by novel pheromones
Neuroregeneration is a dynamic process synergizing the functional outcomes of multiple signaling circuits . Channelrhodopsin-based optogenetics shows the feasibility of stimulating neural repair but does not pin down specific signaling cascades . Here , we utilized optogenetic systems , optoRaf and optoAKT , to delineate the contribution of the ERK and AKT signaling pathways to neuroregeneration in live Drosophila larvae . We showed that optoRaf or optoAKT activation not only enhanced axon regeneration in both regeneration-competent and -incompetent sensory neurons in the peripheral nervous system but also allowed temporal tuning and proper guidance of axon regrowth . Furthermore , optoRaf and optoAKT differ in their signaling kinetics during regeneration , showing a gated versus graded response , respectively . Importantly in the central nervous system , their activation promotes axon regrowth and functional recovery of the thermonociceptive behavior . We conclude that non-neuronal optogenetics targets damaged neurons and signaling subcircuits , providing a novel strategy in the intervention of neural damage with improved precision . Inadequate neuroregeneration remains a major roadblock toward functional recovery after nervous system damage such as stroke , spinal cord injury ( SCI ) , and multiple sclerosis . Extracellular factors from oligodendrocyte , astroglial , and fibroblastic sources restrict axon regrowth ( Liu et al . , 2006; Yiu and He , 2006; Liu et al . , 2011; Lu et al . , 2014; Schwab and Strittmatter , 2014 ) but eliminating these molecules only allows limited sprouting ( Sun and He , 2010 ) , suggesting a down-regulation of the intrinsic regenerative program in injured neurons ( Sun and He , 2010; He and Jin , 2016 ) . The neurotrophic signaling pathway , which regulates neurogenesis during embryonic development , represents an important intrinsic regenerative machinery ( Ramer et al . , 2000 ) . For instance , elimination of the PTEN phosphatase , an endogenous brake for neurotrophic signaling , yields axonal regeneration ( Park et al . , 2008 ) . An important feature of the neurotrophin signaling pathway is that the functional outcome depends on signaling kinetics ( Marshall , 1995 ) and subcellular localization ( Watson et al . , 2001 ) . Indeed , neural regeneration from damaged neurons is synergistically regulated by multiple signaling circuits in space and time . However , pharmacological and genetic approaches do not provide sufficient spatial and temporal resolutions in the modulation of signaling outcomes in terminally differentiated neurons in vivo . Thus , the functional link between signaling kinetics and functional recovery of damaged neurons remains unclear . The emerging non-neuronal optogenetic technology uses light to control protein-protein interaction and enables light-mediated signaling modulation in live cells and multicellular organisms ( Zhang and Cui , 2015; Khamo et al . , 2017; Johnson and Toettcher , 2018; Leopold et al . , 2018; Dagliyan and Hahn , 2019; Goglia and Toettcher , 2019 ) . By engineering signaling components with photoactivatable proteins , one can use light to control a number of cellular processes , such as gene transcription ( Motta-Mena et al . , 2014; Wang et al . , 2017 ) , phase transition ( Shin et al . , 2017; Dine et al . , 2018 ) , cell motility ( Wu et al . , 2009 ) and differentiation ( Khamo et al . , 2019 ) , ion flow across membranes ( Kyung et al . , 2015; Ma et al . , 2018 ) , and metabolism ( Zhao et al . , 2018; Zhao et al . , 2019 ) , to name a few . We have previously developed optogenetic systems named optoRaf ( Zhang et al . , 2014; Krishnamurthy et al . , 2016 ) and optoAKT ( Ong et al . , 2016 ) , which allow for precise control of the Raf/MEK/ERK and AKT signaling pathways , respectively . We demonstrated that timed activation of optoRaf enables functional delineation of ERK activity in mesodermal cell fate determination during Xenopus laevis embryonic development ( Krishnamurthy et al . , 2016 ) . However , it remains unclear if spatially localized , optogenetic activation of ERK and AKT activity allows for subcellular control of cellular outcomes . In this study , we used optoRaf and optoAKT to specifically activate the Raf/MEK/ERK and AKT signaling subcircuits , respectively . We found that both optoRaf and optoAKT activity enhanced axon regeneration in the regeneration-potent class IV da ( C4da ) and the regeneration-incompetent class III da ( C3da ) sensory neurons in Drosophila larvae , although optoRaf but not optoAKT enhanced dendritic branching . Temporally programmed and spatially restricted light stimulation showed that optoRaf and optoAKT differ in their signaling kinetics during regeneration and that both allow spatially guided axon regrowth . Furthermore , using a thermonociception-based behavioral recovery assay , we found that optoRaf and optoAKT activation led to effective axon regeneration as well as functional recovery after central nervous system ( CNS ) injury . We note that most of the previous optogenetic control of neural repair studies were based on channelrhodopsion in C . elegans ( Sun et al . , 2014 ) , mouse DRG culture ( Park et al . , 2015a ) or motor neuron-schwann cell co-culture ( Hyung et al . , 2019 ) . Another study used blue-light activatable adenylyl cyclase bPAC to stimulate neural repair in mouse refractory axons ( Xiao et al . , 2015 ) . These work highlighted the feasibility of using optogenetics to study neural repair but did not pin down the exact downstream signaling cascade mediating neuronal repair . Additionally , most studies focused on peripheral neurons that are endogenously regenerative . Here , we specifically activated the ERK and AKT signaling pathways and performed a comprehensive study of neural regeneration in both the peripheral nervous system ( PNS ) and CNS neurons in live Drosophila . We envision that features provided by non-neuronal optogenetics , including reversibility , functional delineation , and spatiotemporal control , will lead to a better understanding of the link between signaling kinetics and functional outcome of neurotrophic signaling pathways during neuroregeneration . To reversibly control the Raf/MEK/ERK and AKT signaling pathways , we constructed a single-transcript optogenetic system using the p2A bicistronic construct that co-expresses fusion-proteins with the N-terminus of cryptochrome-interacting basic-helix-loop-helix ( CIBN ) and the photolyase homology region of cryptochrome 2 ( CRY2PHR , abbreviated as CRY2 in this work ) . Following a similar design of the optimized optoRaf ( Krishnamurthy et al . , 2016 ) , we improved the previous optogenetic AKT system ( Ong et al . , 2016 ) with two tandom CIBNs ( referred to as optoAKT in this work ) ( Figure 1—figure supplement 1A ) . Consistent with previous studies , the association of CIBN and CRY2 took about 1 s , and the CIBN-CRY2 complex dissociated in the dark within 10 min ( Kennedy et al . , 2010; Zhang et al . , 2014 ) . The fusion of Raf or AKT does not affect the association and dissociation kinetics of CIBN and CRY2 and multiple cycles of CRY2-CIBN association and dissociation can be triggered by alternating light-dark treatment ( Figure 1—figure supplement 1B–1D , Videos 1 and 3 ) . Activation of optoRaf and optoAKT resulted in nuclear translocation of ERK-EGFP ( Figure 1A , Video 2 ) and nuclear export of FOXO3-EGFP ( Figure 1B , Video 4 ) resolved by live-cell fluorescence imaging , indicating activation of the ERK and AKT signaling pathways , respectively . Western blot analysis on pERK ( activated by optoRaf ) in HEK293T cells showed that pERK activity ( Figure 1C ) increased within 10 min blue light stimulation and returned to the basal level 30 min after the blue light was shut off ( Figure 1D ) . There was a slight decrease in pERK activity upon optoRaf activation for over 10 min , likely due to negative feedback , which has been consistently observed in previous studies ( Zhou et al . , 2017 ) . On the other hand , continuous light illumination maintained a sustained activation of pCRY2-mCh-AKT within an onset of 10 min ( Figure 1E ) . The inactivation kinetics of pAKT was 30 min , similar to that of pERK ( Figure 1F and G ) . Note we use only the phosphorylated and total forms of CRY2-mCh-AKT to quantify the light response of optoAKT because the endogenous AKT does not respond to light . Binding of neurotrophins to their receptor activates multiple downstream signaling subcircuits , including the Raf/MEK/ERK and AKT pathways . Delineation of signaling outcomes of individual subcircuits remains difficult with pharmacological assays given the unpredictable off-targets of small-molecule drugs . We hypothesized that optoRaf and optoAKT could delineate signaling outcomes because they bypass ligand binding and activate the intracellular signaling pathway . To test this hypothesis , we probed phosphorylated proteins , including pERK and pAKT with WB analysis in response to light-mediated activation of optoRaf and optoAKT . Results show that optoRaf activation does not increase endogenous pAKT ( Figure 1H and I ) . Similarly , optoAKT activation does not increase pERK or endogenous pAKT ( Figure 1H and I ) . Thus , at the level of ERK and AKT , optoRaf and optoAKT do not show crosstalk activity in mammalian cells . Although activation of both optoRaf and optoAKT bypasses ligand-receptor binding , it remains unclear if other upstream signaling molecules are required to activate optoRaf and optoAKT . Endogenous Raf1 activation requires its membrane translocation mediated by the GTP-bound form of Ras , followed by phosphorylation at several residues , including Ser338 , which is located in the junction region between the regulator domain and the kinase domain ( Mason et al . , 1999 ) . Replacement of Ser338 with alanine abolishes Raf activation ( Xiang et al . , 2002; Goetz et al . , 2003 ) . Note that phosphorylation of Ser338 itself does not activate Raf but is a prerequisite regulatory event for Raf activation ( Diaz et al . , 1997 ) , likely leading to Raf dimerization ( Takahashi et al . , 2017 ) . To determine if Ser338 phosphorylation is involved in optoRaf activation , we probed the phosphorylation state of CRY2-mCh-Raf upon blue light stimulation and found that indeed Ser338 is significantly phosphorylated upon blue light stimulation ( Figure 1—figure supplement 1E , top panel , lane 5–8 ) . We then constructed the S338A mutant for optoRaf and confirmed that no phosphorylation occurs on S338 in optoRaf S338A . Importantly , optoRafS338A mutant significantly reduced the blue-light-mediated pERK activation ( Figure 1—figure supplement 1E , third panel ) . This result suggests that optoRaf activation through membrane translocation requires upstream kinases . Similarly , for optoAKT , it remains unclear if its activation requires upstream PI3K signaling . Full activation of AKT requires phosphorylation on both T308 in the activation loop of the catalytic protein kinase core and S473 in a C-terminal hydrophobic motif ( Manning and Toker , 2017 ) . PH-domain-containing kinases such as PDK1 are essential for AKT activation by phosphorylating AKT on T308 , whereas the mechanistic target of rapamycin ( mTOR ) complex 2 ( mTORC2 ) phosphorylates AKT on S473 . In addition to verifying that phosphorylation of S473 occurs during optoAKT activation ( Figure 1E and Figure 1—figure supplement 1F ) , we probed pT308 for optoAKT upon blue light stimulation ( Figure 1—figure supplement 1F ) . We found that light stimulation indeed enhances the level of pT308 in optoAKT , indicating that upstream kinases ( e . g . PDK1 ) are involved in the activation of optoAKT upon membrane translocation of CRY2-mCh-AKT . We verified that the activation of optoRaf enhances PC12 cell neuritogenesis , which is consistent with previous studies ( Zhang et al . , 2014; Krishnamurthy et al . , 2016 ) . The neuritogenesis ratio is defined as the ratio between the number of transfected cells with at least one neurite longer than the size of the cell body and the total number of transfected cells . Twenty-four hours of blue light stimulation ( 0 . 2 mW/cm2 ) increased the neuritogenesis ratio from the basal level ( 0 . 24 ± 0 . 04 ) to 0 . 52 ± 0 . 03 ( Figure 1J and L ) . Light-mediated activation of optoAKT , on the other hand , did not increase the neuritogenesis ratio ( 0 . 23 ± 0 . 04 in the dark versus 0 . 20 ± 0 . 02 under light ) ( Figure 1K and L ) . A membrane-targeted Raf1 ( Raf1-EGFP-CaaX ) was used as a positive control , which caused significant neurite outgrowth independent of light treatment ( 0 . 65 ± 0 . 01 in the dark versus 0 . 63 ± 0 . 01 under light ) . Expression of CIBN2-EGFP-CaaX ( without CRY2-Raf1 ) , a negative control , did not increase PC12 neurite outgrowth either in the dark ( 0 . 20 ± 0 . 02 ) or under light ( 0 . 14 ± 0 . 01 ) ( Figure 1L ) . To determine the efficacy of the optogenetic tools in vivo , we generated transgenic flies with inducible expression of optoRaf ( UAS-optoRaf ) and optoAKT ( UAS-optoAKT ) . We induced the expression of the transgenes in a type of fly sensory neurons , the dendritic arborization ( da ) neurons , which have been used extensively to study dendrite morphogenesis and remolding ( Gao et al . , 1999; Grueber et al . , 2002; Sugimura et al . , 2003; Kuo et al . , 2005; Williams and Truman , 2005; Kuo et al . , 2006; Williams et al . , 2006; Parrish et al . , 2007 ) . Using the pickpocket ( ppk ) -Gal4 , we specifically expressed optoRaf in the class IV da ( C4da ) neurons , to test whether light stimulation would activate the Raf/MEK/ERK pathway . At 72 hr after egg laying ( h AEL ) , wild-type ( WT ) and optoRaf-expressing larvae were anesthetized with ether and subjected to whole-field continuous blue light for 5 , 10 and 15 min , while as a control , another transgenic group was incubated in the dark ( 0 min ) . The larval body walls were then dissected and immunostained with the pERK1/2 antibody , as a readout of the Raf/MEK/ERK pathway activation . We found that 5 min light stimulation was sufficient to significantly increase the pERK signal in the cell body of C4da neurons in optoRaf-expressing larvae , while 15 min illumination enhanced pERK activation and induced ERK translocation into the nucleus ( Figure 2—figure supplement 1A and B ) . Compared with the optoRaf-expressing larvae incubated in the dark , 15 min light illumination resulted in a more than twofold increase of pERK fluorescence intensity ( Figure 2—figure supplement 1B ) . Similarly , in C4da neurons expressing optoAKT , the 10- and 15 min blue light stimulation significantly increased the fluorescence intensity of phospho-p70 ribosomal S6 kinase ( phospho-p70S6K ) ( Figure 2—figure supplement 1C and D ) , which functions downstream of AKT ( Lizcano et al . , 2003; Miron et al . , 2003 ) . These results collectively demonstrate that optoRaf and optoAKT were robustly expressed in flies and blue light is sufficient to activate the optogenetic effectors in vivo . The phosphorylation of ERK/p70S6K in response to blue light was only observed in C4da neurons but not in other classes of da neurons or epithelial cells ( Figure 2—figure supplement 2 ) , proving they are triggered by optoRaf/optoAKT , which were only expressed in C4da neurons under the control of ppk-Gal4 . Furthermore , we found ERK was not activated in optoAKT-expressing neurons ( Figure 2A , right-most panel ) , nor was phospho-p70S6K in the optoRaf-expressing larvae ( Figure 2—figure supplement 3 , right-most panel ) , confirming that there is no crosstalk between these two systems , at least at the node of pERK and p70S6K . We also examined the inactivation kinetics of ERK/phospho-p70S6K after blue light was shut off ( Figure 2A–D ) . The pERK ( Figure 2C ) and pAKT ( Figure 2D ) activity started to decrease as the light was shut off , although the decay rate of pERK decays appears slower than pAKT . Compared with the transgenic larvae kept in the dark , there was no significant difference in phospho-p70S6K intensity at 15 min after blue light was turned off ( Figure 2D ) . In contrast , a 15-min off time reduces pERK activity , but the level remains higher than the basal level . When the off-time was increased to 45 min , there is still a slightly higher pERK activity than the dark control . The difference in the inactivation kinetics may reflect distinct signaling sensitivity between Raf and AKT in optoRaf and optoAKT , respectively . These results confirmed that the intermittent pattern of light stimulation could modulate the temporal profile of ERK and AKT signaling activities . We next investigated if optoRaf or optoAKT activation would affect neural development such as dendrite morphogenesis . We labeled C4da neurons with ppk-CD4tdGFP and reconstructed the dendrites of the lateral C4da neurons – v'ada . Without light stimulation , the dendrite complexity of neurons in transgenic larvae was comparable to that of WT ( Figure 2F and G ) . However , optoRaf activation resulted in a significant increase in both total dendrite length and branch number , whereas optoAKT activation exhibited a slight reduction in dendritic branching ( Figure 2E–G ) . These results confirm the possibility of independently activating the Raf/MEK/ERK and AKT signaling pathways in flies with our optogenetic tools , prompting us to test the feasibility of their in vivo applications , such as promoting axon regeneration with high spatial and temporal resolution . Administration of neurotrophins to damaged peripheral neurons results in functional regeneration of sensory axons into the adult spinal cord in rat ( Ramer et al . , 2000 ) . Here , our photoactivatable transgenic flies empower precise spatiotemporal control of the neurotrophic signaling in live animals . To test whether light-mediated activation of the Raf/MEK/ERK or AKT signaling subcircuits would also promote axon regrowth , we used a previously described Drosophila da sensory neuron injury model ( Song et al . , 2012; Song et al . , 2015 ) . Da neurons have been shown to possess distinct regeneration capabilities among different sub-cell types , and between the PNS and CNS , resembling mammalian neurons ( Song et al . , 2012; Song et al . , 2015 ) . In particular , the C4da neurons regenerate their axons robustly after peripheral injury , while the C3da neurons largely fail to regrow . Moreover , the axon regeneration potential of C4da neurons is also diminished after CNS injury . First , we asked whether optoRaf or optoAKT activation can enhance axon regeneration in the regeneration-competent C4da neurons in the PNS . We severed the axons of C4da neurons ( labeled with ppk-CD4tdGFP ) with a two-photon laser at 72 hr AEL , verified axon degeneration at 24 hr after injury ( AI ) and assessed axon regeneration at 48 hr AI . At this time point , about 79% C4da neurons in WT showed obvious axon regrowth , and the regeneration index ( Song et al . , 2012; Song et al . , 2015 ) , which refers to the increase in axon length normalized to larval growth ( Figure 3—figure supplement 1A and B , and Materials and methods ) , was 0 . 3810 ± 0 . 06653 ( Figure 3A–C ) . Strikingly , C4da neurons expressing optoRaf or optoAKT showed further enhanced regeneration potential in response to blue light , leading to a significant increase in the regeneration index ( optoRaf: 0 . 7102 ± 0 . 1033; optoAKT: 0 . 7354 ± 0 . 07755 ) , while there was no difference between WT and unstimulated transgenic flies ( Figure 3A–C ) . In order to test the potential synergy between optoRaf and optoAKT , we co-expressed both transgenes in C4da neurons . While there was a slight increase in the regeneration percentage , activation of both ERK and AKT pathways in the same neuron did not further increase the regeneration index ( 0 . 7387 ± 0 . 08390 ) ( Figure 3A–C ) . The light stimulation paradigm used in the aforementioned regeneration experiments was constant blue light applied immediately after the injury . We reason that intermittent light stimulation may provide insights into the signaling kinetics in vivo and finetune axon regeneration dynamics . Therefore , instead of constant blue light illumination , we delivered two sets of programmed light patterns to injured larvae , 15 min on-15 min off or 15 min on-45 min off per cycle for 48 hr ( Figure 3D ) . We found that , for optoRaf-expressing C4da neurons , when the off-time was 15 min , the intermittent light stimulation was sufficient to accelerate axon regrowth , with the regeneration index ( 0 . 6352 ± 0 . 09627 ) significantly increased compared with larvae incubated in the dark ( Figure 3E and F ) . However , when the off-time was 45 min , the intermittent light failed to promote axon regeneration ( Figure 3E and F ) . Considering that pERK activity remains slightly higher than the basal level after 45 min dark incubation ( Figure 2C ) , the regeneration failure at 45 off-time suggests that optoRaf regulates C4da axon regeneration in a threshold-gated manner . On the other hand , C4da neurons expressing optoAKT displayed a graded response: a moderate increase of regeneration index ( 0 . 6278 ± 0 . 09801 ) in response to the 15 min on-15 min off light and a smaller uptick ( 0 . 5312 ± 0 . 06963 ) to the 15 min on-45 min off light; both were less effective than the constant light stimulation ( Figure 3E and F ) . These results suggest that although the higher frequency of light stimulation generally resulted in stronger regeneration potential in the transgenic flies , constant light was not always required for maximum axon regeneration . Moreover , optoRaf and optoAKT differ in their signaling kinetics during regeneration , showing a gated versus graded response , respectively . We next determined whether optoRaf or optoAKT activation would trigger regeneration in C3da neurons , which are normally incapable of regrowth ( Song et al . , 2012 ) . C3da neurons were labeled with 19–12-Gal4 , UAS-CD4tdGFP , repo-Gal80 and injured using the same paradigm as C4da neurons . Compared with WT , which exhibited poor axon regeneration ability demonstrated by the low regeneration percentage and the negative regeneration index ( −0 . 03201 ± 0 . 02752 ) ( Figure 3G–I ) , light stimulation significantly increased the regeneration index in optoRaf- or optoAKT-expressing larvae to 0 . 1298 ± 0 . 04637 or 0 . 1354 ± 0 . 06161 , respectively ( Figure 3G–I ) . Similar to C4da neurons , activation of both Raf/MEK/ERK and AKT pathways failed to further enhance axon regrowth compared to optoRaf or optoAKT activation alone ( Figure 3—figure supplement 2 ) . This result confirms that the actions of optoRaf and optoAKT are not additive in promoting axon regeneration , suggesting that these two subcircuits may share the same downstream components in neuroregeneration ( see Discussion ) . Altogether , these data indicate that optoRaf and optoAKT activation not only accelerates axon regeneration but also converts regeneration cell-type specificity . While C4da neurons are known to possess the regenerative potential , it is unclear whether the regenerating axons navigate correctly . To address this question , we focused on v'ada – the lateral C4da neurons . Uninjured v'ada axons grow ventrally , showing a typical turn and then join the axon bundle with the ventral C4da neurons ( Figure 3—figure supplement 1A ) . We found that their regenerating axons preferentially regrew away from the original ventral trajectory ( Figure 4A and B white bars ) . More than 60% v'ada axons bifurcated and formed two branches targeting opposite directions . In the majority cases in WT , the ventral branch , which extends toward the correct trajectory , regenerated less frequently than the dorsal branch , with 15% v'ada containing only the ventral branch ( Figure 4A and B black bars ) . One possibility is that the ventral branch encounters the injury site , which may retard its elongation . As a result , only a minority of regenerating axons are capable of finding the correct path . The poor pathfinding of regenerating axons was similar among WT and the transgenic larvae , regardless of whether incubated with whole-field light or in the dark ( Figure 4B ) . Thus , proper guidance of the regenerating axons toward the correct trajectory remained to be resolved . We thus investigated whether spatially restricted activation of the neurotrophic signaling using our optogenetic system could guide the regenerating axons . To specifically enhance the regrowth of the ventral branch , we used a confocal microscope to focus the blue light ( delivered by the 488 nm argon-ion laser ) on the ventral branch for 5 min at 24 hr AI . The lengths of both the ventral and dorsal branches were measured at 24 hr AI and 48 hr AI . We subtracted the increased dorsal branch length ( Δdorsal ) from the increased ventral branch length ( Δventral ) , then divided that by the total increased length of these two branches ( Figure 4D ) . This value was defined as the relative regeneration ratio . If the dorsal branch exhibits more regenerative potential , the ratio would be negative; otherwise , it would be positive . Without light stimulation , the relative regeneration ratio of the transgenic larvae ( optoRaf: −0 . 6062 ± 0 . 1453; optoAKT: −0 . 5530 ± 0 . 1011 ) was comparable to that of WT ( −0 . 5786 ± 0 . 08229 ) ( Figure 4C and D ) , confirming preferred regrowth of the dorsal branch . Strikingly , the 5 min local blue light stimulation significantly increased the ratio in optoRaf- or optoAKT-expressing v’ada ( optoRaf: 0 . 04762 ± 0 . 1123; optoAKT: −0 . 1725 ± 0 . 09560 ) , while this transient stimulation resulted in no difference in WT ( −0 . 6018 ± 0 . 1290 ) ( Figure 4C and D ) . This result indicates that a single pulse of local light stimulation was sufficient to lead to preferential regrowth of the ventral branch . Notably , although whole-field light illumination could significantly promote axon regrowth , it failed to increase the relative regeneration ratio in transgenic larvae ( Con . on optoRaf: −0 . 7048 ± 0 . 1015; Con . on optoAKT: −0 . 5517 ± 0 . 09644 ) ( Figure 4D ) , revealing the difference between activating the neurotrophic signaling in a whole neuron and a single lesioned axon branch . On the other hand , while a 5-min local light stimulation did not lead to an overall enhancement of axon regrowth , it provided adequate guidance instructions for the regenerating axons to make the correct choice . Achieving functional axon regeneration after CNS injury remains a major challenge in neural repair research . Motivated by the capacity of optoRaf and optoAKT to accelerate axon regeneration in the PNS , we went on to determine whether they also show efficacy after CNS injury . We focused on the axons of C4da neurons , which project into the ventral nerve cord ( VNC ) and form a ladder-like structure . Each pair of axon bundles correspond to one body segment in an anterior-posterior pattern ( Li et al . , 2020 ) . We injured the abdominal A6 and A3 bundles by laser as previously described ( Song et al . , 2012; Li et al . , 2020; Figure 5—figure supplement 1 ) , and confirmed axon degeneration at 24 hr AI ( Figure 5A ) . At 48 hr AI , we found that axons began to extend from the retracted axon stem and towards the commissure region . We defined a commissure segment as regenerated only when at least one axon extended beyond the midline of the commissure region or joined into other intact bundles ( Figure 5—figure supplement 1 ) . In WT , only 16% of lesioned commissure segments displayed obvious signs of regrowth ( Figure 5A and B ) . To quantify the extent of regrowth , we measured the length of the regrown axons and normalized that to the length of a commissure segment – regeneration index ( Figure 5—figure supplement 1 , Materials and methods ) . After light stimulation , the regeneration indexes of the two transgenic lines ( optoRaf: 5 . 375 ± 0 . 3391; optoAKT: 4 . 765 ± 0 . 4236 ) were significantly increased compared with the WT control ( 2 . 643 ± 0 . 3050 ) , and the percentage of regenerating commissure segments also exhibited a mild increase in both optoRaf- and optoAKT- expressing larvae ( Figure 5A–C ) . On the other hand , there was no significant difference between WT and the unstimulated transgenic flies ( Figure 5A–C ) . This result suggests that both signaling subcircuits reinforce C4da neuron axon regeneration in the CNS . We then tested whether the axon regrowth in the CNS induced by optoRaf or optoAKT activation leads to behavioral improvement . We utilized a recently established behavioral recovery paradigm based on larval thermonociception ( Figure 6A , Materials and methods ) ( Li et al . , 2020 ) . In brief , we injured the A7 and A8 C4da neuron axon bundles in the VNC , which correspond to the A7 and A8 body segments in the periphery . We then assessed the nociceptive behavior in these larvae in response to a 47°C heat probe applied at the A7 or A8 segments at 24 and 48 hr AI . Since C4da neurons are essential for thermonociception , injuring A7 and A8 axon bundles in the VNC would lead to an impaired nociceptive response to the heat probe specifically at body segments A7 and A8 . Indeed , all the injured larvae exhibited diminished response at 24 hr AI , while the total score is approaching three in uninjured WT larvae ( Figure 6B ) . At 48 hr AI , substantial recovery was observed in the two transgenic groups with light stimulation , whereas WT showed a very limited response and a low recovery percentage ( Figure 6B and C ) . Both the response score and the percentage of larvae exhibiting behavioral recovery in these two groups were more than twice as that of the WT , while the unstimulated groups were comparable to WT . Altogether , these results demonstrate that our optogenetic system empowers ligand-free and non-invasive control of the Raf/MEK/ERK and AKT pathways in flies , which not only promote axon regeneration after injury but also benefit functional recovery , suggesting that the regenerated axons may rewire and form functional synapses . Neurotrophins are known to activate Trk receptors and trigger the Raf/MEK/ERK , AKT , and PLCγ pathways which are involved in cell survival , neural differentiation , axon and dendrite growth and sensation ( Bibel and Barde , 2000; Huang and Reichardt , 2001; Chao , 2003; Cheng et al . , 2011; Joo et al . , 2014 ) . Here , we used optogenetic systems to achieve specific and reversible activation of the neurotrophin subcircuits , including the Raf/MEK/ERK ( via optoRaf ) and AKT ( via optoAKT ) signaling pathways . We further verified that optoRaf and optoAKT did not show crosstalk at the level of phosphorylated ERK and AKT proteins , and activation of optoRaf but not optoAKT promoted PC12 cell differentiation . Note that in the canonical growth factor signaling pathways , crosstalk actually occurs between the ERK and AKT signaling pathways , particularly at the upstream signaling node such as Ras . Indeed , the binding of growth factors to their receptors activates the transmembrane receptor tyrosine kinase , which recruits adaptor protein such as Grb2 ( growth factor receptor-bound protein ) and Sos ( son of sevenless ) , a guanine exchange factor ( GEF ) for Ras . Sos then transforms the inactive , GDP-bound Ras to an active , GTP-bound Ras , which then recruits multiple proteins , including Raf and PI3K , an upstream kinase for AKT , to the plasma membrane . Thus , Ras serves as a common signaling node and therefore creates possible signaling crosstalk between the PI3K-AKT and Raf-MEK-ERK pathway . Another possible signaling crosstalk arises from the PI3K-mediated production of phospholipids , which could recruit a number of signaling molecules containing the lipid-binding domain ( e . g . , PH domain ) including Sos , which then affects Ras/Raf activation . Our observation that optoRaf and optoAKT do not crosstalk ( i . e . optoRaf does not activate the AKT downstream effector p70S6K; optoAKT does not activate pERK ) may arise from the fact that both optoRaf and optoAKT bypass the ligand-binding , receptor activation , Ras activation , and phospholipid production signaling steps . Activation of optoRaf and optoAKT does require upstream signaling molecules ( e . g . kinases ) . However , there could be common downstream signaling molecules ( such as transcription factors ) that mediate the effects of neural regeneration by optoRaf and optoAKT . While ongoing efforts aim to elucidate these common signaling effectors , evidence from previous literature ( some from other cell types ) implies several possible candidates such as CREB ( cAMP response element-binding protein ) and FOXO ( forkhead box transcription factors ) . Activation of Raf leads to phosphorylation of CREB , a family of transcription factors that regulate cell survival ( Ginty et al . , 1994 ) . Evidence also suggests that CREB is a regulatory target for AKT ( Du and Montminy , 1998 ) . Besides the positive regulation of CREB by ERK and AKT signaling , their activity could also negatively regulate the function of FOXO transcription factors . FOXO is a family of transcription factors that can directly be phosphorylated by AKT ( Brunet et al . , 1999 ) . Phosphorylated FOXO transcription factors translocate out of the nucleus , and their transcriptional program is attenuated . Interestingly , phosphorylated ERK can downregulate FOXO activity by directly interacting with and phosphorylates FOXO3a at Ser 294 , Ser 344 , and Ser 425 , which leads to FOXO3a degradation via an MDM2-mediated ubiquitin-proteasome pathway ( Yang et al . , 2008 ) . Additional evidence supporting this idea is that inhibition of PI3K/AKT and MEK/ERK pathways both enhance the activation of FOXO transcription factors in pancreatic cancer cells ( Roy et al . , 2010 ) . After spinal cord injury , the synthesis of neurotrophins is elevated to support axon regrowth ( Cho et al . , 1998; Hayashi et al . , 2000; Fukuoka et al . , 2001; Fang et al . , 2017 ) . AKT signaling , which functions downstream of Trk receptors , was reported to accelerate axon regeneration in mammals ( Guo et al . , 2016; Miao et al . , 2016 ) . While NGF family members of neurotrophic factors have only been identified in vertebrates , the AKT pathway has also been shown to promote axon regrowth in flies ( Song et al . , 2012 ) . However , the role of Raf/MEK/ERK signaling during nerve repair is controversial . Although some studies revealed that ERK is involved in axon extension , others suggested that ERK activation impedes axon regeneration and functional recovery ( Markus et al . , 2002; Huang et al . , 2017; Cervellini et al . , 2018 ) . To specifically evaluate the efficacy of Raf/MEK/ERK and AKT signaling in promoting axon regeneration , we generated fly strains with tissue-specific expression of optoRaf or optoAKT and found that light stimulation was sufficient to activate the corresponding downstream components in fly larvae in vivo . Consistent with previous studies ( He and Jin , 2016 ) , we found that AKT activation resulted in significantly increased axon regeneration in C4da neurons as well as the regeneration-incompetent C3da neurons . Interestingly , we found that C4da and C3da neurons expressing optoRaf also exhibited greater regeneration potential in response to light stimulation . This result also corroborates with a previous finding that activated B-RAF signaling enables axon regeneration in the mammalian CNS ( O'Donovan et al . , 2014 ) . We speculate that the differential outcomes of ERK activation on axon regeneration may be due to the different injury models used , and the strength and cell type origin of ERK signaling . The regenerative capacity varies significantly among different neuronal subtypes , as well as the PNS and CNS . Although the administration of neurotrophins enhances axon regeneration in peripheral neurons , its capacity to promote functional regeneration in the CNS is limited , in part due to the inaccessibility of neurotrophins to reach injured axons ( physical barrier ) ( Silver and Miller , 2004; Yiu and He , 2006 ) and innate inactivation of the regenerating program in CNS ( Lu et al . , 2014 ) . OptoRaf and optoAKT could be used to address both issues by direct delivery of light ( rather than ligand ) to reactivate the regenerating program and thereby significantly increase neural regeneration in the CNS as well . We further showed that activation of the Raf/MEK/ERK or AKT subcircuit was capable of improving behavioral performance in fly larvae , suggesting that it may promote synapse regeneration leading to functional recovery . Ineffective functional recovery at least partially results from the inappropriate pathfinding of the regenerative neurons . As shown in this study , the majority of regenerating C4da neuron axons preferentially grew away from their original trajectory . We surprisingly found that delivering a 5 min light stimulation to the ventral branch , which extended toward the correct direction , was sufficient to convey guidance instructions and increase the preferential elongation of the ventral branch against the dorsal branch . Correct guidance cannot be achieved by whole-body administration of pharmacological reagents . Similarly , when casting blue light on the whole transgenic larvae , light stimulation must be given at a high frequency to promote axon regrowth ( there is a threshold for the light off-time ) , and the dorsal branch extension was also dominant in this case . This result highlights the importance and necessity of restricted activation of neurotrophic signaling . Indeed , the strength and location of Raf/MEK/ERK and AKT activation during axon regeneration may be important to the functional consequences . Notably , although the transient restricted stimulation likely affects the decision-making of the growth cone at the branching point , constant light is still required to increase overall axon regeneration . Neurotrophins are engaged in a variety of important cellular processes , and their physiological concentration is essential for the normal function of both neurons and non-neuronal cells ( Rose et al . , 2003; Xiao et al . , 2010; Pöyhönen et al . , 2019 ) . Despite exhibiting substantial efficacy for enhancing nerve regeneration , neurotrophin-based therapeutic applications have been confronted with a number of obstacles such as their nociceptive side effects and lack of strategy for localized signaling activation ( Aloe et al . , 2012; Mitre et al . , 2017; Mahar and Cavalli , 2018; Sung et al . , 2019 ) . OptoRaf and optoAKT aim to improve neurotrophin signaling outcomes by preferentially activating the neuroregenerative program and enabling spatiotemporal control . Our systems offer insights into the ERK and AKT subcircuits and delineate their differential roles downstream of neurotrophin activation , as evidenced by the distinct functional outcomes of Raf/MEK/ERK and AKT signaling in several aspects . First , ERK signaling promoted PC12 cell neuritogenesis , which was not induced by AKT activation . Second , elevated ERK activity significantly increased dendritic complexity , while on the contrary , AKT activation led to decreased dendrite branching . Third , optoRaf and optoAKT displayed different sensitivity in response to light illumination when expressed in Drosophila C4da neurons . Correspondingly , neurons expressing optoRaf and optoAKT responded differently to intermitted light stimulation after injury , suggesting that the strength and activation duration of optoRaf and optoAKT is differentially gauged during axon regeneration . These collectively suggest that , since Raf could be activated by membrane translocation as well as dimerization , CRY2 oligomerization could further lead to a more potent Raf . This multimodal activation mechanism may render that a threshold of optoRaf can be reached so that a saturated ERK activation could be achieved . On the other hand , AKT activation does not depend on dimerization and may display a graded response . As a result , optoAKT activates the AKT pathway in a dose-dependent manner and may not recapitulate the maximum activation of AKT . This work provides a proof-of-concept to use optogenetics to accelerate and navigate axon regeneration in mammalian injury models . Besides spatiotemporal control of the neurotrophic signaling , optoRaf and optoAKT allow for finetuning of the signaling activity with programmed light pattern during axon regeneration . Follow-up studies are warranted to determine how Raf/MEK/ERK and AKT subcircuits are involved in each process of nerve repair , including lesioned axon degeneration , regenerating axon initiation and extension , and the formation of new synapses and remyelination in mammals . Understanding the machinery will , in turn , allow better utilization and development of the optogenetic systems . Recently , Harris et al . succeeded in directing axon outgrowth with optogenetic tools in zebrafish embryos ( Harris et al . , 2020 ) . Although intact optogenetics in larger mammals is limited by the poor penetration depth of blue light ( less than 1 mm ) , we are excited to witness the rapid progress in implantable , wireless µLED devices ( Jeong et al . , 2015; Park et al . , 2015b ) and the integration of optogenetics with long-wavelength-responsive nanomaterials such as the upconversion nanoparticles ( He et al . , 2015; Wu et al . , 2016; Chen et al . , 2018 ) , both of which would facilitate precise delivery of light stimulation . HEK293T cells were provided by Dr . Linfeng Chen , BHK21 cells were provided by Dr . Xianlin Nan , PC12 cells were a gift from Dr . Tobias Meyer . Confirmation of cell line authentication was done in the Cancer Center at Illinois at the University of Illinois at Urbana-Champaign . Mycoplasma contamination was done by a PCR-based protocol . All cell lines were kept at low passages in order to maintain their health and identity . Cell lines used in this work are not among the commonly misidentified cell lines maintained by the International Cell Line Authentication Committee . The plasmid of FOXO3-EGFP generated by cloning the human FOXO3 gene-containing plasmid ( a gift from Prof . Anne Brunet at Stanford University ) into the pEGFP-N1 backbone using overlap PCR with the ( forward primer: cggactcagatctcgacgccaccatgtacccatacgatgttccggattacgc and the reverse primer: ccatggtggcgaccggtggatccccctgcttagcaccagt ) . 19–12-Gal4 ( Xiang et al . , 2010 ) , repo-Gal80 ( Awasaki et al . , 2008 ) , ppk-CD4-tdGFP ( Han et al . , 2011 ) , and ppk-Gal4 Han et al . , 2011 have been previously described . To generate the UAS-optoRaf and UAS-optoAKT stocks , we cloned the entire coding sequences into the pACU2 vector , and the constructs were then injected ( Rainbow Transgenic Flies , Inc ) . Randomly selected male and female larvae were used . Analyses were not performed blind to the conditions of the experiments . The experimental procedures have been approved by the Institutional Biosafety Committee ( IBC ) at the Children's Hospital of Philadelphia . Da neuron axon lesion and imaging in the PNS was performed in live fly larvae as previously described ( Song et al . , 2012; Stone et al . , 2014; Song et al . , 2015 ) . VNC injury was performed as previously described ( Song et al . , 2012; Li et al . , 2020 ) . In brief , A3 and A6 axon bundles in the VNC were ablated with a focused 930 nm two-photon laser and full degeneration around the commissure junction was confirmed 24 hr AI . At 48 hr AI , axon regeneration of these two commissure segments were assayed independently of each other ( Figure 5—figure supplement 1 ) . Quantification was performed as previously described ( Song et al . , 2012; Song et al . , 2015 ) . Briefly , for axon regeneration in the PNS , we used ‘regeneration percentage’ , which depicts the percent of regenerating axons among all the axons that were lesioned; ‘regeneration index’ , which was calculated as an increase of ‘axon length’/‘distance between the cell body and the axon converging point ( DCAC ) ’ ( Figure 3—figure supplement 1A and B ) . An axon was defined as regenerating only when it obviously regenerated beyond the retracted axon stem , and this was independently assessed of the other parameters . The regeneration parameters from various genotypes were compared with that of the WT if not noted otherwise , and only those with significant differences were labeled with the asterisks . For VNC injury , the increased length of each axon regrowing beyond the lesion sites was measured and added together . The regeneration index was calculated by dividing the sum by the distance between A4 and A5 axon bundles ( Figure 5—figure supplement 1 ) . Regeneration percentage was assessed independently of the regeneration index . A commissure segment was defined as regenerated only when at least one regenerating axon passed the midline of the commissure region or joined into other intact bundles ( Figure 5—figure supplement 1 ) . Live imaging was performed as described ( Emoto et al . , 2006; Parrish et al . , 2007 ) . Embryos were collected for 2–24 hr on yeasted grape juice agar plates and were aged at 25°C or room temperature . At the appropriate time , a single larva was mounted in 90% glycerol under coverslips sealed with grease , imaged using a Zeiss LSM 880 microscope , and returned to grape juice agar plates between imaging sessions . The behavioral test was performed to detect functional recovery after VNC injury as described ( Li et al . , 2020 ) . A7 and A8 C4da neuron axon bundles in the VNC , which correspond to the A7 and A8 body segments in the periphery , were injured with laser ( Figure 6A ) . Since C4da neurons are essential for thermonociception , such lesion results in impaired nociceptive response to noxious heat at body segments A7 and A8 . We assessed larva nociceptive behavior in response to a 47°C heat probe at 24 and 48 hr AI . At each time point , the larva was subjected to three consecutive trials , separated by 15 s ( s ) . In each trial , the heat probe was applied at the A7 and A8 body segments for 5 s . If the larva produced head rolling behavior for more than two cycles , it would be scored as ‘1’ , otherwise ‘0’ ( Figure 6A ) . The scores of the three trials were combined and the total score at 24 hr AI was used to determine whether A7 and A8 bundles were successfully ablated . A larva was defined as recovered only when its total score was below 1 at 24 hr AI but increased to 2 or 3 at 48 hr AI . Those failed to exhibit such improvement at 48 hr AI were defined as unrecovered . All the injured larvae exhibited normal nociceptive responses when the same heat probe was applied at the A4 or A5 body segment at 24 hr AI . Third instar larvae or cultured neurons were fixed according to standard protocols . The following antibodies were used: rabbit anti-Phospho-p44/42 MAPK ( Erk1/2 ) ( Thr202/Tyr204 ) ( 4370 , 1:100 , Cell Signaling ) , rabbit anti-Phospho-Drosophila p70 S6 Kinase ( Thr398 ) ( 9209S , 1:400 , Cell Signaling ) and fluorescence-conjugated secondary antibodies ( 1:1000 , Jackson ImmunoResearch ) . Larval body walls were mounted in VECTASHIELD Antifade Mounting Medium . HEK293T cells were cultured in DMEM medium supplemented with 10% fetal bovine serum ( FBS ) , and 1 × Penicillin Streptomycin solution ( complete medium ) . Cultures were maintained in a standard humidified incubator at 37°C with 5% CO2 . For western blots , 800 ng of DNA were combined with 2 . 4 µL Turbofect in 80 µL of serum-free DMEM . The transfection mixtures were incubated at room temperature for 20 min prior to adding to cells cultured in 35 mm dishes with 2 mL complete medium . The transfection medium was replaced with 2 mL serum-free DMEM supplemented with 1 × Penicillin Streptomycin solution after 3 hr of transfection to starve cells overnight . PC12 cells were cultured in F12K medium supplemented with 15% horse serum , 2 . 5% FBS , and 1 × Penicillin Streptomycin solution . For PC12 neuritogenesis assays , 2400 ng of DNA were combined with 7 . 2 mL of Turbofect in 240 mL of serum-free F12K . The transfection medium was replaced with 2 mL complete medium after 3 hr of transfection to recover cells overnight . Twenty-four hours after recovery in high-serum F12K medium ( 15% horse serum + 2 . 5% FBS ) , the cell culture was exchanged to a low-serum medium ( 1 . 5% horse serum + 0 . 25% FBS ) to minimize the base-level ERK activation induced by serum . For western blot analysis , transfected and serum-starved cells were illuminated for different time using a home-built blue LED light box emitting at 0 . 5 mW/cm2 . For PC12 cell neuritogenesis assay , PC12 cells were illuminated at 0 . 2 mW/cm2 for 24 hr with the light box placed in the incubator . The whole optogenetics setup is modified from previous work ( Kaneko et al . , 2017 ) . Larvae were grown in regular brown food at 25°C in 12 hr-12 hr light-dark cycle . At 72 hr AEL , early 3rd instar larvae were transferred from food , anesthetized with ether for axotomy . After recovery in regular grape juice agar plates , larvae were kept in the dark or under blue light stimulation thereafter . A 470 nm blue LED ( LUXEON Rebel LED ) was set over the grape-agar plate for stimulation . The LED was mounted on a 10 mm square coolbase and 50 mm square ×25 mm high alpha heat sink and set under circular beam optic with integrated legs for parallel even light . The light pattern was programmed with BASIC Stamp 2 . 0 microcontroller and buckpuck DC driver ( LUXEON , 700 mA , externally dimmable ) . Local light stimulation was delivered by a 488 nm argon-ion laser using a Zeiss LSM 880 microscope . At 24 hr AI , larvae were anesthetized and C4da neurons were imaged with a confocal microscope . For lesioned axons that bifurcated and formed two branches , we focused the laser beam ( at 15% laser power ) on the ventral branch for 5 min . The larva was then returned to grape juice agar plates and imaged again at 48 hr AI to assess the increased length of each branch . For the light-induced membrane recruitment assay , BHK-21 cells were co-transfected with optoRaf or optoAKT . Fluorescence imaging of the transfected cells was carried out using a confocal microscope ( Zeiss LSM 700 ) . GFP fluorescence was excited by a 488 nm laser beam; mCherry fluorescence was excited by a 555 nm laser beam . Excitation beams were focused via a 40 × oil objective ( Plan-Neofluar NA 1 . 30 ) . Ten pulsed 488 nm and 555 nm excitation were applied for each membrane recruitment experiment . CRY2-CIBN binding induced by 488 nm light was monitored by membrane recruitment of CRY2-mCherry-Raf1 ( for optoRaf ) or CRY2-mCherry-AKT ( for optoAKT ) to the CIBN-CIBN-GFP-CaaX-anchored plasma membrane . The powers after the objective for 488 nm and 555 nm laser beam are approximately 40 µW and 75 µW , respectively . Alternatively , an epi-illumination fluorescence microscope ( Leica DMI8 ) equipped with a 100 × objective ( HCX PL FLUOTAR 100×/1 . 30 oil ) and a light-emitting diode illuminator ( SOLA SE II 365 ) was used for the CRY2-mCherry-Raf1 membrane translocation assay . Neurite outgrowth of PC12 cells was imaged using an epi-illumination fluorescence microscope ( Leica DMI8 ) equipped with 10× ( PLAN 10×/0 . 25 ) and 40× ( HCXPL FL L 40×/0 . 6 ) objectives . Fluorescence from GFP was detected using the GFP filter cube ( Leica , excitation filter 472/30 , dichroic mirror 495 , and emission filter 520/35 ) ; fluorescence from mCherry was detected using the Texas Red filter cube ( Leica , excitation filter 560/40 , dichroic mirror 595 , and emission filter 645/75 ) . Cells were washed once with 1 mL cold DPBS and lysed with 100 µL cold lysis buffer ( RIPA + protease/phosphatase cocktail ) . Lysates were centrifuged at 17 , 000 RCF , 4°C for 10 min to pellet cell debris . Purified lysates were normalized using Bradford reagent . Normalized samples were mixed with LDS buffer and loaded onto 10% or 12% polyacrylamide gels . SDS-PAGE was performed at room temperature with a cold water bath . Samples were transferred to PVDF membranes at 30 V 4°C overnight or 80 V for 90 min . Membranes were blocked in 5% BSA/TBST for 1 hr at room temperature and probed with the primary and secondary antibodies according to company guidelines . Membranes were incubated with ECL substrate and imaged using a Bio-Rad ChemiDoc XRS chemiluminescence detector . Signal intensity analysis was performed by ImageJ . Antibodies used were listed in the Key Resources Table . No statistical methods were used to pre-determine sample sizes but our sample sizes are similar to those reported in previous publications ( Song et al . , 2012; Song et al . , 2015 ) , and the statistical analyses were done afterward without interim data analysis . Data distribution was assumed to be normal but this was not formally tested . All data were collected and processed randomly . Each experiment was successfully reproduced at least three times and was performed on different days . The values of ‘N’ ( sample size ) are provided in the figure legends . Data are expressed as mean ± SEM in bar graphs , if not mentioned otherwise . No data points were excluded . Two-tailed unpaired Student's t-test was performed for comparison between two groups of samples . One-way ANOVA followed by multiple comparison test was performed for comparisons among three or more groups of samples . Two-way ANOVA followed by multiple comparison test was performed for comparisons between two or more curves . Fisher's exact test was used to compare the percentage . Statistical significance was assigned , p<0 . 05 , p<0 . 01 , **p<0 . 001 .	Most cells have a built-in regeneration signaling program that allows them to divide and repair . But , in the cells of the central nervous system , which are called neurons , this program is ineffective . This is why accidents and illnesses affecting the brain and spinal cord can cause permanent damage . Reactivating regeneration in neurons could help them repair , but it is not easy . Certain small molecules can switch repair signaling programs back on . Unfortunately , these molecules diffuse easily through tissues , spreading around the body and making it hard to target individual damaged cells . This both hampers research into neuronal repair and makes treatments directed at healing damage to the nervous system more likely to have side-effects . It is unclear whether reactivating regeneration signaling in individual neurons is possible . One way to address this question is to use optogenetics . This technique uses genetic engineering to fuse proteins that are light-sensitive to proteins responsible for relaying signals in the cell . When specific wavelengths of light hit the light-sensitive proteins , the fused signaling proteins switch on , leading to the activation of any proteins they control , for example , those involved in regeneration . Wang et al . used optogenetic tools to determine if light can help repair neurons in fruit fly larvae . First , a strong laser light was used to damage an individual neuron in a fruit fly larva that had been genetically modified so that blue light would activate the regeneration program in its neurons . Then , Wang et al . illuminated the cell with dim blue light , switching on the regeneration program . Not only did this allow the neuron to repair itself , it also allowed the light to guide its regeneration . By focusing the blue light on the damaged end of the neuron , it was possible to guide the direction of the cell's growth as it regenerated . Regeneration programs in flies and mammals involve similar signaling proteins , but blue light does not penetrate well into mammalian tissues . This means that further research into LEDs that can be implanted may be necessary before neuronal repair experiments can be performed in mammals . In any case , the ability to focus treatment on individual neurons paves the way for future work into the regeneration of the nervous system , and the combination of light and genetics could reveal more about how repair signals work .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2020	Optical control of ERK and AKT signaling promotes axon regeneration and functional recovery of PNS and CNS in Drosophila
Individuals vary in their responses to stroke and trauma , hampering predictions of outcomes . One reason might be that neural circuits contain hidden variability that becomes relevant only when those individuals are challenged by injury . We found that in the mollusc , Tritonia diomedea , subtle differences between animals within the neural circuit underlying swimming behavior had no behavioral relevance under normal conditions but caused differential vulnerability of the behavior to a particular brain lesion . The extent of motor impairment correlated with the site of spike initiation in a specific neuron in the neural circuit , which was determined by the strength of an inhibitory synapse onto this neuron . Artificially increasing or decreasing this inhibitory synaptic conductance with dynamic clamp correspondingly altered the extent of motor impairment by the lesion without affecting normal operation . The results suggest that neural circuit differences could serve as hidden phenotypes for predicting the behavioral outcome of neural damage . Experimental and theoretical studies have shown that individual animals can exhibit similar behaviors while differing substantially in the properties of the neurons and synapses underlying those behaviors ( Prinz et al . , 2004; Goaillard et al . , 2009; Calabrese et al . , 2011; Norris et al . , 2011; Roffman et al . , 2012 ) . The consequences of such hidden variability in neural circuits have not been addressed . One possible consequence is that it impacts the behavioral susceptibility of the animal to trauma . It has been noted that individual people differ from one another to such an extent that it can impair the ability to predict outcomes in cases of traumatic brain injury ( Hukkelhoven et al . , 2005; Lingsma et al . , 2010; Forsyth and Kirkham , 2012 ) or stroke ( Cramer , 2008a ) . Such variability can be hidden under normal conditions but cause differential survival of individuals in the face of critical challenges . In this study , we report that differences in synaptic properties , which were of no consequence under ordinary conditions , caused different outcomes when the circuit was challenged with an injury . With recent advances in detection techniques , there has been a growing awareness that axonal injury in the white matter plays a complex role in disruption of neural networks underlying higher brain functions ( Adams et al . , 2000; Schiff et al . , 2007; Kinnunen et al . , 2011; Squarcina et al . , 2012 ) . However , there are technical difficulties in manipulating specific neural circuit elements and providing precisely controlled lesions in the mammalian brain . In this study , we use a nudibranch mollusc , Tritonia diomedea , in which a neural circuit for rhythmic swimming behavior is widely distributed in the brain . The Tritonia swim central pattern generator ( CPG ) consists of three neuronal types: DSI , C2 , and VSI ( Figure 1A ) , which form a network oscillator circuit that produces the rhythmic bursting activity ( Figure 1B ) underlying production of the rhythmic movements ( Getting , 1981 , 1989b; Katz , 2007a , 2007b , 2009 ) . C2 and VSI both send axons through one of the pedal commissures , Pedal Nerve 6 ( PdN6 ) , which connects the two pedal ganglia ( Figure 1C ) . Previously , we reported that disconnecting this commissure blocks or seriously impairs the swimming behavior and the motor pattern underlying it ( Sakurai and Katz , 2009b ) . In this study , we found substantial individual variability in the synaptic actions of C2 onto VSI , which correlated with variability in the susceptibility of the behavior to disruption following disconnection of PdN6 . Such individual variability in neural circuit elements was hidden under normal conditions , but became functionally relevant only when the system was challenged by injury . 10 . 7554/eLife . 02598 . 003Figure 1 . The Tritonia swim central pattern generator . ( A ) A schematic diagram of the swim central pattern generator ( CPG ) . The CPG consists of three types of interneurons: C2 , cerebral cell 2; DSI , dorsal swim interneuron; VSI , ventral swim interneuron . Based on Getting et al . ( 1980 ) and Getting ( 1983a , 1983b ) . All neurons are electrically coupled to contralateral counterparts , which are not represented here . There are three DSIs , but C2 and VSI are individual neurons . Filled triangles represent excitatory synapses and filled circles represent inhibitory synapses . Combinations of triangles and circles are multi-component synapses . ( B ) An example of the swim motor pattern recorded from an isolated brain preparation . Simultaneous intracellular recordings from the three CPG neurons are shown . The bursting pattern was elicited by electrical stimulation of the left body wall nerve , pedal nerve 3 ( cf . , Figure 3A ) , using voltage pulses ( 8 V , 1 ms ) at 5 Hz for 3 s . Arrows show onset and offset of the nerve stimulation . ( C ) The Tritonia brain and the site where PdN6 was cut in vivo . The body wall above the buccal mass was cut open ( left ) . A schematic drawing shows a dorsal view of the Tritonia brain ( right ) with the locations of the interneurons and their axonal projections . DSI and C2 are located on the dorsal surface of the cerebral ganglion ( CeG ) . VSI is located on the ventral side of the pleural ganglion ( PlG ) . C2 and VSI project axons through the Pedal commissure ( PdN6 ) , which connects the two pedal ganglia ( PdG ) ( Sakurai and Katz , 2009b ) . PdN6 was transected near the right pedal ganglion with scissors . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 003 The escape swim behavior of Tritonia consists of a series of whole body flexions in response to a noxious stimulus ( Getting , 1989b; Katz , 2009 ) . We previously showed that when one of the pedal commissures , PdN6 , was severed ( Figure 1C ) , the swimming behavior of the animal was impaired in that the number of body flexions per swim episode decreased compared to sham-operated controls ( Sakurai and Katz , 2009b ) . With additional data , we further noticed that the extent of the impairment , in terms of the number of body flexions , varied across individuals ( Figure 2 ) . In this study , we use the term ‘impairment’ to mean a decrease in the number of body flexions per swim episode or in the number of VSI bursts per swim motor pattern and the term ‘susceptibility’ for the likelihood of being impaired upon lesion or blockade of a commissure . 10 . 7554/eLife . 02598 . 004Figure 2 . Individual variability in the extent of swim impairment by a lesion . ( A ) Nerve-transected animals were blindly paired with sham-operated animals . Two examples ( Pair 1 and Pair 2 ) show different effects on the number of body flexions during the escape swim behavior for animals in response to PdN6 transection ( gray squares ) compared to sham-operated controls ( white circles ) . In one animal , cutting PdN6 caused a large decrease in the number of body flexions compared to sham ( Pair 1 ) , whereas the same lesion caused a small decrease in other experimental preparation ( Pair 2 ) . ( B ) Mean number of body flexions during the escape swim behavior for animals with PdN6 transected ( gray squares ) and sham-operated controls ( white circles ) . The surgery caused a significant decrease in the number of flexions in both cut and sham animals ( cut animals , F ( 3 , 30 ) = 21 . 0 , p< 0 . 001 , N = 11; sham animals , F ( 3 , 30 ) = 7 . 47 , p< 0 . 001 , N = 11 by One-way Repeated Measures ANOVA ) . A two-way repeated measures ANOVA with post-hoc pairwise comparison revealed a significant difference in the number of flexions between cut and sham animals 2 hr after the surgery ( p<0 . 001 ) . Prior to the cut , there was no significant difference between the test and sham-operated animals in the number of flexions ( 16 hr , p = 0 . 57; −12 hr , p = 0 . 52; −2hr , p = 0 . 89 ) . ( C ) The coefficient of variance ( CoV ) of the number of body flexions for the transected animals ( gray squares ) showed a threefold increase after the cut , but only a slight increase in sham-operated animals ( white circles ) . There is a significant difference in variance between the cut group and the sham group after the surgery ( by Levene median test , N = 19 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 00410 . 7554/eLife . 02598 . 005Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 005 Figure 2A shows two examples that illustrate the individual variability in the effect of cutting the commissure on the swimming behavior . In the first example , commissure transection completely disrupted the swimming behavior compared to its sham-operated paired-control , which showed no change ( Figure 2A , Pair 1 ) . In the second example , the animal with the PdN6 transected had a swim consisting of five flexions compared to six flexions for the sham-operated control ( Figure 2A , Pair 2 ) . On average , the surgery decreased the number of flexions in both cut and sham animals ( Figure 2B ) . The coefficient of variance ( CoV ) showed a three-fold increase after the cut ( 0 . 21–0 . 64 ) , but stayed relatively constant for the sham controls ( Figure 2C ) . Thus , cutting the commissure revealed a greater variability in the population than could be observed under normal conditions , meaning that some animals were more susceptible to the lesion than others . Individual-to-individual variability was also seen in the extent of impairment in the swim motor pattern recorded from isolated brain preparations ( Figure 3 ) . Here , action potential propagation in PdN6 was blocked either by physical transection or by local application of TTX to PdN6 ( Figure 3A , see ‘Materials and methods’ ) . Figure 3B shows the examples of swim motor patterns , consisting of six cycles of rhythmic bursts before application of TTX , from two different animals . After blocking PdN6 by replacing saline in the pipette with the saline containing 0 . 1 mM TTX , the swim motor pattern in Animal 1 was reduced to just two bursts in VSI; whereas in Animal 2 , VSI still produced five bursts ( Figure 3B , right; Figure 4A ) . 10 . 7554/eLife . 02598 . 006Figure 3 . Individuals differed in the extent of motor pattern impairment by disconnection of PdN6 . ( A ) A schematic drawing of the Tritonia brain showing how axonal impulse propagation was blocked in PdN6 either by delivering TTX ( 1 × 10−4M ) into a suction pipette or by physical transection . The stimulus was delivered to the left pedal nerve 3 ( PdN3 ) . The pedal ganglion closer to the VSI cell body was called the proximal pedal ganglion whereas the other pedal ganglion was called the distal pedal ganglion . ( B ) Simultaneous intracellular recordings from C2 and VSI from two representative animals ( Animals 1 and 2 ) . Arrows ( Stim ) indicate the time of PdN3 stimulation . Animal 1 showed a large decrease ( from 6 to 2 ) in the number of VSI bursts after PdN6 was blocked , whereas in Animal 2 the number of VSI bursts was less affected ( from 6 to 5 ) . The boxed insets show overlaid traces of VSI spikes recorded from the soma and the corresponding axonal impulses recorded from PdN6 with an en passant suction electrode during the swim motor program . The traces were triggered at the peak of the somatic action potential and overlaid . The shapes of the impulses show that the action potentials were blocked ( see text for explanation ) . Calibration: 50 mV , 10 ms . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 00610 . 7554/eLife . 02598 . 007Figure 4 . Individual variability in the extent of motor pattern impairment by disconnection of PdN6 . ( A ) The number of VSI bursts per swim motor pattern episode recorded from Animal 1 was more affected by blocking PdN6 than that from Animal 2 ( same individuals as in Figure 3B ) . The swim motor pattern was evoked 3 or 4 times at constant intervals ( approximately 10 min ) , and PdN6 was blocked between the last two swim motor pattern bouts . ( B ) The average number of the VSI bursts decreased significantly after PdN6 block ( p<0 . 001 by one-way repeated measures ANOVA , N = 34 ) . Post-hoc pairwise comparisons ( Tukey test ) show significant differences of the 4th swim test from all other swim tests . ( C ) The coefficient of variance ( CoV ) of the number of bursts increased after PdN6 block . There is a significant increase in variance in the number of VSI bursts after PdN6 disconnection ( p<0 . 05 by Levene median test , N = 34 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 00710 . 7554/eLife . 02598 . 008Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 008 In the experiments with local application of TTX , blockade of axonal impulses was confirmed by examining the change in the impulse waveform ( Figure 3B , overlaid traces in boxes; Sakurai and Katz , 2009b ) . The axonal impulse , recorded en passant by a pipette , was triphasic with an apparent positive deflection between two negative deflections . When the action potential was blocked inside the pipette , the impulses became biphasic with an initial negative deflection followed by a slower positive deflection . This indicates that the impulses came into the pipette but did not exit . This was also accompanied with the disappearance of the VSI's synaptic action onto neurons in the contralateral pedal ganglion ( data not shown , cf . , Sakurai and Katz , 2009b ) . Before PdN6 disconnection in Animal 1 , axonal impulses in Pd6 appeared earlier than the soma spike , indicating that VSI was producing antidromic action potentials . After blocking PdN6 , the axonal impulse appeared after the soma spike , indicating that VSI was now producing orthodromic action potentials . In Animal 2 , VSI was already exhibiting orthodromic action potentials even before blocking PdN6 , showing individual variability in spike initiation zones ( see ‘Direction of VSI spike propagation predicted the extent of motor impairment’ ) . On average , disconnection of PdN6 significantly decreased the number of VSI bursts per swim motor pattern episode ( Figure 4B ) , as shown previously ( Sakurai and Katz , 2009b ) . This decrease was accompanied by an increased CoV from 0 . 16 to 0 . 47 ( Figure 4C ) . Thus , as with the behavior of the animal , individuals also differed in the extent of impairment of the swim motor pattern in an ex vivo preparation after disconnection of PdN6 . C2-evoked excitation of VSI is essential for the production of the swim motor pattern ( Getting , 1983b; Calin-Jageman et al . , 2007; Sakurai and Katz , 2009b ) . Stimulating C2 to fire a train of actions potentials ( 10 Hz for 2 s ) induced a burst of action potentials in VSI in 47 out of 53 preparations with PdN6 intact ( Figure 5 ) . The C2-evoked VSI spikes were mostly antidromic , traveling from the pedal ganglion distal to the VSI soma through PdN6 ( Figure 5A , N = 39 of 47; cf . , Sakurai and Katz , 2009b ) , but their number varied across individuals , ranging from zero to 58 spikes , as seen in the two examples ( Figure 5B , Animal 3 and Animal 4 ) . The number of bursts in the swim motor pattern , under normal conditions , was not correlated with the number of action potentials evoked by C2 when PdN6 was intact ( Figure 5C ) . 10 . 7554/eLife . 02598 . 009Figure 5 . The extent of motor impairment showed little or no correlation with the C2-evoked VSI spiking recorded before blocking PdN6 . ( A ) A schematic illustration showing the stimulus ( C2 ) and recording ( VSI ) microelectrodes , the direction of action potential propagation ( dashed arrows ) in C2 and VSI , and synaptic action ( + , excitatory; − , inhibitory ) of C2 onto VSI before blocking PdN6 . Repetitive square current pulses ( 10 nA , 20 ms ) were injected into the C2 soma to evoke a train of action potentials at a constant frequency ( 10 Hz ) . ( B ) Two examples of swim motor patterns ( Bi , Bii ) and the membrane potential responses ( Biii , Biv ) of VSI to C2 stimulation are shown for two animals ( Animals 3 and 4 ) . With PdN6 intact , Animal 3 showed five VSI bursts ( Bi ) and Animal 4 had six VSI bursts ( Bii ) . The effect of C2 stimulation on VSI varied among individuals; causing an intense burst in VSI of Animal 3 ( Biii ) but only two spikes in Animal 4 ( Biv ) . VSI exhibited antidromic spikes in the majority of preparations ( see text ) that were presumably caused by the C2 excitatory action in the distal terminal of VSI ( Sakurai and Katz , 2009b ) . In Biii and Biv action potentials are truncated to show underlying membrane potential . ( C ) No correlation was detected between the number of VSI bursts per swim episode and the number of C2-evoked VSI spikes with PdN6 intact ( R2 = 0 . 05 , p = 0 . 10 by linear regression , N = 52 ) . Graph symbols in this and Figures 6 and 7 each represent data from the same individuals . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 00910 . 7554/eLife . 02598 . 010Figure 5—source data 1 . Source data for panel C . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 010 Disconnection of PdN6 ( Figure 6A ) decreased the number of VSI bursts per swim episode to different extents in Animals 3 and 4 ( Figure 6Bi , Bii ) . Elimination of antidromic spikes by this procedure revealed the C2-evoked membrane potential change evoked in the region of VSI proximal to PdN6 ( Figure 6Biii , Biv ) ( cf . , Sakurai and Katz , 2009b ) . The synaptic responses of VSI to C2 stimulation were highly variable among individuals . In 47 out of 56 preparations , it was a mix of both depolarization and hyperpolarization ( cf . , Figure 6Biii , Animal 3 ) . In the remaining four preparations , C2 stimulation caused only depolarization with multiple components ( cf . , Figure 6Biv , Animal 4 ) . The response to C2 appeared to contain both monosynaptic and polysynaptic components; during the depolarizing phase , there was a barrage of recruited EPSPs from unknown neurons lasting for about 10 s ( cf . , Figure 6—figure supplement 1 ) . 10 . 7554/eLife . 02598 . 011Figure 6 . The extent of motor impairment showed a strong correlation with C2-evoked VSI depolarization recorded after blocking PdN6 . ( A ) A schematic illustration showing the stimulus ( C2 ) and recording ( VSI ) microelectrodes , the direction of action potential propagation ( dashed arrows ) in C2 , and synaptic action ( + , excitatory; - , inhibitory ) after blocking PdN6 . ( B ) Two examples ( Animals 3 and 4 ) of swim motor patterns ( Bi , Bii ) and the membrane potential responses ( Biii , Biv ) of VSI to C2 stimulation are shown after blocking PdN6 . Animal 3 and 4 are the same animals as in Figure 5B . The effects of blocking PdN6 on the swim motor pattern were different: Animal 3 showed 40% reduction ( 5 to 3 ) in the number of VSI bursts ( Bi ) , whereas Animal 4 showed a 16 . 7% reduction ( 6 to 5 ) ( Bii ) . With PdN6 blocked , C2 stimulation ( 10 Hz , 4 s ) no longer caused VSI to spike in either animal , but instead evoked a complex membrane potential change consisting of both depolarization and hyperpolarization ( Biii , Biv ) . ( C ) After PdN6 disconnection , there was a significant correlation between the number of VSI bursts per swim episode and the amplitude of the C2-evoked VSI depolarization ( R2 = 0 . 53 , p< 0 . 001 by linear regression , N = 50 ) . ( D ) The percent change in the number of VSI bursts caused by PdN6 disconnection showed a significant correlation with the amplitude of the C2-evoked depolarization in VSI ( p<0 . 001 by linear regression , R2 = 0 . 47 , p<0 . 001 , N = 50 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 01110 . 7554/eLife . 02598 . 012Figure 6—source data 1 . Source data for panels C and D . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 01210 . 7554/eLife . 02598 . 013Figure 6—source data 2 . Source data for figure supplement 1 panel B . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 01310 . 7554/eLife . 02598 . 014Figure 6—figure supplement 1 . C2 recruited unidentified neurons to excite VSI . ( A ) C2 stimulation caused a bombardment of small EPSPs that lasted longer than the duration of stimulation ( black trace ) . Recruitment of these polysynaptic EPSPs was minimized by superfusion of high-divalent cation ( Hi-Di ) saline ( gray trace ) . Positions of the two traces were offset for visualization . ( B ) A graph showing that C2 stimulation increased the instantaneous frequency of the recruited EPSPs in VSI . Inset diagram shows C2-evoked polysynaptic recruitment of unidentified neurons making excitatory synapses onto VSI . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 014 The amplitude of C2-evoked depolarization in the proximal region of VSI was predictive of the susceptibility of the swim motor pattern to the lesion ( Figure 6C , D ) . After PdN6 disconnection , the number of VSI bursts per swim episode showed a significant correlation with the amplitude of the C2-evoked depolarization ( Figure 6C ) . The extent of decrease in the number of VSI bursts caused by PdN6 disconnection was also correlated with the amplitude of the C2-evoked depolarization in VSI ( Figure 6D ) . In contrast , the amplitude of C2-evoked VSI depolarization showed no correlation with the number of VSI bursts per swim episode with intact PdN6 ( p = 3 . 2; not shown ) . Thus , the larger the VSI depolarization caused by C2 after PdN6 disconnection , the less impairment there was in the swim motor pattern . This makes intuitive sense; when PdN6 is disconnected , only animals in which C2 still has an excitatory action onto VSI should be capable of swimming because C2 excitation of VSI is necessary for production of the swim motor pattern ( Getting , 1983b; Calin-Jageman et al . , 2007; Sakurai and Katz , 2009b ) . The C2-evoked synaptic response of VSI contained both monosynaptic and polysynaptic components ( Figure 6—figure supplement 1 ) . Even before C2 stimulation , spontaneous excitatory post-synaptic potentials ( EPSPs ) from unidentified neurons continuously occurred in VSI . These EPSPs obscured the direct synaptic action of C2 onto VSI . Therefore , we minimized such polysynaptic actions by applying high-divalent cation ( Hi-Di ) saline , which reduces polysynaptic inputs by raising the firing threshold of neurons ( Sakurai and Katz , 2009a ) . In Hi-Di saline , C2 stimulation evoked a smooth biphasic synaptic potential in VSI , with an initial depolarization and a delayed hyperpolarization phase ( Figure 6—figure supplement 1 , Figure 7A ) . The shapes of these responses were not affected by blocking PdN6 , suggesting they were produced in the proximal VSI region , which was electrotonically detectable from the soma recording site . The hyperpolarizations tended to be larger in amplitude and were more variable ( 2 . 9 ± 1 . 3 mV , N = 32 , CoV = 0 . 66 ) than the depolarizations ( 0 . 8 ± 0 . 36 mV , N = 32 , CoV = 0 . 42 ) ; there was a significant difference in variance between the amplitudes of hyperpolarizations and depolarizations ( p< 0 . 05 by Levene median test , N = 32 ) . 10 . 7554/eLife . 02598 . 015Figure 7 . The extent of motor impairment correlated with the inhibitory component of C2-to-VSI synapse . ( A ) Two examples ( Animals 5 and 6 ) of VSI membrane potential responses to C2 stimulation recorded with PdN6 disconnected in normal saline ( left ) and in high divalent cation ( Hi-Di ) saline ( right ) to decrease the contribution of polysynaptic inputs . ( B ) The impairment , measured as the percent change in the number of VSI bursts , showed a significant correlation with the amplitude of the hyperpolarization phase ( R2 = 0 . 44 , p<0 . 001 by linear regression , N = 26 ) of the C2-evoked synaptic potential in VSI . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 01510 . 7554/eLife . 02598 . 016Figure 7—source data 1 . Source data for panel B . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 01610 . 7554/eLife . 02598 . 017Figure 7—source data 2 . Source data for figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 01710 . 7554/eLife . 02598 . 018Figure 7—figure supplement 1 . The magnitude of C2-evoked depolarization in VSI in normal saline correlated with the amplitude of hyperpolarizing phase of C2-to-VSI synaptic potential . The magnitude of C2-evoked VSI depolarization in normal saline did not correlate to the direct C2-evoked depolarization measured in Hi-Di saline ( A , R2 = 0 . 04 , p=0 . 37 by linear regression , N = 24 ) , but did correlate with the C2-evoked hyperpolarization ( B , R2 = 0 . 43 , p<0 . 001 by linear regression , N = 24 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 018 The percent decrease in the number of VSI bursts per swim episode showed a significant inverse correlation with the amplitude of the C2-evoked delayed hyperpolarization in VSI in Hi-Di saline ( Figure 7B ) . In contrast , we could not detect a correlation between the initial depolarization phase and the number of VSI bursts ( p = 0 . 07 , N = 26; not shown ) . These results indicate that the variability in the susceptibility of the motor pattern to PdN6 disconnection originates at least in part from the difference in the extent to which C2 inhibits VSI; animals in which C2 evoked larger hyperpolarizations in VSI were more vulnerable to having their motor pattern disrupted . Indeed , the amplitude of C2-evoked depolarization in normal saline ( Figure 7—figure supplement 1A , inset next to y axis ) showed no correlation with the amplitude of the depolarizing phase recorded in Hi-Di saline ( Figure 7—figure supplement 1A ) , whereas it showed a significant inverse correlation with the amplitude of the hyperpolarization phase ( Figure 7—figure supplement 1B ) . Thus , the magnitude of C2-evoked polysynaptic excitation of VSI is more likely determined by the amplitude of C2-evoked hyperpolarization phase , which limits the depolarizing effect of the recruited polysynaptic EPSPs in VSI in normal saline . The difference in the synaptic action of C2 onto VSI in the proximal pedal ganglion may cause individual differences in the location of the spike initiation zone in the VSI process . We have shown earlier in this study that in some animals the action potential propagation in VSI axon was orthodromic during the swim motor pattern ( Figure 3B , Animal 2 , PdN6 intact ) . Therefore , we checked the direction of VSI spikes propagation during the swim motor pattern and examined how it correlated with the susceptibility of the motor pattern to PdN6 disconnection . We previously showed that in most preparations , stimulation of C2-induced VSI action potentials that propagated antidromically through the axon in the PdN6 to the cell body ( Sakurai and Katz , 2009b ) . By simultaneously recording VSI spikes from the soma and its axonal impulses from PdN6 ( Figure 8A , B ) , we found that the direction of VSI action potential propagation during the swim motor pattern varies among individuals ( Figure 8C ) . Out of 69 animals , 17 ( 24 . 6% ) exhibited only antidromic spike propagation during the swim motor program in which axonal impulses appeared earlier in PdN6 than in the cell body ( Figure 8Ci ) . In contrast , five animals ( 7 . 2% ) showed only orthodromic spikes in which action potentials were generated in or near the pedal ganglion proximal to the VSI soma ( Figure 8Ciii ) . In 47 animals ( 68 . 1% ) , the direction switched from orthodromic to antidromic , or vice versa , during the swim motor pattern ( Figure 8Cii ) . There was a significant difference in the percentage change in the number of VSI bursts per swim episode after disconnecting PdN6 when comparing animals showing only orthodromic VSI spikes with those having only antidromic VSI spikes; animals with only orthodromic VSI spikes were significantly less impaired than those with only antidromic VSI spikes ( Figure 8D ) . 10 . 7554/eLife . 02598 . 019Figure 8 . The direction of spike propagation in VSI axon was predictive of susceptibility of the swim motor pattern to PdN6 disconnection . ( A ) A schematic diagram showing the recording configuration . VSI action potentials were recorded with an intracellular microelectrode in the soma and an extracellular en passant suction electrode on PdN6 . To initiate a swim motor pattern , the left PdN3 was stimulated via a suction electrode ( see Figure 3A ) . ( B ) Intracellular activity recorded from VSI and the axonal impulses recorded extracellularly from PdN6 during a swim motor pattern . Arrows indicate the time of PdN3 stimulation to initiate the swim program . Each VSI burst is indicated by a number ( 1–5 ) . ( C ) Overlaid spike-triggered impulses for each burst recorded from PdN6 in three individuals show variability in the direction of VSI spike propagation ( Ci , antidromic; Cii , mixed; Ciii , orthodromic ) . Schematic drawings above the traces show the presumptive spike-initiation zones ( yellow explosion symbols ) and the direction of action potential propagation ( arrows ) in the VSI axons . In Ci , all five bursts in the swim program consisted of antidromic VSI spikes ( the nerve impulse appearing earlier than the soma spike ) , whereas in Cii , VSI spike propagation shifted from orthodromic to antidromic during the course of the swim motor pattern . In Ciii , all VSI spikes were evoked near the soma and propagated orthodromically . Traces in Cii were reused from Sakurai and Katz ( 2009b ) . ( D ) The direction of VSI spike propagation in PdN6 was predictive of the extent of impairment after PdN disconnection . The extent of impairment by PdN6 disconnection , shown as the percent change in the number of VSI bursts per swim episode , is plotted in three groups categorized by the direction of VSI spike propagations ( black , all VSI bursts were antidromic; gray , mixed; white , all bursts were orthodromic ) . One-way ANOVA with a post-hoc pairwise comparison ( Holm-Sidak method ) revealed that individuals exhibiting only orthodromic VSI bursts were significantly less impaired than those with only antidromic VSI bursts as indicated by an asterisk ( F ( 2 , 66 ) = 4 . 64 , p = 0 . 015 , N = 5 and 16 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 01910 . 7554/eLife . 02598 . 020Figure 8—source data 1 . Source data for panel D . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 02010 . 7554/eLife . 02598 . 021Figure 8—source data 2 . Source data for figure supplement 1 panels B and C . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 02110 . 7554/eLife . 02598 . 022Figure 8—figure supplement 1 . Inter- and intra-individual variation in the direction of VSI spike propagation during the swim motor pattern . ( A ) Individual bursts were categorized into three groups: ( i ) burst with all antidromic spikes , ( ii ) burst with mixture of antidromic and orthodromic spikes , and ( iii ) burst with all orthodromic spikes . ( B ) More individuals showed antidromic VSI bursts later in the swim motor pattern . For each VSI burst ( 1st through 6th ) , bars represent the percentages of individuals before PdN6 disconnection that exhibited bursts with only antidromic spikes ( black ) , mixed spikes ( gray ) , and only orthodromic spikes ( white ) . The 1st burst , N = 69; the 2nd burst , N = 69; the 3rd burst , N = 69; the 4th burst , N = 56; the 5th burst , N = 31; the 6th burst , N = 7 . ( C ) After disconnecting PdN6 , the swim motor pattern was more likely to be terminated at the VSI burst that had consisted of antidromic spikes . The graph shows the probability of elimination of each VSI burst ( from 1st to 6th ) for the three VSI burst types with PdN6 intact . For example , if the 4th burst consisted of all antidromic VSI spikes ( black triangles ) , then 82% of animals lost that burst after PdN6 disconnection , causing the swim motor pattern to be less than four bursts . In contrast , only 23% of animals lost the 4th burst if it had consisted of all orthodromic spikes ( white triangles ) prior to PdN6 disconnection or 40% if it consisted of mixed spikes ( gray boxes ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 022 VSI tended to show more antidromic spiking later in the swim motor pattern in the majority of animals ( Figure 8—figure supplement 1A , B ) . For individual VSI bursts , there are three types of bursts: bursts with all antidromic spikes , bursts with mixed spikes , and bursts with all orthodromic spikes ( Figure 8—figure supplement 1A ) . In a majority of animals , VSI exhibited all antidromic spiking after the 3rd burst ( Figure 8—figure supplement 1B ) . Upon blocking PdN6 , the motor pattern tended to lose the terminal bursts that contained only antidromic spikes ( Figure 8—figure supplement 1C ) . Thus , the susceptibility of the motor pattern to lesion was apparently dependent on the location of the primary spike initiation zone in VSI . If spikes originated in the proximal region of VSI , PdN6 disconnection would have less effect on the swim performance . Although we could not provide direct evidence of how the primary spike initiation zone was determined , it likely involves the C2 synaptic action onto VSI . The results above suggest that the strength of the inhibitory component of the C2-to-VSI synapse does not affect the function of the intact swim circuit under normal conditions , but may determine its susceptibility to the lesion . To test this , we employed the dynamic clamp technique ( Sharp et al . , 1992 , 1993a , 1993b ) to introduce an artificial C2 to VSI synaptic conductance . The time course of the conductance was based on that from previous models of the Tritonia swim CPG ( Getting , 1989c; Calin-Jageman et al . , 2007 ) . The activation and maximum conductance were adjusted to mimic the synaptic strength observed in that preparation ( see ‘Materials and methods’ ) . The example traces shown in Figures 9 and 10 were obtained from the same preparation , which was slightly susceptible to PdN6 disconnection; it exhibited 5 VSI bursts per swim episode with PdN6 intact ( Figure 9Ai ) and 4 bursts after PdN6 was blocked ( Figure 10Ai ) . Introducing an artificial inhibitory synaptic conductance in VSI corresponding to the times of C2 spikes caused no change in the number of bursts when PdN6 was intact ( Figure 9Aii , Bi ) . When the numbers of bursts recorded with dynamic clamp were plotted against the number of bursts without dynamic clamp , they lined up along the unity slope line ( Figure 9Bii ) . Before dynamic clamping , orthodromic VSI action potentials were detected in 50% of the preparations during the swim motor pattern ( N = 8 of 16 ) . When the artificial inhibitory synaptic conductance was added to the soma by dynamic clamping , all VSI action potentials became antidromic during the swim motor pattern in each of the 16 preparations . This indicates that enhanced hyperpolarization in the soma suppresses orthodromic spiking in VSI , but the distal VSI terminal is still able to generate antidromic bursts . 10 . 7554/eLife . 02598 . 023Figure 9 . An artificial synaptic conductance created a hidden circuit change that caused no motor impairment with PdN6 intact . A ) Recordings of a five-cycle swim motor pattern with PdN6 intact ( Ai ) . Introduction of an artificial synaptic conductance from C2 to VSI using dynamic clamp ( D . C . ) had no effect on the number of VSI bursts ( Aii ) . The artificial synapse is represented as a dotted blue line with a filled blue circle in the schematic . VSI displayed unnaturally large hyperpolarizations on each burst because the currents were injected at the site of the electrode impalement in the soma instead of occurring in the neuropil . The spikes rode on the hyperpolarizing phase of the burst , indicating that they were antidromic spikes , generated in the distal pedal ganglion ( green arrows in schematic ) . When the inhibitory synaptic conductances were subtracted by applying negative conductances of the same amount as in Aii , the number of bursts increased ( Aiii ) . B , With PdN6 intact , addition of synaptic inhibition with dynamic clamp ( D . C . ) did not change the number of VSI bursts compared to control ( Cont ) ( Bi , p = 0 . 16 by paired t-test , N = 18 ) . A plot of the number of bursts under dynamic clamp vs the number of bursts without dynamic clamp has a slope close to one ( Bii ) . C , With PdN6 intact , subtraction of synaptic inhibition significantly increased the number of VSI bursts 26 . 6% compared to control . ( p = 0 . 01 by paired t-test , N = 9 ) ( Ci ) . All of the preparations either produced the same number of bursts or increased by up to 2 bursts ( Cii ) . Graph symbols in Figures 9 and 10 each represent data from same specimens . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 02310 . 7554/eLife . 02598 . 024Figure 9—source data 1 . Source data for panels B and C . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 02410 . 7554/eLife . 02598 . 025Figure 10 . With PdN6 disconnected , an artificial synaptic conductance reduced the number of VSI bursts . ( A ) Recordings from the same preparation as Figure 9 , but with PdN6 disconnected . PdN6 disconnection reduced the number of VSI bursts from five to four bursts per swim episode ( Ai ) ( Compare with Figure 9Ai ) . Addition of an artificial inhibitory synaptic conductance using dynamic clamp ( D . C ) further decreased the number of VSI bursts to one ( Aii ) . Subtraction of the inhibitory synaptic conductance with the dynamic clamp restored the number of VSI bursts to five ( Aiii ) . ( B ) With PdN6 blocked , addition of synaptic inhibition with dynamic clamp significantly decreased the number of VSI bursts ( Bi , PdN6 blocked vs D . C . , p<0 . 0001 by paired t-test , N = 20 ) . PdN6 disconnection decreased the number of VSI bursts by 24 . 7 ± 20 . 8% from control . Addition of an artificial synaptic conductance using dynamic clamp decreased the number of VSI bursts further to 57 . 7 ± 22 . 0% . Comparison of the number of VSI bursts with dynamic clamp to control shows the points falling below the unity line ( Bii ) . ( C ) With PdN6 blocked , subtracting the synaptic inhibition restored the motor pattern . With dynamic clamp , number of VSI bursts was increased from −21 . 9 ± 20 . 7% below control to −3 . 1 ± 30 . 0% ( Ci , p = 0 . 003 by paired t-test , N = 19 ) . For most preparations , the effect of dynamic clamp was to increase the number of VSI bursts ( Cii ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 02510 . 7554/eLife . 02598 . 026Figure 10—source data 1 . Source data for panel B . DOI: http://dx . doi . org/10 . 7554/eLife . 02598 . 026 We next tested the effects of subtracting the delayed inhibitory component onto the swim motor pattern by setting the inhibitory conductance to a negative value ( see ‘Materials and methods’ ) . Subtraction of the C2-evoked inhibition of VSI significantly increased the number of VSI bursts per swim episode in 6 out of 9 preparations ( Figure 9Aiii , Ci ) . The swim motor pattern was lengthened approximately 20% ( Figure 9Ci , D . C . ) , causing an upward shift in the plot comparing the number of VSI bursts with and without dynamic clamp ( Figure 9Cii ) . Under such conditions , all VSI bursts contained orthodromic action potentials in all 16 preparations . Thus , with PdN6 intact , additional synaptic inhibition from C2 to VSI did not affect the number of bursts in the motor pattern; it acted like a hidden change in the circuit . In contrast , subtraction of the inhibitory conductance extended the motor pattern by causing more orthodromic spiking . When PdN6 was blocked , addition of the artificial C2 to VSI synaptic inhibition further decreased the number of VSI bursts per swim episode , making the preparation more susceptible to disconnection of PdN6 . When the dynamic clamp was turned on , it decreased the number of VSI bursts from four to one ( Figure 10Aii ) . The number of C2 bursts was largely unaffected , probably because the artificial conductance was applied to only one of bilateral VSI pair and C2 is electrically coupled to its contralateral counterpart . On average , without dynamic clamp , blocking PdN6 caused a 25% decrease in the number of VSI bursts ( Figure 10Bi , PdN6 blocked ) . When the dynamic clamp was turned on , this became an approximate 60% decrease compared to control ( Figure 10Bi , D . C . ) . Thus , after PdN6 disconnection , addition of inhibitory synaptic current significantly decreased the number of bursts . Plotting the numbers of VSI bursts with dynamic clamp against the numbers of bursts without dynamic clamp shows that all but two of the 18 preparations fell below the unity slope line ( Figure 10Bii ) . Thus , artificial enhancement of C2-to-VSI inhibition made the swim motor pattern more susceptible to disruption caused by PdN6 disconnection . If the extent of motor impairment by PdN6 disconnection were proportional to the extent of C2-evoked synaptic inhibition of VSI , then decreasing the delayed inhibitory component through subtraction ( see ‘Materials and methods’ ) ought to prevent the motor impairment . Indeed , subtracting the inhibition from C2 to VSI mitigated the effect of disconnecting PdN6 . When the dynamic clamp was turned on , the decreased number of VSI bursts was restored from 27 . 7% below the control swim motor pattern to 3 . 3% above the control , a significant shift ( Figure 10Ci ) . The effect of subtracting the inhibition tended to be more effective in the more vulnerable preparations , which increased the number of bursts by one ( Figure 10Cii ) . Thus , subtracting the inhibitory component of the C2 to VSI synapse restored the impaired swim motor pattern after PdN6 disconnection . The variation of susceptibility to a neural lesion appeared to arise from differences in the synaptic action from neuron C2 to VSI in the swim CPG . The difference was hidden under normal conditions . C2-evoked excitation of VSI is thought to be essential for initiating the ventral phase of each swim cycle during the swim motor pattern ( Getting , 1989a; Calin-Jageman et al . , 2007 ) . Getting ( 1989a ) suggested that C2 excites VSI via direct synaptic action , but we found that the excitation of the proximal VSI process was mainly caused by a bombardment of recruited EPSPs that overrode the direct synaptic action of C2 onto VSI ( Figure 6—figure supplement 1 ) . However , the polysynaptic recruitment did not appear to play a major role in causing the individual differences in the extent of motor impairment after PdN6 disconnection . Rather , it was the inhibitory component of the direct synaptic action by C2 onto VSI that was more significant in predicting the susceptibility to the lesion ( Figure 7 ) . In Hi-Di saline , the C2-evoked synaptic potential in the proximal region of VSI can be detected from the cell body as a slow membrane potential change with an initial depolarization and a later hyperpolarization ( Getting , 1989c; Calin-Jageman et al . , 2007; Sakurai and Katz , 2009b ) . Stronger inhibition would counteract the depolarizing effect of recruited polysynaptic EPSPs in VSI and hence make the system more susceptible to loss of the distal spike initiation zone . Synaptic recruitment and concurrent actions of excitatory and inhibitory synapses have been shown to provide flexibility in neural network outputs in both vertebrates and invertebrates ( Berg et al . , 2007; McLean et al . , 2007 , 2008; Sasaki et al . , 2009; Dougherty and Kiehn , 2010; Kiehn , 2011; Petersen et al . , 2014 ) . Artificially increasing the extent of inhibition from C2 to VSI in the pedal ganglion , using the dynamic clamp technique , caused the motor pattern to become more susceptible . Importantly , it did not affect the production of the motor pattern with PdN6 intact . This shows that this site of variability is not critical to the animal's behavior under normal conditions , but it becomes critical when challenged by lesion . Thus , this synaptic difference serves as a hidden phenotype that predicts the susceptibility of the neural circuit to disruption . Subtracting the C2 to VSI inhibition using dynamic clamp partly rescued vulnerable preparations from the effects of the lesion . This further supports the role of this synapse in determining the susceptibility . Subtraction also caused an increase in the number of bursts per swim episode prior to the lesion . This is likely because artificial reduction of the inhibitory synaptic conductance would enhance C2-evoked depolarization through bombardment of recruited EPSPs , which would make the proximal spike initiation zone of VSI more excitable . This enhanced VSI bursting may have a cascading effect in swim motor pattern generation; by spiking more , VSI would decrease the burst duration in C2 and DSI , which may lead toward generating more burst cycles by releasing less serotonin in each burst ( Fickbohm and Katz , 2000; Katz et al . , 2004 ) . The difference in the synaptic action in the pedal ganglia may affect the direction of spike propagation between them . VSI has two spike initiation zones , one in each pedal ganglion , where C2 synaptic actions take place ( Sakurai and Katz , 2009b ) . The proximal and distal spike-initiation zones each are capable of initiating spikes; however , firing of either zone produces spikes in the same axon but with different directions of propagation ( Figure 8; Sakurai and Katz , 2009b ) . This resembles crayfish central interneurons that integrate segmental mechanosensory inputs ( Hughes and Wiersma , 1960; Calabrese and Kennedy , 1974 ) , locust lobular giant movement detector neurons that generate spikes in both end of the axon ( O'Shea , 1975 ) , and leech heart interneurons ( Calabrese , 1980 ) . In leech heart interneurons , a hyperpolarizing current injection near the dominant spike-initiation site revealed the secondary spike initiation site ( Calabrese , 1980 ) . It is not unusual for neurons to have multiple spike initiation zones with multiple synaptic sites . Since the early description of the crustacean cardiac ganglion ( Bullock and Terzuolo , 1957 ) , there have been numerous reports on multiple spike initiation zones in single neurons . Such neurons include crustacean stomatogastric ganglion neurons ( Moulins et al . , 1979; Meyrand et al . , 1992; Combes et al . , 1993; Smarandache and Stein , 2007; Thuma et al . , 2009 ) , molluscan giant neurons ( Tauc and Hughes , 1963; Zecevic , 1996; Antic et al . , 2000 ) , and insects sensory neurons ( Heitler and Goodman , 1978; Killian et al . , 2000 ) . In these neurons , individual spike-initiation zones can independently fire , which can also be the target for neuromodulation ( Daur et al . , 2009; Bucher and Goaillard , 2011 ) . Local interactions between multiple spike initiation zones are commonly seen in vertebrate cortex ( Llinas et al . , 1968; Schiller et al . , 1997; Stuart et al . , 1997; Larkum and Zhu , 2002 ) and olfactory mitral cells ( Andreasen and Lambert , 1998 ) . Recently , axonal spike back-propagation was shown to play an important role in memory consolidation ( Bukalo et al . , 2013 ) . In this study , we directly showed that the presence of multiple spike-initiation zones made the circuit more resilient to a lesion by providing a backup spike-initiation zone after loss of the primary spike-initiation zone when the commissure was disconnected . To our knowledge , there have been no studies on whether there is individual variability in controlling multiple spike initiation zones and how an injury to such systems would impair important brain functions . The present results may provide insights of how individual differences in spike initiation could covertly reside in a neural circuit where neurons have multiple spike-initiation zones and how partial damage could therefore differentially impair performance . The findings of hidden variability in well-characterized neural circuits may have profound implications for understanding how neural circuits function . In the mammalian brain , there is a general acknowledgement of variability . Individual variability in the susceptibility to neural damage may be attributed to substantial variation in the brain function and anatomy ( Cramer , 2008a ) . For example , the map of motor cortex is actually the population mean; among individuals there is substantial variation in the relationship between brain function and brain anatomy ( Whitaker and Selnes , 1976; Rademacher et al . , 1993; Van Essen et al . , 2012; Cramer et al . , 2003; Cramer , 2008a , 2008b; Smith et al . , 2014; Walhovd et al . , 2014 ) . Endophenotypes such as brain morphology ( Prasad and Keshavan , 2008; Brown and Thompson , 2010; Nenadic et al . , 2012 ) , dendritic spine morphology ( van Spronsen and Hoogenraad , 2010; Penzes et al . , 2011 ) , and other synaptic markers have been sought for neuropathology such as schizophrenia and Alzheimer's disease . Luebke and Foster ( 2002 ) showed that differences between animals in an acetylcholine receptor subunit might account for differences in susceptibility to noise damage . The inability to predict the outcome of neuronal damage is a serious impediment towards developing effective treatments or prophylactic measures . There is also growing concern regarding inter-individual variability in the efficacy and reliability of brain stimulation ( Kim et al . , 2014; Lopez-Alonso et al . , 2014 ) . In contrast , neural circuits in invertebrate systems are composed of relatively small numbers of neurons . In such systems , individual identified neurons often play critical roles and researchers have taken advantage of the simplicity and reproducibility of the connectivity . The number , location , and anatomy of individual neurons are very similar among individuals and their synaptic connections are often drawn by connecting one neuron to the next . However , recent results in several labs have shown that there is considerably more nuance to this view of uniformity than initially appreciated . For example , in the snail respiratory CPG , it was shown there is individual variability at a specific synapse due to distinct activation of two different dopamine receptors ( Magoski and Bulloch , 1999 ) . In the stomatogastric nervous system of crabs , it was suggested that similar neuronal firing patterns could be obtained from neurons with different compositions of ion channels and that different network configurations could produce similar patterns of activity ( Golowasch et al . , 2002; Prinz et al . , 2004 ) . More recently , it was shown that the properties of neurons vary substantially even though the overall behavior of the network is constant ( Goaillard et al . , 2009 ) . In the leech heartbeat CPG , it was found that there is considerable variability among individuals in the strength of synaptic connections among interneurons ( Norris et al . , 2011 ) . Despite such animal-to-animal variability , the CPGs produce the same or similar functional outputs . Interestingly , it has been recently shown in the crab stomatogastric nervous system that the bursting activities of circuit neurons showed different susceptibility to extremely high temperature , but such differences were hidden at temperatures within the physiological range ( Tang et al . , 2012 ) . A modeling study further demonstrated that different sets of network parameters could underlie differential sensitivity of neural circuit to extreme temperatures , but are “good enough” to produce robust bursting activities under normal conditions ( Rinberg et al . , 2013 ) . In the present study , we have shown that in Tritonia , the C2 to VSI synapse serves as a hidden phenotype that helps predict the outcome of this particular lesion . Thus , it is possible that a better understanding of the nature of variability in neural circuits could lead to the discovery of a wider range of hidden phenotypes that predict a variety of conditions and allow them to be treated effectively . Variability even in well-defined circuits indicates that there are many solutions to generating a simple function . These solutions might not all be equivalent; under different circumstances , such as injury , one solution might be superior to another . The causes for the individual synaptic differences in Tritonia are not known . It is possible that the synapses differ because of prior experience or that the synapses change autonomously over time . Such fluid neuronal network structure has been suggested in many studies on both mammals and invertebrates . In the mammalian cortex and spinal cord , trial-to-trial variability in the size and/or pattern of the activated neuronal population has been reported ( Shadlen and Newsome , 1998; Bair et al . , 2001; Cai et al . , 2006; Hansen et al . , 2012 ) . This indicates that continuous changes in functional connectivity within the motor pools and in their activation patterns might underlie such variability ( Cai et al . , 2006; Cramer , 2008a ) . In invertebrates , Magoski and Bulloch ( 2000 ) showed that specific synapses in the snail respiratory CPG constantly change in sign and that even the valence of synaptic transmission can be modulated by environmental and neurohumoral conditions . It is equally possible that the differences in synaptic properties between individuals are caused by genetic differences in this wild-caught population . If the differences are genetic , then they could form the raw material for natural selection to act upon while the behavior remains constant . Such variability , which is neither useful nor injurious , can be hidden in normal life but cause differential survival of individuals when challenged by new conditions ( Darwin , 1876 ) . Although the experimental procedure of blocking the brain commissure is far from a natural stress to the nervous system , our results shed light on how neural circuits can contain hidden variability , which may lead to differential survival under changing conditions including injury . Specimens of the nudibranch , Tritonia diomedea , were obtained from Living Elements Ltd . ( Delta , BC , Canada ) and Friday Harbor Laboratories ( San Juan , WA , USA ) . For behavioral tests of lesions , we employed the same procedure as described previously ( Sakurai and Katz , 2009b ) . Animals were placed for 1 hr in ice-chilled artificial seawater ( Instant Ocean Salt Water , Miami , USA ) containing 0 . 1% 1-Phynoxy-2-propanol ( Redondo and Murray , 2005; Wyeth et al . , 2009 ) . A 2-cm incision was made in the skin between the rhinophores to expose the brain . In the experimental animals , the pedal commissure , which is also known as Pedal Nerve 6 or PdN6 ( Willows et al . , 1973 ) , was cut near the right pedal ganglion with fine scissors , whereas in the sham controls , PdN6 was exposed but not cut . The skin incision was stitched with silk thread and sealed with ‘cyanoacrylate glue’ ( Ethyl Cyanoacrylate , WPI , Sarasota , USA ) . Animals with lesions were paired with sham-operated animals for the behavioral assay . The observer was blind to the condition of each animal in a trial . The swim was induced by applying 5 M NaCl solution ( 0 . 5 ml ) to the dorsal body surface , and the number of swim cycles was counted from complete ventrally directed body flexions . A retraction of the body without a body flexion in response to the stimulus was considered as a swim failure . The number of body flexions during the escape behavior was measured three times before the surgery ( 16 ± 1 , 12 ± 1 and 2 ± 1 hr prior to the commissure transection ) and once after the surgery ( 2 ± 1 hr ) . Before dissection , the animal was chilled to 4°C in the refrigerator . The brain , consisting of the fused cerebropleural and pedal ganglia , was removed from the chilled animal and immediately pinned to the bottom of a Sylgard-lined chamber ( 1 ml ) where it was superfused with saline at 4°C . Physiological saline composition was ( in mM ) : 420 NaCl , 10 KCl , 10 CaCl2 , 50 MgCl2 , 10 D-glucose , and 10 HEPES , pH . 7 . 4 . The cell bodies of the neurons were exposed by removing the connective tissue sheath from the surface of the ganglia ( Willows et al . , 1973 ) . The preparation was left >3 hr superfused in saline at 8–10°C before the electrophysiological experiment . Suction electrodes made from polyethylene tubing were placed on both left and right Pedal Nerves 3 ( PdN3 ) . A suction electrode fabricated from a pulled , fire-polished , borosilicate glass tube ( i . d . , 1 . 0 mm; o . d . , 1 . 5 mm ) was placed in the en passant configuration on PdN6 . In this study , the other pedal commissure , Pedal Nerve 5 ( PdN5 , Willows et al . , 1973 ) was cut in all isolated brain preparations during dissection . Cutting it after the dissection had no apparent effect on the swim motor pattern ( N = 6 ) . Neurons were identified by soma location , electrophysiological monitoring of axonal projection ( cf . , Figure 1C ) , coloration , synaptic connectivity , and activity pattern at rest and during the swim motor program as previously described ( Getting , 1981 , 1983b ) . There are three types of CPG neurons: Dorsal Swim Interneurons ( DSIs , http://www . neuronbank . org/Tri0001043 ) , Cerebral Neuron 2 ( C2 , http://www . neuronbank . org/Tri0002380 ) , and Ventral Swim Interneuron-B ( VSI-B , http://www . neuronbank . org/Tri0002436 ) . For simplicity , we will refer to VSI-B as VSI for this paper . C2 and DSI have cell bodies on the dorsal surface of the cerebral ganglion and project their axons toward the contralateral pedal ganglion , whereas VSI has its cell body on the ventral side of the pleural ganglion and projects its axon toward the ipsilateral pedal ganglion ( Figure 1C ) . To record from both C2 and VSI , the brain was twisted around the cerebral commissure as described by Getting ( 1983b ) . The axons of C2 , DSI and VSI exit the pedal ganglion through one of two commissural nerves ( PdN5 and PdN6 ) and reach the other pedal ganglion ( Newcomb et al . , 2006; Hill and Katz , 2007 ) . To identify neurons , the swim motor program was evoked by stimulating body wall nerve PdN3 with a train of voltage pulses ( 5–15 V , 1 . 5 msec ) at 5 Hz for 3 s via a suction electrode . Unilateral electrical stimulation of PdN3 is sufficient to elicit the bilaterally symmetric swim motor pattern ( Figure 1B ) . Electrical stimuli were given at intervals of greater than 10 min to avoid habituation of the swim motor pattern ( Frost et al . , 1996 ) . In the isolated brain preparation , PdN6 was functionally disconnected by either physical transection or by blocking impulse propagation with TTX ( 1 × 10−4M ) in a suction pipette that contained the commissure . There was no statistical difference between cutting and pharmacological disconnection in either the number of bursts or the intra-burst spike frequency ( Sakurai and Katz , 2009b ) . In this study , both procedures were referred to as PdN6 disconnection . When TTX was used , blockade of axonal impulses in PdN6 was confirmed by examining the change in the impulse waveform recorded on the commissure ( Figure 3B , insets ) . Under control conditions , the impulses were triphasic , with two downward phases ( arrowheads ) and a positive deflection between them ( Figure 3B , PdN6 intact ) . When the action potentials were blocked inside the pipette , the impulses became biphasic with a downward deflection followed by an upward deflection , suggesting that the impulses came into the pipette and were blocked inside it ( Figure 3B , PdN6 blocked; see Sakurai and Katz , 2009b ) . In some experiments , the bathing medium was switched to saline containing a high concentration of divalent cations ( Hi-Di saline ) , which raises the threshold for spiking and reduces spontaneous neural firing . The composition of the Hi-Di saline was ( in mM ) : 285 NaCl , 10 KCl , 25 CaCl2 , 125 MgCl2 , 10 D-glucose , and 10 HEPES ( pH 7 . 4 ) ( Sakurai and Katz , 2003 ) . For all experiments the ganglia were superfused at 2 ml/min at 10°C . Neurons were impaled with glass microelectrodes filled with 3 M potassium chloride ( 12–44 MΩ ) . To test C2-evoked synapses , a standard stimulation was used; C2 made to fire at 10 Hz using repeated injection of 20 ms current pulses . Axoclamp-2B amplifiers ( Molecular Devices , Sunnyvale , USA ) were used for all electrophysiological experiments . Recordings were digitized at 2–6 kHz with a 1401plus A/D converter from Cambridge Electronic Design ( CED , Cambridge , UK ) . Data acquisition and analysis were performed with Spike2 software ( CED ) and SigmaPlot ( Jandel Scientific , San Rafael , CA ) . A cluster of two or more action potentials with intervals of less than 1 s was considered as a burst . VSI often exhibited a few spikes during nerve stimulation; they were not counted as a burst . To measure the polysynaptic action of C2 onto VSI , the amplitude and the frequency of spontaneous EPSPs in VSI were measured for 10 s after the onset of the stimulation . Care was taken not to include stimulus artifacts . EPSPs smaller than 0 . 1 mV were excluded from the analysis . No polysynaptic IPSPs were seen in VSI when C2 was stimulated . We did not distinguish between electrical or chemical synapses in this study . To measure the direct synaptic action of C2 onto VSI , Hi-Di saline was used to remove polysynaptic input . The amplitude of depolarization was measured from the basal resting potential to the maximal peak whereas the amplitude of hyperpolarization was measured from peak to trough . Dynamic clamp software StdpC ( Kemenes et al . , 2011 ) was used to mimic the inhibitory component of the direct synapse of C2 onto VSI . The conductance was based on the Tritonia swim CPG model designed by Getting ( 1989c ) and modified by Calin-Jageman et al . ( 2007 ) . The current injected into the postsynaptic VSI , Isyn , is calculated in each dynamic clamp cycle using a first order kinetics model of the release of neurotransmitter ( Destexhe et al . , 1994; Sharp et al . , 1996; Kemenes et al . , 2011 ) : ( 1 ) Isyn=gsyn S ( t ) [Vsyn−Vpost ( t ) ] , where S ( t ) is the instantaneous synaptic activation , gsyn is the maximum synaptic conductance , Vsyn is the reversal potential ( −80 mV ) of the synapse , and Vpost was fixed to the resting membrane potential of VSI . When subtracting a conductance , a negative value of the same magnitude was used for gsyn . The instantaneous activation , S ( t ) is given by the differential equation: ( 2 ) ( 1−S∞ ( Vpre ) ) τsyndSdt= ( S∞ ( Vpre ) −S ( t ) ) , where , ( 3 ) S∞ ( Vpre ) ={tanh[Vpre−VthreshVslope] if Vpre>VThresh0 otherwise S∞ is the steady state synaptic activation and τsyn is the time constant for synaptic decay . We included two inhibitory components with different τsyn values , 700 ms for the faster component and 1300 ms for the slow component . Vpre is the presynaptic membrane potential and Vthresh is the threshold potential for the release of neurotransmitter; it was set to the level of 50% height of the smallest C2 action potentials . Vslope is the synaptic slope parameter of the activation curve . gsyn ( 1000 ± 100 nS ) and Vslope ( 20 ± 5 mV ) were adjusted before experiment to match the time course of natural inhibition amplitude seen in VSI in that preparation . All synaptic parameters were set identically in each experiment . Because of the locations of the neurons ( C2 is on the dorsal side of the brain and VSI is on the ventral side ) , we were able to manipulate only one C2/VSI pair at a time . Statistical comparisons were performed by using SigmaPlot ver . 12 ( Jandel Scientific , San Rafael , CA ) for Student's t-test , linear regression , Levene median test , One-way ANOVA , and One-way or two-way repeated measures ANOVA followed by all pairwise multiple comparison ( Holm–Sidak method or Tukey test ) . In all cases , p< 0 . 05 was considered significant . Results are expressed as the mean ±standard deviation .	The outcome of a traumatic brain injury or a stroke can vary considerably from person to person , making it difficult to provide a reliable prognosis for any individual person . If clinicians were able to predict outcomes with better accuracy , patients would benefit from more tailored treatments . However , the sheer complexity of the mammalian brain has hindered attempts to explain why similar damage to the brain can have such different effects on different individuals . Now Sakurai et al . have used a mollusc model to show that the extensive variation between individuals could be caused by hidden differences in their neural networks . Crucially , this natural variation has no effect on normal behavior; it only becomes obvious when the brain is injured . The experiments were performed on a type of sea slug called Tritonia diomedea . When these sea slugs encounter a predator they respond by swimming away , rhythmically flexing their whole body . This repetitive motion is driven by a specific neural network in which two neurons—called a cerebral 2 ( C2 ) neuron and a ventral swim interneuron—play important roles . Both of these neurons are quite long and they run alongside each other in the brain , with the ventral swim interneuron being activated by signals sent from the C2 neuron at multiple ‘synaptic connections’ between the two . Sakurai et al . showed that the strength of the connections between the C2 neuron and the ventral swim interneuron varied substantially between animals . However , despite this variation , the sea slugs still performed the same number of whole-body flexions as they swam . Sakurai et al . then made a lesion to the brain , which removed about half of the connections between the C2 neuron and the ventral swim interneuron . This meant that the response of the sea slugs to predators depended on the strength of the remaining connections between the two neurons . Sakurai et al . found that the responses of some sea slugs were only mildly impaired , whereas others were severely impaired . This showed that although variations in the strength of the individual connections had no effect on swimming behavior of normal sea slugs , the same variations had a substantial effect when the brain was damaged . Moreover , by creating computer-generated synapses between the C2 neuron and the ventral swim interneuron , Sakurai et al . were able to change the level of impairment . These findings suggest that the variability in human responses to brain injury could be due to hidden differences at the neuronal level . In everyday life , these differences are unimportant and individuals are able to function in similar ways in spite of subtle differences in their neuronal configurations . However , when the brain is damaged , the differences become more important . This suggests that certain configurations within neuronal networks are more resistant to brain damage than others .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2014	Hidden synaptic differences in a neural circuit underlie differential behavioral susceptibility to a neural injury
bicoid mRNA localises to the Drosophila oocyte anterior from stage 9 of oogenesis onwards to provide a local source for Bicoid protein for embryonic patterning . Live imaging at stage 9 reveals that bicoid mRNA particles undergo rapid Dynein-dependent movements near the oocyte anterior , but with no directional bias . Furthermore , bicoid mRNA localises normally in shot2A2 , which abolishes the polarised microtubule organisation . FRAP and photo-conversion experiments demonstrate that the RNA is stably anchored at the anterior , independently of microtubules . Thus , bicoid mRNA is localised by random active transport and anterior anchoring . Super-resolution imaging reveals that bicoid mRNA forms 110–120 nm particles with variable RNA content , but constant size . These particles appear to be well-defined structures that package the RNA for transport and anchoring . mRNA localisation is a widely-used mechanism for targeting proteins to the regions of the cell where they are required and is often coupled to translational repression to prevent expression of the encoded protein until after its transcript is localised ( Lécuyer et al . , 2007; Jambor et al . , 2015 ) . This is particularly important during axis formation in organisms such as Drosophila and Xenopus where mRNAs localise during oogenesis to provide the primary patterning signals for the embryo . In Drosophila , the anterior-posterior axis is determined by the microtubule-dependent localisation of bicoid ( bcd ) and oskar ( osk ) mRNAs to the anterior and posterior poles of the oocyte , respectively ( Pokrywka and Stephenson , 1991; Clark et al . , 1994; Roth et al . , 1995 ) . bcd mRNA is translationally repressed during oogenesis and is only translated when the egg is laid , providing a local source of Bcd protein , which diffuses to form a morphogen gradient that patterns the anterior half of the embryo ( Ephrussi and St Johnston , 2004 ) . By contrast , osk mRNA is translated when it reaches the posterior of the oocyte to produce long and short isoforms of Oskar protein ( Markussen et al . , 1995; Rongo et al . , 1995 ) . Long Oskar anchors its own RNA , whereas short Oskar nucleates the polar granules , leading to the posterior recruitment of the germ line determinants and the abdominal determinant , nanos mRNA ( Wang and Lehmann , 1991; Ephrussi and Lehmann , 1992; Vanzo and Ephrussi , 2002 ) . Both bcd and osk mRNAs are transcribed in the nurse cells within the germline cyst and are then transported along microtubules through the ring canals into the oocyte by Dynein ( Clark et al . , 2007; Mische et al . , 2007 ) . The localisation of osk mRNA to the posterior of the oocyte requires the plus end-directed microtubules motor protein , Kinesin-I ( Brendza et al . , 2000 ) . Live imaging of fluorescently-labelled osk mRNA reveals that it forms particles that undergo rapid movements in all directions with a slight posterior bias , indicating that the RNA takes a biased random walk to the posterior cortex , where it is then anchored ( Zimyanin et al . , 2008 ) . Since almost all osk mRNA movements depend on Kinesin-I , the microtubule cytoskeleton appears to be largely disordered , with a small excess of microtubule plus ends pointing posteriorly . This is consistent with measurements of the direction of growing microtubule plus ends , which reveal that most grow from the anterior/lateral cortex and extend in random directions with a weak orientation bias that is stronger close to the posterior pole ( Parton et al . , 2011 ) . How bcd mRNA is targeted to the anterior of the oocyte at mid-oogenesis is less well understood . Disrupting the Dynein/Dynactin complex by over-expressing Dynamitin causes either a posterior spreading or complete delocalisation of bcd mRNA , suggesting that the RNA is localised by Dynein-dependent , minus end-directed transport along microtubules ( Clark et al . , 1997; Duncan and Warrior , 2002; Januschke et al . , 2002 ) . However , as Dynein is also required for bcd mRNA transport into the oocyte , it is hard to distinguish direct from indirect effects . Furthermore , injected naïve bcd mRNA accumulates at the nearest region of the anterior/lateral cortex to its site of origin and not specifically at the anterior , consistent with the observation that microtubules ends are anchored or nucleated from all of the cortex except the very posterior pole ( Cha et al . , 2001 ) . Only pre-treatment with nurse cell cytoplasm renders in vitro transcribed RNA competent to localise specifically to the oocyte anterior , and this conditioning requires the the pseudonuclease Exuperantia ( Exu ) ( Cha et al . , 2001 ) . The role of Exu in the localisation of bcd mRNA requires its homo-dimerisation and RNA binding ( Lazzaretti et al . , 2016 ) . Finally , computer simulations of the microtubule network in the anterior region of the oocyte suggest that it has little orientation bias , making it unlikely that RNA movement towards microtubules minus ends can account for the rapid anterior accumulation of the RNA ( Trong et al . , 2015 ) . The number of genes required for bcd mRNA localisation increases as oogenesis proceeds , suggesting distinct mechanisms localise the RNA at different stages . exu is required at all stages of localisation , whereas swallow , the γ-tubulin ring complex ( γ-TURC ) , staufen and the ESCRT-II complex are only needed from stage 10b onwards ( Berleth et al . , 1988; St Johnston et al . , 1989; Ferrandon et al . , 1994; Irion and St Johnston , 2007; Schnorrer et al . , 2000 , 2002; Weil et al . , 2006 ) . The γ-TURC forms a new microtubule organising centre ( MTOC ) in the middle of the anterior cortex at this stage and this coincides with the re-localisation of bcd mRNA from an anterior ring into a central disc adjacent to the MTOC . Furthermore , live imaging of stage 10b-12 oocytes reveals that bcd mRNA particles move towards the anterior in a Dynein-dependent manner ( Weil et al . , 2006 , 2008 ) . Since the RNA localisation is labile , it has been proposed that it is not anchored at the anterior at this stage and continually diffuses away , to be re-localised by Dynein-mediated transport ( Weil et al . , 2008 ) . Here we use fast live imaging , fluorescence recovery after photo-bleaching , photo-conversion and super-resolution microscopy to investigate the mechanism of the initial localisation of bcd mRNA to the anterior at stage 9 . Our results indicate that at this stage the mRNA is not localised by continual directed transport , but by random active transport and anterior anchoring . Fast , high magnification wide-field imaging of bcdGFP in stage 9 oocytes revealed many small RNA particles that moved at speeds of up to 2 . 2 μm/sec ( Figure 1C–D , Video 1 ) . All movements were abolished by treatment with the microtubule-depolymerising drug , Colcemid , whereas the actin depolymeriser , Cytocholasin D , caused premature cytoplasmic streaming but had no effect on particle motility ( Figure 1—figure supplement 1A , Video 2 , data not shown ) . These results are consistent with the observation that bcd mRNA localisation at all stages of oogenesis is disrupted by microtubule-depolymerising drugs , and supports the view that particle movements play a role in delivering the mRNA to the oocyte anterior ( Pokrywka and Stephenson , 1991; Weil et al . , 2006 ) . 10 . 7554/eLife . 17537 . 005Video 1 . ( related to Figure 1C ) – bcd mRNA assembles into particles that undergo fast active transport . High-magnification , wide-field live imaging of bcdGFP in stage 9 oocytes . The right panel shows the fast moving RNA particles as coloured tracks . Images were acquired at a rate of 0 . 64 s/frame . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 00510 . 7554/eLife . 17537 . 006Video 2 . ( related to Figure 1—figure supplement 1C ) – bcd mRNA particles undergo microtubule-dependent active transport . High-magnification , wide-field live imaging of bcdGFP in stage 9 oocytes , with and without depolymerisation of microtubules . Left – mock; Right – Colcemid ( 400 μg/ml ) . The fast moving RNA particles are shown as coloured tracks . Images were acquired at a rate of 0 . 18 s/frame . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 006 The bcd mRNA particles in oocytes with one copy of bcdMS2 moved with an average velocity of 0 . 64 μm/sec , which is significantly faster than osk mRNA particles ( 0 . 47 μm/sec ) imaged under equivalent conditions ( Zimyanin et al . , 2008 ) ( Table 1 , Table 1—source data 1 , Figure 1D ) . This difference was even more marked when we imaged egg chambers expressing two copies of bcdMS2 , with the mean velocity increasing to 0 . 78 μm/s ( Table 1 , Table 1—source data 1 , Figure 1D ) . This increase is presumably because the signal from the fastest particles is spread across more camera pixels per frame ( 6 pixels for particles moving at 2 μm/s , imaged for 0 . 25 s ) , making them harder to detect . Doubling their brightness therefore increases the efficiency of detection of the fastest particles . 10 . 7554/eLife . 17537 . 007Table 1 . Parameters of fast bcd mRNA in wild-type and mutant oocytes . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 00710 . 7554/eLife . 17537 . 008Table 1—source data 1 . Tracking of bcdGFP particles in wild-type and mutant stage 9 oocytes . Includes the data in: Table 1; Figure 1 panels D–F; Figure 2 , panels A–E; Figure 2—figure supplement 1 , panel C . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 008Genotype ( 2x bcdGFP ) Tracks ( n ) Movies ( n ) Oocytes ( n ) Anterior Mov ( %/n ) Binomial P-valuea ) Speed ±S . E . M . ( µm/s ) Wilcoxon P-valueb ) Mixed-effects P-valuec ) Track distance ±S . E . M . ( µm ) Anterior displacement ±S . E . M . ( µm/s ) Wilcoxon P-valued ) Wild-type118132952 . 6 / 6210 . 0400 . 78 / 0 . 011 . 41 / 0 . 030 . 02 / 0 . 020 . 200Dhc6-10/8-16697550 . 8 / 3400 . 3500 . 50 / 0 . 01<2 . 2E-16*6 . 90E-101 . 61 / 0 . 050 . 01 / 0 . 010 . 500exu121517453 . 0 / 1140 . 2070 . 47 / 0 . 02<2 . 2E-163 . 10E-091 . 39 / 0 . 070 . 03 / 0 . 020 . 230Genotype ( 1x bcdGFP ) Tracks ( n ) Movies ( n ) Oocytes ( n ) Anterior Mov ( %/n ) Binomial P-valuea ) Speed ±S . E . M . ( µm/s ) Wilcoxon P-valueb ) Mixed-effects P-valuec ) Track distance ±S . E . M . ( µm ) Anterior displacement ±S . E . M . ( µm/s ) Wilcoxon P-valued ) Wild-type450221354 . 4 / 2450 . 0330 . 64 / 0 . 010 . 98 / 0 . 050 . 02 / 0 . 020 . 135Khc27 GLC92114653 . 7 / 4950 . 0130 . 94 / 0 . 01<2 . 2E-16*1 . 60E-061 . 55 / 0 . 040 . 01 / 0 . 020 . 272Khc17 GLC639211253 . 2 / 3400 . 0570 . 89 / 0 . 02<2 . 2E-162 . 00E-041 . 60 / 0 . 050 . 02 / 0 . 020 . 401Khc23 GLC61215553 . 9 / 3300 . 0291 . 01 / 0 . 02<2 . 2E-16*2 . 20E-061 . 67 / 0 . 060 . 06 / 0 . 030 . 044Dhc8-1/+14174- - -- - -0 . 38 / 0 . 023 . 61E-16*3 . 00E-041 . 15 / 0 . 080 . 06 / 0 . 02- - -Dhc8-1 GLC43167- - -- - -0 . 25 / 0 . 02<2 . 2E-163 . 10E-070 . 72 / 0 . 06-0 . 05 / 0 . 03- - -a ) Binomial test for the frequency of anterior-directed movements being >50% ( one-tailed ) b ) Wilcoxon rank sum test for speed comparisons - comparison to wild-type ( 2x bcdGFP or 1x bcdGFP ) c ) Mixed-effects linear model ( LMER ) test for speed comparisons – comparison to wild-type ( 2x bcdGFP or 1x bcdGFP ) . Fixed Effect: Genotype; Random Effects: Variability between oocytes and moviesd ) Wilcoxon 1-sample test for the net anterior displacement . Null hypothesis: mean=0 ( two-tailed ) p<0 . 05; p<0 . 01; p<0 . 001- - - Not applicable / Not done The observation that bcd mRNA particles move at nearly twice the average speed of osk mRNA particles suggests that they are transported by different motor proteins . The most likely candidate for a motor that transports bcd mRNA is cytoplasmic Dynein , since putative microtubule minus end markers are enriched anteriorly ( Clark et al . , 2007 , 1997 ) . It is not possible to test null mutations in Dynein components , as these block oocyte determination . We therefore used a combination of hypomorphic alleles of the Dynein heavy chain , Dhc6-10/Dhc8-1 , that has previously been shown to reduce the speed of mRNA movement towards the microtubule minus ends in the embryo ( Bullock et al . , 2006 ) . bcd mRNA particles moved with a mean velocity of 0 . 50 μm/sec in Dhc6-10/Dhc8-1 compared with 0 . 78 μm/sec in wild-type oocytes ( Table 1 , Table 1—source data 1 , Figure 1E , Figure 1—figure supplement 1C ) . We observed a similar reduction in the velocities of bcd mRNA particles in Dhc8-1/+ heterozygotes ( 59% of wild-type; Table 1 , Table 1—source data 1 , Figure 1E , Figure 1—figure supplement 1C ) , suggesting that Dhc6-10 has little effect on motor speed and that Dhc8-1 has a dominant negative effect . Consistent with this , homozygous mutant germline clones of Dhc8-1 showed an even greater reduction in particle velocity to 39% of the wild-type speed ( Table 1 , Table 1—source data 1 , Figure 1E , Figure 1—figure supplement 1C ) . Thus , this allele produces a functional motor protein that still moves , but significantly more slowly than the wild-type protein . The strong reduction in the speed of bcd mRNA particles in Dhc8-1 homozygotes indicates that the majority are transported by Dynein . The Dhc8-1 mutant also significantly reduces the amount of bcd mRNA localised to the anterior ( Figure 1—figure supplement 1D ) . This does not seem to be due to reduced frequency of movement because the mobile fraction of bcd mRNA particles is unaffected in Dhc6-10/Dhc8-1 mutants ( 26% in wild-type versus 22% in mutant; Table 2 , Table 2—source data 1 , Figure 1G ) . Thus , slower Dynein-dependent transport , presumably both from the nurse cells into the oocyte and within the oocyte , impairs the delivery of bcd mRNA to the oocyte anterior . 10 . 7554/eLife . 17537 . 009Table 2 . Mobile fraction of bcd mRNA particles in wild-type and mutant oocytes . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 00910 . 7554/eLife . 17537 . 010Table 2—source data 1 . Mobile fraction of bcdGFP particles in wild-type and mutant stage 9 oocytes . Includes the data in: Table 2; Figure 1 , panel G . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 010GenotypeOocytesMobile fraction/5 sT-test P-valuea ) bcdGFP/+50 . 22 ± 0 . 03bcdGFP/+;Khc27 GLC50 . 47 ± 0 . 020 . 0004**bcdGFP/+;Khc23 GLC40 . 25 ± 0 . 020 . 624bcdGFP/bcdGFP;Dhc6-10/8-140 . 26 ± 0 . 050 . 556bcdGFP/bcdGFP;exu140 . 07 ± 0 . 010 . 006*a ) T-test for comparison of mobile fractions ( two-tailed ) - comparisons to bcdGFP/+ ( wild-type ) p<0 . 01; p<0 . 001 Null mutations in the Kinesin heavy chain ( Khc ) also disrupt the localisation of bcd mRNA , with the majority of oocytes showing spreading of the RNA along the anterior and lateral cortex ( Januschke et al . , 2002 ) ( Figure 1—figure supplement 1E ) . It is unclear whether this phenotype arises because Kinesin-I plays a direct role in the transport and/or anchoring of bcd mRNA . Kinesin-I transports Dynein to the oocyte posterior , indicating that the two motors can associate in the same complex , and Kinesin-I could therefore affect bcd mRNA indirectly , for example by recycling Dynein to the oocyte posterior for further rounds of minus end-directed transport , or by modulating the activity of Dynein ( Januschke et al . , 2002; Palacios and St Johnston , 2002 ) . To test the role of Kinesin-I directly , we analysed the movement of bcd mRNA particles in germline clones of the null allele , Khc27 ( Brendza et al . , 2000 ) . Surprisingly , the velocity of bcd mRNA particle movements was significantly increased in the absence of Kinesin-I , with an average speed of 0 . 98 μm/sec , compared to 0 . 64 μm/sec in wild-type ( Table 1 , Table 1—source data 1 , Figure 1E–F , Figure 1—figure supplement 1C ) . This increase could be explained if a fraction of the bcd mRNA particles are transported at low speeds by Kinesin-I , so that its loss raises the average velocity by removing the slow population of moving particles . If so , one would expect the fraction of mobile particles to be reduced in Khc27 . In wild-type , approximately 20% of the bcd mRNA particles move over a five second period ( Table 2 , Table 2—source data 1 , Figure 1G ) , which is nearly twice the proportion observed for osk mRNA ( Zimyanin et al . , 2008 ) . In Khc27 mutants , the mobile fraction more than doubled to 47% ( Table 2 , Table 2—source data 1 , Figure 1G ) . Thus , both the speed and frequency of bcd mRNA particle motility are increased in the absence of Kinesin-I , making it highly unlikely that this motor is responsible for a significant proportion of the movements . Furthermore , bcd mRNA particles showed virtually no reversals of movement in either wild-type or Khc27 mutants ( 1 . 9% versus 1 . 5% , respectively ) , suggesting the RNA does not alternate between transport by motors of opposing polarity . Another way to test the role of Kinesin-I is to analyse two mutants , Khc17 and Khc23 , with single amino acid changes in the motor domain that reduce the speed of Kinesin-I movement without affecting its other properties ( Brendza et al . , 1999; Serbus et al . , 2005; Zimyanin et al . , 2008 ) . Like the null allele , germline clones of Khc17 and Khc23 also caused a significant increase in the mean velocity of bcd mRNA particle movements ( 0 . 89 μm/sec and 1 . 0 μm/sec , respectively ) ( Table 1 , Table 1—source data 1 , Figure 1E ) . This rules out the possibility that Kinesin-I is responsible for the slow movements of bcd mRNA particles , as this would result in a decrease in the average velocity in the mutants . Kinesin-I must therefore act by some other mechanism to reduce the speed of movement by another motor , presumably Dynein , for example by engaging in a tug of war . The slow allele , Khc23 , has little effect on the fraction of mobile particles ( 25% ) , compared to 47% in the null allele ( Table 2 , Table 2—source data 1 , Figure 1G ) , indicating that the effects on speed and frequency are separable . One possible explanation for this difference is that the slow allele still transports the Dynein/Dynactin complex to the posterior pole of the oocyte , albeit more slowly , whereas the null allele does not ( Januschke et al . , 2002 ) ( Figure 1H ) . The absence of Dynein transport to the posterior in the null mutant will therefore increase the concentration of Dynein at the anterior of the oocyte , which could account for the more than doubling of the frequency of particle movement in this region . The localisation of bcd mRNA at all stages depends on Exu protein ( Berleth et al . , 1988 ) and we therefore also examined the behaviour of bcd mRNA particles in the mutant , exu1 , which reduces the affinity of Exu to RNA ( Lazzaretti et al . , 2016 ) . Very few particles were visible in exu1 homozygous oocytes , and these moved significantly less frequently than normal ( 7% ) and at a reduced speed ( Table 1 , Table 1—source data 1 , Table 2 , Table 2—source data 1 , Figure 1E , G , Figure 1—figure supplement 1C ) . Thus , Exu is required for both the formation of bcd mRNA particles and their efficient transport on microtubules , consistent with the dimerisation of Exu ( possibly leading to the dimerisation of bcd mRNA ) and the results obtained from bcd mRNA injections ( Cha et al . , 2001; Lazzaretti et al . , 2016 ) . The residual bcd mRNA motility in the mutant may explain why a small amount of RNA is still diffusely localised at the anterior of exu mutant oocytes ( Figure 1—figure supplement 1F ) . We next assessed the directionality of bcd mRNA particle movements to determine if it could account for the anterior accumulation of the mRNA . We found only a slight excess of movements towards the anterior compared to the posterior over a region up to 40 μm from the anterior of the oocyte ( 52 . 6% versus 47 . 4% , p=0 . 04 ) ( Figure 2A ) . The particles were tracked on near-surface optical sections , where they were better detected , but a similarly weak directional bias was also observed in deeper optical sections ( Figure 2—figure supplement 1A–B ) . The velocity of the movements in each direction was not significantly different , and the net displacement was also not significantly different from zero ( Table 1 , Table 1—source data 1 , Figure 2A–C ) . To test whether this bidirectional movement reflected motors moving in opposite directions along a strongly polarised microtubule cytoskeleton , or mainly unidirectional transport along a weakly polarised cytoskeleton , we measured the velocity in each direction in slow Dynein mutants . All Dhc mutant combinations reduced the velocity of posterior movements to the same extent as of anterior movements ( Figure 2B , D ) . Thus , Dynein is responsible for the majority of particle movements both towards and away from the anterior cortex , and the absence of a strong bias in the direction of bcd mRNA transport is due to the very weak polarisation of the microtubule cytoskeleton in this anterior region . This is in good agreement with tracking of plus-ends of microtubules ( Parton et al . , 2011 ) and computer simulations of the oocyte microtubule network , which predict almost no bias in the orientation of microtubules near the anterior and a stronger bias in the posterior ( Trong et al . , 2015 ) . The exu mutant caused a similar reduction in speed in both directions ( Figure 2B ) , whereas Khc mutants increased the speed in both directions , consistent with the unpolarised nature of the microtubule network at the anterior ( Figure 2D–E ) . 10 . 7554/eLife . 17537 . 011Figure 2 . Fast bcd mRNA particles have little directional bias towards the oocyte anterior . ( A ) Directionality of the fast bcdGFP particles imaged near the cortex of stage 9 oocytes; i ) Windchart of the frequency of movements per angle interval; the upper semi-circle shows all particles whereas the lower semi-circle shows particles according to their distance from the oocyte anterior; ii ) Frequency and average speed of bcdGFP particles moving towards the anterior or posterior of the oocyte; iii ) Frequency table of bcdGFP particles moving in anterior , posterior or lateral directions . ( B–C ) Average speed ( mean ± S . E . M . , 9 oocytes ) ( B ) and speed distribution ( C ) of bcdGFP particles moving towards the anterior ( black bar ) or posterior ( red bar ) of wild-type oocytes . ( D–E ) Average speed ( mean ± S . E . M . , 6 oocytes ) ( D ) and speed distribution ( E ) of bcdGFP particles moving towards the anterior ( black bar ) or posterior ( red bar ) of Khc27 mutant oocytes . ( F–H ) Confocal images of microtubules ( α-tub; F ) , endogenous bcd mRNA ( RNA FISH; G ) or endogenous hts mRNA ( RNA FISH; H ) in wild-type and shot2A2 mutant oocytes . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 01110 . 7554/eLife . 17537 . 012Figure 2—source data 1 . Tracking of bcdTom particles in wild- stage 9 oocytes . Includes the data in: Figure 2—figure supplement 1 , panel B . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 01210 . 7554/eLife . 17537 . 013Figure 2—figure supplement 1 . Fast bcd mRNA particles have little directional bias towards the oocyte anterior . ( A ) Scheme ( i ) and examples of cortical ( ii ) and sub-cortical ( iii ) focal planes of stage 9 oocytes expressing bcdTomato; blue lines show the oocyte margins; arrowheads show individual RNA particles . ( B ) Directionality of fast bcdTomato particles imaged at a deeper , subcortical optical section ( as in Aiii ) ) ; i ) Windchart of the frequency of movements per angle interval; the upper semi-circle shows all particles , whereas the lower semi-circle shows particles according to their distance from the anterior; ii ) Frequency table of particles moving in anterior or posterior directions , or in anterior , posterior or lateral directions . ( C ) Average frequency ( mean ± S . E . M . , 9 oocytes ) of bcdGFP particle movements in anterior , posterior or lateral directions relative to their distance from the oocyte anterior . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 013 We classified the particle movements according to their distance from the anterior cortex to determine if the bias varies across this region ( Figure 2A ) . This revealed that there is a strong anterior bias of 12% in the movements of particles that are more than 10 μm from the anterior cortex ( 39 . 6% versus 27 . 6% ) , but this decreases to 6% in the region 5–10 μm from the anterior and reverses in the 0-5 μm region ( 20 . 8% versus 23 . 9% ) to give a 3% excess of movements away from the anterior ( Figure 2A ) . Similar results were obtained when tracking bcd mRNA particles deeper in the oocyte ( Figure 2—figure supplement 1A–B , Figure 2—source data 1 ) . Our analysis therefore indicates that although Dynein-dependent transport is required for bcd mRNA localisation , it is not sufficient to explain its robust anterior accumulation , because Dynein moves the RNA in and out of the anteriormost region at similar rates . This is incompatible with a model in which bcd mRNA is maintained by continuous anteriorly-directed transport , as has been proposed to occur at later stages of oogenesis ( Weil et al . , 2006 ) . Instead , the data suggest that bidirectional transport facilitates the delivery of bcd mRNA particles to the anterior where they are specifically sequestered by some other mechanism . To further test whether bcd mRNA localisation is independent of a polarised microtubule cytoskeleton , we examined the phenotype of shot2A2 ( Chang et al . , 2011 ) . shot2A2 disrupts the anchoring of microtubules to the anterior/lateral cortex of the oocyte , resulting in a largely unpolarised microtubule network and the failure to localise osk mRNA to the posterior ( Figure 2F ) ( Nashchekin et al . , 2016 ) . bcd mRNA localises normally in shot2A2 mutant oocytes , despite the lack of directional bias in microtubule orientation ( Figure 2G ) . For comparison , we analysed the behaviour of hu-li tai shao ( hts ) mRNA , which also localises anteriorly , but to a somewhat broader region than bcd mRNA and with different genetic requirements ( Ding et al . , 1993 ) . Unlike bcd mRNA , the anterior enrichment of hts mRNA is largely lost in shot2A2 ( Figure 2H ) . Thus , hts mRNA localises by a different mechanism to bcd mRNA that depends on the weakly polarised microtubule cytoskeleton in the vicinity of the anterior . The proposal that there is a mechanism that retains or anchors bcd mRNA once it reaches the anterior predicts that the mRNA should be relatively stable at the anterior , whereas the continual transport model predicts a rapid turn-over of the localised RNA . To distinguish between these possibilities , we performed Fluorescent Recovery After Photobleaching ( FRAP ) experiments on localised bcd mRNA in stage 9 oocytes . The rate of recovery was best fit by two exponential curves , suggesting the existence of fast and slow recovering populations ( Figure 3A , D , Video 3 ) . Furthermore , only the slow population was affected by microtubule-depolymerisation with Colcemid ( Figure 3B , D , Video 3 ) , which abolishes the active transport of bcd mRNA particles ( Figure 1—figure supplement 1A , Video 2 ) . The fast population is therefore likely to correspond to highly-diffusive , nonspecific signal , most likely from autofluorescent background and/or free MCP-GFP . Consistent with this , FRAP on the cytoplasm of nurse cells , which have very low levels of bcd mRNA , or at the anterior of oocytes expressing MCP-GFP alone , yielded very fast recoveries that fit single exponential curves ( Table 3 , Table 3—source data 1 , Figure 3C–D ) . We therefore used the recovery in the nurse cells to fit the FRAP data to a bi-exponential and then removed the nonspecific , fast component ( see Material and Methods ) . The remaining specific signal recovered to 33% over an hour in untreated oocytes , a value that is reduced to 10% by Colcemid treatment ( Table 3 , Table 3—source data 1 , Figure 3E ) . Thus , most bcd mRNA is stably anchored at the anterior cortex at stage 9 , and the limited recovery is predominantly due to microtubule-dependent delivery of mRNA . 10 . 7554/eLife . 17537 . 014Figure 3 . Localised bcd mRNA is anchored at the anterior of the oocyte . ( A–E ) Confocal time-series of FRAP experiments at the anterior of stage 9 oocytes ( A–C ) and the corresponding fluorescence recovery curves ( D–E ) . ( A–B ) Egg chambers expressing bcdGFP were treated with Colcemid ( B , 400 μg/ml ) or control vehicle alone ( A ) 20 min prior to photobleaching . ( C ) Egg chamber expressing only MCP-GFP . ( D–E ) Graphs of FRAP of bcdGFP or MCP-GFP alone , before ( D ) or after ( E ) removal of the fast-recovering , nonspecific component . ( F–K ) Confocal time-series of photo-converted localised bcdDendra2 ( F , I ) , oskDendra2 ( G ) and htsDendra2 ( H ) and the corresponding fluorescence decay graphs after removal of the fast-recovering , nonspecific component ( J–K ) . Dashed lines mark the outline of the oocyte; arrows indicate the photobleached or photo-converted regions and the insets are the corresponding close-ups . ** F-test P value <0 . 0001 . N . S . Statistically not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 01410 . 7554/eLife . 17537 . 015Figure 3—source data 1 . Photo-conversion data for bcdDendra2 and grkDendra2 ( only timepoints 10 min and 60 min after photo-conversion ) . Includes the data in: Figure 3—figure supplement 1 , panel D . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 01510 . 7554/eLife . 17537 . 016Figure 3—figure supplement 1 . Localised bcd mRNA is anchored at the anterior of the oocyte . ( A ) Confocal time-series of photo-conversion at the anterior of stage 9 oocytes expressing just MCP-Dendra2 . ( B ) Graph of the fluorescence decay of photo-converted bcdDendra2 in stage 9 egg-chambers . ( C–D ) Confocal time-series of the photo-conversion of localised grkDendra2 in stage 8 oocytes ( C ) and the corresponding fluorescence decay graph ( D ) . ( E–F ) Confocal time-series of photo-conversion of posteriorly-localised bcdDendra2 in grk2B6/2E12 mutant stage 9 oocytes ( E ) and the corresponding fluorescence decay graph ( F ) . The arrows indicate the photo-converted regions and the insets are the corresponding close-ups . *F-test P value <0 . 0001 . N . S . – Not statistically significant . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 01610 . 7554/eLife . 17537 . 017Table 3 . FRAP kinetics of localised bcd mRNA . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 01710 . 7554/eLife . 17537 . 018Table 3—source data 1 . FRAP data for MCP-GFP and bcdGFP in stage 9 oocytes . Includes the data in: Table 3; Figure 3 , panels D–E . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 018SampleMobile fraction @ 20 minFluorescence Half-time ( min ) Oocyte ( n ) MCP-GFP - St90 . 763 . 76bcdGFP / Nurse cell - St90 . 712 . 08SampleMobile fraction @ 55 minFluorescence Half-time ( min ) Oocytes ( n ) F-test P-value a ) bcdGFP - St9 - Mock0 . 3352 . 29bcdGFP - St9 - Colcemid0 . 1075 . 69<0 . 0001a ) F-test for pairwise comparison of fluorescence recovery curves - comparisons to bcdGFP/+ ( wild-type ) 10 . 7554/eLife . 17537 . 019Video 3 . ( related to Figure 3A–B ) – Anteriorly-localised bcd mRNA has slow and limited turn-over . FRAP of anteriorly-localised bcdGFP in stage 9 oocytes , with and without depolymerisation of microtubules . Left – Control; Right – Colcemid ( 400 μg/ml ) . Confocal images were acquired every 5 min , for 60 min . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 019 The FRAP experiments cannot distinguish whether the recovery is due to the de novo delivery of newly-synthesised bcd mRNA or to the recycling of previously localised RNA from outside the bleached region , as predicted by the continual transport model . To distinguish between these possibilities , we generated a transgene expressing MCP fused to the photo-convertible protein Dendra2 ( MCP-Dendra2 ) so that we could label only the RNA that is already localised ( Gurskaya et al . , 2006; Chudakov et al . , 2007 ) . 63% of photo-converted bcd mRNA ( bcdDendra2 ) remained localised in the same small region over a 55 min period , in good agreement with the FRAP data ( Table 4 , Table 4—source data 1 , Figure 3F , J , Video 4 ) . Moreover , there was very little spreading of the photo-converted RNA along the anterior margin , arguing against continual re-localisation of the RNA ( Figure 3F ) . 10 . 7554/eLife . 17537 . 020Table 4 . Photo-conversion kinetics of localised mRNAs . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 02010 . 7554/eLife . 17537 . 021Table 4—source data 1 . Photo-conversion data for all samples ( MCP-Dendra2; bcdDendra2 , oskDendra2 and htsDendra2 in wild-type stage 9 oocytes; bcdDendra2 in stages 10b , 13 and 14 wild-type oocytes; bcdDendra2 in wild-type stage9 oocytes expressing 2 copies of the bcdMS2 transgene; bcdDendra2 in grk mutant stage9 oocytes; bcdDendra2 in wild-type stage 9 oocytes treated with Colcemid or mock Control ) . Includes the data in: Table 4; Figure 3 , panels J-K; Figure 3—figure supplement 1 , panels B , F; Figure 4 , panel C . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 021SampleMobile fraction @ 20 minFluorescence Half-time ( min ) Oocytes ( n ) bcdDendra2 / Nurse cell - St90 . 933 . 210mRNAMobile fraction @ 55 minFluorescence Half-time ( min ) Oocytes ( n ) F-test P valuebcdDendra2 - St90 . 3726 . 010oskDendra2 - St90 . 20173 . 65<0 . 0001a ) htsDendra2 - St90 . 5931 . 710<0 . 0001a ) bcdDendra2 - St100 . 6416 . 18<0 . 0001a ) bcdDendra2 - St130 . 3924 . 570 . 81a ) bcdDendra2 - St140 . 24138 . 510<0 . 0001a ) bcdDendra2 / grk2E12/2B6 - St90 . 4033 . 2110 . 91a ) bcdDendra2 / 2x bcdMS2 - St90 . 4451 . 450 . 4a ) mRNAMobile fraction @ 55 minFluorescence Half-time ( min ) Oocytes ( n ) F-test P valuebcdDendra2 - St9 - Mock0 . 3527 . 950 . 86a ) bcdDendra2 - St9 - Colcemid0 . 22152 . 15<0 . 0001a ) , b ) a ) , b ) F-test for pairwise comparison of fluorescence recovery curves . Comparisons to ( a ) bcdDendra2 - St9 or ( b ) bcdDendra2 - St9 - Mock10 . 7554/eLife . 17537 . 022Video 4 . ( related to Figure 3F ) – Localised bcd mRNA is stably anchored at the oocyte anterior . Photo-conversion of anteriorly-localised bcdDendra2 in a stage 9 oocyte . Left – Unconverted bcdDendra2 in green , photo-converted in red; Right – Photo-converted bcdDendra2 alone . Confocal images were acquired every 5 min , for 55 min . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 022 We also analysed the behaviour of MS2-tagged hts mRNA labelled with MCP-Dendra2 ( htsDendra2 ) . Unlike bcd mRNA , photo-converted htsDendra2 showed marked spreading along the anterior cortex and was significantly more labile , with less than half ( 41% ) of the signal remaining by the end of the time course ( Figure 3H , I , Video 5 ) . Both the spreading and the lower retention of localised hts mRNA are consistent with the idea that its anterior enrichment depends on continual transport , unlike bcd RNA at stage 9 of oogenesis . 10 . 7554/eLife . 17537 . 023Video 5 . ( related to Figure 3H ) – Localised hts mRNA is less stable at the oocyte anterior and spreads laterally . Photo-conversion of anteriorly-localised htsDendra2 in a stage 9 oocyte . Left – Unconverted htsDendra2 in green , photo-converted in red; Right – Photo-converted htsDendra2 alone . Confocal images were acquired every 5 min , for 55 min . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 023 As bcd RNA has been proposed to be localised by continual active transport beginning at stage 10b , we examined if the retention of localised bcd mRNA varies during oogenesis , by performing similar photo-conversion experiments on stage 10b , stage 13 and stage 14 oocytes . The RNA is significantly more mobile at stage 10b , with only 36% remaining in the ROI after 55 min ( Figure 3K ) , consistent with the results of Weil et al . ( 2006 ) . Stage 13 showed similar levels of retention to stage 9 , but the RNA was significantly more stable at stage 14 , coincident with the formation of larger aggregates ( Figure 3J–K ) ( Weil et al . , 2008 ) . bcd mRNA therefore appears to be localised by a distinct mechanism at stage 10b , when fast cytoplasmic streaming starts and the RNA relocalises from an anterior ring into a central disc ( Theurkauf et al . , 1992; Schnorrer et al . , 2000 , 2002 ) . Since our results suggest that bcd mRNA is anchored in some way at the anterior of the oocyte at stage 9 , we compared its behaviour to that of osk and gurken ( grk ) mRNAs , which are both anchored to the cytoskeleton after their localisation ( Delanoue et al . , 2007; Vanzo and Ephrussi , 2002 ) . Although osk RNA is more stably anchored than bcd mRNA , with 80% of the signal remaining at the end of the experiment ( Figure 3G , I ) , grk and bcd mRNAs showed almost identical retention rates ( Figure 3—figure supplement 1C–D , Figure 3—source data 1 ) . The greater retention of osk mRNA compared to bcd and grk mRNAs could reflect distinct anchoring mechanisms , but might also be due to different conditions at the posterior of the oocyte relative to anterior , where bcd and grk RNAs localise . To directly test of the effects of position within the oocyte , we examined bcd mRNA stability in strong grk mutants , in which bcd mRNA localises to both anterior and posterior poles ( González-Reyes et al . , 1995; Roth et al . , 1995 ) . Photo-converted bcd mRNA at the posterior of grk mutants yielded decay kinetics indistinguishable from those of RNA at the anterior of wild-type oocytes ( Figure 3—figure supplement 1E–F , Table 4 , Table 4—source data 1 ) . Thus , the stability of localised bcd mRNA is intrinsic and not a consequence of the local geometry of the oocyte . We next investigated mechanisms that might retain bcd mRNA at the anterior cortex . grk mRNA is anchored at the dorsal anterior corner of the oocyte by the binding of static Dynein to minus ends of microtubules , a mechanism that also anchors pair rule transcripts apically in the blastoderm embryo ( Delanoue and Davis , 2005; Delanoue et al . , 2007 ) . We therefore tested whether the anterior retention of bcd mRNA is microtubule-dependent by culturing the egg chambers in the presence of Colcemid . Although the microtubules were completely depolymerised after ten minutes , as monitored with the microtubule binding protein , Tau-GFP , bcd mRNA labelled with MCP-Tomato ( bcdTom ) remained tightly localised after 1 hr ( Figure 4A , Video 6 ) . Furthermore , performing the same experiment with photo-converted RNA revealed that the immobile fraction increased from 65% to 78% in the absence of microtubules ( Table 4 , Table 4—source data 1 , Figure 4B–C ) . Thus , the anchoring of bcd mRNA is microtubule-independent , and microtubule-dependent processes enhance its depletion from the anterior cortex . It is notable that bcd mRNA is as stably localised as osk mRNA in the absence of microtubules , with only about 20% loss over the period of 55 min . hts mRNA was more sensitive than bcd mRNA to the depletion of microtubules at stage 9 , with only a small amount of RNA remaining localised ( Figure 4D ) . 10 . 7554/eLife . 17537 . 024Figure 4 . bcd mRNA is not anchored on microtubules at the anterior of stage 9 oocytes . ( A–B ) Confocal time-series of stage 9 egg chambers expressing bcdTOM and Tau-GFP ( A ) or bcdDendra2 ( B ) treated with the microtubule-depolymerising drug , Colcemid ( mock or 400 μg/ml ) ; the arrows indicate anteriorly localised bcdTOM ( A ) or photo-converted bcdDendra2 ( B ) and the insets are the corresponding close-ups . ( A ) Colcemid was added to medium 15 min after the beginning of imaging; Images on the right are maximum intensity projections over the Z-dimension , showing the anterior-lateral ring of bcdTOM . ( B ) Colcemid was added to the medium 20 min prior photo-conversion of localised bcdDendra2 . ( C ) Graph of the fluorescence decay of photo-converted Dendra2 ( B ) , after removal of the fast-recovering , nonspecific component; *F-test p value <0 . 0001 . ( D ) Confocal imaging of endogenous hts mRNA ( FISH ) in stage 9 ( left - single confocal section; right - maximum intensity projection of the full volume of the oocyte ) and stage 10b egg chambers after 90 min treatment with Colcemid ( mock or 400 μg/ml ) ; DNA ( DAPI ) in blue; asterisk indicates the oocyte nucleus . ( E ) Confocal images of bcdTOM in stage 10b , 11 and 13 egg chambers after 90 min treatment with Colcemid ( mock or 400 μg/ml ) ; asterisks indicate the oocyte nucleus . ( F–G ) High magnification wide-field two-colour imaging of bcdGFP and the minus-end microtubule marker , mCherry-Patronin , in stage 9 oocytes ( F ) , and the corresponding Van Steensel co-localisation analysis ( G ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 02410 . 7554/eLife . 17537 . 025Figure 4—source data 1 . Van Steensel co-localisation analyses of bcdGFP and mCherry-Patr . Includes the data in: Figure 4 , panel G . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 02510 . 7554/eLife . 17537 . 026Figure 4—source data 2 . Van Steensel co-localisation analyses of bcdTom and Tau-GFP . Includes the data in: Figure 4—figure supplement 1 , panel B . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 02610 . 7554/eLife . 17537 . 027Figure 4—figure supplement 1 . bcdmRNA is not anchored on microtubules at the anterior of stage 9 oocytes . ( A–B ) High magnification imaging of the microtubule marker , Tau-GFP ( green ) , and bcdTomato ( red ) in stage 9 oocytes , after feeding flies for 2 hr with fresh yeast paste containing Colcemid ( 150 μg/ml ) ( A ) , and the corresponding Van Steensel co-localisation analysis ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 02710 . 7554/eLife . 17537 . 028Video 6 . ( related to Figure 4A ) – Localised bcd mRNA is anchored at the oocyte anterior independently of microtubules . Confocal live imaging of bcdTom and the microtubule marker , Tau-GFP , in stage 9 oocytes , with and without depolymerisation of microtubules . Top – mock control; Bottom – Colcemid ( 400 μg/ml ) . Colcemid was added 15 min after the start of imaging . Confocal images were acquired every 5 min . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 028 By contrast to stage 9 , treating stage 10b egg chambers with Colcemid led to the loss of most bcd mRNA from the anterior , except for a small amount around the oocyte nucleus ( Figure 4E ) . This is consistent with the lower anterior retention of photo-converted bcd mRNA at stage 10b . This reduction in stability coincides with the assembly of new MTOC in the middle of the anterior cortex and the relocalisation of the RNA from an anterior ring to a central disc ( Schnorrer et al . , 2000 , 2002 ) . The microtubule-dependence of bcd mRNA retention is transient , however , and microtubule depolymerisation had little effect on the stability of the localisation at stage 11 and stage 13 ( Figure 4E ) . Although Colcemid depolymerises all detectable microtubules in the oocyte , short 'stumps' of microtubules may persist where the minus ends are attached to the cortex , which could provide anchors for bcd mRNA . We therefore examined the distribution of bcd mRNA particles relative to the microtubule minus ends , using the microtubule minus end-binding protein Patronin as a marker ( Goodwin and Vale , 2010; Hendershott and Vale , 2014 ) . Although both Patronin-labelled microtubule minus ends and bcd mRNA particles are most concentrated at the anterior corners of the oocyte , they very rarely overlap ( Figure 4F ) . Analysing their distributions using the van Steensel method gives a Pearson’s correlation coefficient of 0 . 1 at zero displacement , indicating that only a very small proportion of bcd mRNA particles co-localise with microtubule minus ends ( Figure 4G , Figure 4—source data 1 ) ( van Steensel et al . , 1996 ) . We obtained similar results when using the microtubule-associated protein , Tau , as marker for stable microtubule ends in the presence of Colcemid ( Figure 4—figure supplement 1A–B , Figure 4—source data 2 ) ( Parton et al . , 2011 ) . The small degree of co-localisation makes it unlikely that the RNA is anchored to minus ends by static Dynein . Since microtubules do not appear to anchor bcd mRNA , we turned to cortical actin , which has been implicated in keeping the RNA localised at later stages ( Weil et al . , 2006 , 2010 ) . To explore an earlier role of actin on bcd mRNA localisation , we cultured stage 9 oocytes in the presence of the actin depolymerising drug , Cytochalasin D . The treatment interfered with the cytoplasmic actin mesh , as it induced premature cytoplasmic streaming ( data not shown ) ( Dahlgaard et al . , 2007 ) , but did not significantly affect the cortical actin or the distribution of bcd mRNA ( Figure 5A ) . Since the drug treatment experiment was not conclusive , we applied two-colour Stimulated Emission Depletion ( STED ) super-resolution microscopy to investigate whether bcd mRNA is anchored on actin , but detected virtually no co-localisation between them ( Figure 5B–C ) . These data suggest that cortical actin is unlikely to act as a direct anchor for bcd mRNA during stage 9 . 10 . 7554/eLife . 17537 . 029Figure 5 . bcd mRNA is not directly anchored on cortical actin at the anterior of stage 9 oocytes . ( A ) Confocal imaging of stage 9 egg chambers expressing bcdGFP ( green ) and labelled for actin ( Phalloidin-TRITC , red ) , after 90 min treatment with the actin-depolymerising drug , Cytochalasin D ( mock or 10 μg/ml ) . ( B–C ) STED super-resolution mid-sagittal ( B ) or surface ( C ) images of stage 9 egg chambers expressing bcdGFP ( stained with GFP-Booster-ATTO647N , green ) and labelled for actin ( Phalloidin-ATTO590 , red ) ; spectral unmixing was applied to the images; the blue line indicates the oocyte anterior . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 029 In many cases , RNAs are retained in a specific location by their incorporation into large particles , such as the polar granules at the posterior of the Drosophila oocyte or the P granules in C . elegans ( Little et al . , 2015; Trcek et al . , 2015; Elbaum-Garfinkle et al . , 2015 ) . bcd mRNA has been shown to be associated with P-bodies , which may act to prevent its translation during oogenesis ( Weil et al . , 2012 ) . This raises the possibility that sequestration in P-bodies also plays a role in anchoring bcd mRNA at the anterior . We observed that bcd mRNA particles partially co-localise with P-bodies , particularly at the very anterior of the oocyte . First , van Steensel co-localisation analysis of bcd mRNA ( bcdGFP ) and the P-body component , Trailerhitch ( Tral-mRFP ) , gave Pearson’s correlation coefficients of 0 . 3 and 0 . 2 at the very anterior and adjacent cytoplasm of the oocyte , respectively ( Figure 6A–B , Figure 6—source data 1 ) . Second , two-colour STED super-resolution microscopy revealed that bcd mRNA particles are enriched within P-bodies , particularly in the larger aggregates , but are also found free in the cytoplasm ( Figure 6C ) . P-bodies are ubiquitous in the cytoplasm , however , and are most abundant in the posterior pole plasm ( Figure 6D ) . Thus , incorporation of bcd RNA into P-bodies could play a role in retaining the RNA at the anterior , but there must be some additional mechanism to ensure that the RNA is only sequestered once it has reached its destination . 10 . 7554/eLife . 17537 . 030Figure 6 . bcd mRNA partially co-localise to P-bodies at the anterior of stage 9 oocytes . ( A–B ) High magnification wide-field imaging of bcdGFP ( green ) and the P-body component , Tral-mRFP ( red ) ( A ) , and the corresponding Van Steensel co-localisation analysis ( B ) in stage 9 oocytes . ( C ) STED super-resolution imaging of a stage 9 egg chamber expressing bcdGFP ( GFP-Booster-ATTO647N , green ) and immuno-labelled for the P-body component Me31B ( ATTO590 , red ) ; spectral unmixing was applied to the image; the blue line indicates the oocyte anterior . ( D ) Confocal image of a wild-type stage 9 oocyte expressing bcdGFP ( green ) and Tral-RFP ( red ) . ( E ) Confocal imaging of endogenous bcd mRNA ( RNA FISH ) in wild-type and Ge-1Δ5 mutant stage 9 oocytes . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 03010 . 7554/eLife . 17537 . 031Figure 6—source data 1 . Van Steensel co-localisation analyses of bcdGFP and Tral-mRFP . Includes the data in: Figure 6 , panel B . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 031 To test the role of P-bodies more directly , we examined bcd mRNA localisation in germline clones of a null mutation in the core P-body component , Ge-1 , which has been reported to disrupt P-body structure ( Fan et al . , 2011 ) . The localisation of bcd mRNA appeared normal in Ge-1∆5 clones , however , suggesting that P-body integrity is not required for bcd mRNA anchoring ( Figure 6E ) . The highly structured 3’UTR of bcd mRNA contains dimerisation/oligomerisation domains ( stem-loop III ) that are required for its efficient transport and apical localisation in the syncytial blastoderm embryo ( Ferrandon et al . , 1997; Wagner et al . , 2001; Snee et al . , 2005 ) . This raises the possibility that the RNA is retained at the anterior by aggregating into larger and less diffusive particles . We therefore used STED microscopy to visualise MS2/MCP-labelled RNA particles with high resolution . Imaging of stage 9 oocytes revealed that bcd , grk and hts mRNAs form regular particles ( Figures 5B-C , 6C , 7A–B , data not shown ) . We then used a fluorescence intensity curve fitting method to estimate the sizes of the particles ( see Material and Methods , Figure 7—figure supplement 1A ) . bcd and grk mRNAs particles averaged 112 nm and 106 nm , respectively , whereas hts mRNA particles were significantly smaller , averaging 77 nm ( Table 5 , Table 5—source data 1 , Figure 7A-B , G ) . 10 . 7554/eLife . 17537 . 032Figure 7 . bcd mRNA assembles into stereotypical particles . ( A–F ) STED super-resolution imaging of mRNA particles labeled with GFP ( A–D ) or single molecule fluorescence in situ hybridization ( smFISH ) ( E–F ) . ( A , C , D ) bcdGFP in wild-type stage 9 ( A ) and stage 14 oocytes ( C , confocal mode on the left , STED mode on the right ) , and exu1/exuVL stage 9 oocytes ( D ) . ( B ) htsGFP in wild-type stage 9 oocytes . ( E–F ) smFISH of endogenous bcd mRNA in stage 9 and stage 14 oocytes . The blue lines indicate the oocyte anterior; the insets are close-ups of the dashed boxes , confocal mode on the left , STED mode on the right . ( G–H ) Boxplots of the sizes of RNA particles labelled with MCP-GFP ( G ) or smFISH ( H ) . ( G ) bcdGFP , grkGFP and htsGFP in stage 9 oocytes . ( G’ ) bcdGFP particles from wild-type and exu1/exuVL stage 9 oocytes . ( G’’ ) bcdGFP particles in wild-type oocytes at stage 9 and the isolated particles at stage 14 . ( G’’’ ) bcdGFP particles in stage 9 oocytes expressing one or two copies of the bcdMS2 transgene . ( H ) bcd RNA particles in wild-type oocytes at stage 9 and isolated or clustered particles at stage 14 . ( H’ ) Stage 9 oocytes expressing only endogenous bcd RNA or one or two additional copies of the bcdMS2 transgene . ( I ) Relative amounts of bcd 3’UTR ( RT-qPCR ) in ovaries expressing only endogenous bcd mRNA ( yw ) or one or two additional copies of the bcdMS2 transgene ( 2 primer pairs , mean ± S . E . M . , 3 biological replicates ) . ( J–K ) Scatterplots of particle sizes versus distance from the anterior in stage 9 oocytes . ( J ) MCP-GFP-labelled transgenic bcdMS2 . ( K ) smFISH-labelled endogenous bcd RNA . p<0 . 05; p <0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 03210 . 7554/eLife . 17537 . 033Figure 7—figure supplement 1 . bcd mRNA assembles into stereotypical particles . ( A ) Example of a bcdGFP particle curve-fitted by the Fiji plugin , Profiler . ( B ) Graph of the fluorescence decay of photo-converted localised bcdDendra2 in oocytes expressing one or two copies of bcdMS2; N . S . – F-test p value not statistically significant . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 03310 . 7554/eLife . 17537 . 034Table 5 . Analyses of mRNA particle properties . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 03410 . 7554/eLife . 17537 . 035Table 5—source data 1 . Properties of RNA particles from STED super-resolution imaging . Includes the data in: Table 5; Figure 7 , panels G–K; Figure 8 , panles A–D . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 035GenotypeParticles ( n ) Particle size ±S . E . M . ( nm ) Mixed-effects test P-valuea ) Mixed-effects test P-valueb ) Summed fluorescence ±S . E . M . ( a . u . ) Mixed-effects test P-valuec ) Mixed-effects test P-valued ) MS2-labellingbcdGFP - St9732111 . 9 ± 1 . 1- - -0 . 14- - -- - -- - -grkGFP - St8-9376106 . 5 ± 1 . 60 . 41- - -- - -- - -- - -htsGFP - St928476 . 6 ± 1 . 00 . 002*0 . 56- - -- - -- - -bcdGFP / exu1 - St922497 . 9 ± 1 . 80 . 046- - -- - -- - -- - -bcdGFP - St14 ( isolated ) 165109 . 6 ± 2 . 50 . 56- - -- - -- - -- - -bcdGFP – St9 / 1x bcdMS2 293100 . 2 ± 1 . 50 . 070 . 78155 . 1 ± 5 . 3- - -0 . 11bcdGFP – St9 / 2x bcdMS2 292105 . 3 ± 1 . 60 . 19- - -149 . 3 ± 5 . 10 . 50- - -smFISHbcd - St9endogenous ( yw ) 901125 . 9 ± 1 . 0- - -<0 . 0001**204 . 4 ± 5 . 4- - -0 . 13bcd – St14 ( isolated ) endogenous ( yw ) 749116 . 3 ± 1 . 20 . 044- - -687 . 7 ± 30 . 0<0 . 0001*- - -bcd – St14 ( clustered ) endogenous ( yw ) 125124 . 6 ± 2 . 60 . 84- - -- - -- - -- - -bcd – St9 endogenous + 1x bcdMS2 935128 . 6 ± 1 . 00 . 28- - -264 . 8 ± 7 . 4<0 . 0001- - -bcd – St9 endogenous + 2x bcdMS2 1509124 . 6 ± 0 . 80 . 74- - -328 . 5 ± 7 . 4<0 . 0001*- - -a ) Mixed effects linear model ( LMER ) test for comparison of RNA particle sizes ( FWHM ) . Fixed effect: mRNA / Genotype; Random effect: variability between oocytes . Compared to bcdGFP - St9 or bcd - St9 endogenous ( yw ) b ) Mixed effects linear model ( LMER ) test to analyse the effect of the distance from the anterior on the RNA particle sizes ( FWHM ) . Fixed effect: Distance from anterior; Random effect: variability between oocytes . c ) Mixed effects linear model ( LMER ) test for comparison of the summed fluorescence of RNA particles . Fixed effect: mRNA / Genotype; Random effect: variability between oocytes . Compared to bcdGFP - St9 / 1x bcdMS2 , or bcd - St9 endogenous ( yw ) d ) Mixed effects linear model ( LMER ) test for comparison of the summed fluorescence of RNA particles . Fixed effect: Distance from anterior; Random effect: variability between oocytes- - - Not applicable / Not determineda . u . arbitrary unitsp<0 . 05; p<0 . 01; p<0 . 001 bcd mRNA forms larger aggregates at stage 14 of oogenesis , but super-resolution imaging revealed that these are still composed of small , discrete RNA particles ( Figure 7C ) . As the particles in clusters are about 200 nm apart in the XY dimension ( 224 nm and 202 nm mean distance to nearest neighbour in stage 9 and 14 oocytes , respectively ) , each ~700 nm optical Z-section is likely to include more than one particle . This causes a high and irregular background , leading to overestimation of fluorescence intensities and unreliable curve fittings . Nevertheless , the estimated size of isolated bcd mRNA particles at stage 14 was comparable to those in stage 9 oocytes ( Table 5 , Table 5—source data 1 , Figure 7G ) . This suggests that bcd mRNA particles remain relatively homogeneous in size throughout oogenesis , despite of their clustering into large , semi-ordered aggregates at stage 14 . To confirm these findings , we also performed STED imaging on endogenous bcd RNA labelled by single molecule FISH ( smFISH ) with probes spanning the 3’UTR . This technique also revealed that bcd RNA forms particles that remain approximately the same size throughout oogenesis , although they appear slightly larger than those seen with MS2-GFP labelling , presumably because smFISH labels the entire bcd 3’UTR , not just the MS2 sites in the RNA ( Table 5 , Table 5—source data 1 , Figure 7E–F ) . To explore if particle remodelling plays a role in anchoring bcd mRNA at the anterior , we compared the properties of particles at different distances from the anterior margin of stage 9 oocytes . The average size of bcd mRNA particles measured by both MS2-labelling and smFISH did not change substantially with the distance from the anterior ( Table 5 , Table 5—source data 1 , Figure 7J–K ) , arguing against their coalescence upon localisation . The uniform size of bcd mRNA particles , regardless of their location or the stage of oogenesis , was unexpected and suggests that the RNA is incorporated into a well-defined structure rather than assembling into aggregates of variable size depending on the RNA concentration . We tested this hypothesis by comparing the sizes of the particles formed by the endogenous RNA in wild-type oocytes with those formed in oocytes expressing either one or two additional copies of bcdMS2 , which raises the RNA levels to 1 . 75x and 3 . 25x the endogenous level , respectively ( Figure 7I ) . The size of the bcd RNA particles remained constant with increasing RNA concentration ( Table 5 , Table 5—source data 1 , Figure 7G–H ) , but we observed significantly more particles ( 172% ) in oocytes expressing 2 copies of bcdMS2 compared to just the endogenous RNA alone . Consistent with this , extra bcdMS2 RNA did not affect the decay kinetics in photo-conversion experiments , indicating that the diffusion characteristics of the particles were also constant under differing RNA concentrations ( Figure 7—figure supplement 1B ) . These experiments demonstrate that the size of bcd mRNA particles is insensitive to the concentration of mRNA , supporting the view that the RNA is assembled into a distinct structure of uniform size . Indeed , the only condition that altered the size of the bcd mRNA particles was the exu1 mutant , which strongly reduces the affinity of Exu for RNA ( Lazzaretti et al . , 2016 ) . In this case , the few particles that were detected were slightly , but significantly , smaller ( 98 nm versus 112 nm ) ( Table 5 , Table 5—source data 1 , Figure 7D–E ) . Thus , Exu may provide part of the scaffolding for the assembly of bcd mRNA particles . To determine whether the particles also contain the same amount of RNA throughout oogenesis and at different bcd gene dosages , we used the summed fluorescent intensities of the particles as a measure of their RNA content . The average fluorescent intensity of the particles detected by both MS2-labelling and smFISH did not change with distance from the anterior , reinforcing the conclusion that the particles do not fuse upon localisation and anchoring at the anterior ( Figure 8A–B ) . By contrast , the mean fluorescence intensity of the particles increased when bcd mRNA levels were raised by expressing one or two copies of the bcdMS2 transgene , with 30% and 60% more fluorescence , and thus more RNA , per particle respectively ( Table 5 , Table 5—source data 1 , Figure 8C ) . Furthermore , stage 14 particles contained more than 3 times as much RNA as those at stage 9 ( Table 5 , Table 5—source data 1 , Figure 8D ) . It has previously been shown that most bcd RNA enters the oocyte during nurse cell dumping at stage 10b-12 ( Weil et al . , 2006 ) , suggesting that much of this extra RNA is incorporated into pre-existing RNA particles . Thus , the bcd RNA particles have a variable RNA content , despite their constant size . 10 . 7554/eLife . 17537 . 036Figure 8 . The RNA content of the bcd mRNA particles increases during oogenesis and with higher gene dosage . ( A–B ) Scatterplots of the summed fluorescence intensities of bcd RNA particles versus distance from the anterior at stage 9 . ( A ) GFP-labelled transgenic bcdMS2 . ( B ) smFISH-labelled endogenous bcd RNA ( yw genotype ) . ( C–D ) Boxplots of the summed fluorescence intensities of bcd RNA particles . ( C ) smFISH-labelled bcd RNA particles from stage 9 oocytes expressing only endogenous bcd mRNA ( yw ) or one or two additional copies of the bcdMS2 transgene . ( D ) smFISH-labelled bcd RNA particles from stage 9 and stage 14 ( isolated ) oocytes expressing only endogenous bcd mRNA ( yw ) . *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 036 It has been generally assumed that bcd mRNA is localised by directed transport along a polarised microtubule cytoskeleton ( Wolpert et al . , 2015; St Johnston , 2005 ) . Indeed , studies on the later stages of oogenesis indicate that the RNA is continually transported towards microtubule minus ends by Dynein and then is increasingly more anchored as oogenesis progresses ( Pokrywka and Stephenson 1991; Weil et al . , 2006 , 2008 ) . Here we provide evidence that the RNA is localised by a different mechanism at other stages . Live imaging of fluorescently-labelled bcd mRNA in stage 9 oocytes revealed that the RNA forms particles that undergo frequent active movements along microtubules . The speed was strongly reduced in Dhc mutants , consistent with bcd mRNA being transported predominantly by Dynein . Furthermore , bcd mRNA moved significantly more frequently and faster in a null mutant in Khc . Since Kinesin-I transports Dynein to the oocyte posterior , the more frequent movements in the Khc null mutant may be explained by the higher levels of Dynein at the anterior of the oocyte , which should increase the probability of its binding to bcd RNPs . This cannot account for the increased velocity of bcd mRNA particle movements , however , as this also occurs in the “slow” Khc alleles , which still localise Dynein posteriorly . Thus , Kinesin-I somehow slows down Dynein through a mechanism that depends on its full motor activity . One possibility is that Kinesin-I engages in a tug of war with Dynein and therefore exerts a drag that slows down Dynein movement . Interestingly , inhibition of Dynein increases both the velocity of Kinesin-I-driven ooplasmic streaming and osk mRNA particle transport , indicating that this antagonistic relationship between Dynein and Kinesin-I is reciprocal ( Serbus et al . , 2005; Zimyanin et al . , 2008 ) . osk mRNA is transported to the posterior of the oocyte by Kinesin-I-dependent movements along a weakly polarised microtubule network , in which approximately 15% more of the microtubules have their plus-ends pointing posteriorly than anteriorly ( Zimyanin et al . , 2008; Parton et al . , 2011 ) . We found that bcd mRNA particles show a reciprocal anterior bias near the middle of the oocyte , with 12% more movements towards the anterior cortex than away from it . This supports the idea that bcd mRNA is mainly transported towards microtubule minus ends by Dynein . The bias becomes increasingly weak close to the anterior , however , and even reverses in the region 0-5 μm from the anterior cortex . Since the small bias in the anterior region is unaffected in the Khc null mutant and there are virtually no directional reversals , it is not due to bidirectional transport along a polarised microtubule network . Instead , there seem to be approximately equal numbers of microtubules pointing anteriorly and posteriorly near the anterior , with most microtubules running parallel to the anterior cortex . This fits well with experimental measurements of microtubule polarity in the oocyte , in which the orientation bias decreases from posterior to anterior , and with 3-dimensional computer simulations of the microtubule organisation , which give negligible orientation bias close to the anterior ( Parton et al . , 2011; Trong et al . , 2015 ) . Thus , directed transport cannot account for the localisation of bcd mRNA to the very anterior of the oocyte , although it can deliver the mRNA to a broader anterior region . In light of these observations , we propose that bcd mRNA is localised by rapid , bidirectional Dynein-dependent transport in and out of the anterior region , coupled to some mechanism that specifically retains or anchors the RNA at the anterior cortex ( Figure 9A ) . This random transport and anterior anchoring model predicts that the RNA will only turn over slowly at the anterior cortex . This is confirmed by FRAP and photo-conversion experiments , which show that more than 60% of the RNA remains stably localised at the anterior over a period of 55 min . This turn-over rate is the same as that measured for grk mRNA , which has previously been shown to be specifically anchored at its localisation site above the oocyte nucleus ( Delanoue et al . , 2007; Jaramillo et al . , 2008 ) . Further support for non-directional transport and anterior anchoring comes from the observation that bcd mRNA is more efficiently retained at the anterior when the microtubules are depolymerised , indicating that microtubule-based transport plays a role in removing the RNA from the anterior , as well as delivering it . 10 . 7554/eLife . 17537 . 037Figure 9 . Diagram of the steps in bcd mRNA localisation during oogenesis . ( A ) Stage 9 of oogenesis: bcd mRNA localises to the anterior-lateral margins of the oocyte ( i ) , forming a ring when viewed end on ( ii ) . Close-up: bcd mRNA is assembled into Exu-dependent particles that are actively transported by Dynein along an unpolarised microtubule cytoskeleton; On reaching the anterior , bcd RNA particles are anchored independently of microtubules , possibly by docking to P-bodies . ( B ) Stage 10b of oogenesis: Following the reorganisation of microtubule minus-ends , bcd mRNA re-localises from the anterior-lateral margin to form a disc at the centre of the anterior cortex of the oocyte . ( C ) Stage 14 of oogenesis: bcd mRNA particles cluster into large aggregates at the oocyte cortex . DOI: http://dx . doi . org/10 . 7554/eLife . 17537 . 037 Unlike bcd mRNA , the behaviour of hts mRNA at stage 9 fits well with the predictions of the continual active transport model . It is localised to a broader anterior region than bcd mRNA , turns-over significantly more rapidly than bcd RNA in photo-conversion experiments and spreads along the anterior margin after photoconversion . Furthermore , hts mRNA localisation is strongly reduced after 90 min of Colcemid treatment and in the shot2A2 mutant , in which the oocyte microtubule cytoskeleton is not polarised , whereas bcd mRNA localisation is largely unaffected in both conditions . Thus , the contrast between the two RNAs reinforces the view that bcd mRNA cannot be explained solely by directed transport and must involve an anterior anchoring step . This model may help to explain the observations of Cha et al . ( 2001 ) , who showed that bcd mRNA injected into the oocyte localises to the nearest region of anterior/lateral cortex , whereas RNA that is exposed to nurse cell cytoplasm before injection localises specifically to the anterior . Although the authors proposed that the “nurse cell-conditioned” RNA gains the capacity to discriminate between microtubules emanating from the anterior and lateral cortex , a simpler explanation is that both untreated and conditioned RNA are transported by Dynein along microtubules , but only the latter becomes competent to be retained at the anterior . The untreated RNA therefore concentrates near microtubule minus ends , much like hts RNA ( which is more biased towards the anterior than injected RNA because it enters from the nurse cells ) , whereas the conditioned RNA localises specifically to the anterior . Thus , factors such as Exu loaded on the RNA in the nurse cells may licence the RNA for anterior anchoring ( Figure 9A ) . The retention of bcd mRNA at the anterior varies over the course of oogenesis , with the RNA being much less stably localised at stage 10b . This fits well with the observations of Weil et al . , 2006 , who measured very similar fluorescence recovery rates at stage 10b to those reported here . This decrease in anterior retention coincides with a redistribution of the RNA from an anterior ring to a disc at the centre of the anterior cortex , and with the formation of a new MTOC in this region ( St Johnston et al . , 1989; Schnorrer et al . , 2002; Vogt et al . , 2006 ) . Thus , the anterior anchoring mechanism seems to be specifically inactivated at this stage to allow the remodelling of the RNA distribution ( Figure 9B ) . During this period , bcd mRNA localisation is consistent with continual active transport along the polarised microtubule network formed by the new anterior MTOC . This is only transient , however , as the RNA becomes more stable at the anterior at stage 13 , and is very efficiently anchored at stage 14 , which is important to keep bcd mRNA localised until fertilisation , so that it can act as the source of the Bcd morphogen gradient in the embryo ( Figure 9C ) . The mechanism that retains bcd RNA at the anterior is unclear . We can rule out anchoring by Dynein to microtubule minus ends , as has been reported for grk mRNA in the oocyte and pair-rule transcripts in the embryo ( Delanoue and Davis , 2005; Delanoue et al . , 2007 ) , since the anterior retention of bcd mRNA is not microtubule-dependent and the RNA does not co-localise with microtubules . The mRNA could be tethered to cortical actin , which would be consistent with the anchoring defect in late oocytes seen in swallow mutants , which disrupt the actin cortex ( Weil et al . , 2010 ) . However , bcd RNA does not show a significant co-localisation with F-actin , although this tethering could be indirect . Another possibility is that the RNA is maintained at the anterior by sequestering it in P-bodies ( Weil et al . , 2012 ) ( Figure 9 ) . P-bodies are ubiquitous throughout the oocyte , and there would therefore have to be some mechanism that induces bcd RNA incorporation into these structures specifically at the anterior . Super-resolution imaging revealed that bcd mRNA forms 110–120 nm particles throughout oogenesis , regardless of whether the RNA is localised or not . Even the large aggregates of RNA at stage 14 are still formed of individual particles of similar size , although their protein composition is different from stage 9 , as Staufen and ESCRT-II are recruited to bcd mRNA only at stage 10b ( Martin et al . , 2003; Weil et al . , 2006 ) ( Figure 9 ) . Importantly , over-expression of the mRNA does not alter particle size , but instead leads to more particles , which have higher average RNA content . The same occurs at stage 14 of oogenesis , when the bcd mRNA content of the oocyte is much higher following nurse cell dumping . Thus , bcd mRNA seems to assemble into a structure of defined size , almost like a virus particle , which can incorporate more or less mRNA molecules depending on availability . An exu mutant that affects RNA binding affinity causes a large reduction in the number of detectable bcd mRNA particles and a small , but significant reduction in the size of the few particles that form . This suggests that Exu , which forms homodimers that probably bind two bcd mRNA molecules ( Lazzaretti et al . , 2016 ) , plays a role in scaffolding the assembly of the particles ( Figure 9 ) . Loss of Exu also strongly reduces both the speed and frequency of bcd mRNP movement , as well as its anterior anchoring at all stages of oogenesis . Particle formation may therefore be a prerequisite for all of these processes , explaining the pleiotropic effects of exu mutants . The invariant size of bcd RNA particles make them fundamentally different from other well-characterised RNA granules , such as the P-granules in C . elegans , which form by the aggregation of RNA and proteins into droplets that phase-separate from the surrounding cytoplasm ( Brangwynne et al . , 2009; Saha et al . , 2016 ) . P-granules have variable size that depends on the RNA concentration and readily fuse with each other when juxtaposed . By contrast , bcd RNA particles stay the same size as the RNA concentration increases , even though they incorporate more RNA , and they do not appear to fuse when tightly clustered in aggregates . The exact nature of the particles will require the identification of more of their components , but their behaviour is compatible with a model in which they consist of a rigid protein framework that contains multiple RNA-binding sites . In future , it will be interesting to determine whether other localised RNAs are packaged into similar structures . The bcdMS2 transgene was generated by inserting 11 MS2-binding sites ( C-loop form ) ( Lowary and Uhlenbeck , 1987 ) into the SpeI restriction site at the 5’-terminus of bcd 3’UTR ( FlyBase ID: FBgn0000166 ) , which was cloned downstream of the maternal α4 tubulin promoter . The htsMS2 transgene was generated by cloning the cDNA of the N4 isoform of hts ( Flybase ID: FBgn0263391; Whittaker et al . , 1999 ) , excluding the 5’UTR and start codon , downstream of the maternal α4 tubulin promoter and inserting 10 MS2-binding sites between the SpeI and NotI restriction sites at the 5’-terminus of the 3’UTR . The hsp83-MCP-Dendra2 transgene is identical to the hsp83-MCP-GFP transgene ( Forrest and Gavis , 2003 ) except that the EGFP sequences are replaced by Dendra2 sequences ( Evrogen , Russia ) . The osk-NLS-MCP-Tomato transgene was generated by inserting the cDNA of tdTomato ( Shaner et al . , 2004 ) after NLS-MCP , which was cloned from hsp83-NLS-MCP-GFP ( Forrest and Gavis , 2003 ) . For germline-specific expression , the NLS-MCP-Tomato fusion was cloned downstream of the osk promoter , cloned from an osk rescue construct ( gift from Anne Ephrussi ) . The mRNA-MS2 fusion transgenes were recombined with hsp83-NLS-MCP-GFP , osk-NLS-MCP-Tomato or hsp83-NLS-MCP-Dendra2 . Germline clones ( GLC ) were generated using the ovoD/FLP system by heat-shocking second to third instar larvae for 2 hr at 37°C for 3 consecutive days ( Chou and Perrimon 1996 ) . Other fly strains used were: y1w1 ( Bloomington stock 1495 ) ; osk- ( MS2 ) 10 ( Zimyanin et al . , 2008 ) ; shot2A2 ( Chang et al . , 2011 ) ; cn , exu1 , bw ( [Schüpbach and Wieschaus , 1986] , Bloomington stock 1989 ) ; cn , exuVL , bw ( Hazelrigg et al . , 1990 ) ; FRT42B , c , Khc27 ( Brendza et al . , 2000 ) ; FRT42B , c , Khc17 ( Brendza et al . , 2000 ) ; FRT42B , c , Khc23 ( Brendza et al . , 2000 ) ; Dhc64C6–10 ( [McGrail and Hays , 1997] , Bloomington stock 8747 ) ; Dhc64C8–1 ( Gepner et al . , 1996 ) ; Dhc64C6 . 10 , FRT2A ( Gift from U . Abdu ) ; Tau-GFP ( Micklem et al . , 1997 ) ; grk- ( MS2 ) 12 , MCP-GFP ( grkGFP ) ( Jaramillo et al . , 2008 ) ; Ubq-Dlic-GFP ( Baumbach et al . , 2015 ) ; grk2B6 , grk2E12 ( Neuman-Silberberg and Schüpbach , 1993 ) ; UAS:mCherry-Patr ( Nashchekin et al . , 2016 ) ; nanos:GAL4-VP16 ( [Van Doren et al . , 1998] , Bloomington stock 64277 ) ; Tral-mRFP trap line ( Lowe et al . , 2014 ) ; Me31B-GFP trap-line ( Buszczak et al . , 2007 ) ; FRT2A , Ge-1Δ5 ( Fan et al . , 2011 ) . Microtubules were depolymerised using Colcemid ( Sigma-Aldrich , MO , USA ) . For acute depolymerisation of microtubules , ovaries from 48–72 hr old females were dissected in live imaging medium ( 5 μg/ml insulin and 2 . 5% foetal calf serum in Schneider’s medium ( Sigma-Aldrich , MO , USA; adapted from Bianco et al . , 2007 ) , in a Poly-L-Lysine-coated imaging chamber ( Thistle Scientific , UK ) . Colcemid was then added to 400 μg/ml . The colocalisation between bcd mRNA and microtubule minus ends was examined in flies expressing bcd*Tomato and Tau-GFP that were starved for 2 hr and then fed fresh yeast paste containing 100 μg/ml Colcemid for 2 hr ( Pokrywka and Stephenson , 1995 ) . We depolymerised F-actin by dissecting ovaries in live imaging medium and then adding Cytochalasin D to 10 μg/ml ( Sigma-Aldrich , MO , USA; Emmons et al . , 1995 ) . Confocal imaging was performed on an Olympus IX81 FV1000 laser scanning confocal microscope ( Olympus , Japan ) using 40x UPlanFLN 1 . 3NA or 60x UPlanSApo 1 . 35NA oil immersion objectives ( Olympus , Japan ) and the Olympus Fluoview FV10-ASW software ( Olympus , Japan , RRID:SCR_014215 ) . Ovaries from 48–72 hr old females were dissected directly onto coverslips in 10S Voltalef oil ( VWR International , PA , USA ) . For acute drug treatments , drugs were added for 20 min to ovaries in live imaging medium ( see above ) in Poly-L-Lysine-coated imaging chambers ( Thistle Scientific , UK ) . Ovaries were transferred onto coverslips and finely dissected in 10S Voltalef oil . Imaging was performed on either a wide field DeltaVision microscope ( Applied Precision , WA , USA ) equipped with a Photometrics 512 EMCCD camera ( Photometrics , AZ , USA ) and a 2x magnification tube fitted between the unit and the camera , or on an Olympus IX81 inverted microscope ( Olympus , Japan ) combined with a Yokogawa CSU22 spinning disk confocal imaging system and an iXon DV855 camera ( ANDOR Technology , UK ) . The softWorXs software ( Applied Precision , WA , USA ) was used to acquire and deconvolve images on the DeltaVision system and MetaMorph Microscopy Automation and Image Analysis Software ( Molecular Devices , CA , USA , RRID:SCR_002368 ) was used to acquire images on the spinning-disk microscope . A 100x UPlanSApo 1 . 4 NA oil immersion objective lens ( Olympus , Japan ) was used in both systems . Ovaries from 48–72 hr old females were dissected directly onto coverslips in 10S Voltalef oil ( VWR International , PA , USA ) , except when treated with drugs , in which case the dissections were performed in live imaging medium ( see above ) in a Poly-L-Lysine-coated imaging chamber ( Thistle Scientific , UK ) ; drugs were added to the medium 20 min before imaging . Fluorescence recovery after photobleaching ( FRAP ) and photo-conversion experiments were performed on an Olympus IX81 FV1000 laser scanning confocal microscope , ( Olympus , Japan ) equipped with the Olympus Fluoview FV10-ASW software ( Olympus , Japan , RRID:SCR_014215 ) and either a 60x UPlanSApo 1 . 35 NA oil immersion objective ( Olympus , Japan; for dissections in oil ) or a 60x UPlanSApo 1 . 2 NA water immersion objective ( Olympus , Japan; for dissections in live imaging medium ) . All imaging conditions ( laser power , bleached or photo-converted area , image dimensions , pixel scanning time and time points ) were kept constant in all samples . At least 5 oocytes were analysed per sample type . Super-resolution imaging was performed on a custom STED microscope built around the IX83 Olympus frame ( Olympus , Japan ) . The microscope design is a variant of a STED system described in detail previously ( Bottanelli et al . , 2016 ) . Imaging was performed with either a 100x UPlanSApo 1 . 4 NA oil immersion objective lens ( Olympus , Japan ) or a 100x UPlanSApo 1 . 35 NA silicone oil immersion objective lens ( Olympus , Japan ) over a region of 15 × 15 μm with square pixels of 14 . 6 nm ( 1024 × 1024 pixels ) . Total RNA was extracted from the ovaries of twenty 48–72 hr old females using the RNeasy kit ( Qiagen , Germany ) . 100 ng of total RNA was then reverse-transcribed using the qPCRBIO cDNA Synthesis Kit ( PCR Biosystems , UK ) , using a combination of poly-dT and random hexanucleotide primers . Real-time PCR was performed on the reverse transcribed samples to independently amplify two regions in the bcd mRNA 3’UTR as well as one region in the internal control , DHFR RNA . The primer pairs used were: bcd 3’UTR 1: 5’-GATGTATCTGGGTGGCTGCT-3’ & 5’-CCGAAATGTGGGACGATAAC-3’ bcd 3’UTR 3: 5’-CACTAAAGCCCGGGAATATG-3’ & 5’-TTTCTTGCTGGCTCGGAATA-3’ DHFR: 5-CTGAGCACCACACTTCAGGA-3’ & 5-TGGTAATGTACAGCCGGTGA-3’ Amplifications were performed using the qPCRBIO SyGreen Mix Hi-ROX Kit ( PCR Biosystems , UK ) and the StepOne Plus Real-Time PCR system ( Applied Biosystems , CA , USA ) . The relative amounts ( fold change ) of bcd RNA in samples were then quantified by the comparative CT method ( Schmittgen and Livak , 2008 ) , using the threshold cycles ( CT ) calculated by the inbuilt StepOne Real-Time PCR software ( Applied Biosystems , CA , USA , StepOne Software , RRID:SCR_014281 ) : ( 8 ) Foldchange=2−ΔΔCT=2[ ( CTgeneofinterest−CTinternalcontrol ) sampleA− ( CTgeneofinterest−CTinternalcontrol ) sampleB] Quantitations represent three biological replicates and two technical replicates , and were performed on the softwares Excel ( Microsoft , CA , USA ) and R ( R Project for Statistical Computing , RRID:SCR_001905 ) ( R Core Team , 2013 ) .	Molecules of messenger RNA , or mRNA for short , contain the instructions needed to make proteins . Many mRNAs are only found in certain parts of the cell to ensure that the corresponding proteins are only produced where they are actually needed . The mRNAs are delivered to their final location in the cell by motor proteins that move along tracks made of filaments called microtubules . In female fruit flies , a mRNA called bicoid is transported to front end of a developing egg cell , while another mRNA called oskar is moved to the rear end . When the egg is fertilized , the region that contains bicoid mRNA develops into the head of the embryo , while the other end gives rise to the abdomen . A motor protein called kinesin-1 transports the oskar mRNA to the rear end , but how bicoid mRNA moves to the front end is not clear . Trovisco et al . used microscopy to study how bicoid mRNA moves . The experiments show that another motor protein called Dynein moves bicoid mRNA along microtubules . However , unlike oskar mRNA , bicoid mRNA moves along microtubules in all directions and is not biased towards the front end of the cell . Trovisco et al . hypothesized that when bicoid mRNA reaches the front end of the egg it is trapped there by other factors . Further experiments found that bicoid mRNA is indeed anchored at the front end of the cell . The mRNA does not seem to be trapped at the ends of the microtubules along which it is transported , nor does it form large clumps . Instead , it forms small , well-defined particles that remain the same size as the egg develops . The findings of Trovisco et al . raise the possibility that bicoid mRNA is packaged into these particles in order to be transported and anchored at the front end of the egg cell . Future work is needed to understand how particles containing bicoid mRNA are tethered at the front end of the egg cell and whether other mRNAs are also packaged in a similar manner .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2016	bicoid mRNA localises to the Drosophila oocyte anterior by random Dynein-mediated transport and anchoring
RIG-I is a key cytosolic pattern recognition receptor that interacts with MAVS to induce type I interferons ( IFNs ) against RNA virus infection . In this study , we found that cyclophilin A ( CypA ) , a peptidyl-prolyl cis/trans isomerase , functioned as a critical positive regulator of RIG-I-mediated antiviral immune responses . Deficiency of CypA impaired RIG-I-mediated type I IFN production and promoted viral replication in human cells and mice . Upon Sendai virus infection , CypA increased the interaction between RIG-I and its E3 ubiquitin ligase TRIM25 , leading to enhanced TRIM25-mediated K63-linked ubiquitination of RIG-I that facilitated recruitment of RIG-I to MAVS . In addition , CypA and TRIM25 competitively interacted with MAVS , thereby inhibiting TRIM25-induced K48-linked ubiquitination of MAVS . Taken together , our findings reveal an essential role of CypA in boosting RIG-I-mediated antiviral immune responses by controlling the ubiquitination of RIG-I and MAVS . The innate immune system is the first line of defense against microbial pathogen invasion via the recognition of pathogen-associated molecular patterns ( PAMPs ) with the help of pattern recognition receptors ( PRRs ) ( Barbalat et al . , 2011; Kawai and Akira , 2010 ) . Among these PRRs , RIG-I like receptors ( RLRs ) function as cytoplasmic RNA sensors that recognize viral RNA and activate a signaling pathway , which is essential for the production of type I interferons ( IFNs ) ( Kato et al . , 2011 ) . RIG-I is required for type I IFN production in response to Sendai virus ( SeV ) , Newcastle disease virus ( NDV ) , influenza A virus ( IAV ) , vesicular stomatitis virus ( VSV ) , and Japanese encephalitis virus ( JEV ) ( Goubau et al . , 2013; Kato et al . , 2006; Loo et al . , 2008 ) . Following ligand binding , ubiquitinated RIG-I is recruited to the mitochondria-associated membrane where it binds to MAVS ( also known as IPS-1 , Cardif , and VISA ) to initiate innate immune signaling ( Hou et al . , 2011; Kawai et al . , 2005; Meylan et al . , 2005; Seth et al . , 2005; Xu et al . , 2005 ) . Cyclophilin A ( CypA , encoded by PPIA ) is a peptidyl-prolyl cis/trans isomerase ( PPIase ) that is expressed ubiquitously in all type of cells . It is the major cellular target for the immunosuppressive drug cyclosporin A ( CsA ) and is involved in protein folding , cell signaling , inflammation , and tumorigenesis ( Handschumacher et al . , 1984; Lu et al . , 2007 ) . Moreover , CypA functions as an important host factor that regulates the replication of a number of viruses , including human immunodeficiency virus type I ( HIV-1 ) , hepatitis C virus ( HCV ) , human papillomavirus ( HPV ) , IAV , rotavirus ( RV ) , enterovirus-71 ( EV71 ) virus , and infectious bursal disease virus ( IBDV ) , which expands the role of CypA in virus infection ( Bienkowska-Haba et al . , 2009; Chatterji et al . , 2009; Towers et al . , 2003; He et al . , 2012; Liu et al . , 2012b , 2009; Qing et al . , 2014; Wang et al . , 2015; Xu et al . , 2010 ) . It has been established that CypA interacts directly with viral protein to regulate virus replication . For example , our previous studies showed that CypA-overexpressing transgenic mice exhibited resistance to influenza A virus infection ( Li et al . , 2016 ) . We further found that CypA interacted with influenza A virus M1 protein and inhibited virus replication by accelerating ubiquitin-proteasome degradation of the M1 protein ( Liu et al . , 2012b , 2009; Xu et al . , 2010 ) . Yet several lines of evidence indicate that CypA can also regulate virus replication through modulating host immune responses . For instance , CypA interacted with the newly synthesized HIV-1 CA domain and subsequently activated the transcription factor IRF3 to promote the production of type I IFNs in dendritic cells ( Manel et al . , 2010 ) . CypA inhibited RV replication by facilitating IFN-β production ( He et al . , 2012 ) . However , the molecular mechanism of how CypA regulates virus-mediated type I IFN production is poorly understood . The present study indicates that CypA promotes RIG-I mediated type I IFN production and inhibits viral replication both in vitro and in vivo . We further demonstrate that CypA facilitates IFN responses through promoting K63-linked ubiquitination of RIG-I and inhibiting K48-linked ubiquitination of MAVS . Therefore , our studies identify a previously unknown mechanism that CypA promotes RIG-I-mediated type I IFN production to suppress virus replication , which adds up a new facet of CypA in host antiviral immunity . Our previous studies have demonstrated that CypA inhibits IAV replication both in vitro and in vivo ( Li et al . , 2016; Liu et al . , 2012b , 2009 ) . To further investigate the impact of CypA on the replication of other RIG-I-recognized RNA viruses , such as SeV and VSV , virus growth was monitored in shRNA-based CypA-knockdown 293T cells ( 293T/CypA- ) and wild-type ( WT ) 293T cells ( 293T/CypA+ ) . The hemagglutination ( HA ) titer of SeV and median tissue culture infective dose ( TCID50 ) of VSV in 293T/CypA- cells was strikingly increased compared with that in 293T/CypA+ cells ( Figure 1A , B ) , indicating that CypA plays an inhibitory role in the replication of SeV and VSV . To confirm the role of CypA in antiviral responses in a CypA deficient system , we purchased CypA-deficient ( Ppia−/− ) 129 mice from Jackson Laboratory and crossed them to WT 129 mice . WT , Ppia+/− and Ppia−/− mice were identified by PCR ( Figure 1—figure supplement 1A ) . The absence of Ppia in CypA-deficient bone marrow-derived macrophages ( BMDMs ) was examined by semi-quantitative PCR and Western blotting ( Figure 1—figure supplement 1B and C ) . We further determined the effect of CypA on SeV replication in BMDMs from WT and Ppia−/− mice . Consistent with the results in 293T cells , we found that the mRNA expression level of SeV M gene was higher in Ppia−/− BMDMs than that in WT BMDMs ( Figure 1C ) . Collectively , these data suggested that CypA inhibited the replication of RIG-I-recognized RNA viruses . 10 . 7554/eLife . 24425 . 003Figure 1 . CypA promotes production of type I IFNs against virus infection . ( A ) HA assays of SeV in 293T/CypA+ or 293T/CypA- cells infected with SeV ( MOI = 1 ) for the indicated time periods . ( B ) TCID50 assays of proliferation level of VSV in 293T/CypA+ or 293T/CypA- cells infected with VSV ( MOI = 1 ) for the indicated time periods . ( C ) Quantitative PCR analysis of SeV M mRNA in wild-type ( WT ) or CypA-deficient ( Ppia−/− ) BMDMs infected with SeV for the indicated time periods . ( D ) Luciferase activity of lysates in 293T/CypA+ or 293T/CypA- cells transfected for 24 hr with IFN-β luciferase reporter ( IFN-β-Luc ) , together with Poly ( I:C ) ( TpIC ) or then treated with SeV , VSV , IAV-mut for 6 hr . ( E ) Quantitative PCR analysis of Ifnb1 and Ifna mRNA in WT or Ppia−/− BMDMs infected with SeV , VSV or IAV-mut for 6 hr . ( F ) ELISA of IFN-β and IFN-α production in the supernatants of WT or Ppia−/− BMDMs treated with SeV for 12 hr . ( G ) Quantitative PCR analysis of Ifit1 , Ifit2 , and Ccl5 mRNA in WT or Ppia−/− BMDMs treated with SeV for 6 hr . ( H and I ) Quantitative PCR analysis of IFNB1 , IFNA ( H ) IFIT1 , IFIT2 , or CCL5 ( I ) mRNA in human monocytes transfected with CypA siRNA or scrambled siRNA for 48 hr and then treated with SeV for 6 hr . Data are shown as mean ± SD ( A: n = 5; B-I: n = 3 ) . p<0 . 05 , p<0 . 01 , *p<0 . 001 ( unpaired , two-tailed Student’s t-test ) . Data are from one representative of at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 24425 . 00310 . 7554/eLife . 24425 . 004Figure 1—source data 1 . Quantification of viral replication and type I IFN production for Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 24425 . 00410 . 7554/eLife . 24425 . 005Figure 1—figure supplement 1 . Identification of Ppia-deficient mice . ( A ) PCR analysis of genomic DNA from tail tissue of Ppia-deficient mice . Water was used as a negative control for the PCR reaction . ( B ) Semi-quantitative analysis of Ppia mRNA in BMDMs from WT and Ppia−/− mice . Primers were designed to amplify Ppia or the housekeeping gene GAPDH ( PCR primers: Ppia forward , 5′- ATGGTCAACCCCACCGTGTTC-3′; Ppia reverse , 5′- TTAGAGCTGTCCACAGTCGG-3′; GAPDH forward , 5’-ACCACAGTCCATGCCATCAC-3’; and GAPDH reverse , 5’- TCCACCACCCTGTTGCTGTA -3’ ) . ( C ) Immunoblot analysis of lysates in WT and Ppia−/− BMDMs . DOI: http://dx . doi . org/10 . 7554/eLife . 24425 . 00510 . 7554/eLife . 24425 . 006Figure 1—figure supplement 2 . CypA promotes production of type I IFNs against virus infection in 293T , U937 cells and human monocytes . ( A and B ) Quantitative PCR analysis of IFNB1 ( A ) , IFIT1 , IFIT2 , or CCL5 ( B ) mRNA in 293T/CypA+ and 293T/CypA- cells infected with SeV ( MOI = 1 ) for 6 hr . ( C ) Immunoblot analysis of lysates in U937 or human primary monocytes transfected with CypA siRNA or scrambled siRNA for 48 hr . ( D–G ) Quantitative PCR analysis of IFNB1 , IFNA ( D and F ) IFIT1 , IFIT2 , or CCL5 ( E and G ) mRNA in U937 cells transfected with CypA siRNA or scrambled siRNA for 48 hr and then treated with SeV ( D and E ) or VSV ( F and G ) for 6 hr . ( H and I ) Quantitative PCR analysis of IFNB1 , IFNA ( H ) IFIT1 , IFIT2 , or CCL5 ( I ) mRNA in human monocyte cells transfected with CypA siRNA or scrambled siRNA for 48 hr and then treated with VSV for 6 hr . Data are shown as mean ± SD ( n = 3 ) . p<0 . 05 and *p<0 . 01 ( unpaired , two-tailed Student’s t-test ) . Data are representative of at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 24425 . 00610 . 7554/eLife . 24425 . 007Figure 1—figure supplement 3 . Effect of CypA on RIG-I-independent signaling . ( A and B ) Quantitative PCR analysis of IFNB1 , IFIT1 , IFIT2 , CCL5 ( A ) , or EMCV VP1 mRNA ( B ) in 293T/CypA+ and 293T/CypA- cells infected with EMCV ( MOI = 0 . 5 ) for 6 hr . ( C ) Quantitative PCR analysis of Ifnb1 and Ifna mRNA in WT or Ppia−/− BMDMs cells infected with HSV-1 ( MOI = 10 ) for 6 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 24425 . 00710 . 7554/eLife . 24425 . 008Figure 1—figure supplement 4 . CypA is inducible against virus infection . ( A ) Quantitative PCR analysis of Ppia mRNA ( top ) and immunoblot analysis of CypA ( below ) in WT BMDMs treated with SeV , VSV , IAV-mut or IFN-β for the indicated time points . ( B ) Quantitative PCR analysis of PPIA mRNA ( top ) and immunoblot analysis of CypA ( below ) in 293T/CypA+ cells treated with SeV or transfected with Poly ( I:C ) for the indicated time points . ( C and D ) Quantitative PCR analysis of PPIA in U937 cells ( C ) or human monocytes ( D ) treated with SeV or VSV for 12 hr . Data are shown as mean ± SD ( n = 3 ) . p<0 . 05 and *p<0 . 01 ( unpaired , two-tailed Student’s t-test ) . Data are representative of at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 24425 . 008 Considering RIG-I-recognized RNA viruses can trigger the RIG-I-mediated signaling pathway and promote the production of type I IFN , which in turn inhibits virus replication , we examined the effect of CypA on production of type I IFNs and interferon-stimulated genes ( ISGs , such as Ifit1 , Ifit2 , and Ccl5 ) . We performed IFN-β promoter-driven luciferase assay in 293T cells ( Figure 1D ) , and quantitative PCR ( Figure 1E , G ) and ELISA assays ( Figure 1F ) in BMDMs in response to transfected Poly ( I:C ) or infection of SeV , VSV or IAV with the NS1R38A/K41A mutant ( IAV-mut , which induces high levels of IFN-β , [Donelan et al . , 2003] ) , respectively . Absence of CypA remarkably decreased RIG-I-mediated production of type I IFNs and ISGs . We also obtained similar results with other cell types , such as human primary monocytes ( Figure 1H , I , Figure 1—figure supplement 2C , H , I ) , 293 T cells ( Figure 1—figure supplement 2A , B ) and U937 cells ( Figure 1—figure supplement 2C–G ) triggered by SeV or VSV . Collectively , these findings suggested that CypA positively regulated expression of type I IFNs and ISGs against RIG-I-recognized RNA virus infection . We further examined the effect of CypA on RIG-I-independent signaling , such as encephalomyocarditis virus ( EMCV ) -triggered MDA5 pathway and herpes simplex virus type 1 ( HSV-1 ) -triggered cGAS-STING pathway . CypA promoted IFN-β and ISGs production and inhibited the replication of EMCV ( Figure 1—figure supplement 3A , B ) , but had no impact on HSV-1-triggered cGAS-STING pathway ( Figure 1—figure supplement 3C ) . Altered host cell gene expression is a universal consequence of virus infection . When we investigated the mRNA and protein levels of CypA in BMDMs , 293T cells , U937 cells and human monocytes infected with SeV , VSV or IAV-mut , we found that CypA was highly inducible , whereas Poly ( I:C ) transfection or IFN-β treatment made no difference to CypA expression , ( Figure 1—figure supplement 4A–D ) , suggesting that the induction of CypA can be triggered by viruses and CypA is involved in cellular antiviral response . Taken together , upon virus infection , CypA expression was upregulated , which inhibited the replication RIG-I-recognized RNA virus by enhancing production of type I IFNs . In the following studies , we used SeV as a naturally occurring agent that strongly triggers antiviral immunity via RIG-I-mediated signaling pathway . Having known that CypA promotes type I IFN production and inhibits the replication of RIG-I-recognized RNA virus in vitro , we next sought to determine CypA-mediated antiviral responses in vivo . In a mouse model of SeV infection , all five monitored SeV-infected Ppia−/− mice died at 9 d after infection , whereas three of five ( 60% ) SeV-infected WT mice survived and remained healthy for the duration of the infection study ( Figure 2A ) , indicating deficiency of CypA accelerated SeV infection-induced death of mice . Anatomical analysis showed that the lung indices of SeV-infected mice were increased at 7 d after infection . Notably , SeV-infected Ppia−/− mice exhibited much higher lung indices than that of WT mice ( Figure 2B ) . Consistently , gross lesion of lung in Ppia−/− mice infected with SeV was severer than that in WT mice ( Figure 2C ) . We further performed histopathological examination of lung , nasal turbinate , and trachea tissues at day 2 , 5 and 7 post infection and found that SeV-infected Ppia−/− mice displayed severe bronchopneumonia , interstitial pneumonia , congestion in blood vessels , and dropout of the mucous epithelium , whereas SeV-infected WT mice only displayed slight bronchopneumonia and congestion in blood vessels during the infection ( Figure 2D , Figure 2—figure supplement 1A , B ) . These data showed that tissue damage was aggravated in CypA deficient mice , correlating with a higher viral load in the lungs , as measured by expression of NP and M genes of SeV ( Figure 2E ) . Most importantly , we found deficiency of CypA reduced expression of type I IFNs and downstream ISGs in lungs ( Figure 2F–H ) and spleens ( Figure 2—figure supplement 1C , D ) , suggesting that CypA also promoted type I IFN production in vivo . Collectively , CypA inhibited SeV replication in vivo through augmenting of expression of IFNs and downstream ISGs . 10 . 7554/eLife . 24425 . 009Figure 2 . CypA positively regulates type I IFN production and antiviral responses in vivo . ( A ) Survival of WT and Ppia−/− mice ( n = 5 ) infected with SeV ( 2000 PFU/mouse ) via nasal inoculation and monitored for 14 days . ( B ) Lung index ( 100× lung/body weight ) of WT and Ppia−/− mice ( n = 5 ) infected with SeV for 2 and 7 days . ( C ) Gross lesion of lungs from WT and Ppia−/− mice inoculated with SeV for 7 days . ( D ) H & E stainings of lungs of WT and Ppia−/− mice ( n = 3 ) infected with SeV or mock-infected with PBS for 2 , 5 , and 7 days . Scale bars , 100 μm . ( E ) Quantitative PCR analysis of SeV NP or M mRNA in WT or Ppia−/− mice treated with SeV for the indicated time points . ( F ) ELISA of IFN-β and IFN-α production in lung tissues of WT or Ppia−/− mice treated with SeV for the indicated time points . ( G and H ) Quantitative PCR analysis of Ifnb1 , Ifna ( G ) , Ifit1 , Ifit2 , and Ccl5 ( H ) mRNA in lung tissues of WT or Ppia−/− mice treated with SeV for the indicated time points . Data are shown as mean ± SD ( B: n = 5; E-H: n = 3 ) . p<0 . 05 and *p<0 . 01 ( unpaired , two-tailed Student’s t-test ) . Data are representative of two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 24425 . 00910 . 7554/eLife . 24425 . 010Figure 2—source data 1 . Quantification of survival , lung index , viral replication and type I IFN production for Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 24425 . 01010 . 7554/eLife . 24425 . 011Figure 2—figure supplement 1 . CypA positively regulates type I IFN production and the antiviral responses in vivo . ( A and B ) H&E stainings of nasal turbinate ( A ) and trachea ( B ) tissues of WT and Ppia−/− mice ( n = 3 ) infected with SeV or mock-infected with PBS for 2 , 5 , and 7 days . Scale bars , 100 μm . ( C ) ELISA of IFN-β and IFN-α production in spleen tissues of WT or Ppia−/− mice treated with SeV for the indicated time points . ( D ) Quantitative PCR analysis of Ifnb1 , Ifna , Ifit1 , Ifit2 , and Ccl5 mRNA in spleen tissues of WT or Ppia−/− mice treated with SeV for the indicated time points . Data are shown as mean ± SD ( n = 3 ) . p<0 . 05 and *p<0 . 01 ( unpaired , two-tailed Student’s t-test ) . Data are representative of two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 24425 . 011 It is well established that expression of type I IFN genes is mainly regulated by two transcription factors IRF3 and NF-κB . To investigate the effect of CypA on SeV-induced activation of IRF3 and NF-κB , interferon stimulated response element ( ISRE ) and NF-κB luciferase reporter constructs were co-transfected with CypA or control vector in 293T/CypA- cells . We found that ISRE- and NF-κB-responsive luciferase activity induced by SeV infection were dramatically lower in the absence of CypA ( Figure 3A , B ) , indicating that CypA was involved in both IRF3- and NF-κB-mediated type I IFN expression . In line with this observation , the activated dimer form of IRF3 ( Figure 3C ) and phosphorylation of IRF3 and p65 ( Figure 3D ) were distinctly suppressed in SeV-infected 293T/CypA- cells compared with those in SeV-infected 293T/CypA+ cells . Moreover , CypA deficiency also inhibited phosphorylation of IRF3 , p65 , IKKα/β , and IκBα in cultured BMDMs , accompanied with lower expression levels of RIG-I and MAVS ( Figure 3E ) . These data indicate that CypA is vital for activation of IRF3 and NF-κB signaling pathways . 10 . 7554/eLife . 24425 . 012Figure 3 . CypA deficiency suppresses IRF3 and NF-κB activation . ( A and B ) Luciferase activity of lysates in 293T/CypA- cells transfected for 24 hr with CypA or control vector , together with either ISRE-Luc ( A ) or NF-κB-Luc ( B ) and then treated with SeV for 6 hr . Results are presented relative to the luciferase activity in control cells treated with luciferase reporter and empty vector . ( C ) Native PAGE and immunoblot analysis of IRF3 in dimer or monomer form and phosphorylated IRF3 in 293T/CypA+ and 293T/CypA- cells infected with SeV for 6 hr . ( D ) Immunoblot analysis of the indicated proteins in 293T/CypA+ and 293T/CypA- cells infected with SeV for the indicated time periods . ( E ) Immunoblot analysis of the indicated proteins in WT and Ppia−/− BMDMs from infected with SeV for the indicated time points . Data are shown as mean ± SD ( n = 3 ) . p<0 . 05 and *p<0 . 01 ( unpaired , two-tailed Student’s t-test ) . Data are representative of at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 24425 . 01210 . 7554/eLife . 24425 . 013Figure 3—source data 1 . Quantification of luciferase activity for Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 24425 . 013 In an attempt to identify the target protein of CypA , we initially tested the effect of CypA on different components of the RLR pathway in their activation of relevant promoters in a reporter assay . 293T/CypA- cells were transfected with expression vectors containing RIG-I-N ( CARD domain of RIG-I ) , MDA5-N ( CARD domain of MDA5 ) , MAVS , TBK1 , or IRF3/5D ( activated form of IRF3 ) , CypA or control vector , together with luciferase reporter constructs driven by promoters of genes encoding IFN-β , or the transcription factor NF-κB or ISRE . We found that CypA promoted the activation of the IFN-β promoter ( Figure 4A ) , ISRE ( Figure 4B ) and NF-κB ( Figure 4C ) induced by overexpression of RIG-I-N , MDA5-N , and MAVS , but not induced by TBK1 or IRF3/5D , suggesting that RIG-I , MDA5 , and MAVS were involved in CypA-regulated RLR pathway . Consistent with these findings , CypA increased the dimerization of IRF3 induced by overexpression of RIG-I , MDA5 and MAVS ( Figure 4D ) . We next sought to determine whether CypA could interact with these key components . Coimmunoprecipitation assay did show that CypA interacted with the transfected RIG-I , MDA5 , or MAVS in 293T cells ( Figure 4E ) . We further observed the endogenous CypA-RIG-I interaction and CypA-MAVS interaction in SeV-infected conditions ( Figure 4F , G ) . Confocal microscopy experiments indicated that CypA co-localized with the endogenous RIG-I and MAVS in 293T cells after infection with SeV ( Figure 4H ) , which are consistent with the results of Figure 4F , G , indicating that RIG-I and MAVS were the target proteins of CypA to augment RIG-I-mediated type I IFN production . 10 . 7554/eLife . 24425 . 014Figure 4 . CypA interacts with RIG-I and MAVS to activate RIG-I signaling pathway . ( A–C ) Luciferase activity of lysates in 293T/CypA- cells transfected for 24 hr with luciferase reporter constructs IFN-β-Luc ( A ) , ISRE-Luc ( B ) or NF-κB-Luc ( C ) , plus Flag-MAVS , Flag-RIG-I-N , Flag-MDA5-N , Flag-IRF3/5 , or Myc-TBK1 , along with Myc-CypA or an empty vector . Results are presented relative to the luciferase activity in control cells treated with luciferase reporter and empty vector . ( D ) Native PAGE and immunoblot analysis of IRF3 in dimer or monomer form and phosphorylated IRF3 in 293T/CypA- cells transfected for 24 hr with Flag-MAVS , Flag-RIG-I , Flag-MDA5 , or Myc-TBK1 , along with an empty vector or Myc-CypA . ( E ) Immunoblot analysis of lysates of 293T/CypA+ cells transfected for 24 hr with Flag-MAVS , Flag-RIG-I , Flag-MDA5 , or Flag-IRF3 , along with Myc-CypA , followed by immunoprecipitation with anti-Flag beads . ( F and G ) Immunoblot analysis of lysates in WT BMDMs infected with SeV for 6 hr , followed by immunoprecipitation with control mouse IgG or anti-CypA antibodies . Lysates and immunoprecipitation extracts were probed with CypA and RIG-I ( F ) or MAVS ( G ) antibodies . ( H ) Confocal microscopy of endogenous CypA and MAVS or RIG-I in 293T/CypA+ cells , treated with SeV for 6 hr . Scale bars , 10 μm . Data are shown as mean ± SD ( n = 3 ) . p<0 . 05 and *p<0 . 01 ( unpaired , two-tailed Student’s t-test ) . Data are representative of at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 24425 . 01410 . 7554/eLife . 24425 . 015Figure 4—source data 1 . Quantification of luciferase activity for Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 24425 . 015 Following ligand binding , RIG-I is ubiquitinated by the E3 ligase TRIM25 . We have known that RIG-I is the target of CypA , it is necessary to investigate whether CypA is involved in this process . Therefore , we first investigated the effect of CypA on ubiquitination of RIG-I . In 293T/CypA- cells , transfected CypA enhanced exogenous TRIM25-mediated K63-linked , but not K48-linked ubiquitination of RIG-I ( Figure 5A ) and increased the interaction between TRIM25 and RIG-I ( Figure 5C ) . Furthermore , we observed similar results of endogenous ubiquitination of RIG-I and endogenous interaction between TRIM25 and RIG-I in BMDMs upon SeV infection ( Figure 5B , D ) . It is well known that TRIM25 interacts with RIG-I-N to deliver the Lys 63-linked ubiquitin to the CARDs of RIG-I ( Gack et al . , 2007 ) . We performed coimmunoprecipitation assays to explore the CypA-binding region of RIG-I . We found that the CypA interacted with RIG-I-C ( C-terminal of RIG-I ) ( Figure 5E ) , indicating that CypA and TRIM25 bind to different regions of RIG-I . It is possible that the binding of CypA to RIG-I facilitates the interaction between TRIM25 and RIG-I-N . We next assessed the effect of CypA on recruitment of RIG-I to mitochondria . The cytoplasmic fraction ( Cyto ) , mitochondrial fraction ( Mito ) and whole-cell lysate ( WCL ) from SeV-infected WT and Ppia−/− BMDMs were separated for western blotting analysis . We observed that RIG-I induction was strongly decreased and much less RIG-I was found in mitochondria of BMDMs when CypA was absent ( Figure 5F ) . We also detected the expression and location of endogenous RIG-I in 293T/CypA+ and 293T/CypA- cells that were stained with Mito-Tracker and infected with SeV by using confocal microscopy . The results confirmed that CypA promoted RIG-I induction and facilitated recruitment of RIG-I to mitochondria upon SeV infection ( Figure 5G ) , which are consistent with the results of Figure 5F . We then investigated the effect of CypA on the interaction between RIG-I and MAVS . Coimmunoprecipitation experiment indicated that CypA enhanced RIG-I-MAVS interaction ( Figure 5H ) . More interestingly , aside from the well-known cytoplasmic distribution ( Galat and Metcalfe , 1995 ) , CypA was also detected in mitochondria and its expression level was upregulated both in mitochondria and cytoplasm against SeV infection ( Figure 5F ) , a finding confirmed by an immunofluorescence assay in 293T/CypA+ cells ( Figure 5I ) , suggesting that CypA plays important roles in response to virus infection both in mitochondria and cytoplasm . Taken together , upon SeV infection , CypA increased the interaction between the E3 ubiquitin ligase TRIM25 and RIG-I and promoted K63-linked ubiquitination of RIG-I to facilitate recruitment of RIG-I to MAVS , leading to up-regulation of RIG-I signaling pathway . 10 . 7554/eLife . 24425 . 016Figure 5 . CypA enhances TRIM25-mediated K63-linked ubiquitination of RIG-I and facilitates recruitment of RIG-I to MAVS . ( A ) Immunoblot analysis of lysates in 293T/CypA- cells transfected for 24 hr with Flag-RIG-I , along with HA-K63-Ub , HA-K48-Ub , Myc-TRIM25 , or Myc-CypA , followed by immunoprecipitation with anti-Flag beads . ( B ) Immunoblot analysis of lysates in WT and Ppia−/− BMDMs infected with SeV for 6 hr , followed by immunoprecipitation with control mouse IgG or anti-RIG-I antibodies . Lysates and immunoprecipitation extracts were probed with K63-Ub , RIG-I and CypA antibodies . ( C ) Immunoblot analysis of lysates in 293T/CypA- cells transfected with Flag-RIG-I , Myc-TRIM25 , or Myc-CypA for 24 hr , and immunoprecipitated with anti-Flag beads . ( D ) Immunoblot analysis of lysates in WT and Ppia−/− BMDMs infected with SeV for 6 hr , followed by immunoprecipitation with control mouse IgG or anti-RIG-I antibodies . Lysates and immunoprecipitation extracts were probed with RIG-I , TRIM25 , CypA and antibodies . ( E ) Immunoblot analysis of lysates in 293T/CypA- cells transfected with Myc-CypA and Flag-RIG-I , Flag-RIG-I-C or Flag-RIG-I-N for 24 hr , and immunoprecipitated with anti-Flag beads . ( F ) Immunoblot analysis of lysates in WT and Ppia−/− BMDMs after SeV infection or mock-infection for 6 hr , followed by mitochondrial-cytoplasm extraction . ( G ) Confocal microscopy of endogenous RIG-I in 293T/CypA+ and 293T/CypA- cells stained with Mito-Tracker after SeV infection or mock infection for 6 hr . Scale bars , 10 μm . ( H ) Immunoblot analysis of lysates in 293T/CypA+ and 293T/CypA- cells transfected with Flag-RIG-I and Myc-MAVS for 24 hr , and immunoprecipitated with anti-Flag beads . ( I ) Confocal microscopy of endogenous CypA in 293T/CypA+ cells stained with Mito-Tracker after SeV infection or mock-infection for 6 hr . Scale bars , 10 μm . Data are representative of at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 24425 . 016 Notably , more MAVS was observed in WT BMDMs than in Ppia−/− BMDMs , when we assessed the protein expression levels in WCL ( Figure 3E ) . Thus , we sought to examine whether CypA affects the stability of MAVS . 293T/CypA+ and 293T/CypA- cells were transfected with MAVS , RIG-I or MDA5 and treated with CHX for various times . The western blotting result showed that CypA inhibited the degradation of exogenous MAVS , but had no effect on the stability of exogenous RIG-I and MDA5 ( Figure 6A ) . Also , CypA enhanced the stability of endogenous MAVS without or with SeV infection ( Figure 6B , C ) . These data indicated that CypA plays key a role in stabilizing MAVS . 10 . 7554/eLife . 24425 . 017Figure 6 . CypA suppresses ubiquitin-mediated proteasome degradation of MAVS . ( A ) Immunoblot analysis of lysates in 293T/CypA+ and 293T/CypA- cells transfected with Flag-RIG-I , Flag-MDA5 or Flag-MAVS for 24 hr and then treated with 100 μg/ml CHX for the indicated durations ( top ) . The relative expression levels of RIG-I , MDA5 and MAVS were quantified ( below ) . ( B ) Immunoblot analysis of lysates in 293T/CypA- cells transfected with Myc-CypA or control vector for 24 hr and then treated with 100 μg/ml CHX for the indicated time points ( top ) . The relative expression levels MAVS were quantified ( below ) . ( C ) Immunoblot analysis of lysates in 293T/CypA+ and 293T/CypA- cells incubated with SeV for the indicated times ( top ) . The relative expression levels of MAVS were quantified ( below ) . ( D ) Immunoblot analysis of lysates in 293T/CypA+ and 293T/CypA- cells treated for 6 hr with 100 μg/ml CHX , along with 10 μM NH4Cl , 10 μM MG132 , or DMSO . ( E ) Immunoblot analysis of lysates in 293T/CypA- cells transfected for 24 hr with HA-Ub , along with Flag-CypA or control vector and then treated with 10 μM MG132 for 6 hr . ( F ) Immunoblot analysis of lysates in 293T/CypA- cells transfected for 24 hr with HA-Ub , along with Myc-CypA or control vector and then immunoprecipitated with anti-MAVS antibody . ( G ) Immunoblot analysis of lysates in 293T/CypA- cells transfected for 24 hr with HA-tagged deletion constructs of MAVS ( amino acids remaining , above lanes ) and point substitution constructs containing amino acids 360–540 ( KK-AA , K371A plus K420A ) , along with Myc-CypA or control vector . Data are representative of at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 24425 . 017 Proteasome- and lysosome-dependent pathways are principally responsible for intracellular protein degradation , so we investigated which pathway mediates the degradation of MAVS . In 293T/CypA+ and 293T/CypA- cells , the proteasome inhibitor MG132 significantly inhibited the degradation of MAVS , whereas the lysosome inhibitor NH4Cl did not ( Figure 6D ) , indicating that the degradation of MAVS is controlled by ubiquitin-meditated proteolysis . We further investigated the effects of CypA on MAVS ubiquitination . MAVS ubiquitination was inhibited in the presence of CypA ( Figure 6E ) . Consistently , in coimmunoprecipitation experiments , we also found that CypA significantly decreased the ubiquitination of MAVS ( Figure 6F ) . To further determine the MAVS domain responsible for CypA-regulated MAVS degradation , we tested the degradation of various MAVS truncation constructs as well as a construct of MAVS amino acids 360–540 with the substitutions K371A and K420A ( 360–540 KK-AA , which completely withstood MAVS degradation ) ( You et al . , 2009 ) . The absence of CypA accelerated the degradation of MAVS amino acids 360–540 , while substitutions K371A and K420A completely withstood the degradation ( Figure 6G ) . Together , CypA suppressed ubiquitin-mediated proteasome degradation of MAVS . E3 ubiquitin ligases TRIM25 , Smurf1 , Smurf2 , AIP4 , RNF5 , and RNF125 have been shown to mediate K48-ubiquitination and degradation of MAVS ( Arimoto et al . , 2007; Castanier et al . , 2012; Pan et al . , 2014; Wang et al . , 2012; You et al . , 2009; Zhong et al . , 2010 ) , so it is important to identify the specific ubiquitin ligase involved in CypA-regulated MAVS ubiquitination . We first investigated which E3 ligase affects CypA-regulated type I IFN production . The luciferase assay indicated that CypA increased the expression of IFN-β in the presence of TRIM25 and Smurf1 but not with AIP4 , RNF125 and RNF5 ( Figure 7A ) . We next evaluated whether CypA regulated TRIM25- and Smurf1-mediated MAVS stability and ubiquitination . We found that the level of endogenous MAVS protein was considerably increased in TRIM25 and CypA co-expressing cells compared with cells only transfected with TRIM25 , suggesting that CypA inhibits TRIM25-mediated MAVS degradation ( Figure 7B ) . In accordance with the stability assay , the ubiquitination of MAVS was inhibited in TRIM25 and CypA co-expressing cells compared with that transfected with TRIM25 ( Figure 7C ) . However , CypA almost had no effect on Smurf1-mediated MAVS stability and ubiquitination ( Figure 7B , C ) . Furthermore , we observed that CypA could enhance the TRIM25-mediated K48-linked , but not K63-linked ubiquitination of both exogenous and endogenous MAVS ( Figure 7D , E ) . 10 . 7554/eLife . 24425 . 018Figure 7 . CypA inhibits TRIM25-mediated K48-linked ubiquitination of MAVS . ( A ) Luciferase activity of lysates in 293T/CypA- cells transfected for 24 hr with IFN-β-Luc and Flag-TRIM25 , Flag-Smurf1 , Flag-AIP4 , Flag-RNF125 , or Flag-RNF5 , along with CypA or an empty vector and then treated with SeV for 6 hr . Results are presented relative to the luciferase activity in control cells transfected with luciferase reporter and empty vector . ( B ) Immunoblot analysis of lysates in 293T/CypA- cells transfected with various combinations of plasmids for 24 hr . ( C ) Immunoblot analysis of lysates in 293T/CypA- cells transfected with various combinations of plasmids for 24 hr , followed by immunoprecipitation with anti-MAVS antibody . ( D ) Immunoblot analysis of lysates in 293T/CypA- cells transfected with various combinations of plasmids for 24 hr , followed by immunoprecipitation with anti-Flag beads . ( E ) Immunoblot analysis of lysates in WT and Ppia−/− BMDMs infected with SeV for 6 hr , followed by immunoprecipitation with control mouse IgG or anti-MAVS antibodies . Lysates and immunoprecipitation extracts were probed with K48-Ub , MAVS and CypA antibodies . ( F ) Immunoblot analysis of lysates in 293T/CypA- cells transfected with various combinations of plasmids , followed by immunoprecipitation with anti-Flag beads . ( G and H ) Immunoblot analysis of lysates in 293T/CypA+ cells transfected for 24 hr with Flag-CypA ( G ) or Flag-TRIM25 ( H ) , along with HA-tagged deletion constructs of MAVS , followed by immunoprecipitation with anti-Flag beads . ( I ) Immunoblot analysis of lysates in 293T/CypA+ cells transfected with HA-Ub , plus Myc-MAVS , Myc-MAVS KK-AA ( K371A plus K420A ) , or the double point substitution construct Myc-MAVS KK-RR ( K7R plus K10R ) , along with Flag-TRIM25 or an empty vector , followed by immunoprecipitation with anti-Myc beads . ( J ) Quantitative PCR analysis of IFNB1 mRNA in 293T/RIG-I−/− cells pretreated for 1 hr with BAY 11–7082 ( 5 μM ) or DMSO , and then transfected for 48 hr with Flag-RIG-I , Flag-MAVS or an empty vector , along with scrambled siRNA or CypA siRNA ( top ) . The phosphorylated IRF3 and p65 were detected by immunoblot ( below ) . ( K ) Quantitative PCR analysis of SeV M mRNA in 293T/RIG-I−/− cells pretreated for 1 hr with BAY 11–7082 ( 5 μM ) or DMSO , then transfected for 48 hr with Flag-RIG-I , Flag-MAVS or an empty vector , along with scrambled siRNA or CypA siRNA , and then infected with SeV for 6 hr . Results are presented relative to mRNA level of SeV M in control cells transfected with empty vector and infected with SeV . Data are shown as mean ± SD ( n = 3 ) . p<0 . 05 and **p<0 . 01 ( unpaired , two-tailed Student’s t-test ) . Data are representative of at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 24425 . 01810 . 7554/eLife . 24425 . 019Figure 7—source data 1 . Quantification of luciferase activity , IFN-β production and SeV replication for Figure 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 24425 . 019 We further determined the possible mechanism by which CypA inhibits TRIM25 mediated-MAVS ubiquitination . We speculated that CypA might compete with TRIM25 to interact with MAVS . To test this hypothesis , we performed coimmunoprecipitation experiments to analyze the interaction between TRIM25 and MAVS in the presence or absence of CypA , as well as the interaction between CypA and MAVS in the presence or absence of TRIM25 . CypA-MAVS interaction was reduced in the presence of TRIM25 , and TRIM25-MAVS interaction also reduced in the presence of CypA ( Figure 7F ) . We next explored whether CypA and TRIM25 interacted with the same region of MAVS . Just as we speculated , CypA ( Figure 7G ) and TRIM25 ( Figure 7H ) both bind to MAVS amino acids 360–540 . Thus , our results indicated that CypA and the E3 ligase TRIM25 competitively interacted with MAVS , thereby inhibiting ubiquitin-mediated proteasome degradation of MAVS . It has been reported that TRIM25 targets MAVS at K7 and K10 for ubiquitination and AIP4 mediates MAVS ubiquitination at K371 and K420 ( Castanier et al . , 2012; You et al . , 2009 ) . Here we found that MAVS amino acids 360–540 , which contains K371 and K420 , is the common binding region for CypA and MAVS . Therefore , we investigated whether K371 and K420 were the ubiquitination sites for TRIM25-mediated MAVS ubiquitination . Both the MAVS mutation KK-AA ( with the substitutions K371A and K420A ) and KK-RR ( with the substitutions K7R and K10R ) distinctly reduced MAVS ubiquitination ( Figure 7I ) , suggesting that K371 and K420 , as well as K7 and K10 , are the ubiquitination sites for TRIM25-mediated MAVS ubiquitination . Collectively , CypA and TRIM25 competitively interacted with MAVS , which inhibited TRIM25-mediated MAVS ubiquitination at K371 and K420 . CypA has been shown to interact with p65 and inhibit the ubiquitin-proteasome degradation of p65 , thereby promoting innate immune responses ( Sun et al . , 2014 ) . Our data showed that CypA targeted the upstream RIG-I and MAVS to upregulate RIG-I-mediated signaling pathway . To explore the relevance and contribution of the proposed CypA mechanisms at the level of RIG-I and MAVS , we tested the effect of CypA on IFN-β induction in RIG-I-knockout cells ( 293T/RIG-I−/− ) transfected with Flag-RIG-I or Flag-MAVS , along with CypA siRNA or control siRNA . CypA increased IFN-β expression and IRF3/p65 phosphorylation with the treatment of overexpressed RIG-I and MAVS respectively ( Figure 7J ) . Then we further blocked the downstream p65 using NF-κB inhibitor ( BAY-11–7082 ) . We found that CypA still had impact on IFN-β expression and IRF3 phosphorylation when 293T/RIG-I−/− were transfected with Flag-RIG-I and treated with BAY-11–7082 , or when 293T/RIG-I−/− were transfected with Flag-MAVS and treated with BAY-11–7082 ( Figure 7J ) . Collectively , these data indicate that CypA is able to promote type I IFN production at the level of RIG-I and MAVS , which is independent of the downstream p65 . Accordingly , CypA could inhibit SeV replication by regulating RIG-I- and MAVS-directed type I IFN production ( Figure 7K ) . CypA functions as either the primary intracellular target of the immunosuppressive drug CsA ( Handschumacher et al . , 1984; Liu et al . , 1991 ) , or as a key modulator of some biological processes ( Lu et al . , 2007 ) . However , the roles of CypA in host antiviral immune responses are not well understood . It had been reported that CypA was highly induced in human gastric carcinoma cell line upon H9N2 influenza virus infection by using proteomics analysis ( Liu et al . , 2008 ) . We also found that CypA was inducible in BMDMs , 293T cells , U937 cells and human monocytes infected by SeV , VSV or IAV-mut , indicating that CypA participates in cellular antiviral response . In the present study , we identified CypA as an important host factor that promotes RIG-I-mediated type I IFN production . Deficiency of CypA greatly decreases type I IFN production , which facilitates virus replication . Furthermore , CypA increased the interaction between E3 ubiquitin ligase TRIM25 and RIG-I , promoting K63-linked ubiquitination of RIG-I that facilitated recruitment of RIG-I to MAVS . Finally , CypA and TRIM25 interacted with MAVS in a competitive manner , inhibiting TRIM25-mediated K48-linked ubiquitination of MAVS at K371 and K420 . Our findings demonstrated that CypA inhibited virus replication via enhancing antiviral immune responses , uncovering a different way for CypA to regulate virus infection . The ubiquitin system has been certified to play an essential role in precisely controlling RIG-I-mediated signal transduction ( Heaton et al . , 2016; Maelfait and Beyaert , 2012 ) . Upon RNA virus infection , viral RNA bind to RIG-I followed by the binding of TRIM25 to deliver the K63 ubiquitin chains to RIG-I ( Gack et al . , 2007; Jiang et al . , 2012; Kowalinski et al . , 2011; Peisley et al . , 2014; Sanchez et al . , 2016 ) , then the CARD domains of RIG-I are exposed to interact with the CARD domains of MAVS . In the present study , we found that CypA increased the interaction between RIG-I and TRIM25 , which facilitated K63-linked ubiquitination of RIG-I and recruitment of RIG-I from the cytosol to mitochondrion-associated MAVS , suggesting a positive role of CypA in the production of type I IFN . It has also been reported that another RIG-I binding protein , 14-3-3ε , is essential for the stable interaction of RIG-I with TRIM25 , which facilitates RIG-I ubiquitination and initiation of innate immunity against hepatitis C virus and other pathogenic RNA viruses . ( Liu et al . , 2012a ) . Therefore , just like 14-3-3ε , CypA could be defined as a key mitochondrial targeting chaperone protein that is required for innate antiviral responses . It has been established that MAVS undergoes K48-linked ubiquitination during virus infection , which mediates MAVS degradation ( Liu et al . , 2013; Paz et al . , 2009 ) . A number of studies have showed that some host proteins , including PCBP2 , IRTKS and Ndfip1 , promoted E3 ligase AIP4- or Smurf-mediated MAVS ubiquitination for degradation , leading to suppressed type I IFN production ( Wang et al . , 2012; Xia et al . , 2015; You et al . , 2009 ) . Here we found that CypA inhibited TRIM25-mediated K48-linked ubiquitination of MAVS to slow down the degradation of MAVS , thereby facilitating RIG-I-mediated type I IFN production . All these data indicated that inhibition of MAVS degradation facilitated RIG-I-mediated type I IFN production . However , a previous study suggested that the proteasomal degradation of MAVS was required to release the signaling complex into the cytosol for phosphorylation of IRF3 and subsequent production of IFN-β ( Castanier et al . , 2012 ) . In that study , MG132 was used as an inhibitor of MAVS degradation to test the effect of MAVS stability on IFN production , while MG132 is not the specific inhibitor of MAVS degradation . This is a tempting and interesting hypothesis , which remains to be further studied ( Jacobs and Coyne , 2013 ) . Besides CypA-regulated ubiquitination of RIG-I and MAVS , we have reported that CypA accelerates ubiquitin-proteasome degradation of the M1 protein of influenza virus and then restricts virus replication ( Liu et al . , 2012b ) . Additionally , CypA and another PPIase , Pin1 , both enhance the stability of P65 by blocking the ubiquitin-proteasome degradation , and SOCS-1 is the ubiquitin ligase for P65 ( Ryo et al . , 2003; Sun et al . , 2014 ) . These results support the hypothesis that PPIases have some common function in regulating ubiquitination of proteins . Together , we reveal a ubiquitination-based mechanism by which CypA controls RIG-I-mediated antiviral immune responses . CypA is widely distributed in almost all tissues . Multiple lines of evidence have revealed that CypA interacts with a large number of proteins and plays various biological roles through different mechanisms . We found that CypA-MAVS interaction was reduced in the presence of TRIM25 , and TRIM25-MAVS interaction also appeared to be reduced in the presence of CypA . In addition , both CypA and TRIM25 interacted with a similar stretch of MAVS ( aa 360–450 ) and within this region K371 and K420 were the ubiquitination target sites for TRIM25 as well as the sites that CypA targeted to stabilize MAVS . All these results suggest that CypA competes with TRIM25 for MAVS binding to inhibit TRIM25-mediated K48-linked ubiquitination of MAVS . But on the other hand , we found that CypA promoted the interaction between TRIM25 and RIG-I , which is a quite different mechanism from that at the level of MAVS . As is well known , TRIM25 interacts with RIG-I-N to deliver the Lys 63-linked ubiquitin to the CARDs of RIG-I ( Gack et al . , 2007 ) . We observed that CypA interacted with RIG-I-C , indicating that CypA and TRIM25 bind to different regions of RIG-I . We speculated that the conformation of RIG-I-N might be changed as soon as CypA interacted with RIG-I-C , then the binding site of TRIM25 was exposed , which facilitated the interaction between TRIM25 and RIG-I-N . A detailed structural study is an interesting future direction . In conclusion , our data demonstrated that CypA regulates RIG-I signaling in two ways . On one hand , CypA promotes K63-linked ubiquitination of RIG-I and recruits more RIG-I to MAVS . On the other hand , CypA stabilizes MAVS by suppressing its ubiquitin-mediated proteasome degradation . Hence , our data further expand the biological functions of CypA in RIG-I-mediated antiviral innate immunity and provide a potential novel target for manipulating viral infection . ShRNA-based knockdown of CypA in human embryonic kidney 293T cells ( CRL-3216 , ATCC ) has been described ( Liu et al . , 2012b ) . CRISPR/Cas9-based knockout of RIG-I in 293T cells has been described ( Jiang et al . , 2016 ) . The cell lines were authenticated by immunoblotting with multiple markers and tested for mycoplasma contamination using the MycoAlert Mycoplasma Detection Kit ( Lonza , Switzerland ) . The 293T cells were maintained in Dulbecco’s modified Eagle’s medium ( GIBCO ) supplemented with 10% heat-inactivated fetal bovine serum ( FBS , GIBCO ) . U937cells ( CRL-1593 . 2 , ATCC ) were maintained in 1640 medium ( GIBCO ) with FBS . BMDMs from 129 mice were maintained in 1640 medium ( GIBCO ) with FBS and maintained in macrophage-colony stimulating factor ( M-CSF , 20 ng/ml ) for 5–7 d . Peripheral blood was obtained from healthy donors under clinical protocols . Human peripheral blood mononuclear cells ( PBMCs ) were isolated using Hypaque-Ficoll density gradients by standard techniques . Monocytes were also isolated by elutriation of leukopheresis product and repurifcation using the autoMACS system if needed . Resulting cell preparations were analyzed by staining with CD14 antibodies and analyzed on a BD LSR II system . Monocyte preparations were ≥95% CD14+ . For immunoblot analysis , the following antibodies were used: rabbit polyclonal antibodies to human CypA were generated as previously described , ( 1:2000 , [Liu et al . , 2009] ) , anti-c-Myc ( 1:2000 , C3956 , Sigma , RRID:AB_439680 ) , anti-FLAG M2 ( 1:2000 , F3165 , Sigma , RRID:AB_259529 ) , anti-β-actin ( 1:1000 , sc-1616-R , Santa Cruz , RRID:AB_630836 ) , anti-GAPDH ( 1:1000 , sc-25778 , Santa Cruz , RRID:AB_10167668 ) , anti-MAVS ( 1:1000 , sc-68881 and sc-166583 , Santa Cruz , RRID:AB_1565328 and AB_2012300 ) , anti-IRF3 ( 1:1000 , sc-9082 , Santa Cruz , RRID:AB_2264929 ) , anti-mouse IgG ( 1:1000 , sc-137075 , Santa Cruz , RRID:AB_2285870 ) , anti-p-IRF3 ( Ser396 ) ( 1:1000 , 4947 , CST , RRID:AB_823547 ) , anti-p65 ( 1:1000 , 8242 , CST , RRID:AB_10859369 ) , anti-p-p65 ( Ser536 ) ( 1:1000 , 3033 , CST , RRID:AB_331284 ) , anti-IKKα/β ( 1:1000 , 2682 and 2370 , CST , RRID:AB_331626 and AB_2122154 ) , anti-p-IKKα/β ( Ser176/180 ) ( 1:1000 , 2697 , CST , RRID:AB_2079382 ) , anti-IκBα ( 1:1000 , 4814 , CST , RRID:AB_390781 ) , anti-p-IκBα ( Ser32 ) ( 1:1000 , 2859 , CST , RRID:AB_561111 ) , anti-RIG-I ( 1:1000 , 3743 , CST , RRID:AB_2269233 ) , and anti-MDA5 ( 1:1000 , 5321 , CST , RRID:AB_10694490 ) , anti-VDAC ( 1:1000 , ab14734 , Abcam , RRID:AB_443084 ) , anti-TRIM25 ( 1:1000 , 12573–1-AP , Proteintech , RRID:AB_2209732 ) , anti-Ubiquitin Antibody , Lys48-Specific ( 1:1000 , 05–1307 , Millipore , RRID:AB_1587578 ) , anti-Ubiquitin Antibody , Lys63-Specific ( 1:1000 , 05–1308 , Millipore , RRID:AB_1587580 ) . For immunofluorescence analysis , the following antibodies were used: CypA ( 1:100 , [Liu et al . , 2009] ) , anti-MAVS ( 1:50 , 3993 , CST , RRID:AB_823565 ) , and anti-RIG-I ( 1:50 , MABF297 , Millipore , RRID:AB_2650546 ) . The IFN-β promoter luciferase reporter plasmid ( IFN-β-Luc ) and NF-κB promoter luciferase reporter plasmid ( NF-κB-Luc ) were provided by C . Zheng ( Su Zhou University , China ) . The ISRE-promoter luciferase reporter plasmid ( ISRE-Luc ) , TBK1 and MAVS expression plasmids were provided by R . Lin ( McGill University , Canada ) . The RIG-I-N expression plasmid was provided by T . Fujita ( Tokyo Metropolitan Institute of Medical Science , Japan ) . RIG-I-C was synthesized by GENEWIZ and then cloned into pcDNA3 . 0-Flag vector . The MDA5-N expression plasmid was provided by S . Goodbourn ( University of London , United Kingdom ) . The IRF3/5D expression plasmid was provided by Y . Lin ( National Defense Medical Center , Taiwan ) . HA-tagged deletion constructs of MAVS or point substitution constructs containing amino acids 360–540 KKAA ( K371A plus K420A ) and AIP4 , expression plasmids were provided by Z , Jiang ( Peking University , China ) . MAVS KKAA ( K371A plus K420A ) , MAVS KKRR ( K7R plus K10R ) were synthesized by GENEWIZ and cloned into pCMV-myc vector respectively . The HA-Ub , HA-K48-Ub , HA-K63-Ub , TRIM25 , Smurf1 , RNF125 and RNF5 expression plasmids were provided by X . Ye ( Chinese Academy of Sciences , China ) . CypA-deficient ( Ppia−/− ) 129 mice were purchased from Jackson Laboratory and crossed to WT 129 mice . For analysis of the genotype of each mouse , genomic DNA was isolated from tail tissue and was identified by PCR using the primers Ppia-oIMR3772 , 5′-GCAGTTGTGATTGATCCAGGTCCG-3'; Ppia-oIMR3773 , 5'CACCCTGGAGCACCACTGCCCACC-3'; and Ppia-oIMR3774 , 5'-CCTGATCGACAAGACCGGCTTCC-3' . The animal research was approved by the Research Ethics Committee of Chinese Academy of Sciences ( Permit Number: PZIMCAS2013001 ) , and complied with the Beijing Laboratory Animal Welfare and Ethical Guidelines of the Beijing Administration Committee of Laboratory Animals . For virus infection of cells , the culture medium was removed from the plates , and the cells were washed twice with PBS . Serum-free culture medium containing SeV ( MOI = 1 ) was added for 2 hr , and then the old medium was replaced with 2% FBS culture medium . For virus infection in 8-week-old WT and Ppia−/− mice , SeV ( 2000 PFU/mouse ) was intranasally injected into the mice . The day of virus inoculation was defined as day 0 . Mice were killed at 2 , 5 , or 7 d after infection , and the lung indices ( 100× lung/body weight ) were measured . Lung tissues were then fixed , sectioned at 5 µm and stained with hematoxylin and eosin . For cytokines and SeV replication analyses , mice were killed at 6 , 12 , and 24 hr after infection . Lung tissues were homogenized using a QIAGEN Tissue Lyser II machine ( 30 cycles/s , 4 min ) in 1 ml of cold PBS under sterile conditions . Total RNA was extracted from homogenized lung tissue using TRIzol ( Invitrogen ) to detect the mRNA level of cytokines and viral genes . Then , the remaining homogenates were centrifuged and the supernatants were used to detect the protein levels of IFN-β and IFN-α via ELISA ( PBL Assay Science ) . SeV virus was harvested from the supernatants of infected cells every day . A standardized concentration of chicken red blood cells ( 0 . 5% RBC ) was used . A serial twofold dilution of supernatant was prepared in U-bottomed 96-well microtiter plates with PBS , 50 μl 0 . 5% RBC was added to each well , and the U-bottomed plates were incubated for 30 min at room temperature . Then , the lattice forming parts were counted , and the titer was calculated . MDBK cells were seeded in 96-well plates 24 hr before VSV virus infection . VSV virus supernatants were serially diluted with DMEM and added to each well with eight replicates of each dilution . 24 hr after infection , the 50% TCID50 was calculated by the Reed-Muench method . 293T cells were seeded into 24-well plates . The following day , cells were transfected with 200 ng luciferase plasmid and 100 ng β-Gal plasmid , along with 200 ng to 400 ng plasmids required for different experiments . Twenty-four hours later , cells were lysed in lysis buffer . After centrifugation , the supernatants were stored at −80°C . The luciferase assays were performed with a luciferase assay kit ( Promega , Madison , WI ) . Total RNA was extracted from cells with TRIzol ( Invitrogen ) according to the manufacturer’s instructions . Samples were digested with DNase I and subjected to reverse transcription-PCR ( RT-PCR ) . RNA was reverse-transcribed using an oligo ( dT ) primer . A mock reaction was performed with no reverse transcriptase added . The analysis of the relative gene expression levels was performed using Corbett 6200 and PCR primers: hIFN-β ( IFNB1 ) forward , 5′-AACTGCAACCTTTCGAAGCC-3′; hIFN-β ( IFNB1 ) reverse , 5′-TGTCGCCTACTACCTGTTGTGC-3′; mIFN-β ( Ifnb1 ) forward , 5′-GGAGATGACGGAGAAGATGC-3′; mIFN-β ( Ifnb1 ) reverse , 5′-CCCAGTGCTGGAGAAATTGT-3′; mIFN-α ( Ifna ) forward , 5′-GGCTTGACACTCCTGGTACAAATGAG-3′; mIFN-α ( Ifna ) reverse , 5′-CAGCACATTGGCAGAGGAAGACAG-3′; hISG54 ( IFIT2 ) forward , 5′-TCATTTTGCATCCCATAGGAGGTT-3′; hISG54 ( IFIT2 ) reverse , 5′-GACTTTGGTCCCCCAGCTTT-3′; mISG54 ( Ifit2 ) forward , 5′-ATGAAGACGGTGCTGAATACTAGTGA-3′; mISG54 ( Ifit2 ) reverse , 5′-TGAGGGCTTTCTTTTTCC-3′; hISG56 ( IFIT1 ) forward , 5′-TTCGGAGAAAGGCATTAGA-3′; hISG56 ( IFIT1 ) reverse , 5′-TCCAGGGCTTCATTCATAT-3′; mISG56 ( Ifit1 ) forward , 5′-CAGAAGCAC ACATTGAAGAAGC-3′; mISG56 ( Ifit1 ) reverse , 5′-TGTAAGTAGCCAGAGGAAGGTG-3′; hRantes ( CCL5 ) forward , 5′-TGCCTGTTTCTGCTTGCTCTTGTC-3′; hRantes ( CCL5 ) reverse , 5′-TGTGGTAGAATCTGGGCCCTTCAA-3′; mRantes ( Ccl5 ) forward , 5′-ACTCCCTGCTGCTTTGCCTAC-3′; mRantes ( Ccl5 ) reverse , 5′-ACTTGCTGGTGTAGAAATACT-3′; hCypA ( PPIA ) forward , 5′- CAACCCCACCGTGTTCTTC-3′; hCypA ( PPIA ) reverse , 5′- ACTTGCCACCAGTGCCATTA-3′; mCypA ( Ppia ) forward , 5′-TTTGCAGACGCCACTGTC-3′; mCypA ( Ppia ) reverse , 5′-CAGTGCTCAGAGCTCGAAAG-3′; SeV NP forward , 5′- caagagcccactcttccaggg-3′; SeV NP reverse , 5′-CTGAACGCCTCTAACCTGTTG-3′; SeV M forward , 5′-GTGATTTGGGCGGCATCT-3′; and SeV M reverse , 5′-GATGGCCGGTTGGAACAC-3′ . GAPDH served as an internal control using PCR primers , GAPDH forward 5’-TTGTCTCCTGCGACTTCAACAG-3’ and GAPDH reverse 5’-GGTCTGGGATGGAAATTGTGAG-3’ . Supernatants from cultured cells or sera were collected at the indicated times . Cytokines were analyzed by ELISA kits ( Thermo ) according to the manufacturer’s instructions . Cells were lysed in lysis buffer containing 0 . 5% NP40 , 150 mM NaCl , 20 mM HEPES ( pH 7 . 4 ) , 10% glycerol , 1 mM EDTA , and protease inhibitor cocktail . After centrifugation , the supernatants were incubated with anti-FLAG , anti-Myc beads ( Sigma ) , or Protein A/G PLUS-Agarose beads ( Santa Cruz ) for 4 hr at 4°C . After five washes in washing buffer ( 0 . 5% NP40 , 300 mM NaCl , 20 mM HEPES ( pH 7 . 4 ) , 10% glycerol , and 1 mM EDTA ) , the immunoprecipitates were analyzed by immunoblot analysis . Native PAGE was performed with an 8% acrylamide gel without SDS . The gel was pre-run for 30 min at 40 mA on ice with 25 mM Tris-HCl ( pH 8 . 4 ) and 192 mM glycine with or without 0 . 5% deoxycholate in the cathode chamber and anode chamber , respectively . Samples in the native sample buffer ( 50 mM Tris-HCl , pH 6 . 8 , and 15% glycerol ) were applied on the gel and underwent electrophoresis for 60 min at 35 mA on ice followed by immunoblot analysis . Duplexes of CypA siRNA and negative controls were synthesized by Genepharma ( Shanghai , China ) . SiRNA oligonucleotides are as follows: CypA sense , 5′-GCUCGCAGUAUCCUAGAAUTT-3′; CypA antisense , 5′-AUUCUAGGAUACUGCGAGCTT-3′; negative control sense , 5′-UUCUCCGAACGUGUCACGUTT-3′; negative control antisense , 5′-acgugacacguucggagaaTT-3′ . Transfection of siRNA into cells was performed according to manufacturer’s instructions . U937 cells and human monocytes were transfected with siRNA using Lipofectamine 2000 ( Invitrogen ) . Cells were washed with PBS three times , fixed in 4% paraformaldehyde for 30 min at room temperature , permeabilized with 0 . 5% Triton X-100 in PBS ( PBST ) for 20 min , and stained with appropriate antibodies . Cell nuclei were stained with 5 μg/ml DAPI ( Sigma ) . Following staining , cover slips were analyzed using a Leica SP8 confocal microscope . Cells ( 5 × 107 ) infected with SeV or left uninfected were washed with PBS and lysed by douncing 35 times in 1 . 5 ml homogenization buffer ( ApplyGen ) . The homogenates were then centrifuged at 800 g for 5 min twice . The supernatants were centrifuged at 12 , 000 g for 10 min to precipitate mitochondria . The supernatants from this step ( cytoplasm fraction ) were also collected . The precipitate fraction was washed with 0 . 2 ml homogenization buffer , centrifuged at 12 , 000 g for 10 min and collected as the mitochondria fraction . Cells were treated with 100 μg/ml CHX for various periods of time at 24 hr after transfection . Then , cells were lysed and analyzed by immunoblotting . MG132 ( 10 μM ) and NH4Cl ( 10 μM ) were used at the same time as CHX , and cells were harvested 6 hr after treatment .	In mammals , an enzyme called Cyclophilin A ( CypA ) is found in almost all tissues and plays important roles in many biological processes including the production of proteins and inflammation . Recent work suggests that it also plays a role in fighting virus infections . CypA can interact directly with a protein from viruses to inhibit the virus from multiplying . Several lines of evidence indicate that CypA can also regulate virus replication by stimulating the production of molecules called type I interferons , but it is not clear how this could work . A receptor protein called RIG-I can detect the presence of a virus and interact with another protein called MAVS to stimulate immune responses , leading to the production of type I interferons . Liu , Li et al . used human cells and mice to investigate how CypA affects this process . The experiments show that the levels of CypA in cells increase during virus infection . Cells that lack CypA produce fewer type I interferon molecules , which gives the virus more of a chance to multiply . Further experiments show that CypA alters the ubiquitin-mediated protein modification of RIG-I and MAVS . The findings of Liu , Li et al . identify a new way in which CypA boosts immune responses during virus infections . A future challenge is to develop new drugs that regulate the protein modification of RIG-I and MAVS , which may help to treat virus infections .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "immunology", "and", "inflammation" ]	2017	Cyclophilin A-regulated ubiquitination is critical for RIG-I-mediated antiviral immune responses
This study examined records of 2566 consecutive COVID-19 patients at five Massachusetts hospitals and sought to predict level-of-care requirements based on clinical and laboratory data . Several classification methods were applied and compared against standard pneumonia severity scores . The need for hospitalization , ICU care , and mechanical ventilation were predicted with a validation accuracy of 88% , 87% , and 86% , respectively . Pneumonia severity scores achieve respective accuracies of 73% and 74% for ICU care and ventilation . When predictions are limited to patients with more complex disease , the accuracy of the ICU and ventilation prediction models achieved accuracy of 83% and 82% , respectively . Vital signs , age , BMI , dyspnea , and comorbidities were the most important predictors of hospitalization . Opacities on chest imaging , age , admission vital signs and symptoms , male gender , admission laboratory results , and diabetes were the most important risk factors for ICU admission and mechanical ventilation . The factors identified collectively form a signature of the novel COVID-19 disease . As a result of the SARS-CoV-2 pandemic , many hospitals across the world have resorted to drastic measures: canceling elective procedures , switching to remote consultations , designating most beds to COVID-19 , expanding Intensive Care Unit ( ICU ) capacity , and re-purposing doctors and nurses to support COVID-19 care . In the U . S . , the CDC estimates more than 310 , 000 COVID-19 hospitalizations from March 1 to June 13 , 2020 ( CDC , 2020 ) . Much of the modeling work related to the pandemic has focused on spread dynamics ( Kucharski et al . , 2020 ) . Others have described patients who were hospitalized ( Richardson et al . , 2020 ) ( n = 5700 ) and ( Buckner et al . , 2020 ) ( n = 105 ) , became critically ill ( Gong et al . , 2020 ) ( n = 372 ) , or succumbed to the disease ( n = 1625 ( Onder et al . , 2020 ) , n = 270 [Wu et al . , 2020] ) . In data from the New York City , 14 . 2% required ICU treatment and 12 . 2% mechanical ventilation ( Richardson et al . , 2020 ) . With such rates , the logistical and ethical implications of bed allocation and potential rationing of care delivery are immense ( White and Lo , 2020 ) . To date , while state- or country-level prognostication has developed to examine resource allocation at a mass scale , there is inadequate evidence based on a large cohort on accurate prediction of the disease progress at the individual patient level . A string of recent studies developed models to predict severe disease or mortality based on clinical and laboratory findings , for example ( Yan et al . , 2020 ) ( n = 485 ) , ( Gong et al . , 2020 ) ( n = 372 ) , ( Bhargava et al . , 2020 ) ( n = 197 ) , ( Ji et al . , 2020 ) ( n = 208 ) , and ( Wang et al . , 2020 ) ( n = 296 ) . In these studies , several variables such as Lactate Dehydrogenase ( LDH ) ( Gong et al . , 2020; Ji et al . , 2020; Yan et al . , 2020 ) and C-reactive protein ( CRP ) have been identified as important predictors . All of these studies considered relatively small cohorts and , with the exception of Bhargava et al . , 2020 , considered patients in China . Although it is believed that the virus remains the same around the globe , the physiologic response to the virus and the eventual course of disease depend on multiple other factors , many of them regional ( e . g . population characteristics , hospital practices , prevalence of pre-existing conditions ) and not applicable universally . Triage of adult patients with COVID-19 remains challenging with most evidence coming from expert recommendations; evidence-based methods based on larger U . S . -based cohorts have not been reported ( Sprung et al . , 2020 ) . Leveraging data from five hospitals of the largest health care system in Massachusetts , we seek to develop personalized , interpretable predictive models of ( i ) hospitalization , ( ii ) ICU treatment , and ( iii ) mechanical ventilation , among SARS-CoV-2 positive patients . To develop these models , we developed a pipeline leveraging state-of-the-art Natural Language Processing ( NLP ) tools to extract information from the clinical reports for each patient , employing statistical feature selection methods to retain the most predictive features for each model , and adapting a host of advance machine learning-based classification methods to develop parsimonious ( hence , easier to use and interpret ) predictive models . We found that the more interpretable models can , for the most part , deliver similar predictive performance compared to more complex , ‘black-box’ models involving ensembles of many decision trees . Our results support our initial hypothesis that important clinical outcomes can be predicted with a high degree of accuracy upon the patient’s first presentation to the hospital using a relatively small number of features , which collectively compose a ‘signature’ of the novel COVID-19 disease . The mean age of hospitalized patients was 62 . 3 years ( SD: 18 years ) and 55 . 3% were male . We employed linear and non-linear classification methods for predicting hospitalizations . Non-linear methods included random forests ( RF ) ( Breiman , 2001 ) and XGBoost ( Chen and Guestrin , 2016 ) . Linear methods included support vector machines ( SVM ) ( Cortes and Vapnik , 1995 ) and Logistic Regression ( LR ) ; each linear method used either ℓ1- or ℓ2-norm regularization and we report the best-performing flavor of each model . Results are reported in Table 1 . We report the Area Under the Curve ( AUC ) of the Receiver Operating Characteristic ( ROC ) and the Weighted-F1 score , both computed out-of-sample ( in a test set not used for training the model ) . As we detail under Methods , we used two validation strategies . The ‘Random’ strategy randomly split the patients into a training and a test set and was repeated five times; from these five splits we report the average and the standard deviation of the test performance . The ‘BWH’ strategy trained the models on MGH , FH , NWH , and NSM patients , and evaluated performance on BWH patients . The hospitalization models used symptoms , pre-existing medications , comorbidities , and patient demographics . Laboratory results and radiologic findings were not considered since these were not available for most non-hospitalized patients . Full models used all ( 106 ) variables retained after several pre-processing steps described in Materials and methods . Applying the statistical variable selection procedure described in the Appendix ( specifically , eliminating variables with a p-value exceeding 0 . 05 ) , yields a model with 74 variables . To provide a more parsimonious , highly interpretable , and easier to implement model , we used recursive feature elimination ( see Appendix ) to select a model with only 11 variables . The best model using the random validation approach has an AUC of 88% while the best parsimonious ( linear ) model has an AUC of 83% , being though easier to interpret and implement . Validation on the BWH patients yields an AUC of 84% for the parsimonious model . Table 1 also reports the 11 variables in the parsimonious LR model , including their LR coefficients , and a binarized version of this model as described in Materials and methods . The most important variables associated with hospitalization were: oxygen saturation , temperature , respiratory rate , age , pulse , blood pressure , a comorbidity of adrenal insufficiency , BMI , prior transplantation , dyspnea , and kidney disease . Additionally , we assessed the role of pre-existing ACE inhibitor ( ACEI ) and angiotensin receptor blocker ( ARB ) medications by adding these variables into the parsimonious binarized model , while controlling for additional relevant variables ( hypertension , diabetes , and arrhythmia comorbidities and other hypertension medications ) . We found that while ARBs are not a factor , ACEIs reduce the odds of hospitalization by 3/4 , on average , controlling for other important factors , such as age , hypertension , and related comorbidities associated with the use of these medications . The mean age of ICU admitted patients was 63 . 3 years ( SD: 15 . 1 years ) and 63% were male . The ICU and ventilation prediction models used the features considered for the hospitalization , as well as laboratory results and radiologic findings . For these models , we excluded patients who required immediate ICU admission or ventilation ( defined as within 4 hr from initial presentation ) . This was implemented in order to focus on patients where triaging is challenging and risk prediction would be beneficial . There were 2513 and 2525 patients remaining for the ICU and the mechanical ventilation prediction models , respectively . For the model including 2513 patients ( Table 2 ) , we first developed a model using all 130 variables retained after pre-processing , then employed statistical variable selection to retain 56 of the variables , and then applied recursive feature elimination with LR to select a parsimonious model which uses only 10 variables . The following variables were included: opacity observed in a chest scan , respiratory rate , age , fever , male gender , albumin , anion gap , oxygen saturation , LDH , and calcium . In addition , we generated a binarized version of the parsimonious model . The parsimonious model for all 2513 patients has an AUC of 86% , almost as high as the model with all 130 features . For comparison purposes against well-established scoring systems , we implemented two commonly used pneumonia severity scores , CURB-65 ( Lim et al . , 2003 ) and the Pneumonia Severity Index ( PSI ) ( Fine et al . , 1997 ) . Predictions based on the PSI and CURB-65 scores , have AUCs of 73% and 67% , respectively . We also developed a model for a more restrictive set of patients . Specifically , the number of missing lab values for some patients is substantial . Given the importance of LDH and CRP , as revealed by our models , the more restricted patient set contains 669 patients with non-missing LDH and CRP values . After removing patients who required intubation or ICU admission within 4 hr of hospital presentation , we included 628 patients and 635 patients for the restricted ICU admission and ventilation models , respectively . The best restricted model for the 628 patients ( Table 3 ) is the nonlinear XGBoost model using 29 statistically selected features with an AUC of 83% , with a linear parsimonious LR model close behind ( AUC 80% ) . An RF model using all variables yields an AUC of 77% when tested on BWH data . PSI- and CURB-65 models have AUCs below 59% . The mean age of patients requiring mechanical ventilation was 63 . 3 years ( SD: 14 . 7 years ) and 63 . 6% were male . Again , we excluded patients who were intubated within 4 hr of their hospital admission . For the model including 2525 patients ( Table 4 ) , we used statistical feature selection to select 55 variables , and recursive feature elimination with LR to select a parsimonious model with only eight variables . The following variables were included: lung opacities , albumin , fever , respiratory rate , glucose , male gender , LDH , and anion gap . In addition , we generated a binarized version of the parsimonious model . The best model for all 2525 patients was a nonlinear RF model using the 55 statistically selected variables and yielding an AUC of 86% . The best linear model was the parsimonious LR model with an AUC of 85% . PSI- and CURB-65 models yield AUCs of 74% and 67% , respectively . The best model for the restricted case of 635 patients ( Table 5 ) was the linear parsimonious LR model ( with just five variables ) achieving an AUC of 82% . PSI- and CURB-65 models do not exceed AUC of 58% . Table 6 reports the mean and the median time interval ( in hours ) between hospital admission time and ICU/ventilation outcomes . Specifically , we report statistics for ICU admission or intubation outcomes from the correct ICU/intubation predictions made by our models trained on four hospitals ( MGH , NWH , NSM , FH ) and applied to BWH patients ( both the models making predictions for all patients and the restricted models ) . As we have noted earlier , our models use the lab results closest to admission ( either on admission date or the following day ) . We also report the time interval between the last lab result used by the model and the corresponding ICU/intubation outcome . We developed three models to predict need for hospitalization , ICU admission , and mechanical ventilation in patients with COVID-19 . The prediction models are not meant to replace clinicians’ judgment for determining level of care . Instead , they are designed to assist clinicians in identifying patients at risk of future decompensation . Patient vital signs were the most important predictors of hospitalization . This is expected as vital signs reflect underlying disease severity , the need for cardiorespiratory resuscitation , and the risk of future decompensation without adequate medical support . Older age and BMI were also important predictors for hospitalization . Age has been recognized as an important factor associated with severe COVID-19 in previous series ( Grasselli et al . , 2020; Guan et al . , 2020; Richardson et al . , 2020 ) . However , it is not known whether age itself or the presence of comorbidities place patients at risk for severe disease . Our results demonstrate that age is a stronger predictor of severe COVID-19 than a host of underlying comorbidities . In terms of patient comorbidities , adrenal insufficiency , prior transplantation , and chronic kidney disease were strongly associated with need for hospitalization . Diabetes mellitus was associated with a need for ICU admission and mechanical ventilation , which might be due to its detrimental effects on immune function . For the ICU and ventilation prediction models screening all at-risk ( COVID-19-positive patients ) , opacities observed in a chest scan , age , and male gender emerge as important variables . Males have been found to have worse in-hospital outcomes in other studies as well ( Palaiodimos et al . , 2020 ) . We also identified several routine laboratory values that are predictive of ICU admission and mechanical ventilation . Elevated serum LDH , CRP , anion gap , and glucose , as well as decreased serum calcium , sodium , and albumin were strong predictors of ICU admission and mechanical ventilation . LDH is an indicator of tissue damage and has been found to be a marker of severity in P . jirovecii pneumonia ( Zaman and White , 1988 ) . Along with CRP , it was among the two most important predictors of ICU admission and ventilation in the parsimonious model among patients who had LDH and CRP measurements on admission . This finding is consistent with previous reports identifying LDH as an important prognostic factor ( Gong et al . , 2020; Ji et al . , 2020; Mo et al . , 2020; Yan et al . , 2020 ) . In addition , lower serum calcium is associated with cell lysis and tissue destruction , as it is often seen as part of the tumor lysis syndrome . Elevated serum anion gap is a marker of metabolic acidosis and ischemia , suggesting that tissue hypoxia and hypoperfusion may be components of severe disease . For all three prognostic models , we developed predicting hospitalizations , ICU care , and mechanical ventilation , AUC ranges within 86–88% , which indicates strong predictive power . Interestingly , we can achieve AUC within 85–86% for ICU and ventilation prediction with a parsimonious linear model utilizing no more than 10 variables . In all cases , we can also develop a parsimonious model with binarized variables using medically suggested normal and abnormal variable thresholds . These binarized models have similar performance with their continuous counterparts . The ICU and ventilation models using all patients are very accurate , but , arguably , make a number of ‘easier’ decisions since more than 60% of the patients are never hospitalized . Many of these patients are younger , healthy , and likely present with mild-to-moderate symptoms . To test the robustness of the models to patients with potentially more ‘complex’ disease , we developed ICU and ventilation models on a restricted set of patients . This is the subset of patients who are hospitalized and most of the crucial labs are available for them ( specifically CRP and LDH which emerged as important from our models ) . The best AUC for these models drops , but not below 82% , which indicates robustness of the model even when dealing with arguably harder to assess cases . LDH , CRP , calcium , lung opacity , anion gap , SpO2 , sodium , and a comorbidity of insulin-controlled diabetes appear as the most significant for these patients . Interestingly , the corresponding binarized models have about 10% lower AUC; apparently , for the more severely ill , clinical variables deviate substantially from normal and knowing the exact values is crucial . The models have been validated with two different approaches , using random splits of the data into training and testing , as well as training in some hospitals and testing at a different hospital . Performance metrics are relatively consistent with these two approaches . We also compared the models against standard pneumonia severity scores , PSI and CURB-65 , establishing that our models are significantly stronger , which highlights the different clinical profile of COVID-19 . We also examined how much in advance of the ICU or ventilation outcomes our models are able to make a prediction . Of course , this is not entirely in our control; it depends on what state the patients get admitted and how soon their condition deteriorates to require ICU admission and/or ventilation . Table 6 reports the corresponding statistics . For example , the restricted ICU and ventilation models are making a correct prediction upon admission ( using the lab results closest to that time ) for outcomes that on average occur 38 and 35 hr later , respectively . To further test the accuracy of the restricted ICU and ventilation models well in advance of the corresponding event , we considered an extended BWH test set ( adding 11 more patients ) and computed the accuracy of the models when the test set was restricted to patients whose outcome ( ICU admission or ventilation ) was more than x hours after the admission lab results based on which the prediction was made , with x being 6 hr , or 12 hr , or 18 hr , or 24 hr , or even 48 hr . The ICU model reaches an AUC of 87% and a weighted F1-score of 86% at x = 18 hr . The ventilation model reaches an AUC of 64% and an F1-score of 72% at x = 48 hr . These results demonstrate that the predictive models can indeed make predictions well into the future , when physicians would be less certain about the course of the disease and when there is potentially enough time to intervene and improve outcomes . A manual review of the predictions by the models indicates that they performed well at predicting future ICU admissions for patients who presented with mild disease several days before ICU admission was necessary . Such patients were hemodynamically stable and had minimal oxygen requirements on the floor , before clinical deterioration necessitated ICU admission . We identified several such patients . A typical case is that of a 51-year-old male with a history of hypertension , obesity , and insulin-dependent type 2 diabetes mellitus , who presented with a 3-day history of dyspnea , cough and myalgias . In the emergency department , he was hemodynamically stable , saturating at 96–97% on 2 L of nasal cannula . The patient was admitted to the floor and did well for 3 days , saturating at 93–96% on room air . On the fourth day of hospitalization , he had increasing oxygen requirements and the decision was made to transfer him to the ICU . He was intubated and ventilated for 30 days . Our prediction models accurately predicted at the time of his presentation that he would eventually require ICU admission and mechanical ventilation . This prediction was based on such variables as an elevated LDH ( 241 U/L ) and the presence of insulin-dependent diabetes mellitus . Another such case is that of a 59-year-old male without a significant prior medical history who presented with 2 days of dyspnea , nausea , and diarrhea . At the emergency department , he was tachycardic at 110 beats per minute and saturating at 96% on room air , and the patient was admitted . For 2 days , the patient was hemodynamically stable , saturating at 94–97% on room air . On the third day of hospitalization , he had increasing oxygen requirements , eventually requiring transfer to the ICU . He was intubated and ventilated for the next 14 days . Our prediction model predicted the patient’s decompensation at his presentation , due to elevations in LDH ( 348 U/L ) and CRP ( 102 . 3 mg/L ) . We also considered the role of ACEIs and ARBs and their potential association with the outcomes . It has been speculated that ACEIs may worsen COVID-19 outcomes because they upregulate the expression of ACE2 , which the virus targets for cell entry . No such evidence has been reported in earlier studies ( Kuster et al . , 2020; Patel and Verma , 2020 ) . In fact , a smaller study ( Zhang et al . , 2020 ) ( n = 1128 vs . 2566 in our case ) reported a beneficial effect and ( Rossi et al . , 2020 ) warn of potential harmful effects of discontinuing ACEIs or ARBs due to COVID-19 . Our hospitalization model suggests that ACEIs do not increase hospitalization risk and may slightly reduce it ( OR 95% CI is ( 0 . 52 , 1 . 04 ) with a mean of 0 . 73 ) . In the ICU and ventilation models , the role of these two medications is statistically weaker to observe any meaningful association . The models we derived can be used for a variety of purposes: ( i ) guiding patient triage to appropriate inpatient units , ( ii ) guiding staffing and resource planning logistics , and ( iii ) understanding patient risk profiles to inform future policy decisions , such as targeted risk-based stay-at-home restrictions , testing , and vaccination prioritization guidelines once a vaccine becomes available . Calculators implementing the parsimonious models corresponding to each of the Tables 1 , 2 , 3 , 4 , 5 have been made available online ( Hao et al . , 2020 ) . Natural Language Processing ( NLP ) was used to extract patient comorbidities ( see Appendix for details ) , pre-existing medications , admission vital signs , hospitalization course , ICU admission , and mechanical intubation . The categorical features were converted to numerical by ‘one-hot’ encoding . Each categorical feature , such as gender and race , was encoded as an indicator variable for each category . Features were standardized by subtracting the mean and dividing by the standard deviation . Several pre-processing steps , including variable imputation , outlier elimination , and removal of highly correlated variables were undertaken ( see Appendix ) . After completing these procedures , 106 variables for each patient remained to be used by the hospitalization model . For the ICU and ventilation prediction models , we added laboratory results and radiologic findings . We removed variables with more than 90% missing values out of the roughly 2500 patients retained for these models; the remaining missing values were imputed as described above . These pre-processing steps retained 130 variables for the ICU and ventilation models . We employed nonlinear ensemble methods including Random forests ( RF ) ( Breiman , 2001 ) and XGBoost ( Chen and Guestrin , 2016 ) . We also employed ‘custom’ linear methods which yield interpretable models; specifically , support vector machines ( SVM ) ( Cortes and Vapnik , 1995 ) and Logistic Regression ( LR ) . In both cases , the variants we computed were robust to noise and the presence of outliers ( Chen and Paschalidis , 2018 ) , using proper regularization . LR , in addition to a prediction , provides the likelihood associated with the predicted outcome , which can be used as a confidence measure in decision making . Further details on these methods are in the Appendix . For each outcome , we used the statistical feature selection and recursive feature elimination procedures described in the Appendix to develop an LR parsimonious model . The LR coefficients are comparable since the variables are standardized . Hence , a larger absolute coefficient indicates that the corresponding variable is a more significant predictor . Positive ( negative ) coefficients imply positive ( negative ) correlation with the outcome . We also developed a version of this model by converting all continuous variables into binary variables , using medically motivated thresholds ( see Appendix ) . We report the coefficients of the ‘binarized’ model and the implied odds ratio ( OR ) , representing how the odds of the outcome are scaled by having a specific variable being abnormal vs . normal , while controlling for all other variables in the model . Model performance metrics included the Area Under the Curve ( AUC ) of the Receiver Operating Characteristic ( ROC ) and the Weighted-F1 score . The ROC plots the true positive rate ( a . k . a . recall or sensitivity ) against the false positive rate ( equal to one minus the specificity ) . We optimized algorithm parameters to maximize AUC . The F1 score is the harmonic mean of precision and recall . Precision ( or positive predictive value ) is defined as the ratio of true positives over true and false positives . The Weighted-F1 score is computed by weighting the F1-score of each class by the number of patients in that class . The data were split into a training ( 80% ) and a test set ( 20% ) . Algorithm parameters were optimized on the training ( derivation ) set using fivefold cross-validation . Performance metrics were computed on the test set . This process was repeated five times , each time with a random split into training/testing sets . In columns labeled as Random in Tables 1 , 2 , 3 , 4 , 5 , we report the average ( and standard deviation ) of the test performance metrics over the five random splits . We also performed a different type of validation . We trained the models on MGH , FH , NWH , and NSM patients , and evaluated performance on BWH patients . These results are reported under the columns BWH in the tables .	The new coronavirus ( now named SARS-CoV-2 ) causing the disease pandemic in 2019 ( COVID-19 ) , has so far infected over 35 million people worldwide and killed more than 1 million . Most people with COVID-19 have no symptoms or only mild symptoms . But some become seriously ill and need hospitalization . The sickest are admitted to an Intensive Care Unit ( ICU ) and may need mechanical ventilation to help them breath . Being able to predict which patients with COVID-19 will become severely ill could help hospitals around the world manage the huge influx of patients caused by the pandemic and save lives . Now , Hao , Sotudian , Wang , Xu et al . show that computer models using artificial intelligence technology can help predict which COVID-19 patients will be hospitalized , admitted to the ICU , or need mechanical ventilation . Using data of 2 , 566 COVID-19 patients from five Massachusetts hospitals , Hao et al . created three separate models that can predict hospitalization , ICU admission , and the need for mechanical ventilation with more than 86% accuracy , based on patient characteristics , clinical symptoms , laboratory results and chest x-rays . Hao et al . found that the patients’ vital signs , age , obesity , difficulty breathing , and underlying diseases like diabetes , were the strongest predictors of the need for hospitalization . Being male , having diabetes , cloudy chest x-rays , and certain laboratory results were the most important risk factors for intensive care treatment and mechanical ventilation . Laboratory results suggesting tissue damage , severe inflammation or oxygen deprivation in the body's tissues were important warning signs of severe disease . The results provide a more detailed picture of the patients who are likely to suffer from severe forms of COVID-19 . Using the predictive models may help physicians identify patients who appear okay but need closer monitoring and more aggressive treatment . The models may also help policy makers decide who needs workplace accommodations such as being allowed to work from home , which individuals may benefit from more frequent testing , and who should be prioritized for vaccination when a vaccine becomes available .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "medicine", "microbiology", "and", "infectious", "disease" ]	2020	Early prediction of level-of-care requirements in patients with COVID-19
HIV-1 reverse transcriptase utilizes a metamorphic polymerase domain that is able to adopt two alternate structures that fulfill catalytic and structural roles , thereby minimizing its coding requirements . This ambiguity introduces folding challenges that are met by a complex maturation process . We have investigated this conformational maturation using NMR studies of methyl-labeled RT for the slower processes in combination with molecular dynamics simulations for rapid processes . Starting from an inactive conformation , the p66 precursor undergoes a unimolecular isomerization to a structure similar to its active form , exposing a large hydrophobic surface that facilitates initial homodimer formation . The resulting p66/p66' homodimer exists as a conformational heterodimer , after which a series of conformational adjustments on different time scales can be observed . Formation of the inter-subunit RH:thumb' interface occurs at an early stage , while maturation of the connection' and unfolding of the RH' domains are linked and occur on a much slower time scale . HIV reverse transcriptase ( RT ) plays a multifunctional role in the transformation of viral RNA into dsDNA and represents a primary target for treatment of AIDS . Currently , all of the drugs in clinical use target the mature RT p66/p51 heterodimer , however , a single p66 peptide chain functions as the precursor for each subunit of the RT heterodimer , requiring a complex maturation process that includes subunit-selective elimination of a single ribonuclease H ( RH ) domain . The need for such a process is a consequence of a metamorphic polymerase domain that is able to adopt different structures in each RT subunit , allowing it to fulfill two different functional roles . The metamorphic polymerase domain reduces the need for additional coding sequences in the HIV gene , consistent with evolutionary pressures on the size of the RNA viral genome ( Belshaw et al . , 2007 ) , while requiring a more complex structural maturation process . Hypotheses for the formation and maturation of the RT homodimer include proposals in which RH domain proteolysis precedes heterodimer formation ( Srivastava et al . , 2006 ) , models in which p66 forms an initially symmetric homodimer followed by RH domain unfolding leading to an asymmetric homodimer ( Anderson and Coleman , 1992; Abram and Parniak , 2005; Sharaf et al . , 2014 ) , and models in which an initially formed asymmetric homodimer leads to partial RH domain unfolding ( Hostomska et al . , 1991 ) . Until recently , no detailed structural data were available for the p66 monomer and very little structural evidence was available to support or refute any of the above models . Not only does this represent a significant gap in understanding the behavior of an important viral enzyme but also the intermediates involved in heterodimer formation provide potentially useful targets for the development of new interventional strategies . We recently determined a crystal structure for an isolated p51 monomer mutant and obtained NMR data indicating that the p66 monomer adopts a structure similar to the p51 monomer with an additional RH domain linked by flexible residues unraveled from the connection subdomain C-terminus ( Zheng et al . , 2014 ) . The p66 monomer is the substrate for dimerization , and thus , provides the starting point for analysis of p66/p66' dimer formation and subsequent conformational changes . Structural comparisons of the RT subunits and the p51 monomer indicate that the most significant conformational variations are observed for the palm thumb connecting segment ( residues 212–240 ) and for the connection domain . However , information for the connection domain has been particularly limited by the fact that it has not been possible to study it in isolation . The series of studies described here was designed to more fully characterize the transformation from monomer to mature heterodimer . Mutagenesis-based assignments of the isoleucine δ-methyl resonances arising from the connection domains provide a more complete description of the changes taking place in this highly plastic region of the protein . This information also provides insight into the coordinated changes that link conformational maturation of the p66' connection' domain to RH' unfolding . We also report molecular dynamics simulations for some of the early isomerization events not directly accessible to our NMR measurements . Using our recently introduced isomerization-restricted p66 mutant , we also demonstrate subunit-selective labeling , which allows us to the study the conformational maturation of the p66' subunit of RT without additional resonances from the p66 subunit , greatly reducing the resonance overlap problem . Although the selective labeling/NMR detection strategies utilized cannot provide an atomic-level description of the entire conformational maturation process , they provide localized snapshots of the environment of the labeled residues that allow us to evaluate specific models for this process , much as crystal structures provide snapshots corresponding to different stages of an enzyme-catalyzed transformation . These studies provide a more complete description of the complex conformational maturation processes leading to formation of the p66/p51 RT heterodimer . The complexity and degeneracy of the system requires particular attention to the nomenclature required to distinguish between the sequentially identical subunits . The subunit and associated domains that become committed to developing into p51 and the supernumerary RH domain are indicated by primes , for example , p66' , thumb' , RH' , etc . In some instances , we have used the conformation-dependent labeling introduced previously ( Zheng et al . , 2010 ) in which we designate p66M as the monomer conformation; p66E corresponds to the more extended p66 conformation observed in the RT heterodimer; p66C corresponds to the p66 subunit that contains the compact and inactively folded polymerase domain ( p51C ) linked to a separate RH domain . Individual resonances can then be identified as M , E , or C indicating the conformational species to which they correspond . Since the conformation and the associated resonances evolve with time , in a few cases , it was necessary to utilize Ci or Ei for the initially observed resonances associated with the E or C conformations . RT has two functional domains , polymerase and RNase H , with the polymerase made up of fingers , palm , thumb , and connection subdomains . In order to simplify the presentation , the rigorous distinction of domain vs subdomain has been ignored . The basic features of the conformational selection model deduced on the basis of earlier NMR , structural , and kinetic studies ( Venezia et al . , 2009; Braz et al . , 2010; Zheng et al . , 2010 , 2014 ) can be described by the relations given below:[1a]p66M↔domainrearrangementsp66Ei[1b]p66M+p66Ei↔KDp66Ei/p66Ci[1c]p66Ei/p66Ci↔↔→RH′unfoldingp66E/p66C In the above , p66M corresponds to the p66 monomer conformation , p66Ei refers to an initially isomerized structure or ensemble of structures similar to , but not exactly identical with the p66 subunit of RT . The structure of p66Ci is very similar to that of the monomer p66M , probably including only small adjustments , for example , in the β7-β8 loop to facilitate interface formation ( Mulky et al . , 2007 ) . There are subsequently a number of conformational adjustments within the dimer , culminating with irreversible RH' unfolding , that complete the conformational maturation process to produce the mature p66E/p66C homodimer . The p66E/p66C structure is equivalent to an RT heterodimer structure in which all residues on the p66C subunit after ∼430 are disordered , exposing the major proteolysis site as well as additional sites susceptible to HIV-1 PR cleavage . The first two steps of the above process are illustrated schematically in Figure 1 . A key structural feature of the monomer , represented in the upper left hand corner , is the absence of most interface contacts; only the interface between the discontinuous fingers/palm and the connection remains . Thus , the necessary domain rearrangements required for conformational isomerization are more easily accomplished than would be the case if the process began from either the E or C conformational states . The unimolecular isomerization of the p66 monomer depicted in Figure 1 requires only the occasional dissociation of the fingers/palm:connection interface . Another important feature of the initial homodimer is that the inter-subunit RH:thumb' interface is not present . The absence of this interface provides ample room for accommodation of the supernumerary RH' domain that is present in the initial homodimer . A third feature of the process represented in Figure 1 is that the detailed interactions between the two connection domains that are present in the mature RT heterodimer are not yet fully realized in the initial dimer structure . Rather , we suggest that the initial structure is more dependent on non-specific hydrophobic stabilization involving residues on the two connection domains . Since many of the early conformational transitions corresponding to the first two equilibria in Equation 1 are not directly accessible to the NMR methods used in the present study , we utilized our palm loop deletion mutant as well as molecular dynamics simulations to further probe these initial events . 10 . 7554/eLife . 06359 . 003Figure 1 . Schematic diagram showing proposed isomerization and initial p66 homodimer formation . The subunit conformations are color coded ( extended , green; compact , blue ) . Primes are introduced after homodimer formation to allow subunit identification and indicate the subunit destined to be proteolyzed . The palm loop E conformation becomes the primer grip . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 00310 . 7554/eLife . 06359 . 004Figure 1—figure supplement 1 . Ribbon diagram representations of reverse transcriptase ( RT ) monomer and dimer structures . ( A ) The monomer structure of p66 is based on the crystal structure of p51∆PL ( pdb: 4KSE ) and NMR data showing that it also contains a folded ribonuclease H ( RH ) domain linked by residues derived from an unfolded α-helix M . Domains are identified as fingers ( green ) , palm ( blue ) , thumb ( red ) , connection ( magenta ) , and RH ( gray ) . ( B ) Ribbon diagram of the RT heterodimer structure ( pdb: 1S9E , Das et al . , 2004 ) . For this panel , we used an NNRTI complex containing the palm loop , so the position of the p66 thumb domain differs from that in panels C and D . Color coding: p66 subunit: fingers/palm ( red ) , thumb ( orange ) , connection ( pink ) , RH ( magenta ) ; p51 subunit: fingers/palm ( blue ) , thumb ( green ) , connection ( cyan ) , palm loop ( yellow ) . ( C ) Ribbon representation of the p51/p51' homodimer derived from the p66/p51 heterodimer structure ( pdb: 1DLO ) by deletion of the p66 RH domain and replacement of the p51 subunit with the p51∆PL monomer . Color coding: p51 subunit: fingers/palm ( red ) , thumb ( orange ) , connection ( pink ) ; p51' subunit: fingers/palm ( blue ) , thumb ( green ) , connection ( cyan ) . Note that α-helix M' is unfolded in the p51' subunit of the homodimer , as it is in the monomer . ( D ) Initial p66/p66' homodimer structure based on NMR results and modeled from the p66/p51 heterodimer by replacing the p51 subunit with the p51∆PL monomer , and adding an additional , folded RH' domain . The disordered helix M' residues ( 418'–430' ) observed in the crystal structure of p51∆PL have been moved to avoid structural conflict with the p66 subunit and are linked to the supernumerary RH' domain ( purple ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 00410 . 7554/eLife . 06359 . 005Figure 1—figure supplement 2 . Structural comparison of connection domains . ( A ) Overlay of the ribbon diagrams for the connection domains observed in the p66 subunit of RT ( green ) , the p51 subunit of RT ( blue ) , and the monomer ( orange ) . Based on pdb files 1DLO and 4KSE . ( B ) Ribbon diagrams for the RT heterodimer ( p66 , green; RH domain , red; p51 , blue ) , and the connection domain of the p51 monomer ( orange ) in which the p66 connection domain is overlayed with the connection domain of the monomer . As illustrated in B , although an initial domain repositioning of the p66 monomer domains could place the connection domain in position to interact with a second monomer , it would not be in the E conformation characteristic of the mature heterodimer . Formation of additional interfaces within the p66 subunit and with p51 may facilitate the straightening of helix αL . For example , interactions between Glu404 and Lys431 and/or Gln509 on the RH domain may facilitate this conformational change . Stacking of Trp406 with Pro420' can also facilitate the conformational change required for formation of the mature heterodimer . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 00510 . 7554/eLife . 06359 . 006Figure 1—figure supplement 3 . Alternate conformations of helix αM' . The figure shows a ribbon representation of helix αM' from the p51 subunits of two RT structures: pdb: 3QIP ( chain B , green ) and 1SV5 ( chain B , blue ) . Both structures correspond to NNRTI-RT complexes . In 3QIP , helix αM' adopts a more standard geometry , with five strong hydrogen bonds ( ≤3 . 2 Å ) , while the helix in 1SV5 has only one strong hydrogen bond and adopts a more extended conformation in which Tyr427 is at the position of Gln426 in the 3QIP structure . These alternate conformations correspond to an alternate set of interactions within the p51 subunit . A similar conclusion supporting a conformational mixture also results from re-analysis of the electron density for some individual structures ( not shown ) . The ability of the helix to adopt alternate registrations results from the fact that nearly all of the residues are hydrophobic . The ability of αM' to adopt these conformations facilitates transfer of residues from the RH' domain , allowing recruitment of Tyr427' by the connection' domain when thermal fluctuations release it from RH' . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 006 The central question related to the initial dimerization event is whether the major reorganization of the polymerase subdomains that interconverts the two subunits of the RT homodimer occurs prior or subsequent to dimer formation . A prior reorganization leads to a conformational selection model ( Equation 1 ) , while a subsequent reorganization implies an induced fit process that can be described by a version of Equation 2 below:[2a]2p66M↔KDp66M/p66M[2b]p66M/p66M→conformationalmaturationp66E/p66C In order to differentiate between these two models , we utilized a p66 deletion mutant , p66∆PL , lacking palm loop residues 219–230 . The residues deleted in this construct are usually disordered in the C conformation of the polymerase domain but play important structural and functional roles forming the primer grip in the E conformation . Thus , this deletion does not significantly interfere with the monomer ( p66M ) or compact ( p66C ) species but strongly destabilizes the extended ( p66E ) form ( Zheng et al . , 2014 ) . The chromatograms shown in Figure 2 , comparing the behavior of p66 and p66∆PL on a size-exclusion column , demonstrate that under similar conditions , p66 exhibits a ∼75/25 dimer/monomer ratio , while p66∆PL fails to form any observable homodimer . This result follows directly from the conformational selection model , Equation 1 , outlined above , since blocking isomerization will also prevent dimerization . Alternatively , if the conformational maturation of the polymerase domain occurred subsequent to dimer formation as described by Equation 2 , we would expect to observe some dimer species . In principle , the loop deletion might interfere with dimerization by an undetermined mechanism; however , these residues are located just before the thumb domain and are not directly involved in the interface of the mature heterodimer . Thus , the behavior of p66∆PL provides strong support for the conformational selection model . 10 . 7554/eLife . 06359 . 007Figure 2 . Effect of palm loop deletion on dimerization . Gel filtration chromatograms comparing p66 and p66∆PL lacking palm loop residues 219–230 . Chromatogram was obtained at 4°C on a HiLoad 26/60 superdex 200 column for p66 ( black ) and p66∆PL ( red ) eluted with 50 mM Tris–HCl , pH 8 . 0 , 200 mM NaCl . The palm loop deletion , developed to block isomerization , also fails to dimerize . The position of the deleted sequence in p66 is indicated at the bottom of the figure . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 007 The monomer structure provides an intuitive starting point for the spontaneous domain rearrangements that would be required for a conformational selection model . A structural comparison of the fingers/palm in an isolated construct ( RT216 , pdb: 1HAR ) ( Unge et al . , 1994 ) , the p51∆PL monomer ( pdb: 4KSE ) ( Zheng et al . , 2014 ) , the p51 and p66 subunits of RT ( pdb: 1DLO ) ( Hsiou et al . , 1996 ) reveals significant differences . This variation is most conveniently characterized by the angle between the approximately coplanar helices A ( residues 28–43 ) in the fingers and F ( residues 194–211 ) in the palm ( Figure 3 ) . In both the monomer and the p51 subunit of RT , this angle is ∼45° . By comparison , in the isolated RT216 construct or the p66 subunit of RT , the angle is more obtuse , with values of 90°–100° . Importantly , in both of the structures with the more acute angle , there is a large interface between the fingers/palm and the connection domains . In contrast , for both of the structures lacking this interface , the angle defined by helices A–F is much more open . This correlation suggests that the conformation with the more acute fingers/palm angle may be stabilized by the inter-domain interactions between the fingers/palm and the connection domains . 10 . 7554/eLife . 06359 . 008Figure 3 . Alternative conformations and molecular dynamic simulations analysis of the fingers/palm subdomains . ( A ) Overlay of ribbon diagrams for fingers/palm residues 1–216 RT216 ( pdb: 1HAR , gray ) and in the p66 subunit of RT ( pdb: 1DLO , fingers , teal; palm , orange ) . ( B ) Overlay of ribbon diagrams for the fingers/palm in the p51∆PL monomer ( pdb: 4KSE , gray ) with the corresponding region of the p51 subunit of RT ( pdb: 1DLO , fingers , teal; palm , orange ) . The fingers/palm angle defined by helices A and F is indicated , illustrating the more acute values for the monomer and the p51 subunit , compared with an isolated fingers/palm construct and the p66 subunit . ( C ) Time-dependent molecular dynamics simulations of the behavior of the αAF angle for the fingers/palm starting with the p66 conformation ( red ) or with the p51 conformation ( black ) . The simulations utilized residues 1–236 in the p66 and p51 subunits of RT ( pdb: 1DLO ) after removing all other domains at t = 0 , and the missing palm loop residues in the p51 starting structure were introduced as indicated in ‘Materials and methods’ . Residues included in the simulations are defined in the inset . The cartoons on the left illustrate the starting fingers/palm conformations and the proposed role of the fingers/palm:connection interface in constraining the initial αAF angle in the monomer and p51 structures . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 00810 . 7554/eLife . 06359 . 009Figure 3—figure supplement 1 . Additional simulations starting from the p51 monomer and from a structure that includes the p51 palm loop . Additional time-dependent molecular dynamics simulations of the isolated fingers/palm domains starting from a structure that includes palm loop residues 219–230 ( 1S9E , green ) , the p51∆PL monomer in which palm loop residues from the 1DLO p66 subunit have been added back ( 4KSE , red ) , or the p51∆PL monomer taken to include the first 236 residues ( corresponding to residues 1–249 , since the construct is missing residues 219–230 ) ( 4KSE , black ) . In all cases , the palm loop moves from a structure with a more acute bend toward the more extended conformation characterized by an αA-αF angle of ∼ 90° , with the main differences related to the time at which this occurs . The transition took longer to occur with the deletion construct . We note , however , that the segment from 218 to 236 does not attain its conformation in the p66 subunit of RT in any of the simulations . Most probably , this results from the artificiality of terminating the sequence at Pro236 and the absence in the simulations of additional inter-domain interactions . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 009 The above hypothesis was evaluated by performing molecular dynamics simulations on the isolated fingers/palm domains . Starting structures included residues 1–236 for the p66 and p51 subunits of apo RT ( pdb: 1DLO ) . As discussed in ‘Materials and methods’ , the missing p51 segment from 219–230 was modeled by introducing the corresponding segment from the p66 subunit . The fingers/palm αAF angle was determined as a function of time after removal of all other domains . Simulations for the isolated fingers/palm starting with either the p66 or the p51 conformations are shown in Figure 3C . As indicated in the figure , the more open conformation present in the p66 subunit ( red line ) is stable over the time period of the simulation . Alternatively , the simulation beginning with the fingers/palm in the p51 subunit ( black line ) indicates that between ∼35 and 45 ns the αAF angle undergoes a transition from its initial acute value to ∼90° . This result is consistent with an intrinsic preference for the open conformation observed in the crystal structure of the isolated fingers/palm construct , RT216 ( pdb: 1HAR ) . Analogous simulations starting with the monomer structure ( pdb: 4KSE ) or with the p51 subunit of an RT-inhibitor complex containing the missing loop residues ( pdb: 1S9E ) produced qualitatively similar results , with the most significant variation related to the time at which the transition to the more extended conformation occurs ( Figure 3—figure supplement 1 ) . The strategy of utilizing an isolated fingers/palm construct to reveal the intrinsic domain orientation preference is , however , subject to the limitation that inter-domain interactions involving the thumb and connection domains are omitted . Thus , although the final fingers/palm conformation produced by the simulations is similar to that observed in the p66 subunit of RT , the conformations of residues located at the domain boundaries do not agree with those in p66 . The simulations are thus consistent with the general conclusion that the conformations of domain boundary residues depend on inter-domain interactions . This conclusion applies to residues in the palm loop , which fail to form strands of the larger β-sheet formed from palm and thumb residues . Once the fingers/palm:connection subdomains have dissociated , palm loop residues can be recruited to cover exposed hydrophobic patches in the palm domain . The large fingers/palm:connection interface of ∼1470 Å2 in the p51 subunit of RT includes extensive hydrophobic contacts ( Ding et al . , 1994 ) . In the monomer , these contacts include palm residues Leu100 , Val106 , Val108 , Tyr181 , Tyr188 , and Leu234 ( Figure 4A ) . In the active , p66E subunit of RT , this same group of hydrophobic residues in the palm domain interacts directly with residues from the palm loop ( Figure 4B ) , which in the E conformation become part of the functionally important primer grip that positions the primer terminus for catalysis ( Ghosh et al . , 1996 ) . Formation of alternate , intra-domain hydrophobic contacts by residues of the palm loop/primer grip can thus tend to interfere with re-association of the fingers/palm and connection domains , thereby enhancing the availability of the connection domain for intermolecular association with the monomer ( Figure 4C ) . Further , these residues also form part of the binding site for non-nucleoside reverse transcriptase inhibitors ( NNRTIs ) . This is a highly flexible region of the protein in which the NNRTI binding site is not identifiable in the absence of a bound inhibitor , and hence , is likely to be able to rapidly form intra-domain hydrophobic contacts that can inhibit connection domain re-association . 10 . 7554/eLife . 06359 . 010Figure 4 . Role of the palm loop in isomerization of the polymerase domain . ( A ) Ribbon diagram of the p51∆PL monomer ( pdb: 4KSE , green ) with the connection domain shown in orange . Several hydrophobic residues in the palm—Leu100 , val106 , Val108 , Tyr181 , Tyr188 , and Leu234 that interact with the connection domain are annotated . ( B ) Ribbon diagram of the p66 subunit of RT ( pdb: 1DLO ) showing a portion of the fingers/palm domains ( green ) interacting with palm loop residues ( 219–230 , magenta ) of the palm domain . In the p66 subunit ( E conformation ) , the palm loop becomes the primer grip and interacts with many of the same hydrophobic residues that interact with the connection domain in the monomer . ( C ) Schematic diagram illustrating how the intrinsic preference of the fingers palm for a more open conformation facilitates disruption of the fingers/palm:connection interface and repositioning of the palm loop . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 010 To summarize , structural data and molecular modeling simulations indicate that the more bent conformation of the fingers/palm present in the monomer structure does not represent a local minimum for the isolated fingers/palm , but a global minimum for the fingers/palm:connection complex . The fingers/palm apparently has an intrinsic preference for a more extended conformation that probably helps to promote dissociation of the fingers/palm:connection interface ( Figure 4C ) . The inherent flexibility of the palm loop segment is expected to facilitate initial formation of intra-domain hydrophobic contacts that compete with inter-domain palm:connection interactions , reducing the tendency for re-association with the connection domain , and enhancing connection domain availability for dimerization . We previously presented data indicating that the p51/p51' homodimer formed by the isolated polymerase domain exists as a conformational heterodimer ( Zheng et al . , 2010 ) , a result consistent with its demonstrated polymerase activity ( Bavand et al . , 1993; Dufour et al . , 1998 ) . The scheme shown in Figure 1 predicts that the initially formed p66/p66' homodimer should resemble the p51/p51' homodimer both of which lack the inter-subunit RH:thumb' interface . To the extent that this analogy holds , the p51/p51' homodimer should provide a stable model for the transiently formed initial p66/p66' homodimer . A comparison of the 1H-13C heteronuclear multiple-quantum correlation ( HMQC ) spectra obtained for the Ile-labeled p51 monomer and the p51/p51' homodimer obtained under high salt conditions ( Figure 5A , B ) provides unequivocal evidence indicating conformational heterogeneity of the two subunits of the homodimer . In Figure 5B ( see also Figure 5—figure supplement 1 ) , the spectrum of the p51/p51' homodimer ( magenta ) is overlaid with the spectra for p66-labeled RT ( green ) and p51-labeled RT ( blue ) . The overlay demonstrates that the spectrum of p51/p51' contains multiple resonances that are nearly coincident with resonances from both the p66 and the p51 subunits of RT . Thus , the p51/p51' homodimer exists as a conformational heterodimer that is structurally similar to the RT heterodimer and contains both E-like and C-like conformations . 10 . 7554/eLife . 06359 . 011Figure 5 . Spectral comparisons of p51/p51' and p66/p66' homodimers . ( A ) 1H-13C heteronuclear multiple-quantum correlation ( HMQC ) spectrum of the [13CH3-Ile]p51 monomer . ( B ) Overlaid HMQC spectra of the [13CH3-Ile]p51/[13CH3-Ile]p51' homodimer with the spectra for [13CH3-Ile]p66/p51 ( green ) and p66/[13CH3-Ile]p51 ( blue ) . We note the absence of homodimer resonances that overlay the resolved RH domain resonances in p66-labeled RT . ( C ) Overlaid HMQC spectra for the labeled p51 homodimer and the Ile-labeled p66/p66' homodimer obtained during the first 5 . 5-hr accumulation period after initiation of dimerization . The p66 homodimerization studies were performed in 25 mM Tris-HCl-d11 in D2O , pD = 7 . 51 , 100 mM KCl , 0 . 02% NaN3 . In order to stabilize the p51/p51' homodimer , it was necessary to use a high salt buffer containing 800 mM KCl and 20 mM MgCl2 in addition to the other components . The labeling pattern corresponds to the color coding in the cartoons near each spectrum , with white indicating an unlabeled subunit . The assignment in parenthesis is considered tentative . The RH , RH' , and Th' labels in the cartoon indicate the RNase H domain in the p66 subunit , the RNase H domain in the p66' subunit , and the Thumb' domain in the p51' or p66' subunits . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 01110 . 7554/eLife . 06359 . 012Figure 5—figure supplement 1 . Spectral comparison of the p51/p51' and initial p66/p66' homodimers . Direct comparison of the 1H-13C HMQC spectra of [13CH3-Ile]p51/[13CH3-Ile]p51' and [13CH3-Ile]p66/[13CH3-Ile]p66' obtained during the first 5 . 5-hr accumulation period after dimerization conditions were introduced . The p51 spectrum was obtained in D2O buffer , 25 mM Tris-d11 , pD 7 . 5 , 0 . 02 % NaN3 that also contained 800 mM KCl and 20 mM MgCl2 in order to convert most of the p51 to the dimeric form . Some monomer resonances are apparent in both spectra , however , in the p66/p66' sample , these resonances decay , while in the p51/p51' spectrum they are constant . A resonance labeled [375M] likely arises from Ile375 in the monomer , however , the shift at high salt differs from the shift observed in the monomer , so this assignment is only suggested . The spectra are color coded as indicated by the cartoons under each spectrum; RH , RH' , and Th' refer to the RNase H domain in p66 , the RNase H domain in p66' , and the Thumb' domain in p66' , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 01210 . 7554/eLife . 06359 . 013Figure 5—figure supplement 2 . Spectral comparison of the initial p66/p66' homodimer with the selectively-labeled subunits of RT . Comparison of the initial 1H-13C HMQC spectrum obtained for [13CH3-Ile]p66/[13CH3-Ile]p66' obtained during the first accumulation period after dimerization conditions are introduced , with the spectra of [13CH3-Ile]p66/p51 ( green ) and p66/[13CH3-Ile]p51 ( blue ) . Note that in all cases , resonances in the initial homodimer spectrum agree with resonances in the p66-labeled RT . Alternatively , this is not the case for the p66' resonances , for which there is good overlap of some resonances , for example , Ile47C and Ile202C , but weak or missing resonances for other residues , for example , Ile274C , Ile329C , and Ile375C . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 01310 . 7554/eLife . 06359 . 014Figure 5—figure supplement 3 . Assignments of connection domain resonances in the p66 subunit of RT . 1H-13C HMQC spectrum of [13CH3-Ile]p66/p51 . Resonances with new assignments are indicated ( in red ) . We have reassigned Ile375 ( which had been mistakenly attributed to Ile329 ) , Ile382 ( which had been attributed to Ile411 ) , and Ile522 ( which had been attributed to Ile526 ) . In each of these cases , the misassigned resonances corresponded to residues positioned very near the mutated residues . We were unable to identify spectral changes in the Ile411V mutant , possibly due to significant broadening . Resonances for Ile380 and Ile542 are either poorly resolved and/or subject to extreme broadening . The assignment of Ile393 is indicated in parenthesis , since the resonance is relatively weak and not well resolved and is thus considered tentative . As noted in the text , the additional assignments are useful for the studies reported here , but were not utilized in our previous analysis , and the reassignments do not alter any of the conclusions previously presented . A table of mutated residues and some of the spectra obtained for the individual mutations are shown in the following figures . The cartoon on the upper left indicates the labeled subunit . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 01410 . 7554/eLife . 06359 . 015Figure 5—figure supplement 4 . Table of mutated residues . In contrast with the assignment procedures used for the isolated fingers/palm ( RT216 ) , thumb , and RH domains , we were unable to express a stable connection domain on which to perform NMR assignment studies . We thus utilized extensive site-directed mutagenesis as the basis for making these assignments in the p66 subunit of the RT heterodimer . Although this method provided a basis for many of the assignments , it also proved problematical for assigning some of the closely positioned methyl resonances such as those arising from Ile329 and Ile375 . Overall , in addition to the Ile → Val mutants , we also utilized several ‘nudge’ mutations , which helped to resolve assignment ambiguities . These included S379C ( to assign Ile341 ) , Y342H ( to assign Ile326 ) , and H361Y ( to assign RH domain resonances Ile522 , Ile505 , and Ile526 located near the connection:RH domain interface . Ultimately , the strategy used was aimed at obtaining a set of self-consistent assignments allowing us to overcome the limitations of individual substitutions , which in a few cases led to ambiguous results . The positions of the additional assigned resonances are indicated in Figure 5—figure supplement 3 , corresponding to [13CH3-Ile]p66/p51 . Some of the spectra obtained for the mutants in the table are shown in the following Figure 5 supplements . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 01510 . 7554/eLife . 06359 . 016Figure 5—figure supplement 5 . Resonance perturbations in [13CH3-Ile]p66 ( I341V ) /p51 . The 1H-13C HMQC spectrum of [13CH3-Ile]p66 ( I341V ) /p51 ( red ) is overlayed with spectrum for [13CH3-Ile]p66/p51 ( green ) . Perturbed resonances are annotated in red . The I341V substitution eliminates the Ile341 resonance and perturbs the shifts of Ile382 and Ile375 . The δ-methyl-δ-methyl distances relative to the mutated residue based on structure 1DLO are also shown . The four most strongly perturbed resonances are the closest in the structure . Additional resonances annotated in black are included to facilitate spectral comparisons . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 01610 . 7554/eLife . 06359 . 017Figure 5—figure supplement 6 . Resonance perturbations in [13CH3-Ile]p66 ( I382V ) /p51 . Overlay of the 1H-13C HMQC spectra of [13CH3-Ile]p66 ( I382V ) /p51 ( red ) and [13CH3-Ile]p66/p51 ( green ) . The I382V substitution eliminates the Ile382 resonance and perturbs the shifts of the resonances indicated . The δ-methyl-δ-methyl distances of the perturbed residues relative to the mutated residue are also shown . The small magnitude of the Ile341 perturbation is surprising , but not necessarily unreasonable . If the mutated and observed residues are not in direct contact , the longer range perturbation will depend on the structural effect of the residue change . We note as well that resonances at the base of the thumb , such as Ile270 , show extreme shift sensitivity to multiple mutations . This may result from the inherent structural flexibility of this region of the protein , also indicated by the binding of non-nucleoside reverse transcriptase inhibitors . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 01710 . 7554/eLife . 06359 . 018Figure 5—figure supplement 7 . Resonance perturbations in [13CH3-Ile]p66 ( I270V ) /p51 . Overlay of the 1H-13C HMQC spectra of [13CH3-Ile]p66 ( I270V ) /p51 ( red ) and [13CH3-Ile]p66/p51 ( green ) . In addition to the disappearance of the Ile270 resonance , the I270V substitution strongly perturbs Ile94 ( 5 . 1 Å ) and weakly perturbs Ile274 ( 8 . 1 Å ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 01810 . 7554/eLife . 06359 . 019Figure 5—figure supplement 8 . Resonance perturbations in [13CH3-Ile]p66 ( Y342H ) /p51 . Overlay of the 1H-13C HMQC spectra of [13CH3-Ile]p66 ( Y342H ) /p51 ( red ) and [13CH3-Ile]p66/p51 ( green ) . The Y342H substitution was introduced following a nudge mutation strategy in order to assign Ile326 and perhaps other nearby residues ( see structural inset ) . The perturbed residues are annotated ( in red ) , and their position relative to residue Tyr342 is shown in the inset . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 01910 . 7554/eLife . 06359 . 020Figure 5—figure supplement 9 . Resonance perturbations in [13CH3-Ile]p66 ( I167V ) /p51 . Overlay of the 1H-13C HMQC spectra of [13CH3-Ile]p66 ( I167V ) /p51 ( red ) and [13CH3-Ile]p66/p51 ( green ) . The I167V substitution eliminates the Ile167 resonance and mildly perturbs Ile5 . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 02010 . 7554/eLife . 06359 . 021Figure 5—figure supplement 10 . Resonance perturbations in [13CH3-Ile]p66 ( I526V ) /p51 . Overlay of the 1H-13C HMQC spectra of [13CH3-Ile]p66 ( I526V ) /p51 ( red ) and [13CH3-Ile]p66/p51 ( green ) . The I526V substitution eliminates the Ile526 resonance and also significantly broadens the resonances of nearby Ile 522 ( 6 . 8 Å ) and Ile505 ( 15 . 4 Å ) . This effect may be due to destabilization of the connection:RH interface , since these residues are located near this interface . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 02110 . 7554/eLife . 06359 . 022Figure 5—figure supplement 11 . Resonance perturbations in [13CH3-Ile]p66 ( I522V ) /p51 . Overlay of the 1H-13C HMQC spectra of [13CH3-Ile]p66 ( I522V ) /p51 ( red ) and [13CH3-Ile]p66/p51 ( green ) . The I522V substitution eliminates the Ile522 resonance and selectively shifts Ile526 ( 6 . 8 Å ) , Ile521 ( 7 . 2 Å ) , and Ile505 ( 8 . 7 Å ) . This effect may result in part from destabilization of the connection:RH interface , since these residues are located near this interface . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 02210 . 7554/eLife . 06359 . 023Figure 5—figure supplement 12 . Resonance perturbations in [13CH3-Ile]p66 ( H361Y ) /p51 . Overlay of the 1H-13C HMQC spectra of [13CH3-Ile]p66 ( H361Y ) /p51 ( red ) and [13CH3-Ile]p66/p51 ( green ) . The connection domain H361Y nudge mutation allows further assignment of residues located at the RH interface . A schematic figure showing the relative positions of H361 , Ile505 , Ile522 , and Ile526 is shown in the lower right hand corner of the spectrum . Ile522 , located 3 . 5 Å from His361 imidazole C1 and N2 , is not observed , due either to broadening or to a shift to the densely populated spectral region , and Ile526 and Ile521 exhibit small shift perturbations . We note further that the assignment of Ile505 is consistent with the perturbation of this resonance observed previously in the presence of an active site RH domain ligand ( see Figure 4b of Zheng et al . , 2012 ) , and its relative proximity to the RH active site , as well as with the upfield shift of this resonance predicted by SHIFTX analysis . Not unexpectedly , the resonance perturbations resulting from the H361Y mutation are very similar ( although not identical ) to the perturbations resulting from the I522V mutation , shown in the previous supplemental figure . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 02310 . 7554/eLife . 06359 . 024Figure 5—figure supplement 13 . Resonance perturbations in [13CH3-Ile]p66 ( S379C ) /p51 . Overlay of the 1H-13C HMQC spectra of [13CH3-Ile]p66 ( S379C ) /p51 ( red ) and [13CH3-Ile]p66/p51 ( green ) . The connection domain S379C nudge mutation helps to resolve the assignments of Ile375 and Ile329 and also supports assignments of Ile341 and Ile382 . The distances , calculated using structure 1DLO , correspond to the oxygen sidechain in Ser379 and the δ-CH3 carbon in each Ile residue . The perturbation of the Ile270 resonance is most probably due to a conformational effect . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 02410 . 7554/eLife . 06359 . 025Figure 5—figure supplement 14 . Resonance perturbations in [13CH3-Ile]p66 ( I375V ) /p51 . Overlay of the 1H-13C HMQC spectra of [13CH3-Ile]p66 ( I375V ) /p51 ( red ) and [13CH3-Ile]p66/p51 ( green ) . The I375V mutation eliminated two connection domain resonances , consistent with the 3 . 4 Å separation of the δ-methyl groups of Ile375 and Ile329 , indicating that Ile329 also experiences a major shift perturbation . We have now been able to make more specific assignments to Ile375 and Ile329 based on the effects of other nearby mutations , particularly I341V ( Figure 5—figure supplement 5 ) , which perturbs the closer Ile375 resonance without affecting the Ile329 resonance , as well as the relative perturbations produced by S379C ( Figure 5—figure supplement 13 ) . This corrects the previous incorrect assignment of Ile375 . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 02510 . 7554/eLife . 06359 . 026Figure 5—figure supplement 15 . 1H-13C HMQC spectrum of [13CH3-Ile]p66 ( I375V ) / [13CH3-Ile]p66 ( I375V ) ' mature homodimer . In the study shown above , the labeled p66 ( I375V ) mutant , observed as a heterodimer in Figure 5—figure supplement 14 , was converted into a labeled p66 ( I375V ) /p66 ( I375V ) ' homodimer , rather than a heterodimer , and allowed sufficient time to mature ( red spectrum ) . The spectrum was then overlayed with that of the mature , Ile-labeled wt p66/p66' homodimer ( green ) . As is apparent from the figure , two new sets of resonances become evident in the wt homodimer that are absent from the mutant homodimer , which then can be assigned to the same Ile375 and Ile329 residues in the p66' subunit . Analysis of these residues is particularly useful based on their isolation and ease of identification . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 026 A more complete analysis of the 1H-13C HMQC spectrum of the Ile-labeled p51/p51' dimer indicates that it contains resonances that are in close agreement with resonances from the fingers , palm , thumb , and connection domain of the p66 subunit , while lacking resonances attributable to the RH domain . We , thus , conclude that the conformation of the p51 subunit of the p51/p51' homodimer can be characterized as adopting a p51E-like conformation , that is , similar to the p66 conformation of the heterodimer but lacking an RH domain . In contrast , the conformation of the p51' subunit of the homodimer is more difficult to characterize . In some cases , for example , Ile202' and Ile47' , the resonances are in close agreement with those of the p51C subunit of the RT heterodimer ( blue spectrum ) , while in other cases , for example , Ile274' , Ile329' , and Ile375' , resonances near the positions expected for p51C are not observed ( Figure 5B ) . The resonance of Ile274' from the thumb' domain is at the position of p51M rather than p51C , and the connection' Ile329' and Ile375' resonances are not readily observed , as is the case with the monomer . This behavior indicates that formation of the p51/p51' dimer leads to shifts in the fingers'/palm' that are consistent with dimer formation , while several of the p51' thumb' and connection' domain resonances more closely approximate the pattern of the p51 monomer . We conclude that the NMR data support a homodimer model in which the p51 subunit approximates the p66 RT subunit without an RH domain , while the p51' subunit conformation approximates that of the p51 monomer that includes a disordered thumb' and disordered C-terminal αM' residues ( Figure 5B schematic and Figure 1—figure supplement 1C ) . Apparently , the interactions between helix αM' and the thumb' are insufficient to stabilize a p51C conformation similar to that observed in the heterodimer , indicating that the additional interactions with the RH domain are required for this conformation to be significantly populated . Extensive similarities are observed in an overlay of the 1H-13C HMQC spectrum of the Ile-labeled p51/p51' with the spectrum of the p66/p66' homodimer obtained during the first 5 . 5-hr accumulation period after initiation of dimerization ( Figure 5C ) . The p66/p66' spectrum further demonstrates even closer agreement with the spectra of the RT heterodimer ( Figure 5—figure supplement 2 ) , demonstrating the presence of E-like and C-like conformers . This result is in direct conflict with the model recently proposed by Sharaf et al . ( 2014 ) in which the initial p66 homodimer observed by NMR exists as a conformationally symmetric homodimer . Consistent with our previous study ( Zheng et al . , 2014 ) , resolved RH domain resonances indicate that the early p66/p66' homodimer contains two-folded RH domains , one of which exhibits a shift pattern similar to that of the isolated subunit . This behavior is most readily observed for the isolated Ile434 resonances , and considered in greater detail in the following sections . As outlined in Figure 1 ( see also Figure 1—figure supplement 1 ) , the initially formed homodimer lacks an RH:thumb' interface . In order to more directly address the question of when this interface is formed , it was first necessary to determine how interface formation affects the isoleucine resonances in the p66 RH domain , and particularly the shift of Ile434 , which is located in p66 RH near the RH:thumb' interface . The strategy presented below compares the Ile shifts in the p66 subunit of the wt RT heterodimer with the shifts in a mutant heterodimer containing a p51 thumb' mutation positioned at the interface with RH . Specifically , residue Leu289 on the p51 subunit interacts with a hydrophobic pocket on the p66 RH domain , so the non-conservative p51 ( L289K ) mutant should significantly disrupt the structure of this interface in the p66/p51 ( L289K ) heterodimer . A comparison of the NMR spectra obtained for [13CH3-Ile]p66/p51 ( L289K ) with the spectrum obtained for the non-mutated protein will reveal the shift perturbations that result from interface formation . In Figure 6A , we compare the 1H-13C HMQC spectrum of the p51-mutated , p66-labeled heterodimer , [13CH3-Ile]p66/p51 ( L289K ) , with the spectrum obtained for the p66-labeled heterodimer lacking the p51 thumb mutation . In order to overcome the reduced tendency of mutated p51 to dimerize ( Goel et al . , 1993; Zheng et al . , 2010 ) , we utilized a twofold excess of unlabeled p51 ( L289K ) to enhance dimer formation with labeled p66 . The spectra in Figure 6A demonstrate that this strategy was successful; the resonance pattern observed for [13CH3-Ile]p66/p51 ( L289K ) is qualitatively similar to that obtained for wt RT labeled in the p66 subunit ( Zheng et al . , 2014 ) , while resonances with shifts that are characteristic of the p66 monomer , for example , Ile393M and Ile274M are very weak . In addition , resonances characteristic of the p66C conformation of the homodimer , for example , Ile202C and Ile47C , which would be present if p66/p66' containing labeled Ile in both subunits was present , are weak or absent . Thus , nearly all of the label has ended up in the p66E subunit of the RT heterodimer , [13CH3-Ile]p66/p51 ( L289K ) , rather than in a p66 monomer or a p66/p66' homodimer . 10 . 7554/eLife . 06359 . 027Figure 6 . Effect of a p51 thumb' domain mutation on the Ile methyl resonances in the p66 subunit of RT . ( A ) Overlay of the 1H-13C HMQC spectrum of [13CH3-Ile]p66/p51 ( green ) and [13CH3-Ile]p66/p51 ( L289K ) ( black ) . Most of the features of the spectrum are preserved , consistent with the formation of a stable heterodimer . The labeling pattern corresponds to the color-coding in the cartoons below the spectrum . ( B ) Ribbon diagram illustrating the relative position of the mutated residue ( blue sphere ) and perturbed resonances in the RH and connection domains of p66 ( orange spheres ) . Color coding: p51 ( blue ) ; p66 RH domain ( magenta ) ; p66 connection domain ( yellow ) , p66 fingers/palm and thumb domains ( green ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 027 Disruption of the p51 thumb':p66 RH interface by the p51 ( L289K ) mutation alters many of the shifts within the p66 RH domain . The shift differences of the resolved Ile434 and Ile495 resonances that are characteristic of the RT-incorporated RH domain are eliminated . Thus , the 13C shifts of Ile434 ( 16 . 6 ppm ) and Ile495 ( 13 . 7 ppm ) observed in [13CH3-Ile]p66/p51 ( L289K ) are similar to the values in the isolated RH domain , but differ from the values of 17 . 0 and 13 . 6 ppm observed in the wt RT heterodimer . The shift differences summarized above , thus , allow us to determine at what point in the maturation process the RH:thumb' interface is formed . A 13C shift of 17 . 0 ppm for the Ile434 resonance indicates that the inter-subunit RH:thumb' interface has formed , while a shift of ∼16 . 6 ppm , similar to that of the isolated RH domain and also observed in the mutant heterodimer discussed above , indicates that this interface is not present or not well-formed . Since we observe strong intensity for the Ile434 resonance at 17 . 0 ppm during the first 5 . 5-hr accumulation period after conditions favoring the homodimer are introduced ( Figure 5C ) , we conclude that the thumb':RH interface has largely been formed during this initial period . A similar conclusion follows from analysis of the shifts of the Ile495 resonance . Interestingly , the 1H-13C HMQC spectrum of [13CH3-Ile]p66/p51 ( L289K ) exhibits multiple additional shift perturbations of other RH and connection domain resonances . Methyl resonances of Ile522 , located at the connection:RH interface , resonances of Ile329 , Ile375 , and Ile382 located within the p66 connection domain experience significant broadening , and the Ile341 resonance is shifted . Locations of the p66 Ile residues exhibiting these perturbations are illustrated in Figure 6B . Since as shown above , the L289K' perturbation is insufficient to prevent heterodimer formation , we conclude that perturbation of the thumb':RH interface with the L289K' mutation introduces additional perturbations that extend into the RH and connection domains of p66 . These observations highlight the cooperative nature of interface formation in RT . The model shown in Figure 1 describes a conformational selection process in which the predominant monomer ‘selects’ a structurally isomerized p66 molecule in a rare , p66Ei conformation as its initial binding partner . Initial dimer formation probably involves non-specific hydrophobic contacts between the connection domains . A comparison of the connection domains in the monomer , the p66 subunit , and the p51 subunit of RT reveals significant structural variations , particularly in regions involved in interface formation ( Figure 1—figure supplement 2 ) , so that a simple rearrangement of domain positions is insufficient to result in formation of an interface similar to that of the mature heterodimer; additional conformational changes within the connection domain are also required . This requirement is most clearly apparent from an overlay of the connection domain in the monomer with the connection domain on the p66 subunit of the heterodimer ( Figure 1—figure supplement 2 ) . Among the various structural changes that must occur , straightening of helix αL in the E conformation alters multiple intra- and inter-domain contacts facilitating inter-subunit interface formation . Consequently , initial dimer formation involving the connection domains prior to this conformational change must include many non-specific hydrophobic contacts . A comparison of the 1H-13C HMQC spectra obtained for the initial [13CH3-Ile]p66/[13CH3-Ile]p66' homodimer with the spectra obtained for the subunit-labeled heterodimer ( Figure 5—figure supplement 2 ) indicates that all of the resolved Ile resonances of residues in the extended p66 conformation , for example , Ile329E , Ile341E , Ile375E , and Ile382E are readily observed . These resonances characteristic of the connection domain in the E conformation are not present in the monomer or in the spectra of labeled p66C . Thus , the conformational changes required to alter the connection domain from its monomer to its p66E conformation have largely been completed during the first accumulation period . In addition to the connection domain resonances , resonances attributed to residues in the fingers ( Ile47 ) , palm ( Ile202 ) , thumb ( Ile274 ) , and RH domain ( Ile434 ) also are in agreement with resonances in the p66-labeled RT spectrum ( Figure 5—figure supplement 2 ) . These observations are consistent with the results summarized in the previous section , indicating that the RH:thumb' interface has largely been formed during the first accumulation period . Time-dependent intensity data for connection domain resonances assigned to Ile329E , Ile375E , and Ile382E , summarized in Table 1 , give time constants of ∼2–3 hr , shorter than the 5 . 5-hr accumulation used for the first spectrum . Thus , the p66 subunit has evolved from an initial conformation involving non-specific hydrophobic contacts to a form that closely approximates its mature , p66E conformation during the initial accumulation period . 10 . 7554/eLife . 06359 . 028Table 1 . Apparent time constants—homodimerization studyDOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 028ResidueMean ± S . E . 329E3 . 3 ± 0 . 5375E2 . 6 ± 0 . 7382E2 . 4 ± 0 . 2329C5 . 8 ± 0 . 3375Cb5 . 9274C8 . 9 ± 0 . 6Fitted parameters are averages ±standard error for three separate studies . bFor Ile375C , one data set was obscured by a spectral artifact , so the tabulated value is the average of two measurements . Illustrative data fits of individual data sets are shown in Figure 7—figure supplement 1 . In contrast with the behavior of the p66 subunit summarized above , resonances arising from p66' support a more complex interpretation . Fingers/palm resonances Ile47' and Ile202' are at the expected Ile47C and Ile202C positions characteristic of the mature dimer . For Ile47' , there is a significant shift difference between the monomer and the dimer , so that this result supports the conclusion that the region of the interface near Ile47' is structurally similar to that of the mature heterodimer . In contrast , connection' domain resonances Ile329' and Ile375' are weak or absent , that is , more similar to their behavior in the monomer . We have assigned two resonances to Ile274': a more intense peak with a shift close to the monomer ( Ile274Ci ) and a second weaker peak with a shift close to position of the mature heterodimer ( Ile274C ) . Based on intensity comparisons with the Ile393M resonance , the Ile274Ci peak is attributed mostly to an immature dimer species with a monomer-like shift , while the weaker Ile274C resonance is attributed to the p66' subunit of the conformationally mature p66/p66' homodimer . Importantly , the evidence outlined in the previous section indicates that the thumb':RH interface is largely formed during the first accumulation period; however , the Ile274' resonance is mostly at the monomer position in the first p66/p66' spectrum . This difference may indicate that the base of the thumb’ undergoes a slow conformational maturation process that is separate from formation of the thumb':RH interface . Alternatively , Ile274' is sufficiently close to the connection' domain so that its time-dependent shift behavior may be sensitive to changes that are occurring in the connection' domain , and particularly to the formation of helix αM' . Ile residues located at or near the subunit interface include: Ile159 , Ile380 , Ile382 , Ile411 , and Ile542 on p66 , and Ile135 on p66' . However , due to broadening and/or resolution limitations , only Ile382 provides a useful probe for dimer formation ( Figures 5B , C and 6 ) . In the heterodimer structure , pdb: 1DLO ( Hsiou et al . , 1996 ) , the Ile382 δ-methyl is positioned 5 . 6 Å from the sidechain carbonyl oxygen of Asn136 on the p51 subunit . The Asn136 residue on p51 and the loop containing it have been shown to play an important role in dimerization ( Balzarini et al . , 2005; Mulky et al . , 2007; Upadhyay et al . , 2010 ) . Based on the behavior of the Ile382E resonance , this interface is formed at a sufficiently early stage so that it is largely present during the first 2D 1H-13C HMQC accumulation period of 5 . 5 hr . Analysis of the time-dependent data gave a time constant of 2 . 4 ± 0 . 2 hr ( Table 1 ) , consistent with a relatively early formation of this portion of the interface involving the connection and fingers' domains . This conclusion also follows from the time-dependent behavior of the Ile47' resonance discussed above . In summary , dimerization is occurring on a scale too rapid for direct NMR observation , however , comparisons of resonance shifts with values in the monomer and heterodimer , as well as structural comparisons with the monomer , indicate that several conformational steps are largely completed during the initial accumulation period . These include maturation of the dimer interface so that the p66 connection domain matures from its monomer to its extended ( E ) conformation and formation of the RH:thumb' domain interface . Maturation of the connection' proceeds on a slower time scale . We previously proposed that the supernumerary RH' domain initially present in the p66' subunit of the homodimer is destabilized and unfolds as a result of transfer of residues near Tyr427' that develop into helix αM' in the connection' domain of the mature p66' subunit . This model was supported by the decay of several resonances that could be assigned specifically to the RH' domain ( Zheng et al . , 2014 ) . The more complete assignments of the connection domain included with the present study ( Figure 5—figure supplements 3–15 ) provide further substantiation of this hypothesis . The Ile329 and Ile375 resonances are particularly useful for analysis of connection domain conformational processes since they are well resolved and give unique signals characteristic of the E and C conformations . These resonances are also not readily observed in the monomers , probably as a result of exchange broadening ( although a broad resonance in the general region of Ile375 may correspond to this residue ) . The I375V mutation eliminates both the Ile375 and Ile329 resonances as a consequence of the proximity of these two residues ( δCH3 ( Ile329 ) -δCH3 ( Ile375 ) = 3 . 4 Å in 1DLO ) ( Figure 5—figure supplement 14 ) . The spectrum of the mature p66 ( I375V ) /p66' ( I375'V ) homodimer ( Figure 5—figure supplement 15 ) shows the same two missing resonances arising from the p66 subunit and also identifies two additional perturbed resonances that we assign to the corresponding residues in the p66' subunit of the homodimer . Figure 7A shows four 1H-13C HMQC spectra of Ile-labeled p66 at successive 5 . 5-hr time periods after dimerization conditions are introduced , for a spectral region containing the Ile329 , Ile375 , and Ile434 resonances . Consistent with the behavior summarized above , the three resonances assigned to residues in the p66 subunit: Ile329E , Ile375E , and Ile434E are approaching their equilibrium intensities during the first NMR accumulation . During the subsequent accumulation periods , the Ile434C resonance , which contains contributions from both the p66' subunit of the homodimer and from the overlapping Ile434M resonance of the monomer , decays almost completely . The Ile329C and Ile375C resonances arising from the connection' domain of the p66' subunit of the homodimer show gradual intensity gains over this same time period . We attribute these changes to the simultaneous destabilization of RH' and the conformational maturation of the connection' as residues derived from RH' are incorporated into helix αM' . The temporal linkage of these events is consistent with a model in which they are functionally coupled processes . These occur on a much slower time scale than the conformational processes described in the previous section that include isomerization of the monomer to an E-type conformation , initial formation of the immature homodimer , and formation of the RH:thumb' interface . 10 . 7554/eLife . 06359 . 029Figure 7 . Slow time-dependent changes of connection and RH domain resonances . ( A ) An expanded spectral region of the [13CH3-Ile]p66/[13CH3-Ile]p66' homodimer obtained at successive time intervals after introduction of conditions favoring dimerization . The selected region includes connection and connection' Ile329 and Ile375 resonances as well as RH and RH' Ile434 resonances . ( B ) The time-dependent changes of the Ile274 resonances during the same time period . ( C ) A schematic diagram illustrating the conformational changes in the connection' and RH' domains that are related to the observed resonance changes . The labeled subunits are indicated in gray . The RH , RH' , and Th' labels in the cartoon indicate the RNase H domain in the p66 subunit , the RNase H domain in the p66' subunit , and the Thumb' domain in the p66' subunit . Data supporting the assignments of the connection and connection' domain Ile329 and Ile375 resonances are presented in Figure 5—figure supplements 4 , 13 , 14 , and 15 . Dimerization was initiated at t = 0 , and the spectra were obtained at 35°C . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 02910 . 7554/eLife . 06359 . 030Figure 7—figure supplement 1 . Illustrative fits of time-dependent intensity data . Illustrative fits of the time-dependent intensities of Ile382E ( panel A ) and Ile329E and Ile329C ( panel B ) in the homodimerization study are shown . Since we have observed no stable monomer in the extended ( E ) conformation , the time-dependent intensity of the Ile382E resonance arises from the active ( extended ) p66 subunit of the dimer and is indicative of time-dependent dimerization . The Ile382 time constant of 2 . 4 hr appears to be the earliest dimerization event that can be monitored by NMR . Since this value is significantly below the time required for accumulation of the NMR spectrum , the value is subject to substantial error , and presumably represents an upper limit for the actual time constant describing interface formation . As in the example of Ile382E , the Ile329E intensity also builds up on a short time scale ( Table 1 ) , corresponding to time constants that are shorter than the NMR accumulation period . Thus , the connection domain in the p66E subunit appears to become conformationally defined at a relatively early time after dimerization . The intensity of the Ile329C resonance builds up on a slower time scale , with mean time constant of ∼ 5 . 9 hr ( Table 1 ) similar to the monomer-corrected value of 6 . 5 hr for RH' unfolding previously reported . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 030 Three resonances are assigned to Ile274 , located near the base of the thumb ( Figure 7B ) . As indicated in Figure 5 , the positions approximate the shifts characteristic of the M , C , and E conformations . The behavior of Ile274E is similar to the other resonances assigned to the E conformer , with the intensity nearing its limiting value during the first accumulation period . The intensity of the Ci resonance , closest to the monomer position , decays on a slow time scale , while the intensity of the Ile274C resonance grows over a similar time period . The Ile374 Ci resonance is attributed to the initially formed homodimer rather than to the monomer ( Figure 1—figure supplement 1D ) , since it is much greater than that of the other monomer resonances , for example , Ile393M . This behavior indicates that either the base of the thumb' is experiencing a slow conformational maturation or , more probably , that Ile274' is sufficiently close to the connection' domain , and particularly to αM' , so that the resonance is sensitive to changes occurring in the nearby domain . Time constants determined from the time-dependent intensities of the connection domain Ile329 and Ile375 resonances are summarized in Table 1 , and representative data fits are shown in Figure 7—figure supplement 1 . As noted above , the Ile329E and 375E resonances increase with time constants that are shorter than the length of the first 5 . 5-hr accumulation period , consistent with the model of Figure 1 in which isomerization of the monomer to the extended E conformation is the initial step . Alternatively , the Ile329C and Ile375C resonances in the connection' domain increase with slower time constants of ∼6 hr ( Table 1 ) that are similar to those reported previously for the decay of the RH' Ile434C , Ile495C , and Ile521C resonances ( Zheng et al . , 2014 ) , consistent with the coupled residue transfer model outlined above . The slow forming Ile274C resonance attributed to the p66' thumb' exhibited a somewhat slower time constant of almost 9 hr ( Table 1 ) . This may correspond to an even slower maturation step; however , there are insufficient data to further develop a more specific hypothesis . One of the difficulties of analyzing homodimer maturation by NMR is the presence of isotopic labels in both subunits . Based on the ability of the palm loop deletion to block formation of the p66E conformation , we performed a time-dependent dimerization study of [13CH3-Ile]p66∆PL in the presence of a twofold molar excess of unlabeled p66 in order to facilitate complete conversion of the p66∆PL to the dimer form ( Figure 8 ) . The time-dependent spectral changes were qualitatively similar to those observed in the homodimerization study ( Figure 8—figure supplement 1 ) . The region of the 1H-13C HMQC spectrum shown in Figure 7 that includes several Ile resonances arising from the connection' and RH' domains shows the same time-dependent decay of the RH' Ile434C and Ile521C resonances in parallel with increases in the intensities of the connection' Ile329C and Ile375C resonances in the study using the palm loop deletion ( Figure 8 ) . Thus , as in the homodimerization study , the data demonstrate that formation of the connection' domain is temporally correlated with the disappearance of resonances characteristic of the folded RH' . This observation further supports the maturation of the connection' domain at the expense of the RH' domain . In these studies , none of the resonances uniquely attributed to the E conformation was observed , indicating that the p66∆PL subunit of the pseudo-homodimer does not adopt the E conformation to any significant extent . Thus , consistent with expectations based on the behavior illustrated in Figure 2 , the labeled p66∆PL is unable to form a homodimer or to dimerize with the p66 monomer by adopting the extended ( E ) conformation . 10 . 7554/eLife . 06359 . 031Figure 8 . Dimerization of [13CH3-Ile]p66∆PL with unlabeled p66 . Time-dependent changes are shown for a region of the 1H-13C HMQC spectrum covering a similar spectral region to that shown in Figure 7 . All resonances are attributed to the M or C species; the labeled Ile50M resonance as well as the Ile434C and 521C resonances arising from the labeled RH' decrease as the RH' domain unfolds , while the connection' 329C and 375C resonances increase as the connection' domain matures . The schematic diagram at the bottom illustrates the subunit-selective labeling pattern and the proposed conformational changes that are inferred from the behavior of the resonances . The labeled subunit is indicated in gray . Each spectrum corresponds to a 5 . 5-hr accumulation period at the time periods indicated . Dimerization was initiated at t = 0 , and the spectra were obtained at 35°C . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 03110 . 7554/eLife . 06359 . 032Figure 8—figure supplement 1 . Time-dependent HMQC spectra for dimerization of p66∆PL with excess , unlabeled p66 showing all Ile δ-methyl resonances . ( A ) Time-dependent spectra for [13CH3-Ile]p66∆PL with a twofold excess of unlabeled p66 after dimerization conditions were initiated . The D2O buffer contained 25 mM Tris-DCl-d11 , pD 7 . 5 , 100 mM KCl , 0 . 02% NaN3 . Spectra were obtained on an INOVA 800 at 35 °C . Monomer resonances attributed to Ile50M and to Ile132M that cannot be unequivocally assigned in the homodimerization study can be identified in this study . ( B ) Expanded region 1 showing the details of the labeled thumb' resonances . The inset illustrates the simultaneous decay of the RH' resonance arising from Ile434 . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 03210 . 7554/eLife . 06359 . 033Figure 8—figure supplement 2 . Time-dependent intensity data for monomer decay . Illustrative examples showing the analysis of the time-dependent monomer decay for the Ile47 and Ile393 resonances for the dimerization study illustrated in Figure 8 and two similar studies . In the case of Ile47 , decay of the monomer resonance was accompanied by growth of the dimer peak , corresponding to the C conformation , and the sum of the initial Ile47M and Ile47C resonance intensities was normalized to 1 . 0 . The behavior of the Ile47 resonances indicates that an initial , more rapid dimerization process was followed by a slower process with time constant ∼ 9 hr ( see Table 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 03310 . 7554/eLife . 06359 . 034Figure 8—figure supplement 3 . Time-dependent decay of RH' resonances . Illustrative fits of the time-dependent intensity data for the resolved RH' resonances assigned to Ile434C , Ile495C , and Ile521C . The fits allowed for asymptotic monomer resonance intensities greater than 0 , and the observed decay generally corresponded to ∼ 90% of the total intensity . The time constants were all in close agreement with a mean value of 10 . 7 hr ( Table 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 03410 . 7554/eLife . 06359 . 035Figure 8—figure supplement 4 . Time-dependent growth of connection' resonances . Fits of the time-dependent intensity data for the resolved Ile329C and Ile375C resonances in the study of the dimerization of [13CH3-Ile]p66∆PL with a twofold excess of unlabeled p66 . The Ile375C resonance is significantly weaker than Ile329 , so that the sample to sample variation was generally greater . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 035 Despite qualitative similarity with the homodimerization study , the kinetic behavior exhibits significant differences ( Table 2 ) . Most importantly , the fraction of p66∆PL initially in the monomer form is greater than that observed in the homodimerization study , the decay of the monomer resonances is slower , and the dimerization is incomplete , reaching only 80–90 % based on comparisons of the intensities of multiple resonances ( Figure 8—figure supplement 2 ) . These kinetic differences can be interpreted within the context of the conformational maturation model ( Equation 1 ) as resulting from competition between the unlabeled p66 monomer and p66∆PL for the rare p66E conformation that is only formed by isomerization of p66 . Note that all of the steps except perhaps for the final RH’ unfolding are expected to be fully reversible , so that the NMR observations represent average populations of the observed species that indicate the conformational mixture present during each accumulation period . Unfolding of the RH' domain on the p66' subunit of p66/p66' will further deplete the pool of p66 available to form the p66E species . 10 . 7554/eLife . 06359 . 036Table 2 . Apparent time constants—p66 + [13CH3-Ile]p66∆PLDOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 036ResidueMean ± S . E . 393M9 . 3 ± 1 . 4274M/Ci8 . 4 ± 0 . 547M9 . 1 ± 1 . 7Mean monomer decay TC9 . 0 ± 0 . 647C8 . 8 ± 1 . 2434C†10 . 0 ± 0 . 2495C†10 . 4 ± 0 . 4521C†11 . 6 ± 0 . 3Mean RH resonance decay TC10 . 7 ± 0 . 3329C10 . 3 ± 1 . 6375C11 . 2 ± 2 . 5Mean connection' growth TC10 . 8 ± 1 . 3Errors determined as in Table 1 . Each value represents the mean of three separate studies . †Resonances 434C , 495C , and 521C also contain contributions from overlapping monomer peaks , and no attempt has been made to correct for this overlap . Similarly , the resonance labeled 274M/Ci contains contributions from both the monomer and the initially formed dimer , so that the decay results from both dimerization and conformational maturation of the dimer . Illustrative data fits are shown in Figure 8—figure supplements 2–4 . RH: ribonuclease H; TC = time constant . The presence of a significant monomer concentration affects the kinetic analysis in two ways: 1 ) there is a gradual contribution to the intensities of all labeled dimer resonances as the monomer is converted to dimer and its various incompletely matured forms , and 2 ) in some cases , for example , the RH' resonances , the monomer resonances overlap those in the dimer . Due to the more significant effects of the monomer in this study , we have not attempted to introduce a monomer correction , as was done in our previous analysis ( Zheng et al . , 2014 ) , and instead just fit the data to the simplest mathematical models that provided reasonable approximations . The results , summarized in Table 2 , indicate that: 1 ) the initial monomer concentration has decreased by ∼ 40% after which it decays with a time constant of ∼10 hr based on Ile393M and Ile47M peaks , reaching a limiting level of ∼20% of the total . 2 ) the RH' Ile434' , Ile495' , and Ile521' resonances all decay with similar apparent time constants with a mean value of 10 . 7 hr , and the connection' Ile329'; and Ile375' resonances increase with similar time constants , kinetically linking these two processes . 3 ) the Ile47' resonance intensity is divided between monomer and dimer species , Ile47M and Ile47C , so that the time-dependent behavior results from the monomer→ dimer conversion ( Figure 8—figure supplement 2 ) . In the initial spectra , the dimer species accounts for 30–40 % of the total , after which it increases with a time constant of ∼9 hr , leveling off at about 80% of the total intensity ( Table 2 ) . 4 ) . All of these rates are longer than the 5 . 9 hr time constant observed for maturation of the connection' Ile329' and Ile375' resonances in the homodimerization study ( Table 1 ) . The behavior summarized above , particularly for the Ile47' resonances , suggests that dimerization of labeled p66∆PL with p66E is initially a rapid process but slows down significantly possibly as the pool of p66E monomer becomes depleted due to dimer formation . Subsequent dimerization of p66∆PL may require release of p66 monomers from p66/p66' dimers at various stages of conformational maturation , until most of the p66 and p66∆PL have formed sufficiently stable dimers so that further release of p66M becomes extremely slow . The dimerization process in this study also allows identification of additional intermediate resonances . Additional thumb' resonances for Ile274' and Ile257' ( Figure 8—figure supplement 1 ) exhibit an initial intensity increase and subsequent decrease , consistent with conformational intermediates . Although it is not clear if the two intermediate states are also present in the p66/p66' homodimerization study , a close examination of the same region of the spectrum suggests that similar intermediate species may be present . Given the involvement of the thumb' in both early and late conformational events , this behavior is probably not surprising . In summary , the dimerization of [13CH3-Ile]p66∆PL with p66 is qualitatively similar to p66 homodimerization , but the monomer is significantly more persistent and the time constants are all longer . The Ile47' resonance provides a direct readout of dimer formation that probably is not limited by additional conformational changes . The residues that form helix αM' are almost all hydrophobic; the lone exception is Lys424 , which also can interact hydrophobically with its ( CH2 ) 4 sidechain . This uniformity allows it to adopt alternate registrations in which one hydrophobic residue substitutes for another . This conformational variability is supported by a comparison of multiple crystal structures ( Figure 1—figure supplement 3 ) . The ability of the helix to adopt alternate registrations facilitates its victory in the tug-of-war for residues from RH' . Thus , immature , distorted helical conformations can be present that are more consistent with a folded RH' domain , and the helix is then able to recruit and incorporate Tyr427' from RH' when this residue is released from RH' due to thermal fluctuations . Recruitiment of Tyr427' into αM' results in improved helical geometry and more stable interactions between αM' and other connection' residues . As shown previously , RH' is significantly destabilized by the loss of Tyr427' , facilitating its unfolding and subsequent proteolytic degradation ( Zheng et al . , 2014 ) . The studies presented above support a modified conformational selection process and provide a basis for characterizing some of the steps in Equation 1c ( Figure 9 ) . The structure of the p66 monomer provides perhaps the most compelling support for a conformational selection model , since most of the domain interfaces are abrogated in the monomer without the need for dimer formation to promote this process . Only the fingers/palm:connection needs to dissociate to allow the necessary reorganization of the domains . The inherent preference of the bent fingers/palm domains to adopt a more extended conformation provides some additional impetus for dissociation of this interface ( Figure 3 ) . Further support for this model is derived from the effect of the palm loop deletion in blocking dimerization and the molecular dynamics simulations presented above ( Figure 4 ) . The initially formed homodimer contains two folded RH domains and two immature connection domains . The fingers/palm domains in the two subunits are probably close to their final conformation in the initial homodimer , since the initial isomerization of the monomer to the extended conformation is proposed to be concerted with straightening of the fingers/palm ( Figure 4 ) . The RH:thumb' interface is not initially present . Based on the analysis of Ile methyl resonances of residues distributed throughout the molecule , slower processes that nevertheless are largely accomplished during the first HMQC accumulation period include: 1 ) formation of the inter-subunit RH:thumb' interface; 2 ) maturation of the connection:RH interface; 3 ) maturation of the p66 connection and thumb subdomains . These conformational steps appear to be cooperative . 10 . 7554/eLife . 06359 . 037Figure 9 . Schematic illustration of the maturation of the p66/p66' homodimer . This figure illustrates the more rapid and the slower time-dependent changes occurring subsequent to initial isomerization/dimerization . The subunit conformations are color coded as in Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06359 . 037 Substantial NMR evidence indicates that several other conformational processes occur on a much slower time scale . These include: 1 ) transfer of residues from RH' to the connection' domains of p66' , 2 ) conformational maturation of the p66' connection' and thumb' domains , including formation and/or extension of helix αM' , and 3 ) unfolding of the destabilized RH' domain . Resonance changes related to these slower processes are readily observed in successive NMR spectra ( Figures 7 and 8 and Tables 1 and 2 ) and result from the gradual transfer of residues from RH' to helix αM' in the connection' domain , which affect primarily the resonances of residues located in the RH' , thumb' , and connection' domains of the p66' subunit . The time constants describing RH' unfolding are similar to the 7-hr lifetime estimated for the HIV virion based on mathematical models ( Perelson et al . , 1996; Perelson and Nelson , 1999 ) . Additional stabilization of the initial p66/p66' dimer species likely results from complex formation with tRNALys , 3-annealed viral genome , which is present in the virion core ( Kleiman et al . , 2010 ) . Formation of such a complex is not expected to impact the maturation pathway described above , but to accelerate the process by eliminating dissociation of immature dimers and introducing additional stabilizing interactions for the E conformer . Preliminary NMR studies demonstrate that addition of dsDNA can promote dimerization and facilitate maturation . In summary , the metamorphic polymerase domain of RT can be considered as a puzzle with two alternate solutions . The monomer structure corresponds to a partially disassembled version of the puzzle , with only the fingers/palm:connection interface remaining , and is thus primed to undergo a unimolecular reorganization into either the compact or extended forms , facilitating dimer formation and followed by conformational maturation . All protein expression and purification followed the protocols described in our previous study ( Zheng et al . , 2014 ) . All mutations used for resonance assignments of connection and RH domain were carried out by the QuickChange XL site-directed mutagenesis kit ( Stratagene ) and confirmed by DNA sequence analysis . Labeled proteins were prepared by growth on M9 minimal medium in 99% D2O supplemented with 50 mg/L [4-13C , 3 , 3-2H2]2-oxobutyrate 1 hr prior to induction as described previously ( Tugarinov and Kay , 2003; Zheng et al . , 2014 ) . The [U-2H , δ-13CH3-Ile] labeling pattern that is produced using this approach is abbreviated as [13CH3-Ile] throughout the manuscript and in the Supplementary figures . Mutants used for site-specific assignments are summarized in Figure 5—figure supplement 4 . The 1H-13C HMQC spectra were obtained using Agilent's gChmqc experiment in Biopack ( Agilent , Santa Clara , CA ) . The NMR data were collected on a UNITY INOVA 800 MHz spectrometer equipped with a 5-mm Varian 1H[13C , 15N] triple-resonance cryogenically cooled probe at 25°C or 35°C . In the 1H dimension , 1024 complex points were acquired with a sweep width of 14 ppm using a relaxation delay of 2 s . In the indirect 13C dimension , 96 complex points were acquired with a spectral width of 10 ppm , and the 13C offset was set to 13 ppm . A WURST-80 decoupling sequence was used for 13C-decoupling during the acquisition period ( Kupce and Freeman , 1995 ) . The residual water peak was suppressed using the WET sequence at the end of the relaxation delay ( Smallcombe et al . , 1995 ) . All NMR data were processed by NMRPipe ( Delaglio et al . , 1995 ) and analyzed with NMRViewJ ( Johnson and Blevins , 1994 ) . The NMR samples were concentrated to 270 µL using Amicon ultracentrifugal filters with a 30 kDa cutoff , into the D2O NMR buffer: 25 mM Tris-HCl-d11 , pD 7 . 5 , 50–100 mM KCl , and 10–30 µM DSS as a chemical shift reference . The purified p66 and p66∆PL were analyzed on the HiLoad 26/60 superdex 200 column separately . The running buffer was 50 mM Tris–HCl , pH 8 . 0 , 200 mM NaCl , 1 mM ethylenediaminetetraacetic acid ( EDTA ) at a flow rate of 0 . 5 ml/min on an Akta FPLC system at 4 °C . The elution profiles recorded the absorbance at 280 nm . For the time-dependent NMR studies of the dimerization of unlabeled p66 with [13CH3-Ile]p66∆PL , we mixed the labeled p66∆PL with a twofold excess of unlabeled p66 , concentrated the sample , and exchanged it into the NMR buffer: 25 mM Tris-HCl-d11 , pD 7 . 5 , 100 mM KCl , and 0 . 02% NaN3 , with Amicon Ultra Centrifual Filters ( 30 Kda cut-off ) . The final 275 µL sample contained 45 µM [13CH3-Ile]p66∆PL and 90 µM of unlabeled p66 . Successive 1H-13C HMQC spectra were obtained in 5 . 5-hr increments , as described in our previous study ( Zheng et al . , 2014 ) . To prepare the labeled p51/p51' homodimer , we concentrated [13CH3-Ile]p51 and exchanged it into 25 mM Tris-HCl-d11 in D2O ( pD = 7 . 5 ) , 800 mM KCl , 20 mM MgCl2 , and 0 . 02% NaN3 to a final concentration of 150 µM [13CH3-Ile]p51 . It was necessary to use the higher salt conditions to compensate for the weak homodimerization constant of p51 . ( Venezia et al . , 2006; Marko et al . , 2013 ) The 1H-13C HMQC spectra indicate that the sample is ∼90% dimer/10% monomer . For the studies of the mutated heterodimer , [13CH3-Ile]p66/p51 ( L289K ) , we mixed a twofold excess of unlabeled p51 ( L289K ) with [13CH3-Ile]p66 and exchanged the sample into 25 mM Tris–HCl in D2O ( pD = 7 . 5 ) , 50 mM KCl , and 0 . 02% NaN3 to get 45 µM [13CH3-Ile]p66/p51 ( L289K ) samples . In our previous study ( Zheng et al . , 2014 ) , we utilized constructs of the isolated fingers/palm , RH , and thumb domain to assign many of the isoleucine δ-CH3 resonances in RT . Several preliminary connection domain assignments were also derived from site-directed mutants . In the present study , we report more complete assignments for the connection and RH domain resonances based on extensive mutagenesis studies ( Figure 5—figure supplements 3–15 ) . In a few cases , these resulted in assignment changes . The analysis presented previously was not dependent on these assignments , and the earlier conclusions are unaffected by the reassignments . Molecular dynamics simulations were performed on the isolated fingers/palm domain , defined to include residues 1–236 , starting with either subunit of the RT heterodimer , pdb: 1DLO ( Hsiou et al . , 1996 ) . Since the segment from 219–230 is missing in the p51 subunit of 1DLO , the missing residues were modeled by using the corresponding segment from the p66 subunit . The ends of the palm loop are separated by 20 Å in p66 compared with 7 . 2 Å in p51 , so that this insertion leads to a localized perturbation . However , the initially increased separation of residues 218 and 231 required for the segment transplant decayed during the first 10 ns equilibration period , and the time-dependent simulations shown in the figures begins at the end of this period . The structures were solvated in a box of water ( p51 with 24 , 721 and p66 with 26 , 635 water molecules , respectively ) , after missing protons were added to each of these structures . Prior to equilibration , all systems were subjected to ( i ) 100-ps belly dynamics runs with fixed peptide , ( ii ) minimization , ( iii ) low-temperature constant pressure dynamics at fixed protein to assure a reasonable starting density , ( iv ) minimization , ( v ) stepwise heating molecular dynamics at constant volume , and ( vi ) constant volume molecular dynamics for 5 ns . All final unconstrained trajectories were calculated at 300 K under constant volume ( 100 ns , time step 1 fs ) using the PMEMD module of Amber ( Case et al . , 2005 ) to accommodate long-range interactions . The parameters were taken from the FF10 force field of Amber ( Case et al . , 2005 ) . An additional 300-ns trajectory for the p51 system was calculated with a different set of starting velocities . Similar calculations were also performed starting with a p51 subunit of a structure that included the palm loop residues , pdb: 1S9E ( Das et al . , 2004 ) , and starting with the monomer , pdb: 4KSE ( Zheng et al . , 2014 ) . Since in the monomer construct residues 218 and 231 are directly bonded , two alternate procedures were used to introduce the missing palm loop residues . Either the segment from the p66 subunit was introduced , analogous to the procedure described above , or the artificial bond was left in place and 13 additional residues were included to maintain the same total number of residues ( 237 ) . The results of these simulations are shown in Figure 3—figure supplement 1 . Time-dependent intensity data obtained in studies of p66/p66' and p66/p66∆PL were analyzed using the non-linear least squares feature of Mathematica ( Wolfram Research ) . Time-dependent intensity data were fitted to growing or decaying exponential functions that also allowed for variable limiting values for data sets that could not be well approximated by a transition between fractional probabilities of 0 and 1 ( Tables 1 and 2 ) . For Ile47 , the t = 0 intercepts were normalized to total 1 . 0 , and the fits demonstrated that the Ile47M + Ile47C summed intensity was nearly constant . Although for some of the resonances analyzed in the p66/p66∆PL dimerization study , there is significant overlap between the monomer and dimer species , no correction for this overlap was utilized .	Proteins are made up of long chains of building blocks called amino acids . These chains can twist and fold in numerous ways to adopt the specific three-dimensional shapes that enable each protein to perform its role . In recent years , researchers have identified several proteins that can adopt different shapes from the same sequence of amino acids . These are known as metamorphic proteins and each shape may carry out a different role . HIV is a virus that causes AIDS , an illness that leads to progressive failure of a person’s immune system . The virus uses an enzyme called “reverse transcriptase” to copy its genetic material . The enzyme consists of two metamorphic protein subunits that are both derived from the same precursor protein called “p66” . One p66 subunit adopts an extended shape that enables it to carry out enzymatic activities . The second is processed into a smaller p51 subunit that is inactive but provides structural integrity to the enzyme . Zheng et al . have now used nuclear magnetic resonance and other state-of-the-art techniques to analyze the different stages of the conversion of the p66 protein into the mature reverse transcriptase enzyme . The analysis revealed the shape of a single p66 protein molecule , and showed that occasional changes in shape allow one p66 molecule to bind to a second . This means that an immature version of reverse transcriptase contains two p66 subunits with different shapes . The shapes of each of the two subunits then undergo further changes with time . In one of the subunits , competing interactions lead to a molecular tug-of-war that prevents part of the protein from adopting its folded shape . This part subsequently unravels and is later destroyed by another HIV enzyme ( called HIV protease ) to form the smaller p51 subunit . Since HIV needs reverse transcriptase in order to multiply and cause infection , drugs that prevent this enzyme from working are used to treat patients with AIDS . Current drugs target the mature form of the enzyme , but are of limited use because mutations can lead to drug-resistant forms of the proteins . The findings of Zheng et al . now fill a major gap in our understanding of the intermediate steps that lead to the formation of mature reverse transcriptase . These findings are expected to guide future work aimed at developing new drugs that interfere with maturation instead of blocking activity of the mature enzyme .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2015	Asymmetric conformational maturation of HIV-1 reverse transcriptase
The ability to revise one’s certainty or confidence in a preceding choice is a critical feature of adaptive decision-making but the neural mechanisms underpinning this metacognitive process have yet to be characterized . In the present study , we demonstrate that the same build-to-threshold decision variable signal that triggers an initial choice continues to evolve after commitment , and determines the timing and accuracy of self-initiated error detection reports by selectively representing accumulated evidence that the preceding choice was incorrect . We also show that a peri-choice signal generated in medial frontal cortex provides a source of input to this post-decision accumulation process , indicating that metacognitive judgments are not solely based on the accumulation of feedforward sensory evidence . These findings impart novel insights into the generative mechanisms of metacognition . The ability to detect errors is an essential feature of adaptive behavior , providing the basis for adjusting or countermanding ongoing actions and optimizing future decision-making ( David et al . , 2012; Fernandez-Duque et al . , 2000; Fleming et al . , 2012 ) . Establishing the neurocomputational principles underpinning this key metacognitive function is therefore a major imperative . Categorical choices are thought to be made by integrating evidence over time into a decision variable that triggers action upon reaching a criterion ( Gold and Shadlen , 2007; Kelly and O'Connell , 2014; Shadlen and Kiani , 2013; Smith and Ratcliff , 2004 ) . Most theoretical models of metacognition propose that the same decision variable makes a key contribution to internal representations of choice accuracy ( de Martino et al . , 2013; Heath , 1984; Kiani et al . , 2014; Kiani and Shadlen , 2009; Link , 2003; Moran et al . , 2015; Pleskac and Busemeyer , 2010; Ratcliff and Starns , 2013; Yu et al . , 2015 ) . However , there is considerable uncertainty regarding the precise nature of this contribution . Initial efforts to model metacognitive performance centered on the proposition that our confidence in a choice is based on a read-out of the level the decision variable has reached at the time of choice commitment ( Heath , 1984; Kiani et al . , 2014; Kiani and Shadlen , 2009; Link , 2003 ) . However , the specification that the decision variable reaches a fixed threshold prior to commitment means that these models cannot account for the fact that human participants can retrospectively categorize certain choices as erroneous even in the absence of external feedback ( Rabbitt and Vyas , 1981; Rabbitt , 1966; Yeung et al . , 2004 ) . To take account of this kind of observation , alternative theoretical models have been proposed in which metacognitive judgments can exploit additional evidence that is accumulated after first-order commitment ( Moran et al . , 2015; Pleskac and Busemeyer , 2010; Yu et al . , 2015 ) . However , a definitive neurophysiological demonstration of post-decisional evidence accumulation has yet to be definitively provided in humans or other animals . Moreover , it is unclear whether such a process would take the form of a simple continuation of first-order evidence gathering ( Moran et al . , 2015; Pleskac and Busemeyer , 2010; Resulaj et al . , 2009; Yu et al . , 2015 ) or might also be crucially dependent on higher-order representations of error likelihood ( Yeung et al . , 2004 ) . While the last two decades have seen intensive research on the neural signatures of decision formation in the non-human primate ( Gold and Shadlen , 2007; Shadlen and Kiani , 2013 ) , these questions have proven difficult to address because the pre-motor neurons that have been the focus of much of this work ( e . g . in area LIP ) fall silent upon initiation of the decision-reporting action , thus precluding measurement of post-decision activity and verification of its potential influence on metacognition . Although post-commitment neural signatures have been described in humans ( Falkenstein et al . , 1990; Gehring et al . , 1993 ) and other animals ( Ito et al . , 2003; Narayanan et al . , 2013; Pardo-Vazquez et al . , 2008 ) which show clear sensitivity to choice accuracy and , in some cases , to the quality of metacognitive judgments ( Boldt and Yeung , 2015; Nieuwenhuis et al . , 2001; O'Connell et al . , 2007; Steinhauser and Yeung , 2010 ) , these signals have yet to be conclusively identified with post-decisional evidence integration . In the present study , we exploited the uniquely supramodal nature of a recently characterized build-to-threshold decision variable signal in the human brain ( Kelly and O'Connell , 2013; O'Connell et al . , 2012; Twomey et al . , 2015 ) to investigate the influence of post-decisional evidence accumulation on the timing and accuracy of explicit error detection . Through a combination of electrophysiological data analysis and computational modeling , we demonstrate that neural evidence accumulation does persist after commitment to the first-order decision and that its rate determines the probability and timing of error detection . Additionally , we show that the rate of post-decision accumulation is not solely the product of feedforward sensory inputs but is influenced by the output of medial frontal structures implicated in performance monitoring and executive control . We analyzed 64-channel electroencephalographic ( EEG ) data [originally collected as part of Murphy et al . ( 2012 ) ; see Materials and methods] , from 28 human subjects performing a Go/No-Go response inhibition task ( Hester et al . , 2005; Figure 1a ) . Subjects viewed a serial sequence of color words , each presented for 0 . 4 s , with the congruency between font color and semantic content varied across trials . The primary task was to execute a right-handed button press as quickly as possible when the semantic content of the word and its font color were incongruent ( Go trial ) , and to withhold this response when either the word presented on the current trial was the same as that presented on the previous trial ( ‘repeat’ No-Go ) or when the meaning of the word and its font color matched ( ‘color’ No-Go ) . Performance on paradigms of this nature can be readily understood in terms of a race-to-threshold between two competing accumulation processes representing the evidence in favor of a ‘Go’ or ‘Don’t Go’ decision ( Gomez et al . , 2007; Logan and Cowan , 1984 ) . 10 . 7554/eLife . 11946 . 003Figure 1 . Go/No-Go task and associated behavior . ( a ) Subjects’ primary task was to make a speeded manual response ( ‘A’ ) to all incongruent color/word stimuli and to withhold from responding to congruent stimuli or when the same word was presented on consecutive trials . Following any commission errors , they were instructed to signal error detection as quickly as possible by pressing a secondary response button ( ‘B’ ) . ( b ) Distribution of trial types averaged across subjects . C = correct withhold , E = error , D = detected error , U = undetected error . Error bars = s . e . m . ( c ) Histograms representing stimulus onset and detection response time ( RTd ) distributions on detected error trials , aligned to error commission and pooled across all subjects . DOI: http://dx . doi . org/10 . 7554/eLife . 11946 . 00310 . 7554/eLife . 11946 . 004Figure 1—figure supplement 1 . ‘Repeat’ and ‘color’ No-Go stimuli did not differentially affect primary RT , RTd or decision signal morphology on detected error trials . ( a ) Group-averaged median primary RT ( left ) and RTd ( right ) for detected errors to repeat and color No-Go stimuli . There was no effect of stimulus-type on either metric ( both p > 0 . 3 ) . ( b ) FCθ power aligned to the primary response on detected error trials , separately for both No-Go stimulus-types . ( c ) Detected-error CPP waveforms aligned to the primary response for both No-Go stimulus-types . Only subjects with ≥8 trials per stimulus-type were included in b and c ( n = 24 ) . Gray running markers at bottom indicate significant stimulus-type effect ( p < 0 . 05 , paired t-test for repeat vs color ) . Error bars and shaded error regions = s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 11946 . 004 If subjects failed to withhold the pre-potent response to either type of No-Go stimulus , they were instructed to signal detection of this error as quickly as possible by pressing a secondary response button . The timing of this error detection report was measured relative to the initial erroneous action ( detection response time; RTd ) . There were no significant differences in primary RT , RTd or electrophysiological signal morphology between repeat and color No-Go trials ( Figure 1—figure supplement 1 ) ; therefore , we collapsed across both No-Go trial-types in all analyses . Subjects successfully withheld from responding on an average of 56 . 6% ( ±13 . 0 ) of No-Go trials , and 68 . 0% ( ±16 . 7 ) of erroneous presses were followed by an error detection report ( Figure 1b ) . Primary RT and RTd were not correlated across detected error trials ( mean within-subject β = −0 . 05 , s . e . m . = 0 . 04; t27 = −1 . 4 , p = 0 . 2 ) . The median primary RTs for detected errors ( 457 ± 21 ms ) were faster than correct Go RTs ( 511 ± 22 ms; t27 = 5 . 5 , p < 1 x 10−4 ) , whereas undetected errors were significantly slower than Go RTs ( 543 ± 28 ms; t27 = 2 . 9 , p = 0 . 007; detected vs . undetected errors: t27 = 5 . 8 , p < 1 x 10−4 ) . A series of recent studies of perceptual decision making have established that a decision variable signal can be isolated in the human event-related potential ( ERP ) over centro-parietal scalp sites ( Kelly and O'Connell , 2013; O'Connell et al . , 2012; Twomey et al . , 2015 ) . This signal exhibits the same decision-predictive dynamics that have been reported in single-unit recordings from a variety of brain areas during perceptual decision formation ( Gold and Shadlen , 2007; Shadlen and Kiani , 2013 ) , including an evidence-dependent rate of rise and a threshold-crossing relationship with reaction time . Another important feature of this signal is that it represents the evolving decision in a domain-general fashion that is independent of motor requirements and indeed traces the decision even when no overt decision-reporting action is required ( O'Connell et al . , 2012 ) . Here , we observed that the same centro-parietal positivity ( CPP ) was elicited by Go and No-Go stimuli ( Figure 2—figure supplement 1 ) . In what follows , we demonstrate that this signal encodes consecutive build-to-threshold processes that determine both first-order choices and second-order error detection decisions on our task . Readers familiar with the human ERP literature will note that these two processing stages incorporate stimulus- and response-evoked activity usually attributed to the ‘P300’ and ‘Pe’ components , respectively ( see Discussion ) . We persist here with the label ‘CPP’ because it is less prescriptive about signal latency , eliciting conditions or measurement technique . To probe the dynamics of the CPP and its relationship to the timing of the first-order decision process , we split each subject’s Go-trial RT distribution into equal-sized fast , medium and slow bins and plotted the average waveforms aligned to action execution for each bin . Consistent with previous observations ( Kelly and O'Connell , 2013; O'Connell et al . , 2012; Twomey et al . , 2015 ) , the CPP exhibited a gradual build-up with a rate that was inversely proportional to RT , and reached a stereotyped amplitude at the time of first-order choice commitment ( Figure 2a ) . Thus , first-order performance on the Go/No-Go task was reliant on the same fundamental neural dynamics as have been reported for conventional perceptual decision-making paradigms ( Kelly and O'Connell , 2013 ) . 10 . 7554/eLife . 11946 . 005Figure 2 . A centro-parietal decision signal for first- and second-order decision-making . ( a ) Go-trial CPP waveforms aligned to primary task response and sorted by RT into three equal-sized bins , and associated scalp topography . ( b ) CPPs again aligned to primary response , but separately for Go trials and detected and undetected errors; gray running marker indicates significant detection effect ( p < 0 . 05 , paired t-test for detected vs . undetected ) . Topography illustrates scalp distribution of error detection effect . ( c ) Time-course of detection-predictive activity estimated as the area under the ROC curve . Permutation mean and significance threshold ( 1 . 96 s . d . ) are marked as solid and dashed gray lines , respectively . ( d ) Detected-error CPP , aligned to primary task response and subsequent error detection report; waveforms were sorted and binned by RTd . ( e ) Single-trial surface plot showing temporal relationship between the CPP and RTd ( curved black line ) ; waveforms were pooled across subjects , sorted by RTd and smoothed over bins of 50 trials with Gaussian-weighted moving average . Vertical dashed lines in a and d represent median RTs . Gray markers at bottom of these plots indicate time points when linear regression of RT on signal amplitude reached significance ( p < 0 . 05 ) ; black markers indicate center of 150 ms time windows in which regression of RT on signal slope reached significance ( p < 0 . 05; one-tailed predicting steeper slope for faster RTs ) . Shaded gray areas show latencies of all associated scalp topographies . All traces were baselined to pre-stimulus period . Shaded error bars = s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 11946 . 00510 . 7554/eLife . 11946 . 006Figure 2—figure supplement 1 . The CPP aligned to stimulus onset , separately for Go and correctly withheld No-Go trials . The plot illustrates that the first-order CPP traces the emerging decision regardless of its outcome . The delayed peak latency and large amplitude of the CPP on correct withhold trials ( green ) are consistent with the biased decision-making context created by our task design: No-Go trials were rare relative to Go trials ( ratio of approximately 1:8 ) which , in models of bounded evidence accumulation , manifests in slower decision times and a greater distance-to-threshold for committing to a No-Go decision . Shaded error regions = s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 11946 . 00610 . 7554/eLife . 11946 . 007Figure 2—figure supplement 2 . Early stimulus-evoked deflections affect first-order CPP amplitude on fast Go trials . ( a ) Grand-average stimulus- ( left ) and response-aligned ( right ) CPPs derived from all Go trials and sorted per-subject by primary RT into three equal-sized bins . Vertical dashed lines show median RTs per bin . Gray markers at bottom indicate time points at which a linear regression of RT on signal amplitude reached significance ( p < 0 . 05 ) ; black markers indicate the center of 150 ms time windows in which a regression of RT on signal slope reached significance ( p < 0 . 05; one-tailed based on prediction of steeper slope for faster RTs ) . First-order CPP amplitude appeared to increase as a function of RT in the response-aligned waveforms . Shaded error bars = s . e . m . ( b ) Scalp map showing the topographic distribution of the single-trial relationship between response-aligned signal amplitude ( latency of measurement indicated by shaded gray region in a , right ) and primary RT on Go trials ( within-subjects robust regressions; effect size = regression coefficient/s . e . m . ) . The distribution of the RT/amplitude effect was very similar to that observed for visual-evoked potentials , thus suggesting contamination of decision signals on fast trials by these early deflections and motivating our exclusion of Go trials with RT <350 ms from analyses reported in the main manuscript . DOI: http://dx . doi . org/10 . 7554/eLife . 11946 . 007 Despite the fact that the median RTd was executed 560 ms after stimulus offset , precluding the continued integration of feedforward sensory evidence in the period preceding error detection , the CPP exhibited persistent post-commitment build-up on a subset of trials , continuing its positive trajectory prior to an error detection report but gradually diminishing in amplitude following Go decisions and undetected errors ( Figure 2b; see Figure 3—figure supplement 1 for explicit comparison of signals on Go and undetected error trials ) . Receiver operating characteristic ( ROC ) curve analysis ( see Materials and methods ) conducted in discrete temporal windows along the entire signal time course revealed that second-order performance could be reliably classified as early as 120 ms prior to error commission ( Figure 2c ) . We next sought to characterize the relationship between the CPP and the timing of error detection by splitting each subject’s RTd distribution into three equal-sized bins and plotting the bin-averaged waveforms aligned to both error commission and the error detection report . Mirroring our observations for the first-order responses , the build-up rate of the second-order CPP was steeper on trials characterized by faster error detection , and it again reached a fixed amplitude immediately prior to the error detection report ( Figure 2d , e ) . Taken together , these findings indicate that subjects engaged in evidence accumulation even after the primary task response had been executed and leveraged the new information gained by this process to make judgments about the accuracy of their preceding choices . However , CPP dynamics during the post-commitment interval were qualitatively distinct from those observed during the first-order decision process . While the positive build-up of the first-order CPP was invariant to trial-type , tracing the emerging decision irrespective of whether the evidence favored a Go or No-Go choice ( Figure 2—figure supplement 1; see also Kelly and O’Connell , 2013 ) , the second-order CPP only increased after detected errors and not , on average , after correct Go responses or undetected errors . Similarly , in a recent study that examined the post-commitment ‘Pe’ signal preceding graded confidence judgments ( without interrogating this signal for accumulation-to-bound characteristics ) , a monotonic relationship between signal amplitude and confidence was observed whereby amplitude was greatest when subjects indicated very low confidence in the primary choice ( Boldt and Yeung , 2015 ) . Thus , rather than reflecting a continuation of the first-order decision process , these data suggest that error detection decisions were based on the selective accumulation of internal evidence that the previous choice was incorrect . In the following set of analyses , we investigated whether a higher-order neural signal that has known sensitivity to choice accuracy might influence the error detection process . A substantial literature incorporating several species and modes of neurophysiological measurement has established that the posterior medial frontal cortex ( pMFC ) is highly responsive to variations in performance accuracy ( Carter et al . , 1998; Ito et al . , 2003; Narayanan et al . , 2013; Ridderinkhof et al . , 2004 ) and its activity predicts neural and behavioral adaptation following error commission ( Cavanagh et al . , 2009; Danielmeier et al . , 2011; Ebitz and Platt , 2015; Narayanan et al . , 2013; Sheth et al . , 2012 ) . These characteristics are thought to reflect the pMFC’s role in coordinating the activity of task-relevant regions in the presence of increasing conflict between incompatible actions ( Botvinick et al . , 2001; Carter and van Veen , 2007; Cavanagh et al . , 2009; Kerns et al . , 2004 ) , while the conflict signal generated in this brain region has also been proposed to provide a reliable basis for subsequent error detection ( Yeung et al . , 2004 ) . Accordingly , we investigated the influence of pMFC on error detection decisions by interrogating a prominent oscillatory signature that provides a proxy for pMFC activity , fronto-central theta power ( FCθ; 2–7 Hz; Cavanagh and Frank , 2014; Cavanagh et al . , 2011; Cohen , 2014; Cohen and Donner , 2013; Narayanan , et al . , 2013 ) . Consistent with previous reports ( Cavanagh et al . , 2009; Cohen and Donner , 2013; Narayanan et al . , 2013 ) , we observed a clear increase in FCθ power following stimulus onset that peaked after primary response execution ( Figure 3a ) . The time-course of FCθ did not distinguish between correct Go responses and undetected errors ( Figure 3—figure supplement 1 ) but underwent a sharp positive deflection at approximately the time of errors that were subsequently detected ( Figure 3b ) . ROC analysis applied to single-trial FCθ waveforms revealed strong detection-predictive activity that , as with the CPP , achieved significant detection classification up to 120 ms before initial error commission ( Figure 3c ) . This sensitivity to error detection was not apparent in the error-related negativity , a commonly investigated error-evoked ERP over fronto-central scalp ( Falkenstein et al . , 1990; Gehring et al . , 1993; Figure 3—figure supplement 2 ) , and was not due to condition-related differences in primary RT ( Figure 3—figure supplement 3a ) . 10 . 7554/eLife . 11946 . 008Figure 3 . Fronto-central θ-band ( 2–7 Hz ) oscillatory power predicts the accuracy and timing of error detection reports . ( a ) Time-frequency plot of fronto-central power , aligned to the primary task response and averaged across detected error , undetected error and RT-matched go trials; black lines enclose regions of significant power change relative to a pre-stimulus baseline ( p < 0 . 01 , paired t-test ) . Scalp topography shows θ power averaged over all trial types . ( b ) Response-aligned FCθ waveforms separately for each trial type; gray trace is the condition-averaged ERP ( arbitrarily scaled ) . Topography illustrates scalp distribution of error detection effect . ( c ) Time-course of FCθ detection-predictive activity quantified by the area under the ROC curve . ( d ) Go-trial FCθ waveforms and associated scalp topography , aligned to primary task response; single-trial waveforms were sorted by primary RT and divided three equal-sized bins . ( e ) Detected-error FCθ waveforms and topographies , aligned to both the primary task response ( left ) and the subsequent error detection report ( right ) ; waveforms were sorted and binned by RTd . ( f ) Single-trial surface plot showing the temporal relationship between the FCθ power and RTd ( curved black line ) . Conventions for b–f are the same as in Figure 2 . All traces were again baselined to the pre-stimulus period . Plots in a were calculated via wavelet convolution; all other plots show filter-Hilbert transformed data ( see Materials and methods ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11946 . 00810 . 7554/eLife . 11946 . 009Figure 3—figure supplement 1 . Comparison of response-aligned CPP ( a ) , fronto-central ERP ( b ) and FCθ ( c ) signals on correct Go trials and undetected errors . Fronto-central ERPs were averaged over electrodes FCz , F1 and F2 ( see also Figure 3—figure supplement 2 ) . Gray running markers at bottom indicate latencies of significant amplitude differences , while black markers indicate center of 150ms time windows within which signal slope differed between trial-types ( both comparisons were paired t-tests , p < 0 . 05 , two-tailed ) . Shaded error regions = s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 11946 . 00910 . 7554/eLife . 11946 . 010Figure 3—figure supplement 2 . Broadband ERPs averaged over fronto-central electrodes FCz , F1 and F2 and aligned to the primary response , separately for Go trials and detected and undetected errors . The prominent peri-response negativity corresponds to the error-related negativity ( ERN ) component . Shaded gray area shows latency of associated scalp topography , collapsed across trial-types ( inset ) . Selected channels for ERN analysis accorded with the topographic location of the fronto-central minimum . Gray running markers at bottom indicate latencies of significant error detection effects ( p < 0 . 05 , paired t-test for detected vs undetected errors ) . Shaded error regions = s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 11946 . 01010 . 7554/eLife . 11946 . 011Figure 3—figure supplement 3 . Robustness of the FCθ effects to primary RT differences and baselining regimes . ( a ) Error-aligned FCθ waveforms for detected and undetected errors using subsets of trials that were matched for primary RT; persistent of error detection effect indicates robustness to condition-related differences in RT . ( b ) Error-aligned FCθ waveforms without application of any baseline; combined with Figure 3b of the main manuscript , indicates robustness of error detection effect to different baselining regimes . ( c ) Detected-error FCθ waveforms baselined at the single-trial level , and aligned to both error commission ( left ) and subsequent error detection report ( right ) after sorting and binning by RTd . Compare to Figure 3e of main manuscript . ( d ) Scatterplot illustrating the linear relationship between FCθ power and RTd when trial-by-trial variation in baseline power is either intact ( darker purple ) or removed ( lighter purple ) ; data were z-scored within subjects , pooled across subjects and grouped into 20 five-percentile bins . ( e ) Estimated effect sizes from within-subjects robust regressions of RTd on FCθ using both baselining regimes depicted in d; effect was qualitatively weaker when variation in baseline power was eliminated . Shaded regions in a-c and vertical lines in d and e = s . e . m . Gray running markers at bottom of a-c indicate latencies of significant error detection effects ( p < 0 . 05 , paired t-test for detected vs undetected errors ) in a and b and latencies of at which a linear regression of RTd on signal amplitude reached significance ( p < 0 . 05 ) in c . Black markers in c indicate center of 150ms time windows in which regression of RTd on signal slope reached significance ( p < 0 . 05; one-tailed predicting steeper slope for faster RTs ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11946 . 011 We repeated the trial-binning analyses that were previously applied to the CPP , this time to explore the relationship between the FCθ signal and the timing of first- and second-order decision-making . The amplitude of the FCθ signal consistently distinguished between RT bins throughout both decision intervals ( Figure 3d , e , f ) . In keeping with previous observations ( Cavanagh et al . , 2011; Cohen and Donner , 2013 ) , pre-response FCθ power was consistently lower on trials with faster first-order RTs ( Figure 3d ) . A reliable relationship also existed between pre-detection FCθ on detected errors and RTd , but here fast error detections were preceded by greater FCθ power ( Figure 3e , f ) . The latter observation is consistent with the notion that peri-response pMFC activity serves as a form of input to the second-order decision process and thereby influences the probability and timing of error detection ( Yeung et al . , 2004 ) . Next , we further explored this possibility by examining the trial-by-trial prediction of RTd by different features of the FCθ and CPP signals . We quantitatively compared the independent contributions of the FCθ and CPP signals to variation in RTd via single-trial , within-subjects robust regressions , leveraging single-trial amplitudes , build-up rates and peak latencies of the signals as predictors in successive models ( see Materials and methods; see Figure 4 for measurement approach ) . As expected , single-trial FCθ power was strongly negatively related to RTd ( t27 = –7 . 2 , p < 1 x 10−6 ) . Topographically , this effect was maximal over fronto-central scalp ( Figure 4a ) . In contrast , the pre-detection amplitude of the second-order CPP did not reliably correlate with RTd ( t27 = −1 . 9 , p = 0 . 08; Figure 4—figure supplement 1a ) , consistent with the aforementioned observation of a threshold-crossing effect prior to error detection . Statistical comparison of the associated regression weights indicated that FCθ power was a better predictor of RTd than CPP amplitude ( t27 = 3 . 3 , p = 0 . 003 ) . The opposite pattern was apparent for the build-up rates and peak latencies of both signals: CPP build-up rate ( t27 = -−5 . 0 , p < 1 x 10−4 ) and latency ( t27 = 4 . 9 , p < 1 x 10–4 , derived by permutation testing; see Materials and methods ) robustly predicted RTd with both effects maximal over centro-parietal scalp ( Figure 4b , c ) , whereas these features of the FCθ signal did not reliably account for trial-by-trial variance in RTd ( build-up rate: t27 = –1 . 9 , p = 0 . 07; latency: t27 = –0 . 7 , p = 0 . 5; Figure 4—figure supplement 1b , c ) . Thus , although a relationship between FCθ build-up rate and the timing of error detection was clearly observed in the trial-averaged waveforms ( Figure 3e ) , single-trial regressions revealed that this effect was only marginally significant when the contribution of CPP build-up rate was accounted for . Formal comparisons of the regression weights confirmed that CPP build-up rate and peak latency were superior predictors of RTd compared to the counterpart metrics derived from FCθ ( build-up rate: t27 = 3 . 0 , p=0 . 006; latency: t27 = 4 . 3 , p = 0 . 0002 ) . 10 . 7554/eLife . 11946 . 012Figure 4 . Variance in the timing of error detection reports is explained by distinct single-trial metrics of FCθ and CPP morphology . ( a ) Schematic depicting the single-trial measurement windows for FCθ power ( purple ) and CPP amplitude ( green; left ) . Bar graphs ( middle ) show estimated effect sizes from within-subjects robust multiple regressions of RTd on both amplitude metrics ( error bars = s . e . m . ) , and the associated topography ( inset right ) indicates the scalp distribution of the theta power effect . Scatterplot ( right ) illustrates the linear relationship between FCθ power and RTd; points and error bars are mean ± s . e . m . of data that were z-scored within subjects , pooled across subjects and grouped into 20 five-percentile bins . ( b ) Measurement approach , regression-estimated effect sizes , topographic distribution and scatterplot representing the single-trial relationships between RTd and signal build-up rates . ( c ) Similar plots representing the relationships between RTd and peak signal latencies . Points in the effect size plot ( middle ) indicate observed effect sizes , solid and dashed lines highlight the mean and 95% confidence intervals , respectively , of permuted distributions used for significance testing ( see Materials and methods ) . Topographies are thresholded at p < 0 . 005 ( a , b ) and p < 0 . 0001 ( c ) for effect visualization . The solid lines in all scatterplots are simple linear robust regression fits to the unbinned data . p < 0 . 01; p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11946 . 01210 . 7554/eLife . 11946 . 013Figure 4—figure supplement 1 . Scatterplots illustrating the linear relationships between RTd and second-order CPP amplitude ( a ) , FCθ build-up rate ( b ) and FCθ peak latency ( c ) . Points and error bars are mean ± s . e . m . of data that were z-scored within subjects , pooled across subjects and grouped into 20 five-percentile bins . Solid lines are simple linear robust regression fits to the unbinned data . DOI: http://dx . doi . org/10 . 7554/eLife . 11946 . 013 Two additional features of our data point to an active role for FCθ in the error detection process . First , FCθ power began to reliably predict the speed of error detection approximately 350 ms before the CPP did ( compare Figures 2d and 3c ) . Thus , internal fluctuations in FCθ power that were independent of the first-order evidence accumulation process reflected in the CPP were predictive of subsequent error detection behavior . Second , we tested whether the single-trial relationship between FCθ and RTd ( Figure 4a ) was formally mediated by a direct effect of FCθ on the rate of second-order evidence accumulation by constructing a three-variable path model with FCθ power as the predictor , RTd as the outcome and CPP build-up rate as the mediator variable ( see Materials and methods ) . FCθ was a reliable predictor of CPP build-up rate in this model ( p = 0 . 0007 ) and the mediation effect was significant ( p = 0 . 0009 ) , indicating that the rate of second-order evidence accumulation partially mediated the observed relationship between medial frontal signaling and the speed of error detection ( Figure 5 ) . 10 . 7554/eLife . 11946 . 014Figure 5 . Mediation analysis . Path diagram depicts the relationships between nodes in a mediation model that tested whether CPP build-up rate mediated the negative relationship between FCθ power and RTd . Lines are labeled with path coefficients , with s . e . m . shown in parentheses . FCθ power ( predictor , left ) predicted CPP build-up rate ( mediator , middle ) , which in turn predicted RTd ( outcome , right ) controlling for FCθ power . The upper-middle coefficient indicates the formal mediation effect . The significant direct path between FCθ power and RTd , calculated controlling for the mediator , indicates partial mediation: CPP build-up rate did not explain all of the shared variance between FCθ and RTd . *p < 0 . 001 , bootstrapped . DOI: http://dx . doi . org/10 . 7554/eLife . 11946 . 01410 . 7554/eLife . 11946 . 015Figure 5—figure supplement 1 . Effects of trial binning on CPP build-up rate mediation effect . Plot depicts effect size ( mediation coefficient/s . e . m . ) of the mediation of the relationship between FCθ and RTd by second-order CPP build-up rate , across a range of bin sizes . The strength of the effect strongly increased as bin size increased initially , indicating the desired enhancement in signal-to-noise by trial binning . Red marker indicates bin size used for the analysis depicted in Figure 5 of the main manuscript . DOI: http://dx . doi . org/10 . 7554/eLife . 11946 . 015 Informed by the above observations , we modeled error detection as a one-choice diffusion process ( Ratcliff and Van Dongen , 2011 ) by which noisy evidence that an error has been committed is accumulated over time ( at mean drift rate v with between-trial standard deviation η ) . Error detection is achieved in the model once the evidence tally passes a threshold ( a ) whereas the temporal integration process terminates if this threshold is not reached by a time deadline , thereby resulting in an undetected error ( Figure 6a; Materials and methods ) . The decision to model second-order performance in isolation was based on the observation that higher-order signals ( FCθ ) play a prominent role in determining second-order decision-making , thus indicating that the input to the primary decision process is not necessarily the same as the input to the second-order process ( see Discussion ) . For simplicity , we here focused on decomposing error detection behavior in isolation . This approach served to further identify our electrophysiological signals with distinct features of post-commitment evidence accumulation ( evidential input , temporal integration ) without requiring speculative assumptions about the relationship between first- and second-order decision processes . 10 . 7554/eLife . 11946 . 016Figure 6 . Diffusion modelling of error detection behavior . ( a ) Schematic representation of the one-choice drift diffusion model . Noisy error evidence accumulates over time at mean drift rate v until a threshold a is reached ( light red traces ) or a deadline on detection expires ( gray traces ) . Drift rate is normally distributed across trials with standard deviation η , and non-decision-related processing time is captured by tnd . Within-trial noise s is fixed at 0 . 1 ( not shown ) . ( b ) Group-level model fit . Points from left to right represent the 0 . 1 , 0 . 3 , 0 . 5 , 0 . 7 and 0 . 9 RTd quantiles , estimated from the data ( black ) and generated by the model fit ( gray ) . ( c ) Observed versus fitted error detection accuracy . Diagonal line is the identity line . ( d ) Group-average decision variables reflecting the accumulation of error evidence , simulated using the best-fitting model parameters for each subject and overlaid on the grand-average second-order CPP signals . Detected error traces are sorted by RTd into three equal-sized bins . CPPs are baselined to the 50 ms preceding error . ( e ) Between-subjects relationship between FCθ power and model-estimated drift ratio ( v/η ) . Bar graph ( inset ) shows standardized regression coefficients from a multiple regression that included CPP amplitude as an additional predictor . ( f , g ) Scatterplots and bar graphs highlighting between-subjects relationships between drift ratio and signal build-up rates ( f ) and peak latencies ( g ) . Although one extreme data point is apparent in f , this did not exert disproportionate influence over the fitted regression line ( studentized deleted residual < 1 ) and the correlation remained marginally significant when it was removed ( r = 0 . 37 , p = 0 . 06 ) . Error bars = s . e . m . p < 0 . 05; *p < 0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 11946 . 01610 . 7554/eLife . 11946 . 017Figure 6—figure supplement 1 . One-choice diffusion model fits for single subjects . Points from left to right represent the 0 . 1 , 0 . 3 , 0 . 5 , 0 . 7 and 0 . 9 RTd quantiles , estimated from the data ( red lines and crosses ) and predicted by the model ( blue circles ) . Text at inset of each plot indicates subject number and χ2 error term for that subject . DOI: http://dx . doi . org/10 . 7554/eLife . 11946 . 01710 . 7554/eLife . 11946 . 018Figure 6—figure supplement 2 . Between-subjects relationship between FCθ power and model-estimated drift ratio ( v/η ) when a trial-averaged baseline was applied to FCθ . Compare to Figure 6e of main manuscript . DOI: http://dx . doi . org/10 . 7554/eLife . 11946 . 018 The model fit the error detection data well , capturing the shapes of the group-level and single-subject RTd distributions ( Figure 6b; Table 1; Figure 6—figure supplement 1 ) as well as the considerable heterogeneity in individuals’ capacities for accurate second-order evaluation ( Figure 6c ) . We then used the best-fitting model parameters for each subject to generate simulated second-order decision variable trajectories for detected and undetected errors . Despite minimal data-driven constraint on the simulation process ( see Materials and methods ) , the temporal evolution of the simulated time-series closely traced that of the second-order CPP ( Figure 6d ) . By allowing a proportion of trials to assume negative drift rate , the model additionally provides a parsimonious account of the apparent downward trajectory of the CPP in the averaged waveforms after undetected errors . 10 . 7554/eLife . 11946 . 019Table 1 . Parameter estimates and goodness-of-fit of error detection diffusion model . DOI: http://dx . doi . org/10 . 7554/eLife . 11946 . 019atndvηχ2Mean0 . 210 . 200 . 450 . 452 . 46s . d . 0 . 120 . 080 . 300 . 302 . 33χ2 degrees of freedom = 1 , critical value = 5 . 024 . 24 of 28 χ2 values were below the critical value . The fitted model parameters were also correlated with key FCθ and CPP signal characteristics across subjects . We employed the per-subject ‘drift ratio’ ( v/η ) as a model-based estimate of each individual’s error evidence strength ( Ratcliff and Van Dongen , 2011; Materials and methods ) and found that this quantity was positively correlated with FCθ power ( r = 0 . 51 , p = 0 . 007; Figure 6e ) . A partial correlation analysis verified that this effect was still present ( r = 0 . 52 , p = 0 . 007 ) when θ power over bilateral posterior electrodes ( P7/P8 ) was included as a covariate , indicating that it is not driven by spurious global differences in oscillatory power ( Cohen and Donner , 2013 ) . Moreover , a multiple regression to quantify the independent contributions of FCθ and second-order CPP amplitude to variance in drift ratio revealed a significant effect only for the former ( βFCθ = 0 . 47 , p = 0 . 013; βCPP = 0 . 21 , p = 0 . 24 ) . Conversely , drift ratio was correlated with the build-up rate ( r = 0 . 47 , p = 0 . 013; Figure 6f ) and peak latency ( r = −0 . 52 , p = 0 . 005; Figure 6g ) of the second-order CPP , whereas no such relationships were observed for FCθ build-up rate ( βFCθ = −0 . 12 , p = 0 . 53; βCPP = 0 . 46 , p = 0 . 017 ) or latency ( βFCθ = −0 . 10 , p = 0 . 64; βCPP = −0 . 46 , p = 0 . 042 ) . We also note that neither FCθ power nor the build-up rate nor peak latency of the second-order CPP were correlated across subjects with primary task behavior ( withhold accuracy and primary RT on detected error trials; all p > 0 . 1 ) , thus highlighting the particular sensitivity of these metrics to second-order decision-making . Additionally , these electrophysiological measures were not correlated with other parameters of the computational model ( all p > 0 . 1 ) . Several lines of evidence indicate that the close relationship between the second-order CPP and error detection signalling cannot be attributed to motor preparation or execution . First , we have previously shown that the CPP that precedes a first-order perceptual decision is fully dissociable from motor preparation signals and is observed even when no overt decision-reporting action is required ( O'Connell et al . , 2012 ) . Second , we observed no change in signal topography between the pre- and post-choice phases ( Figure 2d ) . Third , the temporally extended and variable build-up of the second-order CPP excludes the possibility that our results can be attributed to the presence of overlapping motor execution potentials . Finally , we asked a new cohort of 12 subjects to perform the same Go/No-Go task but gave no instructions to report errors on half of the task blocks . FCθ and the CPP were clearly observed following these ‘no-report’ errors and , consistent with the fact that no-report blocks contain a mixture of detected and undetected errors , linear contrasts confirmed that FCθ and second-order CPP amplitudes were intermediate between their amplitudes following detected and undetected errors in the condition with self-initiated error reporting ( FCθ , t11 = 4 . 2 , p = 0 . 002; CPP , t11 = 6 . 9 , p < 1 x 10–4; Figure 7 ) . 10 . 7554/eLife . 11946 . 020Figure 7 . Second-order decision signals persist in the absence of error reporting demands . ( a ) Time-courses and topographic distributions of FCθ power from a new cohort that performed half of task blocks without any explicit instruction to self-monitor performance ( ‘no-report’ blocks ) . Left topography is the average topographic distribution of θ-power after pooling all detected and undetected error trials from blocks with self-initiated error detection reporting; right topography is θ distribution averaged across all errors in no-report blocks . ( b ) Time-courses and topographies of the second-order CPP component from the same cohort . Conventions are the same as in ( a ) . Shaded gray areas in left show latencies of associated scalp topographies . Shaded error bars = s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 11946 . 020 Our electrophysiological and model-based analyses demonstrate that neural evidence accumulation continues after decision commitment to facilitate reflections on choice accuracy . This empirical observation has important implications for the ongoing debate on the mechanistic basis of metacognition . A prominent hypothesis is that confidence judgments are based on internal information available at the time of choice commitment ( Heath , 1984; Kiani et al . , 2014; Kiani and Shadlen , 2009; Link , 2003 ) , and models that incorporate this assumption provide a good account of behavior when choice and confidence are interrogated simultaneously . But the manner in which metacognitive evaluations are probed is likely to have a profound impact on how they are constructed in the brain . Whether participating in a laboratory experiment or everyday activity , human decision-makers are most often required to express or act upon their metacognitive evaluations some time after an initial choice . Our results reveal that if the decision-maker is allowed to report delayed , self-initiated judgments of their own performance , these judgments will incorporate new internal information available after the point of commitment , even in the absence of continued sensory input . This process of post-decision evidence accumulation has already been invoked by several computational models to successfully account for delayed confidence judgments and changes-of-mind ( Moran et al . , 2015; Pleskac and Busemeyer , 2010; Resulaj et al . , 2009; Yu et al . , 2015 ) , while a parallel literature on error detection has long debated the mechanistic basis of post-commitment information processing ( Rabbitt and Vyas , 1981; Rabbitt , 1966; Yeung et al . , 2004; Yeung and Summerfield , 2012 ) . Our findings , however , represent the first definitive neurophysiological demonstration of post-decision evidence accumulation . In so doing , we show that the critical neural dynamics that give rise to both first- and second-order decisions are captured by a single brain signal , opening up new avenues for basic and clinical investigations . Comparison of pre- and post-decisional CPP dynamics highlighted an important qualitative distinction between the first- and second-order decision processes invoked by our paradigm . The first-order task required subjects to discriminate between the two possible stimulus categories ( i . e . Go versus No-Go ) , a process that can be understood in terms of competitive accumulation of evidence in favor of each choice alternative ( Gold and Shadlen , 2007; Gomez et al . , 2007 ) . We have previously shown that the CPP builds in response to evidence favoring each of the choice alternatives in such contexts ( Kelly and O’Connell , 2013; O’Connell et al . , 2012 ) and accordingly , we observed here that the same signal exhibited a gradual build-up prior to both Go and No-go choices . The second-order task , by contrast , was more comparable to simple signal detection with detected errors translating to hits and undetected errors translating to misses . Evidence accumulation mechanisms have also been invoked to account for signal detection decisions ( e . g . Deco et al . , 2007; Donner et al . , 2009 ) but in this case the neural decision variable seems to increase towards a single detection boundary and misses occur when this boundary is not reached ( Carnevale et al . , 2012; Deco et al . , 2007 ) . Again , previous studies have shown that the CPP in such contexts exhibits a clear ramp-up to threshold for hit decisions that is diminished or absent on miss trials ( O’Connell et al . , 2012 ) , and presently this effect was recapitulated for error detection decisions . Collectively , these findings indicate that human decision-makers are capable of quickly engaging multiple decision processes in succession that , although reliant on the same neural mechanism of evidence integration , are constructed in qualitatively distinct ways . Our results also offer important new insights into the nature of the input to the metacognitive decision process . Several features of our results mark pMFC activation , as indexed by the power of FCθ oscillations , as a likely candidate for furnishing a source of this input . First , peri-response FCθ power was strongly predictive of subsequent error detection . Second , FCθ power predicted error detection speed from a particularly early latency relative to the CPP . Third , this predictive capacity was formally mediated by the effect of FCθ on the rate of second-order evidence accumulation , as indexed by CPP build-up rate . Fourth , FCθ correlated across-subjects with a model-based estimate of the strength of the evidence that fed into the error detection process . This specification of an important role for pMFC signaling in explicit error detection suggests that metacognitive decisions are not solely based on the accumulation of sensory evidence but are also influenced by internally generated higher-order signals . While these observations accord with theoretical proposals that the pMFC provides a critical input to the error detection process ( Botvinick et al . , 2001; Yeung et al . , 2004 ) , they are not decisive about the specific nature of this input . One possibility is that pMFC furnishes a distinct source of abstracted ‘error evidence’ that directly informs second-order judgments . Extensive data suggest that θ-band activity in pMFC signals the degree to which competing action plans are simultaneously activated , commonly known as ‘conflict’ ( Cavanagh and Frank , 2014; Cavanagh et al . , 2011; Cohen , 2014; 2013 ) . Computational models have shown that a post-commitment conflict signal alone would provide a reliable basis for error detection because its magnitude scales with error likelihood and can reliably classify trial-to-trial performance accuracy ( Yeung et al . , 2004; see also Charles et al . , 2014 ) . Such a causal role for pMFC-encoded conflict in second-order decision-making would also provide a mechanistic explanation for the recent finding that manipulating effector-specific premotor activity both before and immediately after perceptual choice has clear effects on subsequent confidence reports ( Fleming et al . , 2015 ) . Alternatively , second-order decisions on our task may receive their sole or primary evidence from visual short-term memory ( Smith and Ratcliff , 2009 ) and pMFC may serve to modulate this process by strategically tuning processing in a global network of task-relevant regions when conflict is detected ( Cavanagh and Frank , 2014; Danielmeier et al . , 2011; Dehaene et al . , 1998; Shenhav et al . , 2013 ) . Consistent with such a non-specific influence on the post-decisional process , we found that the mediation of the FCθ/RTd relationship by CPP build-up rate was partial , indicating that pMFC may have also influenced the speed of error detection independently of its effect on the rate of evidence accumulation – perhaps via effects on other parameters of the decision process like the response threshold ( Cavanagh et al . , 2011 ) or motor execution time . The extent to which pMFC signaling directly modulates the second-order evidence accumulation process can be established in future work by examining the impact of pMFC perturbation on CPP dynamics ( Hayward et al . , 2004; Reinhart and Woodman , 2014; Sela et al . , 2012 ) . In contrast to other sequential sampling accounts of confidence judgments and changes-of-mind which assume that first-order decisions and metacognitive judgments are both based on the same evidence source ( Moran et al . , 2015; Pleskac and Busemeyer , 2010; Resulaj et al . , 2009; Yu et al . , 2015 ) , our simple diffusion model of error detection behaviour is agnostic to the nature of the evidence that drives the second-order decision process . The model thus accommodates suggestions that pMFC , or indeed other neural signals , provide modulatory inputs or additional sources of evidence that are multiplexed in a compound decision variable that determines second-order judgments ( Ullsperger et al . , 2014; 2010 ) . Of course , a corresponding limitation of our modelling framework is that it does not specify the precise nature of the relationship between the first- and second-order decision processes , and thereby fails to provide explicit accounts of some aspects of our observed data . For example , our selective modelling of error detection behaviour does not provide an explanation for the similarity of the post-response CPP signals on undetected errors and correct Go trials . This similarity could perhaps be due to a common paucity of error evidence on both trial-types . On the other hand , our analysis did reveal a post-response difference in frontal ERP morphology between these trial-types ( Figure 3—figure supplement 1 ) which suggests that they are at least partially dissociable in terms of neural dynamics . Such unresolved questions emphasize that a central goal of future research must be to build a unified model of decision-making that not only accounts for the often complex relationships between first- and second-order behavior ( Moran et al . , 2015; Pleskac and Busemeyer , 2010 ) , but is also constrained by neurophysiological characterizations of the post-decision accumulation process . Moreover , a remaining challenge will be to devise innovative ways to manipulate the evidence that feeds into the second-order process in an effort to test such a model and further corroborate our identification of the CPP with post-decisional accumulation . A substantial literature has already investigated error-related electrophysiological signals in human subjects and highlighted in particular a post-response centro-parietal ERP , labeled the Error Positivity ( Pe ) , that reliably discriminates between detected and undetected errors ( Murphy et al . , 2012; Nieuwenhuis et al . , 2001; O'Connell et al . , 2007; Overbeek et al . , 2005; Ridderinkhof et al . , 2009; Wessel et al . , 2011 ) . However , a consensus regarding the precise functional significance of this signal has never been achieved . Although a series of recent studies have demonstrated that Pe amplitude correlates with confidence in perceptual decisions ( Boldt and Yeung , 2015 ) and the criterion that subjects impose on error detection reports ( Steinhauser and Yeung , 2010 ) and have thereby associated this component with the general quality of the metacognitive decision process [see also Steinhauser and Yeung ( 2012 ) ] , these studies could not identify the Pe with a specific neural mechanism . The present study , by contrast , demonstrates that the post-decisional CPP encodes a second-order decision variable that bears precisely the same dynamical properties as have been reported for first-order decision signals in single-unit and population-level neurophysiology , including an RT-predictive build-up rate and a boundary-crossing relationship to response execution – properties that have never been previously reported for the Pe . Our use of a self-initiated error detection report was a critical design feature in this regard because it facilitated the interrogation of signal dynamics leading up to the moment of error detection; this has not been possible in previous studies of the Pe , which enforced either delayed metacognitive reporting or none at all . Where we have previously demonstrated that the pre-decision build-up of the CPP encompasses activity that has traditionally been associated with the classic P300 or ‘P3b’ ( Kelly and O'Connell , 2013; O'Connell et al . , 2012; Twomey et al . , 2015 ) , an important implication of the present findings is that the post-decision build-up of the CPP corresponds to the activity commonly attributed to the Pe . Thus , our data suggest that the P3b and Pe reflect distinct stages of the same neurophysiological process and point to a unifying mechanistic framework for understanding both signals . In conclusion , we have reported the first definitive neurophysiological demonstration that evidence accumulation continues after the point of decision commitment and predicts the timing and accuracy of subsequent error detection . This finding furnishes critical neurophysiological support for theoretical accounts of metacognitive decision-making that have relied on the concept of post-decisional accumulation . Moreover , we have shown that this process is informed by a higher-order neural signal generated in medial frontal cortex , which suggests that metacognitive judgments are not solely based on the feedforward accumulation of sensory evidence but also on representations of conflict or error likelihood . Collectively , these results shed significant new light on the generative mechanisms of metacognition and furnish new evidence that error detection , confidence judgments and their neural substrates can be understood in terms of the same mechanistically principled framework of evidence accumulation . All subjects were right-handed , had normal or corrected-to-normal vision , no history of psychiatric illness or head injury , reported no color-blindness , and refrained from ingesting caffeine on the day of testing . They provided written informed consent , and all procedures were approved by the Trinity College Dublin ethics committee and conducted in accordance with the Declaration of Helsinki . Subjects received a gratuity of €20 for their participation . Testing was performed in a dark , sound-attenuated room . Stimuli were presented using the ‘Presentation’ software suite ( NeuroBehavioural Systems , San Francisco , CA ) and subjects responded with the thumb of their right hand using a Microsoft ‘Sidewinder’ controller . During task performance , subjects used a table-mounted head-rest which fixed their distance from the display monitor ( 51-cm CRT operating at 85 Hz ) at 80 cm for the entire task . They were instructed to maintain gaze at a centrally-presented white fixation cross on a gray background . Color/word stimuli appeared 0 . 25° above fixation . The first version of the task was administered to thirty-two individuals , four of whom were excluded from all analyses: one due to technical issues with the EEG recording , two with excessively poor task accuracy ( <30% withheld No-Go trials ) , and a further subject with no observable CPP component . Thus , we analyzed a final sample of 28 subjects ( 13 male ) for the primary study , with a mean age of 23 . 5 years ( s . d . = 5 . 8 ) . This pre-planned sample size is consistent with other electrophysiological studies of decision-making from our lab that interrogated similar neural signals and invoked similar analytical methods ( Kelly and O'Connell , 2013; O'Connell et al . , 2012; Twomey et al . , 2015 ) . Subjects were required to respond with a single ‘A’ button press on all Go trials , and to withhold this response on both ‘repeat’ and ‘color’ No-Go trials ( Figure 1 ) . They were instructed to give equal emphasis to the speed of their Go responses and the accuracy of their No-Go withholding . In the event of any failure to withhold the pre-potent response to either type of No-Go stimulus , subjects were required to signal detection of this error as quickly as possible by pressing a second ‘B’ button . They were instructed to execute this error detection report even if they became aware of an error after the onset of the following stimulus . Although such late error detections were rare ( mean = 2 . 5 trials , s . d . = 2 . 4 ) and excluded from analysis , this instruction mitigated against the adoption of a time-dependent decision criterion that may have obscured any threshold-like relationship between second-order decision signals and the latency of error detection . Each subject first completed a brief automated training protocol ( Murphy et al . , 2012 ) , and was then administered at least 8 blocks of the task . Where time constraints allowed , we administered more blocks to increase the number of error trials available for analysis . On average , subjects completed 9 . 5 ± 0 . 8 blocks ( range 8–10 ) . Each block consisted of 224 word presentations , 200 of which were Go stimuli and 24 of which were No-Go stimuli ( 12 repeat , 12 color ) . All stimuli were presented for 0 . 4 s , followed by an inter-stimulus interval of 1 . 6 s . The duration of each block was therefore approximately 7 . 5 minutes . Stimuli were presented in a pseudo-random order with a minimum of three Go trials between any two No-Go trials . To establish the generality of the post-decisional signals observed under the previous task version to situations in which decision-makers are not explicitly instructed to monitor their own performance , we tested an independent cohort of sixteen individuals on a version of the Go/No-go task in which error reporting instructions were manipulated . Four of these subjects were excluded from analysis due to insufficient numbers of undetected errors ( <6 ) for reliable EEG analysis , leaving a final sample of twelve ( 5 male ) with a mean age of 23 . 7 years ( s . d . = 6 . 7 ) . Although this sample size is lower than that employed in the first study , we here limited our analyses to dependent measures with inherently high signal-to-noise ( trial-averaged waveforms ) and this sample size is consistent with a previous study of error detection using the same task employed presently ( Hester et al . , 2005 ) . Task design and procedures were identical to those used for the previous version of the task , with the exception that every subject was administered five task blocks with regular error detection reporting and five ‘no-report’ blocks in which subjects were not instructed to monitor for errors . The five blocks in each condition were administered contiguously , and the order in which each condition was presented was counter-balanced across subjects . Thus , half of the full sample performed the no-report blocks without knowing that they would later be required to signal their errors . The presence of both FCθ and CPP signals in the no-report condition ( Figure 7 ) did not depend on condition order . Continuous EEG was acquired using an ActiveTwo system ( BioSemi , The Netherlands ) from 64 scalp electrodes , configured to the standard 10/20 setup and digitized at 512 Hz . Eye movements were recorded using two vertical electro-oculogram ( EOG ) electrodes positioned above and below the left eye and two horizontal EOG electrodes positioned at the outer canthus of each eye . EEG data were processed in Matlab ( Mathworks ) via custom scripting and subroutines from the EEGLAB toolbox ( Delorme and Makeig , 2004 ) . Eye-blinks and other noise transients were isolated and removed from the EEG data via independent component analysis ( ICA ) . Specifically , continuous data from each block were re-referenced to channel Fz; data were high-pass filtered to 1 Hz , low-pass filtered up to 95 Hz and notch filtered to remove 50 Hz line noise using a two-way least-squares FIR filter; noisy channels were identified by visual inspection of signal variance and removed; data were segmented into temporally contiguous epochs of 1 s duration; epochs containing values that violated amplitude ( ± 250 µV ) and joint probability ( ± 4 . 5 s . d . ; Delorme and Makeig , 2004 ) criteria were rejected; and , the remaining data were subjected to temporal ICA using the infomax algorithm . The ICA weights yielded by this procedure were then back-projected to the original continuous , unfiltered EEG data for the associated block . Next , independent components representing stereotyped artifactual activity such as eye blinks , saccades and individual electrode artifacts were identified by visual inspection and discarded , and the ICA-pruned data were low-pass filtered to 35 Hz . No high-pass filter was applied . Previously-identified noisy channels were then interpolated via spherical spline interpolation , and the data were re-referenced to the common average . Data epochs were extracted from 2 . 5 s before to 3 . 5 s after stimulus onset on each trial ( thus minimizing edge artifacts during spectral analysis ) and baseline-corrected to the 0 . 3 s interval preceding stimulus onset . Subsequent epoch rejection employed a dynamic window with a fixed start time of −0 . 3 s relative to stimulus onset and an end time that depended on the primary RT of each trial: the window ended at RT + 1 . 2 s for go trials and undetected errors , and RT + 0 . 2 s + the slowest RTd of the current participant for detected errors; end time was truncated to +2 s relative to stimulus onset if the window encroached upon onset of the subsequent stimulus on any trial . Epochs were rejected from all further analysis if any scalp channel exceeded ±100 µV at any point within this trial-specific window . Detected error trials on which RTd followed next-trial onset were also excluded . Lastly , all EEG data were converted to current source density ( Kayser and Tenke , 2006 ) to increase spatial selectivity and minimize volume conduction ( Kelly and O'Connell , 2013; Twomey et al . , 2015 ) . A previous paper by our group investigated the relationship between the Pe component ( referred to here as the second-order CPP ) and error detection using the same data and reported a strong correlation between Pe peak latency and RTd ( Murphy et al . , 2012 ) . However , we did not consider the influence of Pe build-up rate on performance , we did not establish functional equivalence between the Pe and CPP and we did not consider the contribution of FCθ to error detection ( see below ) . In fact , in that paper we reported a reliable negative association between RTd and Pe amplitude immediately prior to error detection that is inconsistent with the presently reported results and difficult to reconcile with the proposal that the Pe reflects the accumulation of evidence toward a fixed decision bound . Two critical methodological distinctions between Murphy et al . and the present study account for this discrepancy . First , whereas Murphy et al . implemented a high-pass temporal filter , the present study did not . High-pass filtering is problematic in the current context because it is likely to attenuate CPP amplitude to a greater degree on trials with long RTs since decision-related neural activity is drawn out over a longer time frame on such trials . Second , the present study used a spatial filter to reduce the overlap with other temporally coincident signals because in recent work we demonstrated that boundary-crossing effects at response can be obscured by spatial overlap of the CPP with anticipatory signals emanating from frontal sites ( Kelly and O'Connell , 2013; Twomey et al . , 2015 ) . Addressing these issues enabled us to make a variety of important new observations that extend our previous results in several significant ways: we provide the first demonstration of the critical build-to-threshold relationship between the second-order CPP and RTd that is characteristic of an evolving decision variable; we show that this component interacts with a frontal conflict signal to determine the probability and timing of error detection; and , we leverage computational modelling to further identify this component with the second-order evidence accumulation process . Fronto-central theta ( FCθ; 2–7 Hz ) power was measured in two ways . First , EEG data were decomposed into their time-frequency representation via complex Morlet wavelet convolution ( between 2 and 12 cycles per wavelet , linearly increasing across 90 linear-spaced frequencies from 1 to 30 Hz ) and the resulting power estimates were normalized by the decibel ( dB ) transform ( dB = 10log10[power/baseline] ) . The baseline consisted of across-trial average power during the 0 . 3 s preceding stimulus onset , calculated and applied separately within each trial-type ( Go , undetected error , detected error ) . This approach yielded a condition-averaged time-frequency plot that allowed us to select channels and frequency boundaries for FCθ analysis in a manner that was orthogonal to any potential trial-type effects . Black lines in this plot ( Figure 3a ) enclose regions in which contiguous time-frequency pixels were significantly different from the pre-stimulus baseline at p < 0 . 01 , for at least 400 ms and at least 5 consecutive frequency bins . Second , power estimates for all subsequent FCθ analyses were derived by band-pass filtering the EEG data from 2 to 7 Hz ( using the fir1 Matlab function to construct a narrow two-way least squares FIR filter kernel ) , Hilbert-transforming the filtered data to derive the analytic signal , and then converting to power . Compared to wavelet convolution this filter-Hilbert method affords greater control over the frequency characteristics of the filter , though in practice the results from both methods were qualitatively very similar . As for the previous wavelet-based approach , analyses were conducted on power estimates that were dB-normalized using a condition-specific trial-averaged pre-stimulus baseline , thus leaving within-condition trial-by-trial fluctuations in baseline power intact . Complementarily , the reported between-subjects FCθ correlation ( Figure 6e ) was conducted on unbaselined power estimates , thus leaving between-individual differences in baseline FCθ power intact . Both within- and between-subjects variation in baseline power emerged to be sources of variance contributing to the respective effects ( Figure 3—figure supplement 3c-e; Figure 6—figure supplement 2 ) . We also verified that the main effect of error detection on FCθ remained unchanged when no baseline was applied ( Figure 3—figure supplement 3b ) . First- and second-order trial-averaged CPP signals were measured as the average voltage per m2 from three centro-parietal electrodes centered on the region of maximum component amplitude in the grand-average response-locked topography ( Pz , P1 , P2 ) , and were low-pass filtered up to 10 Hz for analysis and display . FCθ was measured as the average power across six fronto-central electrodes , also centered on the topographic maximum ( FCz , FC1 , FC2 , Cz , C1 , C2 ) . The relationships between RT and signal build-up rates and amplitudes were examined for the CPP on go trials ( Figure 2a ) , the second-order continuation of this signal on detected errors ( Figure 2d ) , and FCθ on both trial-types ( Figure 3d , e ) . Analysis of Go-trial dynamics was restricted to trials with primary RTs > 350 ms ( leading to the exclusion of 17 . 4 ± 12 . 1% of trials per subject ) because the amplitude of the CPP signal was affected by visual-evoked potentials that coincided with the evolution of the decision-related activity on quicker trials ( Figure 2—figure supplement 2 ) . For each participant , single-trial waveforms were sorted into 3 equal-sized bins according to RT ( primary RT for Go trials , RTd for detected errors ) and averaged . To establish the timing of the relationship between signal build-up rate and RT , we measured the temporal slope of each signal in each subject’s bin-averaged waveforms using a sliding window of 150 ms , covering the entirety of both response- and detection-aligned waveforms . Build-up rate was computed as the slope of a straight line fitted to the signal within each temporal window , and a linear contrast was applied to this metric across RT bins . The centers of windows that were characterized by a significant group-level contrast in the expected direction ( linear decrease in build-up rate with increasing RT; p < 0 . 05 , one-tailed ) are marked by a black running line in the associated plots . To establish the timing of the relationship between signal amplitude and RT , we conducted a linear contrast of amplitude as a function of RT bin for each temporal sample . Samples characterized by contrasts that deviated from zero at the group level ( p < 0 . 05 , two-tailed ) are marked by a gray line in the associated plots . This sample-wise approach was also employed to characterize the effect of error detection on the amplitude of both centro-parietal ( Figure 2b ) and FCθ ( Figure 3b ) signals . For FCθ and CPP ROC curve analyses ( Figure 2c; Figure 3c ) , single-trial waveforms for detected and undetected errors were pooled across all subjects and the area under the ROC curve was calculated for the average of each signal within discrete peri-error time windows ( window width of 20 ms , moving in 20 ms increments , from −0 . 4 to +0 . 6 s relative to error ) . Significant deviations in classification accuracy from chance levels were determined via permutation testing ( 1000 iterations with random trial reassignment conserving individual detected versus undetected error proportions ) . All further single-trial analyses of the second-order CPP leveraged waveforms that were low-pass filtered to 6 Hz to increase signal-to-noise . Single-trial amplitude was defined as the mean power from –0 . 1 to +0 . 4 s relative to error commission for FCθ and the mean signal in the 0 . 2 s preceding error detection for the CPP ( Figure 4a ) . Single-trial build-up rate was measured as the slope of a straight line fitted to each waveform using the interval 0 to +0 . 2 s for FCθ and +0 . 1 to +0 . 3 s for the CPP , both relative to error commission ( Figure 4b ) . Single-trial peak latency was measured as the time of maximum signal amplitude relative to error commission within a dynamic measurement window with a start time of −0 . 1 s for FCθ and +0 . 1 s for the CPP , and an end time of the RTd for that trial +0 . 15 s ( Figure 4c ) . In cases where trials were initially assigned the minimum possible latency given the above constraints , the window start time was adjusted to be the earliest latency at which the waveform next became positive-going . The independent contributions of FCθ and the CPP toward RTd were examined via single-trial within-subjects robust regression ( O'Leary , 1990 ) . For signal amplitude , the equation RTd = β0 + β1FCθAmp + β2CPPAmp yielded fitted regression coefficients representing the linear relationships between RTd and both single-trial theta power ( β1 ) and CPP amplitude ( β2 ) . For build-up rate and peak latency , two further models were constructed by replacing the predictor variables where appropriate . All coefficients in a given model were estimated simultaneously via type III sum of squares . RTd was log-transformed and peak signal latency was square root-transformed to normalize their respective distributions before coefficient estimation . Variance inflation factors for all predictors across all models were <1 . 8 , indicating weak multi-collinearity . For the amplitude and build-up rate metrics , effect sizes of the fitted β coefficients ( effect size t = β/s . e . m . ) were tested for group-level statistical significance via one-sample t-test ( H0: effect size = 0 ) and contrasted via paired t-test ( H0: effect size1 − effect size2 = 0 ) . For the latency metric , our use of a dynamic measurement window that was determined by single-trial RTd ensured an arbitrary positive correlation between single-trial signal latency and RT . To test for effect significance , we thus compared the observed effect sizes derived from the above regression model to the expected values of the same effect size metrics computed in the case in which signal latencies were randomly chosen from anywhere within each trial’s measurement window . This process was repeated 1 , 000 times per subject to derive subject-specific permuted distributions of the FCθ and CPP latency effects , against which we tested for statistical significance . The above procedures for amplitude , build-up rate and latency effects were also repeated on an electrode-by-electrode basis to construct topographic representations of where these effects were strongest on the scalp ( Figure 4 ) . For the latency analysis , the trial-by-trial timing of the peak positivity or negativity was extracted for each electrode , depending on whether the trial-averaged amplitude at that electrode in the 0 . 3 s preceding the error detection report was above or below zero , respectively . We employed mediation analysis ( M3 toolbox for Matlab; http://wagerlab . colorado . edu/tools ) to establish whether second-order CPP build-up rate mediated the relationship between FCθ power and RTd ( Figure 5 ) . For this analysis , we measured CPP build-up rate from −0 . 3 to −0 . 1 s relative to error detection report in order to minimize the temporal overlap between the FCθ and CPP measures , though the results were very similar when the original error-aligned measurement window for build-up rate was employed . Single-trial values for each measure were z-scored within-subjects and pooled across-subjects . To mitigate the low signal-to-noise ratio inherent in correlating two noisy single-trial electrophysiological metrics , average values for each of FCθ , CPP build-up and RTd were computed in bins of 12 trials that were grouped after sorting trials in order of increasing RTd . Mediation effects at larger bin sizes were of comparable magnitude ( Figure 5—figure supplement 1 ) . For CPP build-up rate to be considered a significant mediator , it was required to reach statistical significance in three tests: it must be related to the predictor ( FCθ ) , it must be related to the outcome ( RTd ) while controlling for the predictor , and the mediation effect ( evaluating whether some covariance between predictor and outcome can be explained by the mediator ) must be significant . Significance of the associated path coefficients was assessed via bias-corrected bootstrap tests with 10 , 000 samples ( Wager et al . , 2008 ) . Second-order behavioral data ( error detection accuracy and RTd ) were decomposed into latent decision-making parameters via a novel application of the one-choice drift diffusion model ( Ratcliff and Van Dongen , 2011; Figure 6a ) . Noisy evidence for an error was assumed to accumulate over time at drift rate ν until a decision bound a was met , at which point error detection was achieved . The moment-to-moment noise in the evidence is determined by the s parameter , which refers to the standard deviation of a zero-mean Gaussian distribution from which random increments to the deterministic component of the accumulation process ( represented by ν ) are drawn . As is common in fits of the drift diffusion model to data , s was fixed at 0 . 1 in order to scale all other parameters in the model across individuals . Drift rate was assumed to be normally distributed across trials with a standard deviation η , and all non-decision-related processing was assigned to a non-decision time parameter tnd . The inclusion of the η parameter accorded with previous one-choice model fits to first-order behavior which suggested that variability in drift rate is necessary to account for the various observable shapes of hazard function in one-choice data ( Ratcliff and Van Dongen , 2011 ) . We made the additional assumption that the temporal integration process terminated if a was not reached by a time deadline , thereby resulting in an undetected error . A deadline on post-decision accumulation also features in a prominent model of fast changes of mind in decision-making ( Resulaj et al . , 2009 ) and was a free parameter in that study . Here , we estimated the subject-specific detection deadline empirically in order to retain a degree of freedom when assessing model fit: any extreme outliers ( > mean + 3 . 5 s . d . ) were trimmed from each subject’s RTd distribution and the deadline was defined as the slowest remaining RT per subject . This procedure yielded an average deadline of 1 . 17 s relative to initial error commission ( ±0 . 19; range 0 . 81 to 1 . 47 s ) . Two additional assumptions were made when plotting simulated decision variables derived from the best-fitting model parameters ( Figure 6d ) , both of which were informed by characteristics of the second-order CPP . First , 90 ms of each subject’s fitted tnd parameter was allotted to post-threshold response preparation , and any residual tnd determined the length of the delay between error commission and the start of evidence accumulation . Second , the decision variable was subject to a linear decay to baseline over the 300 ms after the decision bound was reached . There is no analytical solution for RT distributions with negative drift rate , which can result from including η in the one-choice model ( Ratcliff and Van Dongen , 2011 ) . The model was therefore implemented as a simulation using a random walk approximation to the diffusion process with 15 , 000 iterations per distribution at 1 ms step size . In order to fit the model to the observed data for each subject , five RT quantiles ( 0 . 1 , 0 . 3 , 0 . 5 , 0 . 7 , 0 . 9 ) were computed from that subject’s RTd distribution and the proportions of all error trials ( detected + undetected ) lying between those quantiles were multiplied by the total number of error trials to yield observed values ( O ) . Thus , the defective cumulative probability distribution of error detection reports was used to derive per-quantile trial frequencies , which allowed the model to simultaneously fit both RTd and error detection accuracy . We then calculated the model-estimated proportions of trials that lay between these RT quantiles , and these were multiplied by the number of actual observations to yield the model-derived expected values ( E ) . A χ2 statistic Σ ( O – E ) 2/E was computed and the parameters of the model were adjusted by a particle swarm optimization routine ( Birge , 2003 ) to minimize this value iteratively ( 30 particles , set at pseudorandom starting points in parameter space ) . Initial attempts at parameter estimation indicated that the particle swarm approach tended to be more robust to local minima than the commonly-used Simplex minimization routine . As described elsewhere ( Ratcliff and Van Dongen , 2011 ) , the one-choice diffusion model suffers from a parameter identifiability problem whereby different estimates of the v , η and a parameters can produce similar goodness of fit but vary in magnitude by as much as 2:1 . However , the ratio of v/η remains invariant across different model fits . We therefore employed this ‘drift ratio’ quantity , which is analogous to d’ in classic signal detection theory ( Ratcliff and Van Dongen , 2011 ) , as our model-based estimate of the quality of the evidence feeding into each subject’s error detection decision process . In the reported between-subjects correlations ( Figure 6e , f , g ) , drift ratio was correlated against summary electrophysiological measures for each subject . Amplitude and peak latency measures were calculated by averaging across single-trial estimates of these metrics on detected error trials , derived via the same measurement windows that were used for the previous within-subjects regression analyses . In an effort to increase the signal-to-noise of the build-up rate metric , per-subject build-up rate for each signal was defined as the slope of a linear fit to the average error-locked waveform on detected error trials , within a window that started at a subject-specific signal onset time obtained by visual inspection and ended at the subject-specific peak signal latency . One outlier data point with an absolute studentized deleted residual value > 4 in all bivariate correlations was excluded from the reported analyses , though all relationships remained at least marginally significant when this subject was included ( all p < 0 . 06 ) .	Reflecting on our previous choices and accurately representing our confidence in their accuracy allows us to detect , correct and learn from our errors . Yet , it remains poorly understood how such “metacognition” , or thoughts about thoughts , occurs in the human brain . In particular , a long-standing debate in this area of research concerns whether metacognitive processes in the brain occur at the same time as those that determine the actual choice , or whether they develop after the choice has been made and rely on different information . Now , Murphy et al . have recorded brain activity in human volunteers who were carrying out a simple task in order to explore metacognition . In short , the volunteers looked at colored words and decided if each word matched its color ( e . g . , is the word ‘RED’ also written in a red font ? ) . At the same time , the volunteers chose whether or not to press a button depending on the specific color/word combination shown , and most importantly reported whenever they noticed that they made an error in the task . This approach allowed Murphy et al . to chart the development of choices and detection of errors as they occurred in the volunteers’ brains . This revealed that the metacognitive judgement about each choice relied on information that was gathered after the point the initial choice was made . Further analysis then suggested that this process relies , at least in part , on a signal generated in a region at the front of the brain . Together , these findings suggest that metacognitive decisions rely on processes that are similar to those behind other decisions , but with a few important differences . Namely , the metacognitive process plays out at a different point in time , and likely incorporates distinct sources of information . Further work should aim to clarify the nature of these sources of information and describe their specific contributions to the process .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Neural evidence accumulation persists after choice to inform metacognitive judgments
To anticipate potential seedling damage , plants block seed germination under unfavorable conditions . Previous studies investigated how seed germination is controlled in response to abiotic stresses through gibberellic and abscisic acid signaling . However , little is known about whether seeds respond to rhizosphere bacterial pathogens . We found that Arabidopsis seed germination is blocked in the vicinity of the plant pathogen Pseudomonas aeruginosa . We identified L-2-amino-4-methoxy-trans-3-butenoic acid ( AMB ) , released by P . aeruginosa , as a biotic compound triggering germination arrest . We provide genetic evidence that in AMB-treated seeds DELLA factors promote the accumulation of the germination repressor ABI5 in a GA-independent manner . AMB production is controlled by the quorum sensing system IQS . In vitro experiments show that the AMB-dependent germination arrest protects seedlings from damage induced by AMB . We discuss the possibility that this could serve as a protective response to avoid severe seedling damage induced by AMB and exposure to a pathogen . Seeds are remarkable structures promoting plant dispersal by preserving the plant embryo in a desiccated and highly resistant state . Their appearance in the course of land plant evolution is regarded as a cornerstone of the striking spread and diversification of angiosperms among terrestrial plants . Seed imbibition with water is the necessary first step to permit germination , transforming the embryo into a fragile juvenile seedling . However , upon imbibition , the seed is also exposed to potentially fatal environmental conditions for the future seedling . To avoid premature death , plants have evolved control mechanisms that block germination under unfavorable conditions to maintain the highly protected embryonic state ( Lopez-Molina et al . , 2001; 2002 ) . Historically , studies focused on how seeds respond to abiotic stresses such as high temperature , canopy light -unfavorable for photosynthesis- or high salinity ( Baskin and Baskin , 1998; Reynolds and Thompson , 1971; Negbi et al . , 1968 ) . In Arabidopsis , perception of abiotic factors leads to changes in seed endogenous levels of gibberellic acid ( GA ) and abscisic acid ( ABA ) , two phytohormones playing a central role to control germination ( Cutler et al . , 2010; Lau and Deng , 2010; Nonogaki , 2014; Davière and Achard , 2016 ) . Under favorable conditions , GA synthesis increases , which is necessary to initiate germination . GA induces proteolysis of DELLA factors repressing germination , which are encoded by a family of five genes: RGL2 , GAI , RGA , RGL1 and RGL3 . DELLA factors collectively repress germination by promoting the accumulation of ABA and ABA-response transcription factors ( TFs ) ultimately repressing seed germination such as ABI3 and ABI5 ( Lee et al . , 2002; Penfield et al . , 2006; Lee et al . , 2010a; Lopez-Molina and Chua , 2000; Lopez-Molina et al . , 2001; 2002; Finkelstein and Lynch , 2000; Piskurewicz et al . , 2008; 2009 ) . RGL2 can play a prominent role among DELLAs to stimulate ABA signaling and thus repress germination . Indeed , only rgl2 mutants can germinate when seeds are treated with a GA synthesis inhibitor ( Lee et al . , 2002; Piskurewicz et al . , 2008 ) . This is likely due to the positive regulation of RGL2 mRNA levels by ABA , which generates a positive feedback loop sustaining high RGL2 accumulation relative to other DELLAs ( Piskurewicz et al . , 2008 , 2009 ) . How DELLA factors stimulate ABA signaling in seeds remains to be understood . DELLAs are unable to directly bind to DNA and thus are more likely to interact with TFs or other factors regulating ABA signaling . Recent work has shown that DELLA protein activity is regulated through phosphorylation , SUMOylation , O-GlcNAcylation , or O-fucosylation ( Conti et al . , 2014; Qin et al . , 2014; Zentella et al . , 2016 , 2017 ) . Along with abiotic factors , seeds are also continuously confronted to bacteria , fungi and animals ( e . g . nematodes ) present in soil and potentially acting as pathogens or commensals ( Silby et al . , 2011 ) . Little is known about whether biotic factors released by non-plant organisms induce seed germination responses in plants . To address this question in the model plant Arabidopsis , we considered the case of Pseudomonas , a genus of Gram-negative bacteria having pathogenic and commensal interactions with both animals and plants ( Silby et al . , 2011 ) . We confronted Arabidopsis seeds to different Pseudomonas species and found a strong germination repressive activity ( GRA ) released by Pseudomonas aeruginosa . Strikingly , P . aeruginosa did not repress the germination of Arabidopsis mutant seeds lacking DELLA factors or ABA signaling components . Metabolomic and bioguided biochemical fractionation approaches led us to identify the oxyvinylglycine L-2-amino-4-methoxy-trans-3-butenoic acid ( also referred as methoxyvinylglycine or AMB ) as the main GRA released by P . aeruginosa . AMB production and release is dependent on the five-gene operon ambABCDE controlling the newly identified quorum-sensing IQS in P . aeruginosa ( Lee et al . , 2010b , 2013a ) . Using synthetic AMB , we provide genetic evidence that the activity of DELLAs to stimulate ABA-dependent responses is enhanced in seed exposed to AMB . Whether this reflects a mechanistic link between AMB and DELLAs is not known . Furthermore , AMB induces severe developmental defects in juvenile seedlings . In contrast , germination-arrested seeds are capable to produce viable plants when no longer exposed to AMB or P . aeruginosa . Oxyvinylglycines are known to inhibit irreversibly pyridoxal phosphate ( PLP ) -dependent enzymes ( Berkowitz et al . , 2006 ) . Furthermore , AMB is a methionine analog . Our results suggest that AMB does not block germination by interfering with ethylene , auxin or methionine synthesis . Rather , our results indicate that AMB interferes with an unknown and GA-independent mechanism promoting DELLA-dependent germination arrest , which involves DELLA-dependent stimulation of ABA signaling . We discuss possible interpretations of our findings , including the possibility that the germination arrest could serve as a protective response to avoid severe seedling damage induced by AMB and exposure to a pathogen . We explored whether Pseudomonas bacteria release compounds inhibiting Arabidopsis seed germination as follows: ( 1 ) individual Pseudomonas species were propagated for 3 days on germination agar medium supplemented with a carbon source; ( 2 ) thereafter Arabidopsis seeds were sown on the germination medium at various distances from the bacteria; ( 3 ) germination was scored after culturing seeds for 3 days ( Materials and methods , Figure 1—figure supplement 1A ) . Germination was not markedly repressed by P . fluorescens , P . putida , P . syringae or Escherichia coli ( used as a Gram-negative non-Pseudomonas species control ) ( Figure 1A and B ) . In marked contrast , germination was strongly repressed by P . aeruginosa ( WT strains PAO1 and PA14 ) ( Figure 1A and B ) . The percentage of germination increased with increasing distance separating seeds and P . aeruginosa ( Figure 1A and Figure 1—figure supplement 1B ) . Furthermore , germination arrest did not occur in absence of germination medium separating bacteria and seeds ( Figure 1—figure supplement 1C ) . From these observations , we conclude that P . aeruginosa releases a germination repressive activity ( GRA ) that diffuses in the germination medium . Upon imbibition seeds respond to unfavorable abiotic conditions by blocking GA synthesis , which promotes DELLA factors accumulation . In turn , DELLA factors collectively repress seed germination by stimulating ABA signaling including the accumulation of the germination repressor TF ABI5 ( Piskurewicz et al . , 2008 , 2009 ) . We studied whether GA and ABA signaling pathways play a role to repress germination in seeds exposed to P . aeruginosa . GA signaling mutant seeds , lacking one or several DELLA factors and particularly ∆della mutant seeds lacking all five DELLA factors , had a higher percentage of seed germination in presence of WT P . aeruginosa ( PAO1 ) ( Figure 2A ) . Furthermore , Arabidopsis mutant seeds deficient in ABA synthesis ( aba1 ) or signaling ( abi3 , abi5 ) also had a higher percentage of seed germination in presence of P . aeruginosa ( PAO1 ) ( Figure 2B ) . We also exposed seeds to lyophilized extracts of P . aeruginosa liquid culture medium ( Materials and methods , Figure 2—figure supplement 1A ) . These bacteria-free extracts ( thereafter referred as ‘extracts’ ) also elicited DELLA-dependent germination arrest responses ( Figure 2—figure supplement 1B , C and D ) . We monitored the accumulation of the DELLA factor RGL2 and the ABA response TF ABI5 in seeds exposed to P . aeruginosa . In absence of bacteria , RGL2 and ABI5 were detectable during the first 44 hr upon seed imbibition but their accumulation decreased thereafter , which was associated with increasing germination percentage , consistent with previous reports ( Figure 2C; Figure 2—figure supplement 2A and B ) ( Lee et al . , 2002; Piskurewicz et al . , 2008 ) . However , WT seeds exposed to P . aeruginosa ( PAO1 ) persistently accumulated RGL2 and ABI5 up to 148 hr upon imbibition , which was associated with low germination percentage ( Figure 2C ) . In contrast , ∆della mutant seeds exposed to P . aeruginosa extracts did not maintain ABI5 accumulation over time upon imbibition , which was associated with increasing germination percentage ( Figure 2D ) . These observations indicate that the GRA released by P . aeruginosa represses germination by promoting DELLA-dependent increase of the germination repressor ABI5 , as previously shown with seeds unable to synthesize GA upon imbibition ( Piskurewicz et al . , 2008 , 2009 ) . However , and surprisingly , exogenous GA ( 10 µM ) did not promote germination of WT seeds exposed to P . aeruginosa extracts ( Figure 3A ) . This was associated with persistent accumulation of RGL2 and ABI5 ( Figure 3B ) . In contrast , and expectedly , exogenous GA promoted the germination of WT seeds exposed to paclobutrazol ( PAC ) , an inhibitor of GA synthesis , which was associated with downregulation of RGL2 and ABI5 protein levels , consistent with previous results ( Figure 3A , Figure 3—figure supplement 1 and Figure 5B ) . Thus , the GRA could elicit a DELLA-dependent germination arrest irrespective of GA levels in seeds . To further study how the GRA triggers a DELLA-dependent germination arrest , we compared the early transcriptome of WT and ∆della seeds exposed to PAC or P . aeruginosa ( PAO1 ) extract ( Materials and methods , Figure 3C and D ) . We also included in the analysis the effect of exogenous GA ( Materials and methods ) . The expression of as many as 1103 genes was changed in PAC-treated WT seeds relative to untreated WT seeds ( MS vs PAC , Figure 3C , Supplementary file 1 ) . In contrast , only three genes had their expression affected in PAC-treated Δdella seeds relative to untreated Δdella seeds ( MS vs PAC , Figure 3C , Supplementary file 1 ) . Furthermore , only 36 and 2 genes had their expression changed when WT and Δdella seeds were treated with both PAC and GA ( 10 µM ) , respectively ( MS vs PAC + GA , Figure 3C , Supplementary file 1 ) . Thus , DELLA factors influence gene expression mainly when GA synthesis is compromised , consistent with previous reports ( Cao et al . , 2006 ) . In contrast to PAC-treated WT seeds , only 130 genes had their expression changed in WT seeds exposed to PAO1 extract relative to non-exposed WT seeds ( MS vs PAO1 extract , Figure 3D , Supplementary file 2 ) . Among the 130 genes , the expression of 87 genes was DELLA-dependent since it did not change in Δdella seeds exposed to PAO1 extract ( Figure 3D , Supplementary file 2 ) . Strikingly , about a quarter of the 87 genes ( 24 genes ) had their expression unchanged when GA was included in the medium together with PAO1 extract ( MS vs PAO1 extract +GA , Figure 3D ) . Altogether , these results show that the GRA released by P . aeruginosa is able to elicit gene expression responses in a DELLA-dependent and GA-independent manner . We next sought to identify the GRA by first identifying P . aeruginosa genes that are necessary to its release . Bacteria have a quorum sensing ( QS ) system regulating gene expression to mount coordinated behavioral responses when a bacterial population reaches high densities ( Ng and Bassler , 2009 ) . The P . aeruginosa QS has four QS subsystems influencing each other: 1 ) las , 2 ) rhl , 3 ) pqs and 4 ) the recently discovered IQS ( for ‘Integrating the QS network’ ) . Each subsystem consists of genes or operons producing signaling molecules , referred as autoinducers , which interact with and activate cognate transcription factors controlling bacterial coordinated behavior ( Lee and Zhang , 2015; Papenfort and Bassler , 2016; Moradali et al . , 2017 ) . We asked whether the GRA released by P . aeruginosa is under QS control . We exposed WT seeds to P . aeruginosa mutant strains deficient in ( 1 ) the LASI and RHLI operons ( ΔlasIΔrhlI double mutants ) , which control the production of the homoserine lactones autoinducers , ( 2 ) the PQS operon ( ΔpqsA and ΔpqsH mutants ) , which controls the production of the quinolones autoinducers and ( 3 ) the AMB operon ( ΔambE mutant ) , which is necessary for the production of the autoinducer 2- ( 2-hydroxyphenyl ) -thiazole-4-carbaldehyde ( IQS ) ( Lee et al . , 2013a ) . We observed markedly lower GRAs released by ΔlasIΔrhlI and particularly with ΔambE mutants ( Figure 4A , Figure 4—figure supplement 1 , Figure 2—figure supplement 1C ) . Lower GRAs were also observed with ΔambA-D mutants , each affected in individual AMB operon genes ( Figure 4B ) . Furthermore , and strikingly , transgenic WT P . aeruginosa bacteria overexpressing the AMB operon ( AMBox ) released a higher GRA relative to the parental PAO1 line ( Figure 4—figure supplement 1 ) . Altogether , these results show that the GRA released by P . aeruginosa is dependent on QS activity and particularly that of IQS . The importance of IQS is further suggested by the fact that P . aeruginosa is the only Pseudomonas species used in this study having the IQS QS ( Figure 1A ) ( Rojas Murcia et al . , 2015; Lee et al . , 2013b ) . We undertook an unbiased metabolomic approach complemented with biochemical purification procedures to identify the GRAs present in P . aeruginosa extracts ( Figure 4—figure supplement 2 ) . In the metabolomic approach , untargeted UHPLC-HRMS2 ( Ultra-High-Performance Chromatography hyphenated to High-Resolution Tandem Mass Spectrometry ) data were acquired from extracts of P . aeruginosa strains releasing a GRA ( PAO1 and ∆pqsa ) , not releasing a GRA ( ∆ambE and ∆lasI/rhlI ) or releasing a higher GRA ( AMBox ) ( Materials and methods , Figure 4—figure supplements 3 , 4 , Supplementary file 3 ) . These analyses provided a list of biomarkers responsible for the metabolic differences between strains ( Supplementary file 4 ) . The top discriminant feature was the mass spectrometry signal corresponding to oxyvinylglycine L-2-amino-4-methoxy-trans-3-butenoic acid ( also referred as methoxyvinylglycine or AMB ) ( Figure 4—figure supplement 5 , Supplementary file 3 and 4 ) . We next used chromatographic methods to fractionate WT PAO1 , AMBox and ∆ambE extracts , which led to the identification of a GRA present in a polar fraction of WT PAO1 and AMBox extracts but absent in that of ∆ambE extracts ( Materials and methods , Figure 4—figure supplement 6A ) . AMBox extracts were used to further fractionate the GRA-containing polar fraction ( Figure 4—figure supplement 6B ) . This led to the isolation of a purified fraction containing AMB ( as confirmed by NMR ) ( Figure 4—figure supplement 6B , Appendix 1—figures 1 and 2 ) and the GRA ( Figure 4—figure supplement 7 ) . AMB present in this purified fraction is hereafter referred as ‘AMBi’ . Altogether , these results point to AMB as being the main GRA . Consistent with this hypothesis synthetic AMB dose response for germination inhibition confirmed that AMB inhibits WT seed germination ( Figure 4C ) . Furthermore , mutant seeds deficient in GA signaling ( ∆della ) , ABA biosynthesis ( aba1 ) or ABA signaling ( abi3 , ) had a higher percentage of seed germination in presence of synthetic AMB ( Figure 4D ) . A concentration of 50 μM synthetic AMB had a GRA equivalent to that 0 . 8 mg/ml PAO1 extract in germination plates ( see dashed red line in Figure 4C E ) . However , we estimated that this extract concentration provides only 16 μM of natural AMB ( quantification by targeted UHPLC-HRMS2 approach , Figure 4—figure supplement 8 ) . This discrepancy could be explained as follows: ( 1 ) there are other compounds present in PAO1 extracts enhancing the GRA of AMB , ( 2 ) AMB is not the only compound released by P . aeruginosa having a substantial GRA . To test the first possibility , we supplemented ∆ambE extracts , lacking AMB , with synthetic AMB so as to match the amount of AMB naturally present in PAO1 extracts . The GRA of the AMB-supplemented ∆ambE extract was similar to that present in PAO1 extracts ( Figure 4E ) . Furthermore , we also purified the natural AMB present in Pseudomonas extracts ( Materials and methods , Figure 4—figure supplement 6B ) . ∆ambE extracts supplemented with AMBi or synthetic AMB contained the same GRA ( Figure 4—figure supplement 9 ) . Taken together , these results strongly suggest that the GRA of synthetic AMB can be enhanced by the presence of other compounds present in P . aeruginosa extracts . We also tested the GRA of aminoethoxyvinylglycine ( AVG ) , another oxyvinylglycine . As much as 200 μM of AVG of did not noticeably repress Arabidopsis seed germination , consistent with previous reports , indicating that AMB is an oxyvinylglycine specifically repressing Arabidopsis seed germination ( Figure 4—figure supplement 10 ) ( Wilson et al . , 2014 ) . AVG is known to inhibit ethylene synthesis , suggesting that AMB does not block germination by inhibiting ethylene synthesis ( further discussed below ) ( Adams and Yang , 1979; Huai et al . , 2001; Capitani et al . , 2002 ) . We next verified whether AMB could recapitulate the effects of P . aeruginosa cells or WT PAO1 extracts on RGL2 and ABI5 protein accumulation . WT seeds exposed to AMB persistently accumulated RGL2 and ABI5 over time , unlike unexposed seeds ( Figure 5A ) . Persistent high ABI5 accumulation was not observed in AMB-treated Δdella mutant seeds ( Figure 5A ) . Furthermore , beyond 24 hr of imbibition in presence of AMB , PAC and GA , RGL2 protein levels persisted in seeds despite the presence of GA , unlike seeds treated with PAC and GA only ( Figure 5B ) . This was associated with persistent ABI5 accumulation and absence of seed germination ( Figure 5B ) . Thus , AMB induces changes in RGL2 and ABI5 accumulation in a manner similar to that observed with P . aeruginosa or WT PAO1 extracts ( Figures 2C , D and 3B ) . Altogether , these results conclusively show that AMB is the main GRA released by P . aeruginosa repressing germination in a DELLA- and ABA-signaling-dependent manner . To better understand the genetic requirement of functional DELLA genes for the AMB-dependent germination arrest , we focused on the DELLA factors RGL2 , GAI and RGA ( Piskurewicz et al . , 2008; 2009 ) . We asked whether GA-dependent DELLA protein degradation was perturbed in AMB-treated seeds . WT seeds were imbibed under normal ( MS ) conditions or in presence of PAC , AMB or ABA . As expected , RGL2 , GAI and RGA protein levels markedly increased in presence of PAC , consistent with the notion that low GA stabilizes DELLA proteins ( Figure 6A , B and C , Figure 2—figure supplement 2C ) . In contrast , no marked increase in RGL2 , GAI and RGA protein levels was observed in presence of AMB or ABA despite the fact that both treatments arrested seed germination as in PAC-treated seeds ( Figure 6A , B and C ) . We also monitored the extinction of RGL2 , GAI and RGA protein accumulation upon exposure to GA in absence or presence of AMB ( Figure 6D , Figure 6—figure supplement 1A and B ) . WT seeds were first treated for 30 hr with PAC to allow for DELLA protein stabilization and high accumulation . After 30 hr seeds were further treated with AMB for 12 hr to ensure AMB presence within seed tissues prior to adding GA . Upon addition of GA to the medium RGL2 , RGA and GAI protein levels decreased over 28 hr in a similar manner to that observed in control plates lacking AMB ( Figure 6D , Figure 6—figure supplement 1A and B ) . Penetrance of AMB within seeds was confirmed by the absence of seed germination of AMB-treated seeds despite the presence of GA ( Figure 6D ) . Altogether , these results strongly suggest that GA-dependent DELLA protein degradation is not affected in AMB-treated seeds . RGL2 persistently accumulated at late time points upon seed imbibition in presence of AMB despite the presence of GA ( Figure 5B ) . Furthermore , ABA signaling in seeds promotes RGL2 mRNA accumulation as well as that of ABI5 ( Lopez-Molina et al . , 2001; Piskurewicz et al . , 2008; 2009 ) . We therefore hypothesized that the DELLA activity promoting ABA signaling in seeds is enhanced in AMB-treated seeds ( Lopez-Molina et al . , 2001; Piskurewicz et al . , 2008; 2009 ) . In turn , this would explain the persistent RGL2 and ABI5 protein accumulation in AMB-treated seeds ( Figures 2C , 4A B ) . To test this hypothesis genetically , we measured RGL2 and ABI5 mRNA levels using total RNA isolated 30 hr and 44 hr upon imbibition from the same seed material used in Figure 5A . We also analyzed the expression of the ABA-responsive genes EM1 and NCED6 ( Lopez-Molina and Chua , 2000; Lopez-Molina et al . , 2001; Lefebvre et al . , 2006; Martínez-Andújar et al . , 2011 ) . Upon 30 hr of imbibition , seed accumulated the same RGL2 levels in absence ( MS ) or presence of AMB ( AMB ) ( Figure 5A ) . Nevertheless , AMB-treated seeds accumulated markedly higher ABI5 , EM1 and NCED6 mRNA levels relative to seeds imbibed in absence of AMB ( Figure 6E ) . In contrast , AMB-treated Δdella mutant seeds accumulated lower ABI5 , EM1 and NCED6 mRNA expression at 30 hr ( Figure 6E ) . Upon 44 hr of imbibition , AMB-treated seeds further increased the expression of ABI5 , EM1 and NCED6 mRNA as well as that of RGL2 ( Figure 6E ) . Collectively , these genetic observations support the notion that the DELLA activity promoting ABA signaling in seeds is enhanced in AMB-treated seeds ( Figure 7E ) . Whether this is the result of a direct interaction between AMB and DELLAs is not known . Interestingly , AMB could still mildly stimulate EM1 , NCED6 and particularly ABI5 mRNA expression in Δdella mutant seeds ( Figure 6E ) . Accordingly , ABI5 protein accumulation , although diminishing , remained higher 44 hr and 68 hr after imbibition of AMB-treated Δdella seeds relative to non-treated seeds ( Figure 5A ) . These genetic observations suggest that a residual activity promoting ABA synthesis or signaling independently of DELLA factors is present in AMB-treated seeds . The relative proportion of germinating seeds relative to non-germinating seeds depended on the distance separating seeds from P . aeruginosa cells or on the concentration of AMB used in the germination plate ( Figure 1A , Figure 1—figure supplement 1 ) . Seeds that germinated in either presence of P . aeruginosa or AMB similarly produced pale seedlings whose growth was severely delayed bearing diminutive roots ( Figure 7A ) . These developmental defects diminished with increased distance between seeds and bacteria or decreased AMB concentrations . Furthermore , they were no longer observed in seeds exposed to ∆ambE P . aeruginosa mutants , unable to produce AMB ( Figure 7A ) . Altogether , these observations show that AMB released by P . aeruginosa severely perturbs seedling development . Remarkably , within a seed population exposed to P . aeruginosa , WT seeds that did not germinate germinated upon transfer to plates lacking P . aeruginosa and produced normal seedlings ( Figure 7B ) . The resulting WT seedlings produced a normal seed yield ( Figure 7C ) . In contrast , WT seeds that had germinated in presence of P . aeruginosa failed to recover ( Figure 7B ) . Furthermore , when WT and Δdella seeds were exposed for 3 days to P . aeruginosa prior to transfer to a bacteria-free medium , WT seeds had a higher survival rate relative to Δdella seeds ( Figure 7D ) . These observations therefore strongly suggest that germination-arrested seeds in presence of AMB are able to retain their vitality unlike newly emerged seedlings . Oxyvinylglycines are a class bacterially produced compounds whose biological function is unclear . Oxyvinylglycines are known to inhibit irreversibly pyridoxal phosphate ( PLP ) -dependent enzymes ( Berkowitz et al . , 2006 ) . Oxyvinylglycines had been previously associated with germination repressive activities ( GRAs ) after exposing graminaceous ( Poaceae ) seeds to bacterial culture filtrates . The best documented cases are those of AVG ( aminoethoxyvinylglycine , produced by Streptomyces sp ) and FVG ( 4-formylaminooxyvinylglycine , produced by Pseudomonas fluorescens strain WH6 ) that were directly and indirectly linked with a GRA , respectively ( McPhail et al . , 2010; Okrent et al . , 2017 ) . Only in the case of AVG a GRA could be established using a pure synthetic compound: 100 µM AVG inhibited the germination of Poa annua seeds . However , we found here that as much as 200 µM AVG did not inhibit Arabidopsis seed germination , consistent with previous reports ( Wilson et al . , 2014 ) . More recently , using Poa seeds , Lee et al . also found a weak GRA in culture filtrates from P . aeruginosa strains overexpressing the AMB operon ( AMBox ) relative to that of AVG and FVG ( Lee et al . , 2013b ) . However , the concentrations of AMB or FVG in the germination assays were unspecified and no synthetic compound was used to directly test their intrinsic GRA . Furthermore , dicot seeds are less responsive to P . fluorescens culture filtrates than graminaceous monocot seed , suggesting that FVG is less active to repress the germination of dicot seeds including seeds of cabbage , which is a brassicaceae ( Banowetz et al . , 2008 ) . We found that AMB-containing WT PAO1 extracts inhibited the germination of brassicaceae seeds , including cabbage seeds , unlike ∆ambE extracts ( Figure 7—figure supplement 1 ) . No such effect was observed with 250 µM AVG ( Figure 7—figure supplement 1 ) . Arrested seeds germinated upon transfer to MS , indicating that the germination arrest is not due to a toxic effect ( Figure 7—figure supplement 2 ) . Altogether , these results indicate that oxyvinylglycines do not affect seed germination in the same manner . They also strongly suggest that AMB is an oxyvinylglycine able to repress the germination of several seed dicot species . Furthermore , brassicaceae seeds that germinated in presence of WT PAO1 extracts developed developmental defects unlike those germinated in presence of ∆ambE extracts ( Figure 7A , Figure 7—figure supplement 3 ) . Germination arrested Capsella seeds in presence of WT PAO1 extracts were protected as they produced normal seedlings upon transfer to a normal medium ( Figure 7—figure supplement 3E and F ) . Since oxyvinylglycines inhibit irreversibly pyridoxal phosphate ( PLP ) -dependent enzymes their proposed GRA was interpreted as a toxic effect killing the seed , rendering these compounds potentially useful as herbicides ( Rando , 1974; Lee et al . , 2013b ) . However , whether oxyvinylglycines repress seed germination in a manner requiring functional DELLA genes , as shown here , was not previously investigated . It should be noted that a signaling role for oxyvinylglycines is not incompatible with toxic effects at high concentrations as in the case of auxin whose synthetic derivative 2 , 4-D is widely used as a systemic herbicide ( Grossmann , 2007 ) . AVG is a well-known inhibitor of 1-aminocyclopropane-1-carboxylic acid ( ACC ) synthase , a PLP-dependent enzyme catalyzing the synthesis of ACC , a precursor of ethylene , from S-adenosylmethionine ( Adams and Yang , 1979; Huai et al . , 2001; Capitani et al . , 2002 ) . AVG was also shown to inhibit auxin synthesis ( Soeno et al . , 2010 ) . AMB may also inhibit ACC synthase as suggested by a report showing that AMB decreases ethylene levels in apples; however , whether AMB can inhibit ACC synthase in vitro was not shown ( Mattoo et al . , 1979 ) . Wilson et al . previously reported that 5 µM AVG , a widely used inhibitor of ACC synthase , lowers ethylene synthesis in Arabidopsis seeds and does not inhibit germination ( Wilson et al . , 2014 ) . Here , we report that as much as 200 µM AVG did not noticeably inhibit seed germination , further confirming the results of Wilson et al . ( Figure 4—figure supplement 10 ) . Furthermore , heptuple acs mutant seeds , deficient several in ACS biosynthetic genes , germinated similarly to WT seeds ( Figure 7—figure supplement 4A and B ) . Seeds deficient in ACS biosynthetic genes would be expected to respond more strongly to AMB-containing WT PAO1 extracts if AMB represses germination by inhibiting ethylene biosynthesis . However , this is not what we observed ( Figure 7—figure supplement 4A and B ) . We also treated seeds with silver nitrate ( AgNO3 ) , which induces ethylene insensitivity , and did not observe an inhibition of seed germination ( Figure 7—figure supplement 4C ) ( Rodríguez et al . , 1999; McDaniel and Binder , 2012 ) . Altogether , these experiments show that inhibition of ethylene biosynthesis upon seed imbibition is not sufficient to block germination and therefore are not consistent with the hypothesis that AMB blocks germination because it blocks ethylene biosynthesis . AMB also inhibits numerous PLP-dependent enzymes in vitro including aspartate aminotransferase and tryptophan synthase ( Rando , 1974; Rando et al . , 1976; Miles , 1975 ) . AMB could also inhibit methionyl-transfer RNA synthetase indicating that it could act as a methionine antimetabolite ( Mattoo et al . , 1979 ) . This could suggest that the effect of AMB to block germination results from its inhibition of auxin or methionine synthesis . The former possibility is unlikely because ( 1 ) auxin promotes ABA-dependent repression of seed germination and ( 2 ) low auxin levels may facilitate seed germination since ABA-dependent inhibition of radicle elongations involves enhancement of auxin signaling in the radicle elongation zone ( Belin et al . , 2009; Liu et al . , 2013 ) . We found that tir1 mutants , deficient in the auxin receptor TIR1 , germinate normally , consistent with the report of Liu et al . showing that tir1/afb2 and tir1/afb3 double mutants , deficient in the auxin receptors TIR1 and AFB1 or TIR1 and AFB3 , germinate normally and are less dormant ( Figure 7—figure supplement 5A and B , Liu et al . , 2013 ) . Furthermore , WT seeds treated with 500 µM of yucasin , a potent inhibitor of the YUCCA proteins , which are flavin mono-oxygenases oxidizing indole-3–pyruvic acid to indole-3–acetic acid ( auxin ) , did not prevent their germination ( Figure 7—figure supplement 5C ) ( Nishimura et al . , 2014 ) . These data indicate that low auxin signaling or synthesis upon seed imbibition does not prevent seed germination . In addition , tir1 mutant seed germination responses to AMB-containing WT PAO1 extracts was similar to that of WT seeds , indicating that AMB-dependent responses in seeds do not require auxin signaling ( Figure 7—figure supplement 5A and B ) . Concerning methionine , previous reports showed that methionine biosynthesis is essential for seed germination ( Gallardo et al . , 2002 ) . We explored whether AMB could block germination by inhibiting methionine synthesis . DL-Propargylgylcine ( PAG ) is an active site-directed inhibitor of cystathionine γ-synthase , which is necessary for methionine synthesis ( Thompson et al . , 1982 ) . WT seeds treated with 1 mM PAG were unable to germinate , consistent with previous reports ( Gallardo et al . , 2002 ) . As expected and consistent with previous reports , exogenously added methionine in the germination medium fully restored germination in PAG-treated seeds ( Figure 7—figure supplement 6 ) ( Gallardo et al . , 2002 ) . However , exogenous methionine did not rescue the germination of PAO1 treated WT seeds . These observations are not consistent with the hypothesis that AMB prevents germination by limiting methionine synthesis . Altogether , these results are not supporting the view that AMB exerts its DELLA-dependent seed germination arrest by inhibiting ethylene , auxin or methionine synthesis . Here , we unambiguously identify AMB as the main if not only GRA released by P . aeruginosa affecting Arabidopsis seed germination . AMB , unlike AVG , represses Arabidopsis seed germination and this requires functional DELLA genes . Our results strongly suggest that AMB does not interfere with GA synthesis . Rather , they suggest that AMB can regulate seed gene expression in both a DELLA-dependent and GA-independent manner ( Figure 3 and 5B ) . Our results with RGL2 , GAI and RGA suggest that AMB does not interfere with GA-dependent DELLA degradation ( Figures 4B , 6A–D and Figure 6—figure supplement 1 ) . We rather provide genetic evidence that DELLA activity to promote ABA-dependent seed germination arrest is stimulated in presence of AMB ( Piskurewicz et al . , 2009 , 2008 ) . The mechanism through which AMB enhances DELLA activity to promote ABA-dependent responses in seeds remains to be understood . In particular , whether it is the result of a direct interaction between AMB and DELLAs is not known . Exogenous GA cannot overcome germination repression triggered by P . aeruginosa or AMB ( Figures 3 , 4B , 6D and Figure 6—figure supplement 1 ) . Thus , genes whose expression is DELLA-dependent and GA-independent in response to P . aeruginosa or synthetic AMB might provide clues about the potential mode of action of AMB . In this respect , our transcriptome analysis identified 87 genes whose expression is regulated by DELLA factors in seeds exposed to P . aeruginosa extracts ( Figure 3D , Supplementary file 2 ) . Among them , 24 genes had their expression unchanged when GA was included in the medium together with P . aeruginosa extracts ( Figure 3D , Supplementary file 2 ) . Interestingly , the expression of 18 of them was not significantly changed in seeds exposed to paclobutrazol ( PAC ) , an inhibitor of GA synthesis , further suggesting these genes are not regulated by GA . In this set of genes , those related to karrikin signaling were overrepresented ( Nelson et al . , 2009; Waters et al . , 2012 ) . Karrikins are a class of compounds found in the smoke of burning plant material , which are known to promote germination and to break seed dormancy ( Nelson et al . , 2009 ) . Furthermore , seeds lacking the karrikin receptor KAI2 are dormant ( Waters et al . , 2012 ) . The karrikin signaling genes KUF1 and BBX20/STH7 are strongly repressed in WT seeds exposed to P . aeruginosa but not in Δdella mutant seeds , lacking all DELLA factors ( Supplementary file 2 ) . Thus , our data suggest that AMB could repress germination by repressing karrikin signaling through the DELLAs . However , karrikins share common signaling components with strigolactones , a class of plant hormones also promoting germination ( De Cuyper et al . , 2017; Toh et al . , 2012 ) . Strigolactones were recently proposed to regulate GA signaling in rice ( Ito et al . , 2017 ) . SLR1 , a rice DELLA factor , was found to interact with the strigolactone receptor DWARF14 fused to its ligand ( Nakamura et al . , 2013 ) . Thus , AMB could also potentially regulate strigolactone signaling in seeds . Beyond the question of how AMB affects Arabidopsis developmental responses , its biological significance in Pseudomonas has recently attracted much attention . Indeed , genetic experiments have shown that IQS , the autoinducer of the recently discovered quorum sensing ( QS ) subsystem named IQS in P . aeruginosa , is controlled by the five-gene operon ambABCDE ( Lee et al . , 2013a ) . Although the IQS receptor remains unknown it also remains to be determined what is the link between ambABCDE and IQS synthesis . Indeed , Lee et al . first proposed that ambABCDE gene products are directly responsible for IQS synthesis and therefore activity of the quorum sensing IQS ( Lee et al . , 2013a ) . However , the link between ambABCDE and IQS synthesis remains to be clarified . Indeed Lee et al . showed that ambABCDE is also necessary for AMB production and Rojas Murcia et al . proposed that ambABCDE gene products rather synthesize and export AMB ( Rojas Murcia et al . , 2015; reviewed in Moradali et al . , 2017; Lee et al . , 2010b , 2013a ) . In any case , there is no genetic controversy regarding the need of a functional ambABCDE operon for ( 1 ) IQS production and signaling and ( 2 ) AMB synthesis and release by P . aeruginosa . In this study , we further confirm that presence of the ambABCDE operon is necessary for AMB production ( Figure 4—figure supplement 5 , Supplementary file 3 ) . Furthermore , we show that AMB is also abolished in ΔlasIΔrhlI P . aeruginosa mutants lacking the las and rhl QS subsystems ( Figure 4—figure supplement 5A , Supplementary file 3 ) . This is in agreement with a previous report showing that expression of amb operon is strongly downregulated in ΔlasIΔrhlI mutants ( Schuster et al . , 2003 ) . The genes of the ambABCDE operon were also singled out as quorum-dependent genes in chronic cystic fibrosis patients infected with P . aeruginosa ( Chugani et al . , 2012 ) . Thus , these reports leave little doubt that the ambABCDE operon , which is necessary for AMB production , is being intimately linked to the activity of QS in P . aeruginosa . Publicly available genomic sequences show that the ambABCDE operon is not only present in the P . aeruginosa PAO1 strain used in this study but is also present in numerous other P . aeruginosa strains ( Figure 4—figure supplement 11 ) . This raises the question of the biological significance of the AMB-dependent germination arrest involving DELLA factors described here . The biological significance can be divided in two broad categories . Firstly , AMB could exert its effect fortuitously , that is in an accidental manner that bears no ecological or evolutionary significance . This does not preclude the potential biological interest of the effect of AMB on seeds . Indeed , we provided evidence that the AMB- and DELLA-dependent germination arrest cannot be readily explained by an AMB-dependent inhibition of ethylene , auxin or methionine synthesis . Furthermore , AMB does not appear to prevent GA-dependent DELLA degradation . Given that oxyvinylglycines were reported to inhibit PLP-dependent enzymes and that AMB is a methionine analog , our results could indicate that AMB interferes with an unknown mechanism present in Arabidopsis that is linking PLP-dependent enzymes or amino acid metabolism with DELLA factors . Alternatively , AMB could interfere with an unknown and GA-independent mechanism involving DELLA factors to control germination . Secondly , the effect of AMB on seeds could indeed be ecologically and evolutionary significant . We hereafter discuss this possibility . Arabidopsis produces high seed numbers , which could contradict the need of evolving protective germination arrest responses since one successful germination event is sufficient to maintain the size of the population . However , Arabidopsis is not a long distance seed dispersal species and the majority of seeds is expected to fall in the vicinity of the mother plant . Furthermore , Arabidopsis seedlings are small and fragile and poor dispersion would increase their chance of being killed at the same time whenever faced by a local threat . It is therefore expected that early development is tightly regulated in Arabidopsis to enhance plant survival . Indeed , Arabidopsis has evolved elaborate germination arrest control mechanisms that are widely regarded as being protective . These include seed dormancy , believed to prevent germination out of season , and control of seed germination of non-dormant seeds in response to abiotic factors , which is also considered to protect the plant ( Kami et al . , 2010; Penfield and King , 2009 ) . Here , we described laboratory conditions where the AMB-dependent germination arrest protects the plant from the potentially fatal effect of AMB on seedlings ( Figure 7 ) . This could indicate that this response has evolved as an adaption to counteract damage induced by biotic harmful compounds similarly to what is proposed for abiotic stresses . More generally , it is also consistent with the notion that it could correspond to a protection mechanism against pathogenic bacteria in the environment . Given the link between AMB and QS activity in P . aeruginosa , it is tempting to speculate that evolving a germination arrest response to AMB could be doubly advantageous: ( 1 ) it could protect the plant from the AMB toxin and ( 2 ) it could reveal the presence of the plant pathogen P . aeruginosa . However , presently these considerations remain highly speculative . Indeed , the GRAs released by P . aeruginosa reported here are observed after culturing P . aeruginosa to high densities that trigger IQS QS activity . Whether bacteria such as P . aeruginosa proliferate to such high densities in the rhizosphere is unclear . The number of P . aeruginosa cells building up in the environment is subject to controversy . Green et al . were able to detect P . aeruginosa in 24% of soil samples studied and reported that it multiplied in lettuce and bean under conditions of high temperature and high humidity ( Green et al . , 1974 ) . On the other hand , Deredjian et al . reported that P . aeruginosa has low occurrence in agricultural soils . However , they were able to detect them in high amounts in various manures , consistent with previous reports ( Deredjian et al . , 2014 ) . These results could suggest that P . aeruginosa could only be found in high densities in the rhizosphere where food is available , including near decaying fruit or animal droppings , together with the proper moisture or temperature conditions . This could limit the ecological significance of controlling seed germination responses to biotic factors . On the other hand , high densities of bacteria in the rhizosphere may not be obligatory to elicit seed germination responses . The Quorum Sensing is usually invoked to describe situations when high densities of cells trigger coordinated responses after autoinducers reach high concentrations in the environment . However , a given individual bacterium can only detect the autoinducer concentration present in its immediate proximity , which can include the autoinducer molecules that the same bacterium releases . In the rhizosphere , there could be situations limiting autoinducer diffusion , altering autoinducer advection , reducing autoinducer degradation or altering autoinducer spatial distribution , which could lead to high local concentrations of autoinducer even in absence of high densities of bacteria . Thus , a given bacteria cannot distinguish among the various scenarios leading to high autoinducer concentration . These considerations lead to the proposal that autoinducers fulfill a role beyond that of detecting high densities of bacteria ( Redfield , 2002 ) . They could allow bacteria to sense whether diffusion of molecules in their immediate environment is limited ( Diffusion Sensing –DS- ) . In turn , this would allow a given bacteria to determine whether a given effector would diffuse efficiently or not ( Hense et al . , 2007 ) . Clearly , testing the model that the AMB-dependent germination arrest fulfills an adaptive function in plants will require future investigations . These include ( 1 ) a better understanding of the ecology of P . aeruginosa in real field settings , ( 2 ) identifying AMB’s interacting targets in Arabidopsis responsible to convey the AMB-dependent germination arrest and ( 3 ) studying the fitness in the field of Arabidopsis mutants lacking those targets . Undoubtedly , the findings reported here offer a very narrow sample of interactions that could take place between seeds and living organisms in the rhizosphere . All seed batches compared in this study were harvested on the same day from plants growth side by side under the same environmental conditions . Seeds of Arabidopsis thaliana plants were all from Columbia Col-0 background . The Arabidopsis mutants used in this study were aba1-6 ( Barrero et al . , 2005 ) , abi5-3 ( Finkelstein and Lynch , 2000 ) , abi3-8 ( Nambara et al . , 2002 ) , tir1-1 ( purchased on Nottingham Arabidopsis Stock Centre -NASC- N3798 , Ruegger et al . , 1998 ) , Δacs145679 ( acs1-1 acs2-1 acs4-1 acs5-2 acs6-1 acs7-1 acs9-1 , purchased on NASC , N16650 , Tsuchisaka et al . , 2009 ) , rgl2-13 ( Tyler et al . , 2004 ) , Δdella ( rgl2-Sk54 rga-28 gai-t6 rgl1-Sk62 rgl3-3; Park et al . , 2013 ) . We generated rgl2-SK54 rga-28 double mutants and rgl2-SK54 rga-28 gai-t6 triple mutants for this study . Seeds were surface sterilized and sowed on germination plates as described ( Piskurewicz and Lopez-Molina , 2016 ) . Germination plates were incubated in growth chambers ( 22°C , 70% humidity , 100μmol/m2/s , 16 hr/8 h day/light photoperiod ) . The Pseudomonas strains used in this study are listed in Supplementary file 5 . Bacteria were cultured with agitation in 5 ml LB medium for 16 hr at 37˚C ( with the exception of P . fluorescens , grown at 30°C ) . Bacteria density was controlled by OD600 ( Ultrospec 2000 , Pharmabiotec ) and stocks were made for each bacteria strain when OD600 reached 1 . 2 . A volume of 25 μL of liquid culture containing approx . 7 × 108 CFU was streaked on a square plate ( 120 × 120 mm , Huberlab ) containing Murashige and Skoog ( MS ) medium ( 4 . 3 g/L ) , 2- ( N-morpholino ) ethanesulfonic acid ( MES ) ( 0 . 5 g/L ) , 0 . 8% ( w/v ) Bacto-Agar ( Applichem ) and 20 mM succinate . Plates were incubated for 3 days in the dark in plant growth chambers ( 22°C , 70% humidity ) . The resulting plates were used for germination tests as described in Figure 1—figure supplement 1 . All experiments were repeated independently several times with similar results . Bacteria were cultured with agitation in a liquid solution ( 0 . 215 g/L MS , 20 mM succinate ) for 24 hr at 37˚C and the saturated culture was centrifuged at 4˚C for 30 min ( HiCen XL , Herolab ) to pellet bacteria . The supernatant was then filtrated ( 0 . 22 µm filter ) and lyophilized ( Freeze-dryer Alpha 2–4 LD plus , Christ ) as described in Figure 2—figure supplement 2 . The resulting bacteria-free lyophilizate ( also referred as ‘extract’ in the main text ) was stored at −20°C . To perform germination assays , the lyophilizate was resuspended in water at 40 mg/ml and used at different concentrations as shown and described in the figures and figure legends . All experiments were repeated independently several times with similar results . Commercially available chemicals used in this study can be find as following: 5- ( 4-Chlorophenyl ) −2 , 4-dihydro-[1 , 2 , 4]-triazole-3-thione ( named Yucasin , CAS registry number: 26028-65-9 ) was ordered from Santa Cruz biotechnology ( product sc-233161 ) ; DL-Propargylglicine ( named PAG , CAS registry number: 64165-64-6 ) , was ordered from Sigma-Aldrich ( product P7888 ) , ( S ) -trans-2-Amino-4- ( 2-aminoethoxy ) −3-butenoic acid ( named AVG , CAS registry number: 55720-26-8 ) was ordered from Sigma-Aldrich ( product A6685 ) and AgNO3 ( CAS registry number: 7761-88-8 ) was ordered from Sigma-Aldrich ( product S7276 ) . Seed extracts were prepared as previously described ( Piskurewicz et al . , 2008 ) . Polyclonal anti-RGL2 and anti-ABI5 were as previously described ( Piskurewicz et al . , 2008 ) . Polyclonal anti-GAI was produced as described in Piskurewicz et al . ( 2008 ) . Anti-RGA antibody was purchased ( Agrisera , product AS11 1630 , RRID:AB_10749442 ) . A commercial anti-UGPase antibody was used as a loading control ( Agrisera , product AS05 086 , RRID:AB_1031827 ) . Total RNA was extracted as described ( Piskurewicz and Lopez-Molina , 2016 ) . RNAs were treated with RQ1 RNase-Free DNAse treatment ( Promega ) and cDNAs were made from 1 ug of RNA using ImProm-II reverse transcriptase ( Promega ) . Amplification was done using GoTaq qPCR Master mix ( Promega ) and reaction was performed on QuantStudio 5 Real-Time PCR equipment ( Thermo Fisher Scientific ) according to manufacturer instructions . Relative transcript levels were calculated using the comparative Ct method and normalized to PP2A ( AT1G69960 ) gene transcript levels . qPCR experiment were performed in biological triplicate . Primers used in this study are listed in Supplementary file 5 . Surface-sterilized WT ( Col-0 ) and Δdella seeds were sown in germination plates in absence or presence of ( PAC 5 μM ) , P . aerurinosoa ( PAO1 ) extracts ( 0 . 7 mg/mL ) or GA ( 10 μM ) and cultured for 20 hr prior to total seed RNA extraction . cDNA libraries from two independent biological replicates were normalized and sequenced using HiSeq4000 ( Illumina ) with single-end 50 bp reads . Reads were mapped to Col-0 genome ( TAIR10 ) with the TopHat program . Differential gene expression analysis was performed with the Cuffdiff program calculated by pooling the biological replicates . Differentially expressed genes were selected according to their significance in fold-change expression ( false discovery rate , FDR < 0 . 05 ) and a threshold level of at least two-fold change between samples ( log2 ratio ≥1 and ≤−1 ) . All RNAseq analysis were performed on GALAXY website . For data visualization , clustering analysis were done with Gene Cluster 3 . 0 using average linkage method and visualized with Java Treeview version 1 . 1 6r4 , were expression levels were color coded as following: red color for overexpressed , black color for unchanged expression and green color for underexpressed genes . All data are publicly available through the GEO database with accession number GSE115272 . With the aim to obtain preliminary information about the chemical nature of the GRA , bacterial extracts containing or not containing the GRA were analyzed by comprehensive UHPLC-HRMS2 metabolite profiling and data were mined by differential untargeted metabolomics ( see below Appendix 1 , part ‘1 . Metabolomic analysis informations’ for details ) . Data were acquired on extracts from strains releasing ( PAO1 WT and ∆pqsA ) or not releasing a GRA ( ∆ambE and ∆lasI∆rhlI ) . After the appropriate data treatment , all MS signals recorded were gathered as peak lists of individual features ( mass and retention time were annotated as follow: m/z @ RT ) each strain for subsequent multivariate data analysis to identify biomarkers . After alignment , the resulting peak list ( Supplementary file 3 ) was then mined for differential features using unsupervised PCA ( principal component analysis ) and supervised statistical analysis approaches OPLS-DA ( orthogonal partial least squares discriminant analysis ) . The PCA already allowed to clearly separate the four different bacteria extracts according to their MS features ( Figure 4—figure supplement 3A ) . An OPLS-DA was carried constructing two ‘active’ vs . ‘non-active’ groups namely ( PAO and ∆pqsA ( ‘active’ ) vs ∆ambE and ∆lasI/rhlI ( ‘non-active’ ) ( Figure 4—figure supplement 3B ) . This analysis afforded a list of biomarkers responsible for the metabolic differences between strains ( Supplementary file 4 ) . The annotation of the most significant MS features was done using exact mass information and search against a database of natural products taxonomically restricted to the genus Pseudomonas . The bioactive compound was known to be a polar compound since activity was observed to be present in the H2O eluted fraction of all GRA-containing extracts when eluted through reversed phase chromatography ( Figure 4—figure supplement 6 ) . Taking all this information into account one specific MS feature m/z 327 . 12 eluting at a retention time of 0 . 46 min ( m/z = 327 . 12 @ RT 0 . 46 min ) could be highlight when filtering the discriminant loadings of the OPLS-DA analysis for the most polar compounds ( Supplementary file 4 ) . This exact mass corresponded to a molecular formula ( MF ) of C11H22N2O7S , which did not yield any hit when querying the whole Dictionary of Natural Products ( http://dnp . chemnetbase . com/ ) . In order to gain information on this specific feature , a molecular network ( MN ) was constructed based on MS fragmentation similarities between extracts constituents using the untargeted MS/MS data acquired on PAO , ∆pqsA , ∆ambE and ∆lasI∆rhlI mutants and bioactivity data were mapped on this MN . The MN generated allows grouping compounds with structural similarities in clusters ( Wang et al . , 2016 ) . The MN was searched for the feature m/z = 327 . 12 @ RT 0 . 46 min which was found to be related to a cluster of 3 ions ( Figure 4—figure supplement 5 ) . As expected the ion m/z 327 . 12 was only found in the ‘active’ labelled species . Surprisingly , one of the related ion at m/z 196 . 06 ( C6H13NO4S ) was found in both active and non-active samples but also in the culture media . Using the CSI:FingerID in silico fragmentation platform ( https://www . csi-fingerid . uni-jena . de/ , Dührkop et al . , 2015 ) , it was identified as 2- ( N-morpholino ) ethanesulfonic acid ) or MES , a known constituent of the used culture media . Since MES in known to readily form coordination complexes we focused on the mass difference between ion m/z 327 . 12 and m/z 196 . 06 , which was found to correspond to a mass difference of 131 . 1 Da . and a MF of C5H9NO3 . Searching this MF within reported metabolites of Pseudomonas sp . permitted to annotate this compound , possibly responsible for the GRA , as L-2-amino-4-methoxy-trans-3-butenoic acid ( also referred as methoxyvinylglycine or AMB ) . PCA analysis ( Figure 4—figure supplement 4A ) followed by an OPLS-DA was also carried between biological replicates of P . aeruginosa extract ∆ambE and AMBox ( Figure 4—figure supplement 4B ) . This analysis indicated that m/z 327 . 12 was found between the most discriminant features ( Figure 4—figure supplement 5B , Supplementary file 4 ) . Comparison of the extracted MS ion trace intensities of m/z 327 . 12 among the mutants indicated that this ion was indeed over-expressed in AMBox ( ca . 20 fold between AMBox and PAO1 WT , Figure 4—figure supplement 5B ) . Additional informations can be found below in the section Appendix 1 . All raw data have been deposited under the Massive Dataset ID MSV000082463 , available at the following address: ftp://massive . ucsd . edu/MSV000082463 . WT PAO1 and ∆ambE extracts were separately fractionated by a reversed phase semi-preparative HPLC ( High Performance Liquid Chromatography ) into four fractions . The conditions for fractionation were obtained by a gradient chromatographic transfer of the metabolite profiling after optimization at the analytical level ( see below Appendix 1 , part ‘2 . Bioguided biochemical purification analysis’ for details ) . A GRA was found to be present in the polar fraction F1 from WT PAO1 extracts only . The NMR analysis of this fraction allowed detecting the presence of characteristic proton signals of the AMB molecule ( data not shown ) . In order to confirm the structure of this compound and assess its biological properties , the crude extract of the strain AMBox was fractionated at large scale using RP-MPLC ( Reversed Phase Medium Pressure Chromatography , C18 ) followed by semi-preparative HPLC purification using an amide stationary phase for efficient selectivity ( Figure 4—figure supplement 6B ) . The structure of AMB was finally confirmed by extensive 1D and 2D NMR and HRMS analyses ( Appendix 1—figures 1 and 2 ) . For AMB quantification and NMR analysis , see below Appendix 1 part ‘3 . AMB quantification’ and part . 4 ‘Chemical identity of the isolate AMB’ for details .	The plant embryo within a seed is well protected . While it cannot stay within the seed forever , the embryo can often wait for the right conditions before it develops into a seedling and continues its life cycle . Indeed , plants have evolved several ways to time this process – which is known as germination – to maximize the chances that their seedlings will survive . For example , if the environment is too hot or too dark , the seed will make a hormone that stops it from germinating . In addition to environmental factors like light and temperature , a seed in the real word is continuously confronted with soil microbes that may harm or benefit the plant . However , few researchers have asked whether seeds control their germination in response to other living organisms . The bacterium Pseudomonas aeruginosa lives in a wide spectrum of environments , including the soil , and can cause diseases in both and plants and animals . Chahtane et al . now report that seeds of the model plant Arabidopsis thaliana do indeed repress their germination when this microbe is present . Specifically , the seeds respond to a molecule released from the bacteria called L-2-amino-4-methoxy-trans-3-butenoic acid , or AMB for short . Like the bacteria , AMB is harmful to young seedlings , but Chahtane et al . showed that the embryo within the seed is protected from its toxic effects . Further experiments revealed that the seed's response to the bacterial molecule requires many of the same signaling components that repress germination when environmental conditions are unfavorable . However , Chahtane et al . note that AMB activates these components in an unusual way that they still do not understand . The genes that control the production of AMB are known to also control how bacterial populations behave as they accumulate to high densities . It is therefore likely that Pseudomonas aeruginosa would make AMB if it reached a high density in the soil . This raises the possibility that plants have specifically evolved to stop germination if there are enough microbes nearby to pose a risk of disease . This hypothesis , however , is only one of several possible explanations and remains speculative at this stage; further work is now needed to evaluate it . Nevertheless , identifying how AMB interferes with the signaling components that control germination and plant growth may guide the design of new herbicides that could , for example , control weeds in the farming industry .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology" ]	2018	The plant pathogen Pseudomonas aeruginosa triggers a DELLA-dependent seed germination arrest in Arabidopsis
Light-gated chloride channels are emerging as promising optogenetic tools for inhibition of neural activity . However , their effects depend on the transmembrane chloride electrochemical gradient and may be complex due to the heterogeneity of this gradient in different developmental stages , neuronal types , and subcellular compartments . Here we characterized a light-gated chloride channel , GtACR2 , in mouse cortical neurons . We found that GtACR2 activation inhibited the soma , but unexpectedly depolarized the presynaptic terminals resulting in neurotransmitter release . Other light-gated chloride channels had similar effects . Reducing the chloride concentrations in the axon and presynaptic terminals diminished the GtACR2-induced neurotransmitter release , indicating an excitatory effect of chloride channels in these compartments . A novel hybrid somatodendritic targeting motif reduced the GtACR2-induced neurotransmitter release while enhancing the somatic photocurrents . Our results highlight the necessity of precisely determining the effects of light-gated chloride channels under specific experimental conditions and provide a much-improved light-gated chloride channel for optogenetic inhibition . Targeted manipulation of neural activity is a powerful approach in neuroscience that has provided fundamental insights into the roles of specific neurons in nervous system functions . Genetically encoded actuators such as light-gated ion channels or pumps enable control of neural activity with unprecedented spatiotemporal specificity and are transforming neuroscience research ( Boyden et al . , 2005; Han and Boyden , 2007; Li et al . , 2005; Nagel et al . , 2003; Zhang et al . , 2007 ) . Actuators that enable neuronal activation are frequently used , but inhibitory optogenetic tools are increasingly crucial because reversible and temporally precise suppression of neuronal activity is key to revealing the causal roles of specific neurons in network dynamics and behavior . The widely used light-driven inward chloride pumps and outward proton pumps , such as Natronomonas pharaonis halorhodopsin ( NpHR ) and Halorubrum sodomense archaerhodopsin ( Arch ) , can hyperpolarize membrane potentials , independent of the electrochemical gradients , to inhibit action potentials with millisecond precision ( Chow et al . , 2010; Chuong et al . , 2014; Han and Boyden , 2007; Han et al . , 2011; Zhang et al . , 2007 ) . However , their efficacies are limited because only one ion is transported per absorbed photon , and their activation does not decrease membrane resistance . Light-gated chloride channels , such as Guillardia theta anion channelrhodopsin 1 and 2 ( GtACR1 and GtACR2 ) , iC++ , and iChloC , overcome these limitations ( Berndt et al . , 2014; Govorunova et al . , 2015 , 2017b; Wietek et al . , 2014 ) . They are highly sensitive to light , allow multiple ions to cross the membrane per photocycle , and reduce membrane resistance , thereby potently inhibiting action potentials . Thus , light-gated chloride channels are emerging as promising optogenetic tools for suppressing neuronal activity ( Govorunova et al . , 2017a ) . Determining the precise effects of light-gated ion channels or pumps under defined conditions is a prerequisite to use them for interrogating the functions of specific neurons and circuits . This is due to the possibility that these channels or pumps not only modulate membrane potentials but also may affect other processes such as ion homeostasis or neurotransmitter release . For example , activation of a light-driven inward chloride pump , eNpHR3 . 0 , can transiently change the reversal potential of GABAA receptors and alter the inhibitory synaptic inputs ( Raimondo et al . , 2012 ) . Prolonged activation of a light-driven outward proton pump , eArch3 . 0 , can increase presynaptic calcium concentrations and spontaneous neurotransmitter release ( Mahn et al . , 2016 ) . Despite these potential confounds , these inhibitory optogenetic molecules become increasingly important tools for targeted silencing of neuronal populations , as long as these confounds are understood and controlled ( Allen et al . , 2015 ) . Therefore , it is also crucial to thoroughly characterize light-gated chloride channels , as their effect depends on the difference between the membrane potential and the reversal potential for chloride , both of which can vary in different neuronal types and subcellular compartments ( Marty and Llano , 2005; Trigo et al . , 2008 ) . To this end , we investigated the effects of activating GtACR2 in mouse cortical excitatory and inhibitory neurons . Much to our surprise , wide-field light activation of GtACR2 not only inhibited the soma , but also caused neurotransmitter release onto neighboring neurons . A similar phenomenon was observed with GtACR1 , iC++ , and iChloC . We further showed that GtACR2 activation in the axon and presynaptic terminals directly depolarized the membrane to induce neurotransmitter release due to high chloride concentrations in these compartments . These data explain the recent observations that photostimulation of neurons expressing GtACR1 or GtACR2 can paradoxically release neurotransmitters or generate action potentials ( Mahn et al . , 2016; Malyshev et al . , 2017 ) . To reduce the excitatory effect of GtACR2 , we screened a panel of somatodendritic targeting motifs to reduce the trafficking of GtACR2 to the axon and presynaptic terminals . We created a hybrid motif ( Kv2 . 1C-linker-TlcnC ) that is most effective in concentrating GtACR2 in the somatodendritic domain . Activation of somatodendritically targeted GtACR2 resulted in larger photocurrents at the soma and less neurotransmitter release than wild type GtACR2 . Thus , restricting the localization of light-gated chloride channels to the somatodendritic domain improves the inhibitory efficacy of these optogenetic tools . Our results also suggest that light-gated chloride channels can be a useful tool for studying the physiological functions of chloride electrochemical gradients in specific brain regions , cell types , or subcellular domains that are otherwise difficult to access , such as small axons and presynaptic terminals . To examine the efficacy of GtACR2 in mouse cortical excitatory neurons , we expressed a GtACR2-EYFP fusion protein ( referred to as GtACR2 below ) together with a red fluorescent protein , tdTomato , in layer 2/3 pyramidal neurons of the mouse visual or somatosensory cortex by in utero electroporation of plasmids at embryonic day 14 . 5–15 . 5 . We obtained acute coronal brain slices from 3 to 8 week-old mice and observed that GtACR2 was present in the soma , dendrites , and axon ( Figure 1A ) . We performed whole-cell patch clamp recordings at the soma of neurons expressing GtACR2 ( GtACR2+ neurons ) with a K+-based pipette solution ( Figure 1B ) . As previously reported ( Govorunova et al . , 2015 ) , activation of GtACR2 by wide-field blue light ( 455 nm ) potently inhibited current-induced spiking in these neurons ( Figure 1C ) . However , when we voltage clamped the neurons to record GtACR2-mediated photocurrents , we unexpectedly found an inward current that was superimposed on the photocurrent . This inward current resembled an excitatory postsynaptic current ( EPSC; Figure 1D ) . To further investigate this phenomenon , we recorded layer 2/3 pyramidal neurons that did not express GtACR2 ( GtACR2– neurons ) with a Cs+-based pipette solution ( Figure 1E ) . Activation of GtACR2 with a short light pulse ( 0 . 5–10 ms ) generated inward currents in all recorded GtACR2– neurons that were voltage clamped at the reversal potential for GABAergic inhibition ( −60 mV ) . The onsets of these inward currents followed the onset of the blue light by 3 . 19 ± 0 . 26 ms ( mean ± s . e . m . , n = 25 ) . These currents were abolished by the glutamatergic receptor antagonists , NBQX and CPP ( Figure 1F ) , or the voltage-gated sodium channel blocker , tetrodotoxin ( TTX; Figure 1G ) , indicating that they were indeed monosynaptic EPSCs caused by the glutamate transmitter released from GtACR2+ neurons . Activation of GtACR2 also produced inhibitory postsynaptic currents ( IPSCs ) in GtACR2– neurons that were voltage clamped at the reversal potential for glutamatergic excitation ( +10 mV ) . These IPSCs were disynaptic because they were abolished by NBQX and CPP ( Figure 1—figure supplement 1 ) , indicating that activating GtACR2 in pyramidal neurons can release sufficient glutamate to recruit inhibitory interneurons . To determine if the phenomenon of GtACR2-induced neurotransmitter release also occurs in GABAergic inhibitory neurons , we expressed GtACR2 in parvalbumin-expressing ( Pv ) neurons by injecting a Flpo recombinase-dependent adeno-associated virus ( AAV ) into the visual cortex of Pvalb-2A-Flpo mice ( PvalbFlpo/+ ) ( Madisen et al . , 2015 ) at postnatal day 1 ( Figure 1H ) . Using acute brain slices from 3 to 6 week-old mice , we found that activation of GtACR2 in Pv neurons generated IPSCs in all recorded GtACR2– layer 2/3 pyramidal neurons , and the IPSCs were abolished by Gabazine , a GABAA receptor antagonist , or TTX ( Figure 1I–K ) . The onsets of the IPSCs followed the onset of the blue light by 2 . 47 ± 0 . 15 ms ( mean ± s . e . m . , n = 17 ) , indicating that they were monosynaptic IPSCs caused by the GABA transmitter released from GtACR2+ Pv neurons . We next determined how repetitive or prolonged activation of GtACR2 would affect neurotransmitter release . We activated GtACR2 in layer 2/3 pyramidal neurons or Pv neurons with a high-frequency train of light pulses and found that each light pulse produced reliable EPSCs or IPSCs , respectively ( Figure 1—figure supplement 2 ) . Interestingly , continuous activation of GtACR2 in layer 2/3 pyramidal neurons or Pv neurons with a long pulse of light ( e . g . , 2 s ) transiently generated large EPSCs or IPSCs , respectively , followed by various amounts of smaller synaptic responses ( Figure 1—figure supplement 3A , B , E , F , G ) . Typically , more than 50% of the neurotransmitter release occurred within the first 100–200 ms of the light pulse ( Figure 1—figure supplement 3C , D , H , I ) . The finding of GtACR2-induced neurotransmitter release was unexpected , because the Nernst equilibrium potential of chloride becomes lower than the action potential threshold in rodent cortical neurons after the second postnatal week , and activation of chloride channels should inhibit , not promote , neurotransmitter release onto neighboring cells ( Ben-Ari , 2002; Owens et al . , 1996 ) . We thus sought to identify the cause of this paradoxical neurotransmitter release and considered three possibilities . First , GtACR2 may conduct cations to depolarize neurons . Second , an increase in the intracellular chloride or strong hyperpolarization induced by GtACR2 activation may lead to rebound spikes . Third , GtACR2-mediated chloride currents may be excitatory . First , it was reported that GtACR2 did not conduct physiological cations ( Govorunova et al . , 2015 ) , but some other light-gated chloride channels retained certain cation conductance ( Berndt et al . , 2014; Wietek et al . , 2014 ) . Thus , we sought to verify that in cortical neurons , GtACR2 has a similar reversal potential as a known chloride channel . To accomplish this , we determined the reversal potential of GtACR2-mediated photocurrents in comparison with that of IPSCs mediated by the endogenous GABAA receptors . We used in utero electroporation to express GtACR2 in layer 2/3 pyramidal neurons and a Cre recombinase-dependent AAV to express a red light-gated cation channelrhodopsin , ReaChR , in Pv neurons of Pvalb-2A-Cre mice ( PvalbCre/+ ) ( Madisen et al . , 2010 ) . We performed whole-cell voltage clamp recordings at the soma of a GtACR2+ layer 2/3 pyramidal neuron and a nearby GtACR2– pyramidal neuron simultaneously ( Figure 2—figure supplement 1A ) . In the GtACR2+ neuron , we sequentially recorded the IPSCs induced by activating ReaChR in Pv neurons via 617 nm light and the GtACR2-mediated photocurrents activated by 455 nm light . Both IPSCs and photocurrents were recorded at different membrane potentials to determine their reversal potentials in the same neuron ( Figure 2—figure supplement 1B , C ) . 617 nm light does not activate GtACR2 ( Govorunova et al . , 2015 ) , whereas 455 nm light partially activates ReaChR ( Lin et al . , 2013 ) . Thus , to avoid Pv neuron-mediated IPSCs contaminating GtACR2-mediated photocurrents , we monitored the IPSCs in the GtACR2– neuron at the membrane potential of +10 mV to ensure that the intensity of the 455 nm light was not sufficient to activate Pv neurons and generate IPSCs ( Figure 2—figure supplement 1B ) . We found that the reversal potentials of GtACR2-mediated photocurrents and GABAergic IPSCs were similar for each neuron and well below the action potential threshold ( Figure 2—figure supplement 1D ) , indicating that GtACR2 does not conduct cations to cause neurotransmitter release . Second , we tested if GtACR2-induced neurotransmitter release could be due to rebound depolarization . An increase in the intracellular chloride caused by GtACR2 activation may trigger rapid efflux of chloride after the blue light illumination terminates . However , this possibility is unlikely because when GtACR2 was activated by a long pulse of blue light , neurotransmitter release occurred before the light illumination ended ( see examples in Figure 1F , G and Figure 1—figure supplement 3 ) . Another possibility is that the strong hyperpolarization induced by GtACR2 activates hyperpolarization-activated Ih currents , which may depolarize the membrane potential above the action potential threshold . However , pharmacological inhibition of Ih currents slightly increased the amplitudes of GtACR2-induced EPSCs ( Figure 2—figure supplement 2A , B ) , most likely because inhibiting Ih currents increases neuronal membrane resistances ( Robinson and Siegelbaum , 2003 ) . Thus , GtACR2-induced neurotransmitter release is not caused by rebound depolarization . Third , although GtACR2-mediated photocurrents are inhibitory at the soma , it is possible that the chloride concentrations are higher in some other cellular compartments , such that the electrochemical gradient causes chloride to exit the cell upon GtACR2 channel opening , resulting in depolarization of the membrane potential . To test this hypothesis , we pharmacologically inhibited the activity of Na+-K+-2Cl– cotransporter 1 ( NKCC1 ) with bumetanide ( 50 or 100 µM ) to decease the intracellular chloride concentrations , as NKCC1 is responsible for transporting chloride into neurons ( Ben-Ari , 2017 ) . When we activated GtACR2 in layer 2/3 pyramidal neurons , the resulting EPSCs in GtACR2– pyramidal neurons were diminished by bath application of bumetanide ( Figure 2A , B ) , indicating that GtACR2-induced neurotransmitter release requires high concentrations of intracellular chloride . An alternative interpretation of this result would be that bumetanide blocks GtACR2 itself . To test this possibility , we simultaneously recorded the photocurrents and EPSCs in GtACR2+ and GtACR2– pyramidal neurons , respectively . While bumetanide diminished the EPSCs in GtACR2– neurons , it had no effect on the photocurrents in GtACR2+ neurons ( Figure 2—figure supplement 3A , B ) , thereby ruling out the possibility that bumetanide affects GtACR2 itself . Furthermore , when a cation channel , channelrhodopsin-2 ( ChR2 ) ( Boyden et al . , 2005; Li et al . , 2005; Nagel et al . , 2003 ) , was expressed and activated in layer 2/3 pyramidal neurons , the resulting EPSCs in ChR2– neurons were not affected by bumetanide ( Figure 2C , D ) , indicating that reducing the intracellular chloride concentration only affects chloride-mediated , and not cation-mediated , EPSCs . Finally , bumetanide also diminished the IPSCs resulting from GtACR2 activation in Pv neurons ( Figure 2E , F ) . Together , these results demonstrate that light activation of GtACR2 generates an excitatory chloride conductance in certain neuronal compartments to trigger neurotransmitter release . Since we expressed GtACR2 by in utero electroporation or neonatal AAV injection , we sought to determine if the excitatory effect of GtACR2 was caused by the long-term expression of GtACR2 throughout development that somehow altered the chloride homeostasis . To selectively express GtACR2 in adult neurons , we in utero electroporated a Flpo-dependent plasmid into layer 2/3 pyramidal neurons that will express GtACR2 only if Flpo is present ( Figure 2—figure supplement 2C , D ) . A Flpo-expressing AAV was then injected into the electroporated mice at postnatal week 4 or 9 to turn on the GtACR2 expression . We obtained acute coronal brain slices 1–3 weeks after injecting Flpo-expressing AAV and found that light activation of GtACR2 , again , produced bumetanide-sensitive EPSCs in GtACR2– neurons ( Figure 2—figure supplement 2E ) . Thus , it is unlikely that GtACR2 expression during neuronal development alters the chloride homeostasis to render GtACR2 excitatory , as acute expression of GtACR2 in mature neurons had the same effect . To determine if activation of other light-gated chloride channels can trigger neurotransmitter release , we examined iC++ and iChloC , two engineered blue light-gated chloride channels that were converted from cation channelrhodopsins ( Berndt et al . , 2016; Wietek et al . , 2015 ) , and GtACR1 , another natural anion channelrhodopsin from Guillardia theta ( Govorunova et al . , 2015 ) . We in utero electroporated plasmids to express iC++ in layer 2/3 pyramidal neurons of the visual cortex and obtained acute coronal brain slices from 3 to 9 week-old mice . Similar to GtACR2 , light activation of iC++ generated EPSCs in iC++– layer 2/3 pyramidal neurons that were abolished by NBQX and CPP ( amplitude reduced by 97 . 3 ± 0 . 9% , mean ± s . e . m . , n = 4 ) . Bumetanide diminished the iC++-induced EPSCs ( Figure 2G , H ) without affecting the iC++-mediated photocurrents ( Figure 2—figure supplement 3C , D ) . We also expressed iC++ in Pv neurons by injecting a Cre-dependent AAV into PvalbCre/+ mice and found that light activation of iC++ caused bumetanide-sensitive IPSCs in iC++– layer 2/3 pyramidal neurons ( Figure 2I , J ) . Similarly , when we expressed iChloC in Pv neurons , light activation of iChloC resulted in IPSCs in 10 out of 17 recorded iChloC– layer 2/3 pyramidal neurons ( 183 ± 42 pA , mean ± s . e . m . , n = 10 ) , presumably because iChloC generated smaller photocurrents than iC++ ( iChloC , 292 ± 62 pA , n = 7; iC++ , 2182 ± 291 pA , n = 15; recorded at the membrane potential of +10 mV; mean ± s . e . m . , p<0 . 0001 , t test with Welch’s correction ) . Finally , activation of GtACR1 in layer 2/3 pyramidal neurons produced EPSCs onto all recorded GtACR1– pyramidal neurons ( 149 ± 47 pA , mean ± s . e . m . , n = 11 ) . Altogether , these results demonstrate that activation of different light-gated chloride channels in neurons can trigger neurotransmitter release . We hypothesized that the most likely neuronal compartments rendering GtACR2 excitatory were the distal axon and presynaptic terminals because of the following previous findings . First , activation of presynaptic GABAA or glycine receptors enhanced neurotransmitter release at several synapses of the hippocampus , cerebellum , and brainstem ( Jang et al . , 2006; Pugh and Jahr , 2011; Ruiz et al . , 2010; Stell et al . , 2007; Turecek and Trussell , 2001; Zorrilla de San Martin et al . , 2017 ) . Second , the chloride concentrations were 4–5 times higher in the presynaptic terminals of the Calyx of Held than the parent soma ( Price and Trussell , 2006 ) . Third , there appeared to be an axo-somato-dendritic gradient in which the reversal potentials of GABA from the axon to the soma and dendrites of cortical neurons become progressively more negative ( Khirug et al . , 2008 ) . To test our hypothesis , we expressed GtACR2 in layer 2/3 pyramidal neurons of the visual cortex in one hemisphere as described above and obtained acute coronal slices from the contralateral hemisphere ( Figure 3A ) . GtACR2 was present in the long-range callosal projections in the contralateral hemisphere ( Figure 3—figure supplement 1A ) , which enabled us to activate GtACR2 in the axon and presynaptic terminals that were severed from their parent somas . Light activation of GtACR2 in the callosal projections generated EPSCs in layer 2/3 pyramidal neurons of the contralateral cortex , which were diminished by TTX ( Figure 3—figure supplement 1B , C ) or bumetanide ( Figure 3B ) . These results demonstrate that activation of GtACR2 in the axon and presynaptic terminals is sufficient to trigger neurotransmitter release . If GtACR2-mediated chloride currents are excitatory in the presynaptic terminals , then GtACR2 should be similar to ChR2 , whose activation can directly depolarize the presynaptic membrane in the absence of action potentials to trigger neurotransmitter release ( Petreanu et al . , 2009 ) . To test this prediction , we recorded EPSCs or IPSCs in GtACR2– neurons while activating GtACR2 in layer 2/3 pyramidal neurons or Pv neurons , respectively ( Figure 3C , E ) . As described above , bath application of TTX abolished the EPSCs and IPSCs . However , when we further blocked voltage-gated potassium channels by 4-aminopyridine ( 4-AP ) and tetraethylammonium ( TEA ) to prolong membrane depolarization ( Petreanu et al . , 2009 ) , the EPSCs and IPSCs were partially recovered ( Figure 3D , F ) . These results indicate that in the absence of action potentials , light activation of GtACR2 is sufficient to depolarize the presynaptic membrane to open voltage-gated calcium channels and trigger neurotransmitter release . We further tested if GtACR2-induced axonal depolarization could evoke antidromic action potentials by performing extracellular loose-patch or whole-cell current clamp recordings at the somas of GtACR2+ pyramidal neurons . We observed antidromic spikes in 9 out of 88 neurons recorded in loose-patch configuration and 21 out of 56 neurons recorded in whole-cell configuration in response to blue light stimulation ( Figure 3G , H ) . In the whole-cell current clamp recordings , although the chloride concentration in the patch pipette solution sets the Nernst equilibrium potential of chloride around −85 mV ( see Materials and methods ) , blue light induced a depolarization following the initial hyperpolarization . This observation is consistent with the notion that the depolarization antidromically propagated from the distal axon to the soma ( Figure 3H ) . The antidromic spikes were not affected by NBQX and CPP , but were abolished by TTX ( Figure 3H ) , indicating that the spikes were generated within the GtACR2+ neurons , rather than by excitatory inputs from other neurons . Antidromic spikes were only observed in a subset of neurons , likely because the GtACR2 expression levels are heterogeneous in different neurons , and the hyperpolarization initiated at the soma can orthodromically propagate to counteract the antidromic spikes . Similarly , TTX-sensitive antidromic spikes were observed in a subset of GtACR2+ Pv neurons ( 3 out of 11 neurons in loose-patch configuration and 7 out of 17 neurons in whole-cell configuration , Figure 3I , J ) . These results show that GtACR2-induced axonal depolarization can be sufficient to elicit antidromic action potentials . Activating GtACR2 and other light-gated chloride channels inhibits the soma but depolarizes the presynaptic terminals to release neurotransmitters . This dichotomic effect can confound the utilization of these channels as inhibitory optogenetic tools for suppressing neuronal activity . We reasoned that reducing the trafficking of light-gated chloride channels into the axon and presynaptic terminals should reduce or eliminate their depolarizing action . Thus , we sought to restrict GtACR2 within the somatodendritic domain of neurons by fusing GtACR2 with a number of reported somatodendritic targeting motifs including a 26-amino acid Myosin Va-binding domain of Melanophilin ( MBD ) ( Lewis et al . , 2009 ) , a 32-amino acid cytoplasmic C-terminal motif of Neuroligin 1 ( Nlgn1C ) ( Rosales et al . , 2005 ) , a 16-amino acid dileucine-containing motif of potassium channel Kv4 . 2 ( Kv4 . 2LL ) ( Rivera et al . , 2003 ) , the N-terminal 150 residues of kainate receptor subunit 2 ( KA2N ) ( Shemesh et al . , 2017 ) , the C-terminal 17 residues of Telencephalin ( TlcnC ) ( Mitsui et al . , 2005 ) , and a 65-amino acid cytoplasmic C-terminal motif of potassium channel Kv2 . 1 ( Kv2 . 1C ) ( Lim et al . , 2000; Wu et al . , 2013 ) . Each of these GtACR2 variants ( Table 1 ) , along with tdTomato , were expressed in layer 2/3 pyramidal neurons of the visual cortex by in utero electroporation ( Figure 4A ) . Since GtACR2 was tagged with EYFP or EGFP , we compared the EYFP or EGFP fluorescence in layer 5 , which only contains the axons of layer 2/3 pyramidal neurons , with the EYFP or EGFP fluorescence in layer 2/3 to estimate the distribution of GtACR2 between the axon and somatodendritic domain . We normalized the EYFP or EGFP fluorescence ratio between layer 5 and layer 2/3 by the tdTomato fluorescence ratio between layer 5 and layer 2/3 to control for variations in the collateral axons . Among tested motifs , TlcnC and Kv2 . 1C were most effective in targeting GtACR2 to the soma and dendrites ( Figure 4D ) . As these two motifs may engage different trafficking mechanisms , we combined them to create two hybrid motifs , Kv2 . 1C-TlcnC and Kv2 . 1C-linker-TlcnC . Kv2 . 1C-linker-TlcnC turned out to be the best in restricting GtACR2 within the somatodendritic domain ( Figure 4A , D ) . Interestingly , GtACR2-EYFP-Kv2 . 1C and GtACR2-EYFP-Kv2 . 1C-linker-TlcnC showed less intracellular aggregation than wild type GtACR2 ( Figure 4B ) , suggesting that the somatodendritic targeting motifs also enhance the surface expression of GtACR2 . Finally , the EYFP fluorescence of the callosal projections in the contralateral hemisphere was also reduced for GtACR2-EYFP-Kv2 . 1C and GtACR2-EYFP-Kv2 . 1C-linker-TlcnC as compared to GtACR2-EYFP ( Figure 4C , E ) , demonstrating that both somatodendritic targeting motifs decreased the trafficking of GtACR2 into the distal axon . To determine how targeting GtACR2 to the somatodendritic domain affects its photocurrent and ability to trigger neurotransmitter release , we compared the somatodendritically targeted GtACR2 variants , GtACR2-EYFP-Kv2 . 1C and GtACR2-EYFP-Kv2 . 1C-linker-TlcnC , with wild type GtACR2 in the same litters of mice . We first recorded GtACR2+ layer 2/3 pyramidal neurons and found that the blue light-activated photocurrents of GtACR2-EYFP-Kv2 . 1C and GtACR2-EYFP-Kv2 . 1C-linker-TlcnC were 2 . 1–2 . 4 and 2 . 7–3 . 5 folds of GtACR2-EYFP photocurrents , respectively ( Figure 5A–C and Figure 5—figure supplement 1A , C , E , G ) . We then recorded the EPSCs in GtACR2– layer 2/3 pyramidal neurons in response to different strengths of blue light stimulation . The EPSCs evoked by activating GtACR2-EYFP-Kv2 . 1C or GtACR2-EYFP-Kv2 . 1C-linker-TlcnC were reduced by 52–60% or 65–77% , respectively , as compared to those evoked by activating GtACR2-EYFP ( Figure 5D–G and Figure 5—figure supplement 1B , D , F , H ) . In a subset of experiments with GtACR2-EYFP-Kv2 . 1C or GtACR2-EYFP-Kv2 . 1C-linker-TlcnC , the lowest light stimulation strength could no longer evoke EPSCs in GtACR2– neurons , but still induced robust photocurrents in GtACR2+ neurons ( Figure 5—figure supplement 1E–H ) . Furthermore , when photostimulating the callosal projections , the EPSCs in layer 2/3 pyramidal neurons of the contralateral cortex were reduced by 54% and 73% for GtACR2-EYFP-Kv2 . 1C and GtACR2-EYFP-Kv2 . 1C-linker-TlcnC , respectively , as compared to GtACR2-EYFP ( Figure 5H , I ) . These results show that the somatodendritic targeting motifs , especially Kv2 . 1C-linker-TlcnC , shift GtACR2 from the axon towards the soma and dendrites , thereby reducing the excitatory action in the axon and presynaptic terminals while enhancing the inhibitory currents at the soma and dendrites . Optogenetic suppression of neuronal activity and synaptic outputs is an essential approach for dissecting the roles of specific neurons in brain functions . Light-gated chloride channels , particularly GtACR1 and GtACR2 , are increasingly used due to their large photocurrents and high sensitivity to light ( Forli et al . , 2018; Mardinly et al . , 2018; Mauss et al . , 2017; Mohamed et al . , 2017; Mohammad et al . , 2017 ) . To use these tools to their full potentials , it is necessary that we understand their function and , importantly , their limitations . Our experiments reveal that wide-field activation of these channels in cortical neurons suppresses action potentials at the soma but also triggers neurotransmitter release at the presynaptic terminals , thereby voiding inhibition of neuronal activity . As demonstrated , the excitatory action of chloride channels in the axon and presynaptic terminals is due to the high intracellular chloride concentrations that create a depolarizing chloride electrochemical gradient . The depolarizing effect of presynaptic chloride channels , GABAA or glycine receptors , has been documented at a few synapses in the hippocampus , cerebellum , and brainstem through application of agonists or antagonists ( Jang et al . , 2006; Pugh and Jahr , 2011; Ruiz et al . , 2010; Stell et al . , 2007; Turecek and Trussell , 2001; Zorrilla de San Martin et al . , 2017 ) . Our results in cortical excitatory and inhibitory neurons expand previous findings and indicate that a presynaptic depolarizing chloride electrochemical gradient is likely a general property across brain regions and neuronal types . More importantly , our experiments directly demonstrate that activation of presynaptic chloride channels is in fact excitatory , as it can elicit action potentials with short latency , which was difficult to explicitly demonstrate by agonist application ( Pugh and Jahr , 2011 ) . Thus , light-gated chloride channels can be targeted to different brain regions , cell types , and subcellular compartments to study chloride electrochemical gradients with unprecedented temporal precision . Although the excitatory effect of light-gated chloride channels is undesired for neuronal silencing , a potential utilization of their dual actions at the presynaptic terminals and soma is to activate specific projections of neurons while minimizing the effect of antidromic spikes . For example , long-range projection neurons often target multiple brain areas , and sometimes it is desired to selectively excite the axonal terminals projecting to one particular area . If ChR2 is used , local activation of ChR2-expressing axonal terminals may generate antidromic spikes , which will affect other projections . However , if GtACR2 is used , one can simultaneously activate GtACR2 in the soma and axonal terminals . Axonal depolarization will result in neurotransmitter release , but the antidromic spikes will be reduced or suppressed by the hyperpolarization originating from the somatodendritic domain , which reduces the likelihood of activating other projections . To create a better inhibitory optogenetic tool , we tested a number of somatodendritic targeting motifs to confine GtACR2 in the soma and dendrites . We generated a hybrid motif , Kv2 . 1C-linker-TlcnC , that was more effective than the widely used Kv2 . 1C ( Baker et al . , 2016; Mardinly et al . , 2018; Wu et al . , 2013 ) . The Kv2 . 1C motif was recently used in a bioRxiv preprint ( Mahn et al . , 2017 ) to target GtACR2 to the somatodendritic domain . This GtACR2 variant achieved a greater reduction of neurotransmitter release from the contralateral callosal projections of medial prefrontal cortical neurons than what we observed with our GtACR2-EYFP-Kv2 . 1C and GtACR2-EYFP-Kv2 . 1C-linker-TlcnC in visual cortical neurons . This quantitative difference is likely due to different expression levels , light stimulation strengths , or neuronal types ( see discussion below ) , which reiterates the importance of precisely characterizing the effect of light-gated chloride channels , specific to each experimental design , when using them to manipulate neuronal activity . While GtACR2-EYFP-Kv2 . 1C-linker-TlcnC can still traffic to the axon to cause neurotransmitter release , it is thus far the most improved light-gated chloride channel for optogenetic inhibition . Since it can generate much larger photocurrents in the somatodendritic domain than what is necessary to suppress action potentials , one approach to use this tool is to reduce the overall GtACR2 expression level to further decrease its presence in the axon while still generating sufficient inhibitory photocurrents at the soma . Supporting this idea , post hoc analysis of our experiments indicates that , for a given experiment , it is possible to identify a light stimulation strength that is low enough not to cause neurotransmitter release but still produce large inhibitory photocurrents in GtACR2+ neurons ( Figure 5—figure supplement 1 ) . However , this stimulation strength must be empirically determined for every experiment . Furthermore , since during continuous activation of GtACR2 , large and synchronous neurotransmitter release occurs at the early phase of the light stimulation ( Figure 1—figure supplement 3 ) , one may also take advantage of the late phase when neurotransmitter release is reduced but photocurrent is still robust for optogenetic inhibition experiments . Another approach is to selectively photostimulate GtACR2 at the soma by two-photon excitation ( Forli et al . , 2018; Mardinly et al . , 2018 ) . However , it is difficult to apply this photoactivation approach to freely moving animals or deep brain areas . Therefore , it is imperative that we further engineer light-gated chloride channels to eliminate their excitatory action in the axonal terminals . Future strategies include generating more effective somatodendritic targeting motifs , creating outwardly rectifying channels , and the combination of both strategies . The GtACR2-EYFP-Kv2 . 1C-linker-TlcnC reported here enhances the available toolkit for optogenetic inhibition and will serve as the foundation for future improvement . All procedures to maintain and use mice were approved by the Institutional Animal Care and Use Committee at Baylor College of Medicine . Mice were maintained on a 14 hr:10 hr light:dark cycle with regular mouse chow and water ad libitum . Experiments were performed during the light cycle . ICR ( CD-1 ) female mice were purchased from Baylor College of Medicine Center for Comparative Medicine or Charles River Laboratories . C57BL6/J , Pvalb-2A-Cre , and Pvalb-2A-Flpo mice were obtained from Jackson Laboratory ( stock numbers 000664 , 012358 , and 022730 , respectively ) . Both male and female mice were used in the experiments . The mice were used at the age of 3–9 weeks for experiments , except for the conditional expression of GtACR2 in adults , where mice were used at the age of 10–12 weeks . Plasmids pLenti-UbiC-GtACR2-EYFP ( Addgene #67877 ) and pLenti-UbiC-GtACR1-EYFP ( Addgene #67795 ) were obtained from Dr . John Spudich , pAAV-CaMKIIα-iC++-TS-EYFP and pAAV-EF1α-DIO-iC++-TS-EYFP from Dr . Karl Deisseroth , pAAV-EF1α-DIO-iChloC-T2A-mCherry from Drs . Matthew Caudill and Massimo Scanziani , pCAG-tdTomato from Anirvan Ghosh , and pCAG-Cre from Addgene ( #13775 ) . Plasmid pCAG-Flpo ( Addgene #60662 ) was previously described ( Xue et al . , 2014 ) . All other plasmids were generated and deposited at Addgene as below . pCAG-hChR2 ( H134R ) -EYFP ( Addgene #114367 ) was created by replacing the EGFP in pCAG-EGFP ( Addgene #11150 ) with the hChR2 ( H134R ) -EYFP from pAAV-EF1α-DIO-hChR2 ( H134R ) -EYFP ( Addgene #20298 ) . pAAV-EF1α-DIO-ReaChR-P2A-dTomato ( Addgene #114368 ) was created by replacing the oChIEF ( E163A T199C ) in pAAV-EF1α-DIO-oChIEF ( E163A T199C ) -P2A-dTomato ( Addgene #51094 ) with the ReaChR from pAAV-hSyn-FLEX-ReaChR-Citrine ( Addgene #50955 ) . pAAV-EF1α-FRT-FLEX-GtACR2-EYFP ( Addgene #114369 ) and pAAV-EF1α-FRT-FLEX-GtACR1-EYFP ( Addgene #114370 ) were created by replacing the mNaChBac-T2A-tdTomato in pAAV-EF1α-FRT-FLEX-mNaChBac-T2A-tdTomato ( Addgene #60658 ) with the GtACR2-EYFP and GtACR1-EYFP from pLenti-UbiC-GtACR2-EYFP and pLenti-UbiC-GtACR1-EYFP , respectively . Motifs MBD and TlcnC were generated by PCR primers . Motifs Nlgn1C and Kv4 . 2LL were generated by PCR from pAAV-CAG-post-mGRASP-2A-dTomato ( Addgene #34912 ) and a Kv4 . 2-expressing plasmid , respectively . Motif Kv2 . 1C was obtained from pAAV-EF1α-DIO-hChR2 ( H134R ) -EYFP-Kv2 . 1C ( Wu et al . , 2013 ) . Motifs Kv2 . 1C-TlcnC and Kv2 . 1C-linker-TlcnC were generated by PCR from Kv2 . 1C and TlcnC . All motifs were then added to the C-terminus of the GtACR2-EYFP to create pAAV-EF1α-FRT-FLEX-GtACR2-EYFP-MBD ( Addgene #114371 ) , pAAV-EF1α-FRT-FLEX-GtACR2-EYFP-Nlgn1C ( Addgene #114372 ) , pAAV-EF1α-FRT-FLEX-GtACR2-EYFP-Kv4 . 2LL ( Addgene #114373 ) , pAAV-EF1α-FRT-FLEX-GtACR2-EYFP-TlcnC ( Addgene #114374 ) , pAAV-EF1α-FRT-FLEX-GtACR2-EYFP-Kv2 . 1C ( Addgene #114375 ) , pAAV-EF1α-FRT-FLEX-GtACR2-EYFP-Kv2 . 1C-TlcnC ( Addgene #114376 ) , and pAAV-EF1α-FRT-FLEX-GtACR2-EYFP-Kv2 . 1C-linker-TlcnC ( Addgene #114377 ) . pAAV-EF1α-FRT-FLEX-GtACR2-KA2N-EGFP ( Addgene #114378 ) was created by replacing the EYFP in pAAV-EF1α-FRT-FLEX-GtACR2-EYFP with the KA2N-EGFP from pAAV-hSyn-soCoChR-EGFP ( Addgene #107708 ) . Female ICR mice were crossed with male C57BL6/J , Pvalb-2A-Cre , or Pvalb-2A-Flpo mice to obtain timed pregnancies . In utero electroporation was performed as previously described ( Xue et al . , 2014 ) with a square-wave pulse generator ( Gemini X2 , BTX Harvard Bioscience ) . To express GtACR2 , GtACR1 , iC++ , or ChR2 in layer 2/3 pyramidal neurons , pLenti-UbiC-GtACR2-EYFP , pLenti-UbiC-GtACR1-EYFP , pAAV-CaMKIIα-iC++-TS-EYFP , or pCAG-hChR2 ( H134R ) -EYFP ( all 2 µg/µl ) was used , respectively . In a few experiments , pAAV-EF1α-FRT-FLEX-GtACR2-EYFP ( 2 µg/µl ) with pCAG-Flpo ( 0 . 2 µg/µl ) , pAAV-EF1α-FRT-FLEX-GtACR1-EYFP ( 2 µg/µl ) with pCAG-Flpo ( 0 . 2 µg/µl ) , or pAAV-EF1α-DIO-iC++-TS-EYFP ( 2 µg/µl ) with pCAG-Cre ( 0 . 2 µg/µl ) was used to express GtACR2 , GtACR1 , or iC++ , respectively . To express somatodendritically targeted GtACR2 variants in layer 2/3 pyramidal neurons and compare them with wild type GtACR2 , the pAAV-EF1α-FRT-FLEX constructs described above were used ( all 2 µg/µl ) with pCAG-Flpo ( 0 . 2 µg/µl ) . pCAG-tdTomato ( 0 . 1 µg/µl ) was included in all experiments . The plasmid concentrations stated above were final concentrations in the plasmid mix . Transfected pups were identified by the transcranial fluorescence of tdTomato with a MZ10F stereomicroscope ( Leica ) 1–2 days after birth . All recombinant AAV serotype 9 vectors were produced by the Gene Vector Core at Baylor College of Medicine except AAV9-hSyn-Flpo ( Addgene #60663 ) , which was produced by the Penn Vector Core ( Xue et al . , 2014 ) . To express GtACR2 , ReaChR , iC++ , or iChloC in Pv neurons , 200–250 nl of the following recombinant AAV serotype 9 vectors at their respective titer were injected into the visual cortex of PvalbFlpo/+ ( for GtACR2 ) or PvalbCre/+ ( for ReaChR , iC++ , or iChloC ) mice at postnatal day 1 as previously described ( Xue et al . , 2014 ) : AAV9-EF1α-FRT-FLEX-GtACR2-EYFP ( 3 . 8 × 1013 genome copies/ml ) , AAV9-EF1α-DIO-ReaChR-P2A-dTomato ( 7 . 0 × 1013 genome copies/ml ) , AAV9-EF1α-DIO-iC++-TS-EYFP ( 3 . 71 × 1013 genome copies/ml ) , and AAV9-EF1α-DIO-iChloC-2A-mCherry ( 3 . 7 × 1014 genome copies/ml ) . To conditionally express GtACR2 in juvenile and adult neurons , mice previously electroporated with plasmid pAAV-EF1α-FRT-FLEX-GtACR2-EYFP into layer 2/3 pyramidal neurons were injected with 200 nl of AAV9-hSyn-Flpo ( 1 . 2 × 1012 genome copies/ml ) at postnatal day 23 , 60 , or 64 . Injection was performed as previously described ( Xue et al . , 2014 ) with an UltraMicroPump III and a Micro4 controller ( World Precision Instruments ) . Mice were anesthetized by an intraperitoneal injection of a ketamine and xylazine mix ( 80 mg/kg and 16 mg/kg , respectively ) and transcardially perfused with cold ( 0–4°C ) slice cutting solution containing 80 mM NaCl , 2 . 5 mM KCl , 1 . 3 mM NaH2PO4 , 26 mM NaHCO3 , 4 mM MgCl2 , 0 . 5 mM CaCl2 , 20 mM D-glucose , 75 mM sucrose and 0 . 5 mM sodium ascorbate ( 315 mosmol , pH 7 . 4 , saturated with 95% O2/5% CO2 ) . Brains were removed and sectioned in the cutting solution with a VT1200S vibratome ( Leica ) to obtain 300 μm coronal slices . Slices were incubated in a custom-made interface holding chamber containing slice cutting solution saturated with 95% O2/5% CO2 at 34°C for 30 min and then at room temperature for 20 min to 10 hr until they were transferred to the recording chamber . Recordings were performed on submerged slices in artificial cerebrospinal fluid ( ACSF ) containing 119 mM NaCl , 2 . 5 mM KCl , 1 . 3 mM NaH2PO4 , 26 mM NaHCO3 , 1 . 3 mM MgCl2 , 2 . 5 mM CaCl2 , 20 mM D-glucose and 0 . 5 mM sodium ascorbate ( 305 mosmol , pH 7 . 4 , saturated with 95% O2/5% CO2 , perfused at 3 ml/min ) at 30–32°C . For whole-cell recordings , a K+-based pipette solution containing 142 mM K+-gluconate , 10 mM HEPES , 1 mM EGTA , 2 . 5 mM MgCl2 , 4 mM ATP-Mg , 0 . 3 mM GTP-Na , 10 mM Na2-phosphocreatine ( 295 mosmol , pH 7 . 35 ) or a Cs+-based pipette solution containing 121 mM Cs+-methanesulfonate , 1 . 5 mM MgCl2 , 10 mM HEPES , 10 mM EGTA , 4 mM Mg-ATP , 0 . 3 mM Na-GTP , 10 mM Na2-Phosphocreatine , and 2 mM QX314-Cl ( 295 mosmol , pH 7 . 35 ) was used . Membrane potentials were not corrected for liquid junction potential ( experimentally measured as 12 . 5 mV for the K+-based pipette solution and 9 . 5 mV for the Cs+-based pipette solution ) . Neurons were visualized with video-assisted infrared differential interference contrast imaging , and fluorescent neurons were identified by epifluorescence imaging under a water immersion objective ( 40× , 0 . 8 numerical aperture ) on an upright SliceScope Pro 1000 microscope ( Scientifica ) with an infrared IR-1000 CCD camera ( DAGE-MTI ) . Data were low-pass filtered at 4 kHz and acquired at 10 kHz with an Axon Multiclamp 700B amplifier and an Axon Digidata 1550 Data Acquisition System under the control of Clampex 10 . 7 ( Molecular Devices ) . Data were analyzed offline using AxoGraph X ( AxoGraph Scientific ) . For the photostimulation of GtACR2- , iC++- , iChloC- , or ChR2-expressing neurons , blue light was emitted from a collimated light-emitting diode ( LED ) of 455 nm , whereas for the photostimulation of GtACR1- or ReaChR-expressing neurons , red light was emitted from a LED of 617 nm . The LEDs were driven by a LED driver ( Mightex ) under the control of an Axon Digidata 1550 Data Acquisition System and Clampex 10 . 7 . Light was delivered through the reflected light fluorescence illuminator port and the 40× objective . Synaptic currents and photocurrents were recorded in the whole-cell voltage clamp mode with the Cs+-based patch pipette solution . Only recordings with series resistance below 20 MΩ were included . EPSCs and IPSCs were recorded at the reversal potential for IPSCs ( −60 mV ) and EPSCs ( +10 mV ) , respectively , unless stated otherwise . Photocurrents were recorded at +10 mV unless stated otherwise . For short light pulse stimulation , pulse duration ( 0 . 5–10 ms ) and intensity ( 2 . 3–23 . 6 mW/mm2 ) were adjusted for each recording to evoke small ( to minimize voltage-clamp errors ) but reliable monosynaptic EPSCs or IPSCs . Disynaptic IPSCs were evoked using the same light pulses that were used for evoking the corresponding monosynaptic EPSCs . Light pulses were delivered at 30 s interstimulus intervals . For long light pulse stimulation , blue light was delivered for 2 s at 60 s interstimulus intervals . To obtain the cumulative charge transfer curves , the EPSC traces were high-pass filtered at 0 . 3 Hz to minimize the effect of slow changes in the baselines , except for two neurons that were filtered at 0 . 5 Hz or 1 . 3 Hz . It was not necessary to high-pass filter the IPSC traces . The traces were baselined before the light onset and then integrated to calculate the cumulative charge transfers . Antidromic spikes in GtACR2+ neurons were recorded with the K+-based patch pipette solution in whole-cell current clamp mode or with ACSF as the patch pipette solution in the loose-patch current clamp mode . For pharmacology experiments , the baseline synaptic currents were recorded for at least 3 min in the absence of any drug . The drugs were then added to the ACSF at the following concentrations: TTX ( 1 μM ) , NBQX ( 10 μM ) , ( RS ) -CPP ( 10 μM ) , SR95531 ( Gabazine , 10 μM ) , ZD7288 ( 20 μM ) , bumetanide ( 50 or 100 μM ) , TEA ( 1 . 5 mM ) , and 4-AP ( 1 . 5 mM ) . The synaptic currents were recorded for at least 3 min in the presence of drugs . For ZD7288 , which did not inhibit GtACR2-induced neurotransmitter release , the efficacy of the drug was monitored by examining the Ih current of cortical layer 5 pyramidal neurons . Fluorescent images were taken from live brain slices , except for the conditional expression of GtACR2 in adults , where images were taken from fixed brain slices . Live brain slices were prepared as described for slice electrophysiology . For the fixed brain slices , mice were anesthetized by an intraperitoneal injection of a ketamine and xylazine mix ( 80 mg/kg and 16 mg/kg , respectively ) and transcardially perfused with phosphate buffered saline ( PBS , pH 7 . 4 ) followed by 4% paraformaldehyde in PBS ( pH 7 . 4 ) . Brains were further fixed overnight in 4% paraformaldehyde , cryoprotected with 30% sucrose in PBS , and frozen in optimum cutting-temperature medium until sectioning . A HM 450 Sliding Microtome ( Thermo Scientific ) was used to section the brains to obtain 30–50 mm coronal slices . Images were acquired on an Axio Zoom . V16 Fluorescence Stereo Zoom Microscope ( Zeiss ) and processed using National Institutes of Health ImageJ . To determine the EYFP ( or EGFP ) fluorescence ratio between layer 5 and layer 2/3 , one or two 350 µm-wide rectangular regions that were perpendicular to the pia and spanned all 6 cortical layers were selected in the most transfected regions of each slice . The mean EYFP ( or EGFP ) fluorescence was measured for layer 5 and layer 2/3 within the selected area . The mean tdTomato fluorescence was measured similarly for layer 5 and layer 2/3 . The mean background fluorescence was measured from a nearby rectangular region ( 140 . 5 µm by 90 . 8 µm ) where no cellular EYFP ( or EGFP ) and tdTomato fluorescence was present . The normalized EYFP ( or EGFP ) fluorescence ratio between layer 5 and layer 2/3 was calculated by Layer 5EYFP-BackgroundEYFPLayer 2/3EYFP-BackgroundEYFP/Layer 5tdTomato-BackgroundtdTomatoLayer 2/3tdTomato-BackgroundtdTomato . To determine the ratio of EYFP fluorescence to tdTomato fluorescence in the callosal projections , one or two rectangular regions that contained the tdTomato-labeled axons were selected in each slice to measure the mean EYFP and tdTomato fluorescence . The mean background fluorescence was measured in a nearby cortical area spanning the same layers . The ratio of EYFP fluorescence to tdTomato fluorescence was calculated by FluorescenceEYFP-BackgroundEYFPFluorescencetdTomato-BackgroundtdTomato . All reported sample numbers ( n ) represent biological replicates that are the numbers of recorded neurons for electrophysiology or the numbers of analyzed regions of interest ( ROI ) for fluorescent images . Statistical analyses were performed with Prism 7 ( GraphPad Software ) . We first determined whether the data were normally distributed by performing the D’Agostino and Pearson test , Shapiro-Wilk test , and KS test . If all data within one experiment passed all three normality tests , we then performed the statistical test that assumes a Gaussian distribution . Otherwise , we performed the statistical test that assumes a non-Gaussian distribution . All statistical tests were two-tailed with an alpha of 0 . 05 . Wilcoxan matched-pairs signed rank test was used for Figure 1F , K; Figure 2B , D , F , J; Figure 2—figure supplement 2B , E; Figure 2—figure supplement 3B ( EPSCs ) ; Figure 3B; and Figure 3—figure supplement 1C . Paired t test was used for Figure 1G , J; Figure 2H; Figure 2—figure supplement 1D; and Figure 2—figure supplement 3B ( photocurrents ) ; Figure 2—figure supplement 3D . t test with Welch’s correction was used for Figure 5C ( WT vs . Kv2 . 1C ) . Mann-Whitney test was used for Figure 5C ( WT vs . Kv2 . 1C-linker-TlcnC ) ; Figure 5F , G , I; Figure 5—figure supplement 1C , D . Repeated measures one-way ANOVA with Greenhouse-Geisser correction and Tukey multiple comparisons with multiplicity adjusted P values was used for Figure 3D , F . Kruskal-Wallis test with Dunn’s multiple comparisons with multiplicity adjusted P values was used for Figure 4D . Ordinary one-way ANOVA with Dunnet’s multiple comparison test with multiplicity adjusted P values was used for Figure 4E . The details of all statistical tests , numbers of replicates and mice , and P values were reported in Supplementary file 1 .	One way to study the role of a specific neuron is to activate or inhibit the cell and then observe the consequences . This can be achieved by using optogenetics , a technique that involves introducing ‘light-gated’ ion channels in the outer membrane of a target neuron . When light is shone on the cell , these pore-like proteins open their channels: this allows ions to move into or out of the neuron . Ions flow from high concentration to low concentration areas . Typically , when a neuron is at rest , there are fewer chloride ions inside the cell than outside . Activating a light-gated chloride channel should thus cause these negatively charged ions to enter the neuron . The charge inside of the cell would become more negative relative to the outside: this would inhibit the neuron , making it less likely to fire . Here , Messier et al . looked into using a light-gated chlorine channel called GtACR2 to inhibit the activity of neurons in mouse brain slices , but the results were not as expected . Activating the chloride channel did inhibit the cell body , the area of the neuron that contains the nucleus . Yet , it had the opposite effect in the axon , the structure that carries electrical signals away from the cell body . There , activating GtACR2 caused chloride ions to leave the axon , which resulted in the neuron firing . Testing other types of optogenetic chloride channels produced the same result . Further experiments revealed that the concentration of chloride ions is higher inside the axon than the cell body , explaining the observed effects . Messier et al . then tried to redistribute the channels from the axon to the cell body , where the proteins are inhibitory . This was accomplished by fitting GtACR2 with a molecular tag that acts like an address label , with the cell body as the target destination . Overall , when these modified channels were activated , the neuron was more strongly inhibited . Ultimately , the GtACR2 channel designed by Messier et al . is a powerful new inhibitory optogenetic tool . In addition , this tool could be used to study chloride gradients in brain regions , cell types and areas of cells that are otherwise difficult to access .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Targeting light-gated chloride channels to neuronal somatodendritic domain reduces their excitatory effect in the axon
Cyclin D1 is a critical regulator of cell cycle progression and works at the G1 to S-phase transition . Here , we report the isolation and characterization of the novel c-Myc-regulated lncRNA LAST ( LncRNA-Assisted Stabilization of Transcripts ) , which acts as a CCND1 mRNA stabilizer . Mechanistically , LAST was shown to cooperate with CNBP to bind to the 5′UTR of CCND1 mRNA to protect against possible nuclease targeting . In addition , data from CNBP RIP-seq and LAST RNA-seq showed that CCND1 mRNA might not be the only target of LAST and CNBP; three additional mRNAs were shown to be post-transcriptional targets of LAST and CNBP . In a xenograft model , depletion of LAST diminished and ectopic expression of LAST induced tumor formation , which are suggestive of its oncogenic function . We thus report a previously unknown lncRNA involved in the fine-tuned regulation of CCND1 mRNA stability , without which CCND1 exhibits , at most , partial expression . The oncoprotein c-Myc plays a pivotal role in multiple cellular processes , such as cell cycle progression , malignant transformation , differentiation suppression and apoptosis induction , predominantly through its transcription activity ( Seth et al . , 1993; Drayton et al . , 2003; Wei et al . , 2003; Demeterco et al . , 2002; Prendergast , 1999; Amati et al . , 1992; Lee et al . , 1996; Hoffman and Liebermann , 2008 ) . Indeed , as a master transcriptional factor , c-Myc regulates the expression of approximately 10–15% of genes in the genome , including a variety of protein-coding genes ( Lin et al . , 2012; Nie et al . , 2012; Fernandez et al . , 2003 ) , such as CDKN1A , CDKN2B , CCND1 , CCND2 , CDK4 and E2F2 ( Adhikary and Eilers , 2005 ) . Among c-Myc target genes , CCND1 is of particular importance in cell cycle control and is characterized by the dramatic periodicity of the abundance of its protein product cyclin D1 throughout the cell cycle ( Sherr , 1995 ) . Cyclin D1 forms a complex with CDK4 or CDK6 and functions as a regulatory subunit whose activity is required for G1/S transition ( Sherr , 1995; Resnitzky et al . , 1994 ) . Cyclin D1 also interacts with the tumor suppressor pRB1 , which in turn positively regulates cyclin D1 expression ( DeGregori , 2004 ) . Mutation , amplification and overexpression of CCND1 are frequently observed in cancer and have been reported to contribute to tumorigenesis ( Wiestner et al . , 2007; Elsheikh et al . , 2008; Musgrove et al . , 2011 ) . Cyclin D1 is a short-lived protein with a rapid turnover rate ( ~24 min ) due to degradation by the ubiquitin-proteasome system ( Diehl et al . , 1998; Diehl et al . , 1997 ) . While early studies showed that the Skp2 F-box protein is involved in cyclin D1 degradation ( Yu et al . , 1998 ) , a recent study has identified two additional F-box proteins that play important roles in targeting cyclin D1 for proteasome degradation ( Lin et al . , 2006; Okabe et al . , 2006 ) . c-Myc can upregulate or downregulate expression of cyclin D1 in a context-dependent manner . On the one hand , c-Myc , together with Max , a co-transcription factor , activates CCND1 transcription through an E box located at −558 nt in its promoter ( Kress et al . , 2015; Yu et al . , 2005; Guo et al . , 2011 ) . On the other hand , simultaneous overexpression of c-Myc and ZO-2 enhances repression of the CCND1 promoter through the E box in MDCK cells ( Gonzalez-Mariscal et al . , 2009 ) . In addition , c-Myc has been reported to repress the cyclin DI promoter and antagonize USF-mediated transactivation in BALB/c-3T3 , Rat6 and rat embryo fibroblasts ( Philipp et al . , 1994 ) . In addition to c-Myc , multiple transcription factors , including AP-1 , NF-κB , E2F and Oct-1 , can bind to their respective CCND1 promoters and regulate its expression ( Guo et al . , 2011 ) . CCND1 can also be regulated epigenetically through histone modifications; GATA3 cooperates with PARP1 to regulate CCND1 transcription by modulating histone H1 incorporation ( Shan et al . , 2014 ) . Moreover , post-transcriptional mechanisms are also involved in the regulation of CCND1 , as exemplified by MYF5-mediated enhancement in CCND1 mRNA translation , which contributes to early myogenesis ( Panda et al . , 2016 ) . Mutations in CCND1 leading to stable truncated transcripts are associated with increased cell proliferation and shortened survival of cancer patients ( Wiestner et al . , 2007 ) . Long noncoding RNAs ( lncRNAs ) , which are defined as transcripts that are longer than 200 nucleotides and lack protein coding capacity , are emerging as important regulators of biological processes , including regulation of gene expression at multiple levels , such as chromatin remodeling , transcription , and post-transcriptional modulation ( Derrien et al . , 2012; Iorio and Croce , 2012; Bonasio and Shiekhattar , 2014; Wilusz et al . , 2008 ) . Genome-wide studies have shown that c-Myc transcriptionally regulates many lncRNA genes , such as PVT1 , the CCAT family , and MYCLos , whereas a number of lncRNAs have been demonstrated to be important components of the c-Myc-mediated signaling network ( Colombo et al . , 2015; Nissan et al . , 2012; Ling et al . , 2013; Kim et al . , 2015a; Kim et al . , 2015b ) . Nevertheless , whether lncRNAs participate in the regulation of CCND1 remains to be fully studied . Here , we report the isolation and characterization of the novel c-Myc regulated LAST , which acts as a CCND1 mRNA stabilizer and without which CCND1 mRNA becomes unstable and cell cycle arrest occurs in the G1 phase . Mechanistically , LAST cooperates with CNBP , a single-stranded DNA/RNA-binding factor , to bind to the 5’ untranslated region of CCND1 messenger RNA , possibly to protect against nuclease degradation . This report describes a model by which lncRNA stabilizes mRNA post-transcriptionally via 5’-end protection . To identify novel functions of c-Myc-regulated long non-coding RNAs , doxycycline-treated or untreated P493-6 cells carrying a c-Myc tet-off system ( Kim et al . , 2007 ) were used to analyze the lncRNA expression profile via long non-coding RNA microarray analysis ( Supplementary file 1 , GSE106916 ) . We selected five significantly c-Myc-downregulated lncRNAs ( fold change above 8 , P-value below 0 . 01 ) that were identified by the lncRNA microarray . Two of the five lncRNAs , namely , lncRNA-51 and lncRNA-52 , along with CDK4 ( positive control ) were found to be downregulated when c-Myc expression was suppressed by doxycycline treatment ( Figure 1A ) . Of these two c-Myc responsive lncRNAs , lncRNA-52 ( RP11-660L16 . 2 , ENST00000529369 ) was chosen for further investigation because knockdown of this lncRNA ( Figure 1B ) showed a significant reduction in colony formation ( Figure 1C and D ) . lncRNA-52 is located approximately 1 . 8 Mb downstream of the cyclin D1/CCND1 gene in a head-to-tail orientation ( Figure 1—figure supplement 1A ) . As will be shown in the following sections , this lncRNA is able to promote the stability of mRNA transcripts , including CCND1 mRNA; we therefore named it LAST ( LncRNA-Assisted Stabilization of Transcripts ) . To verify the existence of endogenousLAST and to determine its molecular size , northern blot analysis was performed . A band of approximately 700 nt in length was detected in both P493-6 and H1299 cells , but was absent in P493-6 cells treated with doxycycline , which suppresses c-Myc expression ( Figure 1E ) . The apparent size of LAST was the same as predicted by the UCSC ( University of California , Santa Cruz ) Genome Browser , suggesting that LAST is a full-length transcript . To investigate the cellular compartment in which LAST is located , single molecule RNA fluorescent in situ hybridization ( FISH ) was performed in both HCT116 and H1299 cells . As shown in Figure 1F , LAST was predominantly localized in the cytosol . Cytosol localization of LAST was also confirmed by determining the levels of LAST in different sub-cellular fractions ( Figure 1G ) . Moreover , the signal intensity of LAST was reduced in LAST-depleted cells ( Figure 1—figure supplement 1B ) . It has been reported that lncRNAs are often present at relatively low copy numbers; hence , we measured the LAST transcript copy number per cell in various cell lines , including the normal cell lines HAFF , IMR90 , and MCF10A and tumor cell lines HCT116 , MCF7 and H1299 . The LAST copy number was higher in tumor cells than in normal cells ( Figure 1H ) . Lentivirus-mediated gene knockdown of c-Myc decreased whereas ectopic expression of c-Myc increased LAST expression in HCT116 , H1299 and MCF10A cells ( Figure 1I and J ) . Furthermore , the levels of LAST and c-Myc appeared to be notably synchronous during cell cycle progression ( Figure 1—figure supplement 1C ) . In particular , the c-Myc and LAST levels were decreased in G2/M ( lane 2 ) , followed by a rapid increase in the G1 phase ( lane 3 ) . These data suggest that LAST expression is positively regulated by c-Myc . We next explored whether c-Myc regulates LAST expression at the transcriptional level . We first inspected the genomic sequence around the LAST gene using the JASPAR database ( Mathelier et al . , 2016 ) . Six putative c-Myc binding sites ( BS1 , BS2 , BS3 , BS4 , BS5 and BS6 ) were identified ( Figure 1K , upper panel ) . Furthermore , we analyzed the genomic sequence around the LAST gene using the ENCODE database . Three fragments ( F1 , F2 and F3 ) were predicted to be recognized by c-Myc ( Figure 1K , upper panel ) . The chromatin immunoprecipitation ( ChIP ) assay determined the association of c-Myc with chromatin fragments corresponding to the BS2 and BS3 sites ( within F2 fragment ) among all examined fragments ( Figure 1K , lower panel ) . We further evaluated whether BS2 and BS3 conferred c-Myc-dependent transcriptional activity . DNA fragments containing wild-type BS2 and BS3 or their corresponding mutant binding sites were inserted into the promoter region of a firefly luciferase reporter plasmid ( Figure 1L , upper panel ) . Luciferase expression from the reporter containing an individual BS2 or BS3 site was indeed induced by ectopic expression of c-Myc ( Figure 1L , lower panel ) . By contrast , mutant BS2M and BS3M sites showed no response to c-Myc induction ( Figure 1L , lower panel ) . These data demonstrate that c-Myc transactivates LAST . Knockdown of LAST results in reduced colony formation ( Figure 1C ) , indicating that LAST normally promotes cell proliferation . To examine how LAST affects cell growth , the cell cycle phase distribution was analyzed by flow cytometric analysis . Knockdown of LAST caused a decrease in the percentage of cells in the S and G2/M phases and an increase in the percentage of cells in the G1 phase ( Figure 2—figure supplement 1A and B ) , indicating that LAST knockdown prevents cell passage from the G1 phase into S phase . As a result , LAST was shown to promote G1/S phase transition . Cell cycle regulation is controlled by many factors . To define which factor ( s ) were involved in LAST-mediated regulation , the mRNA levels of G1-related cyclins and CDKs genes were selected for comparison in HCT116 cells before and after LAST gene knockdown . Among all of the mRNAs examined , only the CCND1 mRNA level was significantly decreased ( Figure 2A ) . Among all of the cyclins or CDKs examined , only cyclin D1 was shown to be downregulated when LAST was depleted ( Figure 2B ) . The lncRNA PVT1 is known to be a c-Myc regulated lncRNA that is involved in c-Myc stability and activity ( Colombo et al . , 2015 ) . However , unlike PVT1 , we found that LAST , which is also regulated by c-Myc ( Figure 1A ) , does not affect c-Myc stability since knockdown of LAST did not change c-Myc expression at either the mRNA or protein levels ( Figure 2A and B ) . The effect of LAST on cyclin D1/CCND1 was further verified in normal HAFF cells and tumor H1299 and HCT116 cells . Depletion of LAST decreased whereas over-expression of LAST increased cyclin D1/CCND1 expression at both the mRNA and protein levels ( Figure 2C and D ) . To test if the function of LAST is mediated through an effect on the adjacent genes , we checked the shRNA-mediated LAST knockdown effect on the two adjacent genes DHCR7 and NADSYN1 ( Figure 1—figure supplement 1A ) and found that LAST showed no effect on either the mRNA or protein levels of those two genes ( Figure 2—figure supplement 1C ) . Furthermore , we introduced shRNA-resistant LAST into LAST depleted cells , and as shown in Figure 2—figure supplement 1D and E , both the CCND1 mRNA and protein levels were rescued . This result excludes off-target effects of LAST shRNA knockdown . To investigate how LAST affects the CCND1 mRNA level , we first examined whether LAST regulates the CCND1 mRNA transcription process . The levels of both CCND1 pre-mRNA and mature mRNA were examined by primers against CCND1 mRNA intron- or exon-specific regions in HCT116 cells treated with and without LAST knockdown . The levels of CCND1 pre-mRNA containing four intronic regions were found to remain unaltered between control and LAST knockdown cells , whereas the levels of mature spliced CCND1 mRNA containing 5’UTR , CDS ( coding sequences ) and 3’UTR regions were greatly reduced upon LAST depletion ( Figure 3—figure supplement 1A and B ) . These results suggest that LAST may post-transcriptionally regulate CCND1 mRNA . To evaluate the effect of LAST on the stability of CCND1 mRNA , HCT116 cells were treated with actinomycin D , which measures the decay of pre-existing mRNA . Knockdown of LAST resulted in a decrease of the half-life of CCND1 mRNA from 5 hr to 3 hr ( Figure 3A ) , whereas over-expression of LAST increased its half-life from 5 hr to 9 hr ( Figure 3B ) , indicating that LAST stabilizes CCND1 mRNA . To determine whether LAST interacts with CCND1 mRNA , we performed a biotinylated oligonucleotide pull-down assay , and as shown in Figure 3C , endogenous CCND1 mRNA but not CCNB1 mRNA co-precipitated with LAST , indicating an association between LAST and CCND1 mRNA . However , by careful inspection , we found there was no complementary base pairing between LAST and CCND1 mRNA . We therefore hypothesized that some protein ( s ) may mediate this binding . Proteins pulled down by LAST were separated by SDS PAGE , and a unique band with a molecular weight of approximately 20 kDa was revealed and identified as CNBP by mass spectrometry ( Figure 3—figure supplement 1C , left panel ) . CNBP has a preference for binding single-stranded DNA and RNA ( Flink and Morkin , 1995 ) and has been reported to function in the translation of ornithine decarboxylase mRNA ( Sammons et al . , 2010 ) . To validate the MS Spectro result , we performed a LAST pull-down assay . A biotin-labeled antisense DNA probe against LAST pulled down CNBP , but not cyclin D1 ( Figure 3—figure supplement 1C , right panel ) . To further demonstrate that CNBP can bridge CCND1 mRNA and LAST , we first pulled down CCND1 mRNA using a biotinylated antisense DNA probe as bait; both CNBP and LAST were coprecipitated ( Figure 3D ) . The RIP assay further concluded that CNBP interacts with both CCND1 mRNA and LAST ( Figure 3E ) . These data demonstrate that CNBP acts as a mediator for LAST and CCND1 mRNA binding . As shown in Figure 3F , CNBP knockdown in HCT116 led to a decrease in both the mRNA and protein levels of cyclin D1/CCND1 . The effect of CNBP on the stability of CCND1 mRNA was evaluated in HCT116 cells treated with actinomycin D . The half-life of CCND1 mRNA was reduced from 5 hr to 3 hr as CNBP was depleted ( Figure 3G ) , further demonstrating that CNBP prolongs the CCND1 mRNA half-life . Because CNBP predominantly resides in the cytosol , we investigated whether the association of CCND1 mRNA with LAST via CNBP only occurs in the cytosol . Using a LAST and CCND1 mRNA pull-down assay , we found that CNBP was co-precipitated by either LAST or CCND1 mRNA in the cytoplasm , but not the nucleus ( Figure 3—figure supplement 1D ) . HNRNPK was used as a nuclear marker . Moreover , knockdown of CNBP was shown to result in decreased levels of CCND1 mRNA ( Figure 3F ) . Further investigation revealed that knockdown of CNBP affected the level of mature CCND1 mRNA , but not unspliced CCND1 pre-mRNA , indicating that the protective role of CNBP in mature CCND1 mRNA stability occurred in the cytosol since nuclear unspliced CCND1 pre-mRNA was not affected when CNBP was silenced ( Figure 3—figure supplement 1E ) . To further confirm that LAST affects CCND1 mRNA stability through CNBP , we knocked down CNBP in HCT116 cells . As shown in Figure 3—figure supplement 1F , the increased expression of CCND1 mRNA caused by over-expression of LAST was diminished when CNBP was depleted ( lanes 2 vs . 4 ) . Thus , our hypothesis is that LAST affects CCND1 mRNA via CNBP . Knockdown of LAST resulted in no change in CNBP at either the RNA or protein levels ( Figure 3H ) , which suggests that LAST affects CCND1 mRNA not according to the quantity of CNBP , but rather by the association of CNBP and CCND1 mRNA . It was therefore expected that knockdown of CNBP would reduce the association between LAST and CCND1 mRNA . This was indeed the case . An RNA pull-down experiment was performed starting with the same amount of CCND1 mRNA , and less LAST was co-precipitated as CNBP was depleted ( Figure 3I ) . Similarly , when we pulled down the same amount of CCND1 mRNA , less co-precipitated CNBP was detected as LAST was silenced . As a negative control , the RNA-binding protein HuR remained unchanged after LAST was knocked down ( Figure 3J ) . These data suggest that LAST cooperates with CNBP to regulate CCND1 mRNA stability . To describe a detailed CNBP , LAST and CCND1 mRNA binding mechanism , we mapped the LAST and CNBP binding sites on CCND1 mRNA by RNA pull-down using different in vitro biotin-labeled fragments ( Figure 4A , upper panel ) . We found that the 5’UTR but not 3’UTR-1 , –2 and −3 of CCND1 mRNA was able to bind LAST and CNBP ( Figure 4A and B ) , indicating that LAST and CNBP bind to the 5’ region of CCND1 mRNA . To further determine whether LAST and CNBP bind to the CCND1 mRNA 5’UTR to enhance its stability , two CCND1 expression constructs were generated , as shown in Figure 4—figure supplement 1A and B . One construct contained the CCND1 coding region ( CD ) plus the 5’UTR and the other contained the CD alone . The expression plasmid plus LAST and CNBP or plasmid alone was individually transfected into 293T cells . Actinomycin D was added to measure the mRNA decay rate . The half-life of ectopically expressed CCND1 mRNA lacking the 5’UTR was not altered by the presence or absence of LAST and CNBP ( Figure 4—figure supplement 1A ) . By contrast , the half-life of CCND1 mRNA bearing 5'-UTRs was extended from 4 hr in the absence of LAST and CNBP to 9 hr in the presence of LAST and CNBP , indicating that LAST and CNBP enhanced CCND1 mRNA stability via its 5’UTR ( Figure 4—figure supplement 1B ) . In addition , we performed CNBP RIP sequencing , and an enrichment peak at the CCND1 5’UTR was observed ( Figure 4—figure supplement 1C ) . This was consistent with the previous conclusion from Figure 4B . Thus , our hypothesis is that CNBP binds both CCND1 mRNA and LAST . We again examined whether LAST , the 5'-UTR of CCND1 mRNA and CNBP form a ternary complex by using a sequential immuno-precipitation assay ( Figure 4—figure supplement 2A ) . By using an anti-FLAG antibody against FLAG–CNBP , both the biotin-labeled 5'-UTR of CCND1 mRNA and LAST were pulled down in an initial immunoprecipitation assay ( Figure 4C , panel 1 and 3 ) . The immunocomplexes were eluted and were subsequently precipitated by streptavidin sepharose beads against the biotin-labeled 5'-UTR of CCND1 mRNA . LAST and CNBP were present in the streptavidin-biotin precipitates ( Figure 4C , panel 5 and 6 ) , indicating that these three components indeed form a ternary complex . CNBP prefers to bind G-rich motifs , especially the GGAG core ( Armas et al . , 2008; Benhalevy et al . , 2017 ) . We checked the proportion of G-rich motifs in all of the peak sequences from the CNBP RIP samples . Nearly sixty percent of the CNBP enriched sequences contained the GGAG motif , and more than ninety percent of the peak sequences contained a GGR motif ( Figure 4—figure supplement 2B ) . To assess the possible CNBP binding sites on LAST and CCND1 mRNA , the bioinformatics software tool QGRS Mapper was utilized ( Kikin et al . , 2006 ) . Two G-rich sequences containing a GGAG core in the 5’UTR of CCND1 mRNA were identified ( Figure 4D , upper part ) . An electrophoretic mobility shift assay ( EMSA ) was performed , and the results showed that G-rich-1 and G-rich-2 in the CCND1 5’UTR were responsible for the binding of CNBP ( Figure 4D , lower part ) . Among the six predicted G-rich sequences ( G-rich-A to F ) found in LAST ( Figure 4E , upper part ) , four G-rich sequences ( G-rich-A , C , D and F ) were found to contain a GGAG core . G-rich-A , C , D and F from LAST were able to bind CNBP , whereas neither G-rich-B nor G-rich-E was able to bind CNBP ( Figure 4E , lower part ) . Thus , CNBP only interacted with G-rich sequences that contained the GGAG core , but not those lacking the GGAG core . These data suggest that both CCND1 and LAST interact with CNBP via their G-rich motifs containing the GGAG core . Four G-rich regions ( A , C , D and F ) were mutated from GGAG to UUUU with either a single mutation or four combined mutations in LAST . We found that over-expression of LAST containing only one site mutation led to an increase in the CCND1 mRNA level , whereas over-expression of LAST containing four G-rich site mutations nullified its effect on the CCND1 mRNA level ( Figure 4F and G ) . This result indicates that the effect of LAST on CCND1 stability requires at least one of the four functional G-rich motifs . To define which domain of CNBP is responsible for binding LAST and the CCND1 5’UTR , a biotin-labeled RNA pull-down assay and deletion mapping were performed . According to the web site InterPro ( Hunter et al . , 2009 ) , CNBP can be divided into four structural domains based on its zinc-finger arrangement ( Figure 4—figure supplement 2C ) . As shown in Figure 4—figure supplement 2C , we concluded that LAST binds to the CNBP fragment corresponding to amino acids 92–134 ( domain 3 ) , whereas the CCND1 5’UTR binds to the CNBP fragment corresponding to amino acids 29–134 ( domain 2 + domain 3 ) ( Figure 4—figure supplement 2C ) . Determination of the exact mechanism of these associations requires further investigation . To globally identify transcripts that simultaneously meet the following requirements: ( i ) transcripts are downregulated by LAST knockdown and ( ii ) transcripts are able to bind to CNBP , we assembled two unbiased transcriptome profiles using LAST knockdown mRNA-seq and CNBP RIP-seq in HCT116 cells . The intersection of these two arrays is shown in Figure 5A , and 225 overlapping genes were found ( Supplementary file 2 ) . We further narrowed this list down to 75 genes ( Supplementary file 2 , bold part ) based on the criteria that CNBP-enriched genes must be 4-fold above the input control level . Three mRNAs , namely , SOX9 , NFE2L1 and PDF , were also likely to be regulated by LAST , as knockdown of LAST led to a decrease in their levels ( Figure 5B ) . Experimental verification showed that knockdown of LAST decreased ( Figure 5C ) whereas over-expression of LAST increased their half-lives ( Figure 5D ) . In addition , CNBP deletion led to a decrease in the mRNA levels of SOX9 , NFE2L1 and PDF ( Figure 5E and F ) . These data suggest that LAST , together with CNBP , can regulate the stabilization of additional mRNAs , such as SOX9 , NFE2L1 and PDF . To further determine whether LAST regulates tumorigenesis , we used a xenograft mouse model . HCT116 cells stably expressing exogenous LAST or LAST shRNA-1 were injected subcutaneously into the dorsal flanks of nude mice ( left ( control ) and right ( treated ) , n = 7 for each group ) . According to animal care and enforcement , mice were sacrificed when the largest subcutaneous tumor mass on one flank was close to one cubic centimeter . Tumors expressing control shRNA or LAST shRNA-1 were excised after 6 weeks , and tumors expressing control RNA or LAST were excised after 3 weeks . Mice were sacrificed and tumors were excised . Knockdown of LAST decreased the tumorigenicity of HCT116 cells ( Figure 6A and B ) . By contrast , induction of LAST promoted HCT116 cell tumorigenicity ( Figure 6C and D ) . Furthermore , based on the TCGA dataset ( Weinstein et al . , 2013 ) , we found that the LAST expression levels were higher in tumor tissues than normal tissues , including the human bladder , breast , colorectal , esophagus , head and neck , kidney , liver , lung , prostate and stomach . In addition , the CCND1 expression levels were higher in tumor tissues than in normal tissues , including the human bladder , breast , cervix , bile duct , colorectal , esophagus , head and neck , kidney , pancreas , stomach and uterus . In conclusion , both the LAST and CCND1 expression levels were higher in most tumor tissues than in their normal counterparts ( Figure 6E and F , Figure 6—figure supplements 1 and 2 ) . The above results suggest that LAST promotes tumorigenesis . To assess the impact of LAST deficiency on gene expression in HTC116 , we performed unbiased transcriptome profiling using RNA-seq in HCT116 cells . The absence of LAST downregulated expression of 667 genes ( log2 ( fold change ) below - 0 . 58 ) ( Supplementary file 3 ) . We then performed pathway analysis in those genes and found the top 10 significant pathways that were significantly associated with 667 differentially expressed genes . Among these 10 pathways , the majority were associated with tumorigenesis ( Figure 6G ) . Cyclin D1 is a critical regulator of CDK kinase , which regulates cell cycle progression at the G1 to S-phase transition . Pre- or mature CCND1 mRNA is regulated at different hierarchical levels bymultiple protein factors . Multiple classical transcriptional factors , such as c-Myc , E2F1 , OCT1，RELA and c-Jun , have been reported to modulate CCND1 at the transcriptional level ( Guo et al . , 2011 ) . Epigenetic and post-transcriptional mechanisms are also involved in the regulation of cyclin D1/CCND1 ( 16 , 29 , 30 ) . Moreover , Pitx2 and HuR , which belong to the same ribonucleoprotein complex , also control the decay rate of CCND1 mRNA ( Gherzi et al . , 2010 ) . However , whether lncRNA ( s ) is ( are ) involved in the regulation of CCND1 mRNA stability remains largely unaddressed . Very recently , NcRNACCND1 was reported to negatively regulate CCND1 transcription by recruiting TLS to the CCND1 promoter ( Wang et al . , 2008 ) . In this study , we characterized an overlooked mechanism of CCND1 mRNA regulation . c-Myc induced-LAST cooperates with CNBP , by which LAST is guided to the 5’ untranslated region of CCND1 messenger RNA and thus stabilizes CCND1 mRNA ( Figure 6H ) . The detailed mechanism underlying this 5’end protection requires further characterization . Normal growth control depends on the architecture of precise cell cycle control , and disturbing any component of this network could result in neoplastic growth and tumorigenesis . The G1/S transition is a major checkpoint in cell cycle progression , as it is a ‘point of no return’ beyond which cells are committed to dividing . Cyclin D1 , along with its catalytically active partner CDK4 , is a positive cell cycle regulator that advances the cell cycle from G1 to S phase ( McKay et al . , 2002 ) . Instead of protein factors , in this study , we found a novel long noncoding RNA , LAST , that ensures normal cell cycle progression . Lacking this lncRNA causes cell cycle arrest at the G1/S stage due to decreased cyclin D1 and attenuates tumor growth . Both the LAST and CCND1 expression levels are higher in most tumor tissues than in their corresponding normal tissues ( Figure 6—figure supplement 1A ) . Conceivably , LAST could be a potential target for new cancer therapeutics . However , a correlation between the expression levels of CCND1 and LAST in the 15 tumor types examined was not found ( Supplementary file 4 ) . In addition , there was no difference in survival when tumors were divided into those expressing high versus low CCND1 or LAST . These results imply that the regulation of CCND1 is more complicated than we had anticipated , and new functions of LAST need to be characterized . CNBP encodes a nucleic-acid binding protein that has seven zinc-finger domains and a preference for binding single-stranded DNA and RNA ( Flink and Morkin , 1995 ) . Previous studies have shown that CNBP acts on cap-independent translation of ornithine decarboxylase mRNA ( Sammons et al . , 2010 ) and also functions in sterol-mediated transcriptional regulation as well as c-Myc transcription ( Rajavashisth et al . , 1989; Murphy et al . , 2009 ) . In this study , we found that CNBP possesses a new function . CNBP is able to guide lncRNA to bind to the 5’UTR of CCND1 mRNA , acting as a mediator between LAST and CCND1 mRNA . lncRNAs are able to regulate their genomic neighborhoods in cis ( Quinn and Chang , 2016 ) . Examples of cis-acting lncRNAs include enhancer RNAs ( eRNAs ) ( De Santa et al . , 2010 ) , imprinted lncRNAs ( Mancini-Dinardo et al . , 2006; Sleutels et al . , 2002 ) and dosage compensation lncRNAs ( Lee , 2012; Conrad and Akhtar , 2012 ) . Homo sapiens cyclin D1/CCND1 and LAST are both located on chromosome 11 , and the two genes are 1 . 8 Mb apart and in the same transcriptional direction ( +strand ) . It is interesting to note that although CCND1 and LAST are both subjected to positive transcriptional regulation by c-Myc , they do not share the same promoter . Rather , CCND1 and LAST are transcribed separately by c-Myc via their respective promoters ( Figures 1K and 6H ) . The relatively long distance between the CCND1 and LAST genes may preclude their direct interaction . Moreover , LAST was shown to borrow a trans-acting factor , CNBP , as a mediator to connect the 5’UTR of CCND1 mRNA and itself , thus affecting CCND1/cyclin D1 expression in trans . Without CNBP , LAST shows no effect on CCND1 ( Figure 3—figure supplement 1F ) , further supporting the concept that LAST does not regulate CCND1 in cis . Therefore , co-location of CCND1 and LAST on the same chromosome appears to be a random event . In summary , our findings from this investigation have uncovered a novel , c-Myc-induced , long non-coding RNA , LAST . The LAST gene is encoded physically on the same chromosome as CCND1 . Normally , LAST interacts with CNBP , a RNA binding protein , by which it is guided towards the 5’UTR of CCND1 mRNA , leading to the stabilization of CCND1 mRNA , which in turn ensures orderly cell cycle progression . In the case of LAST dysregulation , CCND1 mRNA becomes unstable , resulting in decreased cyclin D1 , inevitably causing cell cycle arrest and stoppage of cell division ( Figure 6H ) . This is a novel mechanism for CCND1 mRNA regulation . Based on the importance of cyclin D1 in proliferative control and its ability to promote oncogenic transformation , this finding provides new insight into the complexity of the regulatory network underlying the mechanistic regulation of cyclin D1/CCND1 . Moreover , this LAST/CNBP regulatory mode can be applied to other genes; three different mRNAs , SOX9 , NFE2L1 and PDF , were identified with half-lives that were prolonged by LAST/CNBP . The lack of similarity between human LAST and transcripts of Mus musculus also precludes using mouse c-Myc-driven tumor models to further clarify the significance of the LAST in c-Myc-mediated cell cycle regulation and tumor growth in vivo ( Figure 6I ) . The following antibodies were used for western blot analysis in this study: anti-c-Myc ( Cell Signaling Technology ) ; anti-GAPDH and anti-β-Actin ( CMC-TAG ) ; normal rabbit IgG , normal mouse IgG2a , anti-HUR , anti-cyclin D1 , anti-CNBP and anti-NADSYN1 ( Santa Cruz ) ; anti-FLAG ( Sigma-Aldrich ) ; anti-cyclin E1 , anti-CDK2 , and anti-CDK4 ( ImmunoWay Biotechnology Company ) ; anti-HNRNPK ( ABclonal ) ; anti-DHCR7 ( ABCAM ) . Anti-c-Myc used for ChIP assay was from Santa Cruz . Thymidine , Nocodazole , Mimosine , EGF , hydrocortisone , Cholera Toxin , insulin and Doxycycline was from Sigma-Aldrich . Actinomycin D was from Solarbio . Strepavidin beads for RNA pull-down assay was from Invitrogen . H1299 , HCT116 , IMR90 , 293T and HAFF cell lines were cultured in DMEM ( Dulbecco's modified Eagle's medium ) medium containing 10% fetal bovine serum . P493-6 and MCF7 cell lines were cultured in RPMI medium 1640 containing 10% fetal bovine serum . MCF10A cell line were cultured in DMEM/F12 medium containing 5% horse serum , 20 μg/mL EGF , 0 . 5 μg/mL hydrocortisone , 100 ng/mL Cholera Toxin and 10 μg/mL insulin . P493-6 cells carrying a c-Myc tet-off system were provided by professor Ping Gao . All other cell lines were purchased from the American Type Culture Collection ( ATCC , Manassas , VA , USA ) . All cells were tested by STR profiling ( GenePrint 10 System kit from Promega and AuthentiFiler PCR Amplification Kit from ThermoFisher ) to authenticate the identity . All cells were tested for mycoplasma contamination by Cell Culture Contamination Detection Kit ( ThermoFisher ) . Western blotting , Northern blotting and real-time RT-PCR were performed as described previously ( Zhang et al . , 2016 ) . HCT116 cells ( 1 × 103 ) expressing control shRNA , lncRNA-52 shRNA-1 , –2 , lncRNA-51 shRNA-1 or −2 were cultured in a six-well plate . Ten days later , cells were fixed , stained with crystal violet and photographed . The exact copy numbers of LAST transcripts per HAFF , IMR90 , MCF10A , HCT116 , MCF7 or H1299 cell were quantified by using quantitative real-time RT-PCR assay . In this assay , serially diluted RT-PCR products of LAST were used as templates to formulate standard curves , and the exact copies of LAST per cell were calculated accordingly . HCT116 cells were crosslinked with 1% formaldehyde for 10 min . The ChIP assay was performed by using anti-c-Myc antibody and the Pierce Agarose ChIP kit ( ThermoScientific , USA ) according to the manufacturer's instructions . Anti-Rabbit immunoglobulin G was used as a negative control . The bound DNA fragments were subjected to real-time PCR using the specific primers ( Table 1 ) . To determine the effect of c-Myc on LAST promoter , either p3xflag-Myc-CMV-24 or p3xflag-Myc-CMV-24-c-Myc was co-transfected into HCT116 cells together with individual pGL3 , pGL3-BS2 , pGL3-BS2M , pGL3-BS3 or pGL3-BS3M construct plus Renilla luciferase reporter plasmid . Twenty-four hours after transfection , firefly and Renilla luciferase activity were measured by a Dual-Luciferase Reporter Assay System ( Promega , Madison , WI , USA ) . The data are represented as mean ± SD of three independent experiments . HCT116 cells were infected with lentiviruses and screened by puromycin , followed by plating into 6 mm dishes . During the proliferative exponential phase ( 50% confluency ) , cells were fixed in 70% ethanol overnight . Cells were then stained with propidium iodide and analyzed by flow cytometry . To detect LAST , RNA FISH was carried out as previously described with in vitro transcribed antisense probes labeled by Nucleic Acid Labeling Kits ( Life technologies , USA ) with Alexa Fluor 488 ( Yin et al . , 2012 ) . The sequence of RNA probe was CGUCUUUUCAGGACACAAAGGCAUGCAGGUGCAUCAUCUCUCUCUAUUAACGGGUCAGCUGGUCGGCAUGGUCAGCUGGUCGGUGGUCUCUUAUUAGGAGAAAGUCACUGAAAUCAGUCUCUUGUCCAAUCACAGCUGCUAUGGCUGAUCGUUUAUGGAGGAUCCUCUUCGCCCCGGGACGUGAGCCCUAGGACCAAGAACUGUGUCUGUUUUGCUCCUUGCGGUGCACCGGCGCCUGGACAUACGCUCCAUCAAUGUGCGUCGCGAGCCGCUGAAGCCCCAUUUGCCGAGGGGGAAACUGAGGCACGAUG . The nuclei were counterstained with PI . HCT116 cells ( 1 × 107 ) were incubated with hypotonic buffer ( 25 mM Tris-HCl , PH 7 . 4 , 1 mM MgCl2 , 5 mM KCl ) on ice for 5 min . An equal volume of hypotonic buffer containing 1% NP-40 was then added , and each sample was left on ice for another 5 min . After centrifugation at 5000 g for 5 min , the supernatant was collected as the cytosolic fraction . The pellets were re-suspended in nucleus resuspension buffer ( 20 mM HEPES , PH 7 . 9 , 400 mM NaCl2 , 1 mM EDTA , 1 mM EGTA , 1 mM DTT , 1 mM PMSF ) , and incubated at 4°C for 30 min . Nuclear fraction was collected after removing insoluble membrane debris by centrifugation at 12000 g for 10 min . RNA immunoprecipitation ( RIP ) was performed as described previously ( Yang et al . , 2014 ) . 1 × 107 cells were lysed in RIP buffer supplemented with RNase A inhibitor and DNase I before centrifugation . Cell lysates were precleared with protein A/G beads ( Pierce ) before they were incubated with protein A/G beads coated with the indicated antibodies at 4°C for 3 hr . After extensive washing , the bead-bound immunocomplexes were eluted using elution buffer ( 50 mM Tris [pH 8 . 0] , 1% SDS , and 10 mM EDTA ) at 65°C for 10 min . To isolate protein-associated RNAs from the eluted immunocomplexes , samples were treated with proteinase K , and RNAs were extracted by phenol/chloroform . Purified RNAs were then subjected to RT-PCR analysis . RIP was performed as described previously ( Xiang et al . , 2014 ) . Briefly , two 10 cm2 dishes of HCT116 cells were washed three times with cold PBS and irradiated at 200 mJ/cm2 at 254 nm in HL-2000 HybriLinkerTM UV Crosslinker . Cells were collected and resuspended in 1 ml RIP buffer . Cells were then homogenized and followed by 3 rounds of sonication on ice . Cell lysates were precleared with protein A/G beads ( Pierce ) before they were incubated with protein A/G beads coated with the indicated antibodies at 4°C for 3 hr . After extensive washing , the bead-bound immunocomplexes were eluted using elution buffer ( 50 mM Tris [pH 8 . 0] , 1% SDS , and 10 mM EDTA ) at 65°C for 10 min . To isolate protein-associated RNAs from the eluted immunocomplexes , samples were treated with proteinase K , and RNAs were extracted by phenol/chloroform . The sequencing was performed and analyzed by KangChen Bio-tech , Shanghai , China . The sequencing data were deposited in the National Center for Biotechnology Information Gene Expression Omnibus database ( GSE106918 ) . Total RNA from HCT116 cells expressing either control shRNA or LAST shRNA-1 was extracted by phenol/chloroform . The mRNA-seq was performed and analyzed by KangChen Bio-tech , Shanghai , China . The sequencing data were deposited in the National Center for Biotechnology Information Gene Expression Omnibus database ( GSE106917 ) . The peak-calling tool MACS2 ( Zhang et al . , 2008 ) ( https://github . com/taoliu/MACS/ ) with default parameter settings was used to call enriched peaks with RIP . bed as input and input . bed as control . A PERL script was written to calculate the proportion of the peaks containing each of the five given motifs ( TGGAGNW , TGGAG , GGAGNW , GGAG and GGR ) in all the RIP peaks . To test the significance of G-rich motif enrichment , another PERL script was used to perform statistical simulations by generating 1000 random samples of DNA sequences with the same size and the same length distribution as that of the RIP peaks . For each given motif , the average proportion ( with standard deviation ) of motif-containing sequences in random samples was calculated . A U-test was performed for each G-rich motif to test the significance of the difference between the proportion of the motif-containing sequences in RIP peaks and that in random DNA samples . All processes were performed in the RNase-free conditions . For antisense oligomer affinity pull-down assay , sense or antisense biotin-labeled DNA oligomers corresponding to LAST or CCND1 mRNA ( 1 μM ) were incubated with lysates from HCT116 cells ( 1 × 107 ) or the cytosolic/nuclear extracts . One hour after incubation , streptavidin-coupled agarose beads ( Invitrogen ) were added to isolate the RNA-protein complex or RNA-RNA complex . For in vitro RNA pull-down assay , 5 μg in vitro-synthesized biotin-labeled RNA was incubated with lysates from HCT116 cells ( 1 × 107 ) for 3 hr . Streptavidin-coupled agarose beads ( Invitrogen ) were then added to the reaction mix to isolate the RNA-protein complex or RNA-RNA complex . Immunocomplexes were then analyzed by real-time RT-PCR or western blotting . The electrophoretic mobility shift assay ( EMSA ) was performed by using an EMSA/gel shift kit ( Beyotime , China ) . Flag-CNBP protein was purified from 293T cells expressing Flag-CNBP . The biotin-labeled RNA fragments ( as shown in Figure 4D and E ) in vitro transcribed by T7 Transcription Kit ( Epicentre , USA ) were used in EMSA . HCT116 cells expressing control RNA or LAST ( 3 × 106 ) were subcutaneously injected into the dorsal flank of 4-week-old male athymic nude mice ( Shanghai SLAC Laboratory Animal Co . Ltd . ) ( n = 7 mice per group ) . After 3 weeks , mice were sacrificed , and tumors were excised and weighed . HCT116 cells expressing control shRNA or LAST shRNA-1 ( 3 × 106 ) were subcutaneously injected into the dorsal flank of 4-week-old male athymic nude mice ( Shanghai SLAC Laboratory Animal Co . Ltd . ) ( n = 7 mice per group ) . After 6 weeks , mice were sacrificed , and tumors were excised and weighed . Mice were randomly assigned to different experimental groups . During testing the tumors' weight , the experimentalists were blinded to the information and shape of tumor tissue masses . Studies on animals were conducted with approval from the Animal Research Ethics Committee of the University of Science and Technology of China ( Permit Number: USTCACUC1701003 ) .	Cell division involves a series of steps in which the cell grows , duplicates its contents , and then divides into two . Together these steps are called the cell cycle , and the transition between each step must be controlled to make sure that events take place in the right order . Any loss of control can cause cells to divide in an unrestrained manner , which may lead to cancer . Proteins called cyclins control progression through the cell cycle . As such , these proteins need to be produced in the correct amounts and at the correct times . Transcription factors are proteins that switch genes on or off to help regulate how much protein is made from those genes . A transcription factor known as c-Myc regulates the expression of the genes that encode the cyclins . Among these genes , one called CCND1 is particularly important because it encodes a protein that controls a crucial transition in the cell cycle: it marks a ‘point of no return’ , beyond which cells are committed to dividing . When a transcription factor switches on a gene , the gene gets copied into a molecule of messenger RNA , which is then translated into protein . But , cells also contain genes that do not code for proteins . Transcription factors can bind to such non-coding genes , leading to the production of so-called long non-coding RNAs ( often abbreviated to lncRNAs ) . Many lncRNAs can affect the expression of other genes . Cao , Zhang et al . have now asked whether any lncRNAs regulate CCND1 in human cells . The analysis revealed that the transcription factor c-Myc promotes the expression of a previously unidentified lncRNA . Cao , Zhang et al . name this lncRNA LAST , which is officially short for LncRNA-assisted stabilization of transcripts , and show thatit makes the CCND1 messenger RNA more stable . In other words , it makes the messenger RNAs ‘last’ longer in the cell . This in turn , ensures that the cell cycle progresses in the correct manner , allowing cells to complete their division . In the absence of LAST , the CCND1 messenger RNA becomes unstable and as a result the cell cycle does not progress . Cao , Zhang et al . then explored the role of LAST in cancer cells . When human colon cancer cells that expressed LAST were implanted into mice , they formed tumors . Yet , reducing the expression of LAST in the colon cancer cells made the tumors grow slower . Future challenges will be to understand how LAST makes messenger RNAs stable and further explore its role in cancer . A better understanding of this molecule could reveal whether it can be used to help doctors diagnose or treat cancers .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "cancer", "biology" ]	2017	LAST, a c-Myc-inducible long noncoding RNA, cooperates with CNBP to promote CCND1 mRNA stability in human cells
Recent functional , proteomic and ribosome profiling studies in eukaryotes have concurrently demonstrated the translation of alternative open-reading frames ( altORFs ) in addition to annotated protein coding sequences ( CDSs ) . We show that a large number of small proteins could in fact be coded by these altORFs . The putative alternative proteins translated from altORFs have orthologs in many species and contain functional domains . Evolutionary analyses indicate that altORFs often show more extreme conservation patterns than their CDSs . Thousands of alternative proteins are detected in proteomic datasets by reanalysis using a database containing predicted alternative proteins . This is illustrated with specific examples , including altMiD51 , a 70 amino acid mitochondrial fission-promoting protein encoded in MiD51/Mief1/SMCR7L , a gene encoding an annotated protein promoting mitochondrial fission . Our results suggest that many genes are multicoding genes and code for a large protein and one or several small proteins . Current protein databases are cornerstones of modern biology but are based on a number of assumptions . In particular , a mature mRNA is predicted to contain a single CDS; yet , ribosomes can select more than one translation initiation site ( TIS ) ( Ingolia et al . , 2011; Lee et al . , 2012; Mouilleron et al . , 2016 ) on any single mRNA . Also , minimum size limits are imposed on the length of CDSs , resulting in many RNAs being mistakenly classified as non-coding ( ncRNAs ) ( Pauli et al . , 2014; Anderson et al . , 2015; Zanet et al . , 2015; Nelson et al . , 2016; Bazzini et al . , 2014; Ji et al . , 2015; Prabakaran et al . , 2014; Slavoff et al . , 2013 ) . As a result of these assumptions , the size and complexity of most eukaryotic proteomes have probably been greatly underestimated ( Andrews and Rothnagel , 2014; Landry et al . , 2015; Fields et al . , 2015; Saghatelian and Couso , 2015 ) . In particular , few small proteins ( defined as of 100 amino acids or less ) are annotated in current databases . The absence of annotation of small proteins is a major bottleneck in the study of their function and their roles in health and disease . This is further supported by classical and recent examples of small proteins of functional importance , for instance many critical regulatory molecules such as the F0 subunit of the F0F1-ATPsynthase ( Stock et al . , 1999 ) , the sarcoplasmic reticulum calcium ATPase regulator phospholamban ( Schmitt et al . , 2003 ) , and the key regulator of iron homeostasis hepcidin ( Nemeth et al . , 2004 ) . This limitation also impedes our understanding of the process of origin of new genes , which are thought to contribute to evolutionary innovations . Because these genes generally code for small proteins ( Carvunis et al . , 2012; Schlötterer , 2015; McLysaght and Hurst , 2016; Sabath et al . , 2012 ) , they are difficult to unambiguously detect by proteomics and in fact are impossible to detect if they are not included in proteomics databases . Functional annotation of ORFs encoding small proteins is particularly challenging since an unknown fraction of small ORFs may occur by chance in the transcriptome , generating a significant level of noise ( Landry et al . , 2015 ) . However , given that many small proteins have important functions and are ultimately one of the most important sources of functional novelty , it is time to address the challenge of their functional annotations ( Landry et al . , 2015 ) . We systematically reanalyzed several eukaryotic transcriptomes to annotate previously unannotated ORFs which we term alternative ORFs ( altORFs ) , and we annotated the corresponding hidden proteome . Here , altORFs are defined as potential protein-coding ORFs in ncRNAs , in UTRs or in different reading frames from annotated CDSs in mRNAs ( Figure 1a ) . For clarity , predicted proteins translated from altORFs are termed alternative proteins and proteins translated from annotated CDSs are termed reference proteins . Our goal was to provide functional annotations of alternative proteins by ( 1 ) analyzing relative patterns of evolutionary conservation between alternative and reference proteins and their corresponding coding sequences; ( 2 ) estimating the prevalence of alternative proteins both by bioinformatics analysis and by detection in large experimental datasets; ( 3 ) detecting functional signatures in alternative proteins; and ( 4 ) testing the function of some alternative proteins . We predicted a total of 539 , 134 altORFs compared to 68 , 264 annotated CDSs in the human transcriptome ( Figure 1b , Table 1 ) . Because identical ORFs can be present in different RNA isoforms transcribed from the same genomic locus , the number of unique altORFs and CDSs becomes 183 , 191 and 54 , 498 , respectively . AltORFs were also predicted in other organisms for comparison ( Table 1 ) . By convention , only reference proteins are annotated in current protein databases . As expected , altORFs are on average small , with a size ranging from 30 to 1480 codons . Accordingly , the median size of predicted human alternative proteins is 45 amino acids compared to 460 for reference proteins ( Figure 1c ) , and 92 . 96% of alternative proteins have less than 100 amino acids . Thus , the bulk of the translation products of altORFs would be small proteins . The majority of altORFs either overlap annotated CDSs in a different reading frame ( 35 . 98% ) or are located in 3’UTRs ( 40 . 09% ) ( Figure 1d ) . 9 . 83% of altORFs are located in repeat sequences ( Figure 1—figure supplement 1a ) , compared to 2 . 45% of CDSs . To assess whether observed altORFs could be attributable solely to random occurrence , due for instance to the base composition of the transcriptome , we estimated the expected number of altORFs generated in 100 shuffled human transcriptomes . Overall , we observed 62 , 307 more altORFs than would be expected from random occurrence alone ( Figure 1e; p<0 . 0001 ) . This analysis suggests that a large number are expected by chance alone but that at the same time , a large absolute number could potentially be maintained and be functional . The density of altORFs observed in the CDSs , 3’UTRs and ncRNAs ( Figure 1f ) was markedly higher than in the shuffled transcriptomes , suggesting that these are maintained at frequencies higher than expected by chance , again potentially due to their coding function . In contrast , the density of altORFs observed in 5’UTRs was much lower than in the shuffled transcriptomes , supporting recent claims that negative selection eliminates AUGs ( and thus the potential for the evolution of altORFs ) in these regions ( Iacono et al . , 2005; Neafsey and Galagan , 2007 ) . Although the majority of human annotated CDSs do not have a TIS with a Kozak motif ( Figure 1g ) ( Smith et al . , 2005 ) , there is a correlation between a Kozak motif and translation efficiency ( Pop et al . , 2014 ) . We find that 27 , 539 ( 15% of 183 , 191 ) human altORFs encoding predicted alternative proteins have a Kozak motif ( A/GNNAUGG ) , as compared to 19 , 745 ( 36% of 54 , 498 ) for annotated CDSs encoding reference proteins ( Figure 1g ) . The number of altORFs with Kozak motifs is significantly higher in the human transcriptome compared to shuffled transcriptomes ( Figure 1—figure supplement 2 ) , again supporting their potential role as protein coding . Next , we compared evolutionary conservation patterns of altORFs and CDSs . A large number of human alternative proteins have homologs in other species . In mammals , the number of homologous alternative proteins is higher than the number of homologous reference proteins ( Figure 2a ) , and nine are even conserved from human to yeast ( Figure 2b ) , supporting a potential functional role . As phylogenetic distance from human increases , the number and percentage of genes encoding homologous alternative proteins decreases more rapidly than the percentage of genes encoding reference proteins ( Figure 2a and c ) . This observation indicates either that altORFs evolve more rapidly than CDSs or that distant homologies are less likely to be detected given the smaller sizes of alternative proteins . Another possibility is that they evolve following the patterns of evolution of genes that evolve de novo , with a rapid birth and death rate , which accelerates their turnover over time ( Schlötterer , 2015 ) . If altORFs play a functional role , they would be expected to be under purifying selection . The first and second positions of a codon experience stronger purifying selection than the third because of redundancy in the genetic code ( Pollard et al . , 2010 ) . In the case of CDS regions overlapping altORFs with a shifted reading frame , the third codon positions of the CDSs are either the first or the second in the altORFs , and should thus also undergo purifying selection . We analyzed conservation of third codon positions of CDSs for 100 vertebrate species for 1088 altORFs completely nested within and conserved across vertebrates ( human to zebrafish ) with their 889 CDSs from 867 genes ( Figure 3 ) . We observed that in regions of the CDS overlapping altORFs , third codon positions were evolving at significantly more extreme speeds ( slow or quick ) than third codon positions of random control sequences from the entire CDS ( Figure 3 ) , reaching up to 67-fold for conservation at p<0 . 0001 and 124-fold for accelerated evolution at p<0 . 0001 . This is illustrated with three altORFs located within the CDS of NTNG1 , RET and VTI1A genes ( Figure 4 ) . These three genes encode a protein promoting neurite outgrowth , the proto-oncogene tyrosine-protein kinase receptor Ret and a protein mediating vesicle transport to the cell surface , respectively . Two of these alternative proteins have been detected by ribosome profiling ( RET , IP_182668 . 1 ) or mass spectrometry ( VTI1A , IP_188229 . 1 ) ( see Supplementary files 1 and 2 ) . We provide two lines of evidence indicating that thousands of altORFs are translated into proteins . First , we re-analyzed detected TISs in publicly available ribosome profiling data ( Michel et al . , 2014; Raj et al . , 2016 ) , and found 26 , 531 TISs mapping to annotated CDSs and 12 , 616 mapping to altORFs in these studies ( Figure 5a; Supplementary file 1 ) . Only a small fraction of TISs detected by ribosomal profiling mapped to altORFs3’ even if those are more abundant than altORF5’ relative to shuffled transcriptomes , likely reflecting a recently resolved technical issue which prevented TIS detection in 3’UTRs ( Miettinen and Björklund , 2015 ) . New methods to analyze ribosome profiling data are being developed and will likely uncover more translated altORFs ( Ji et al . , 2015 ) . In agreement with the presence of functional altORFs3’ , cap-independent translational sequences were recently discovered in human 3’UTRs ( Weingarten-Gabbay et al . , 2016 ) . Second , we re-analyzed proteomic data using our composite database containing alternative proteins in addition to annotated reference proteins ( Figure 5b; Supplementary file 2 ) . We selected four studies representing different experimental paradigms and proteomic applications: large-scale ( Hein et al . , 2015 ) and targeted ( Tong et al . , 2014 ) protein/protein interactions , post-translational modifications ( Sharma et al . , 2014 ) , and a combination of bottom-up , shotgun and interactome proteomics ( Rosenberger et al . , 2014 ) . In the first dataset , we detected 3957 predicted alternative proteins in the interactome of reference proteins ( Hein et al . , 2015 ) , providing a framework to uncover the function of these proteins . In a second proteomic dataset containing about 10 , 000 reference human proteins ( Rosenberger et al . , 2014 ) , a total of 549 predicted alternative proteins were detected . Using a phosphoproteomic large data set ( Sharma et al . , 2014 ) , we detected 384 alternative proteins . The biological function of these proteins is supported by the observation that some alternative proteins are specifically phosphorylated in cells stimulated by the epidermal growth factor , and others are specifically phosphorylated during mitosis ( Figure 6; Supplementary file 3 ) . We provide examples of spectra validation ( Figure 6—figure supplement 1 ) . A fourth proteomic dataset contained 77 alternative proteins in the epidermal growth factor receptor interactome ( Tong et al . , 2014 ) ( Figure 5b ) . A total of 4872 different alternative proteins were detected in these proteomic data . The majority of these proteins are coded by altORFCDS , but there are also significant contributions of altORF3’ , altORFnc and altORF5’ ( Figure 5c ) . Overall , by mining the proteomic and ribosomal profiling data , we detected the translation of a total of 17 , 371 unique alternative proteins . 467 of these alternative proteins were detected by both MS and ribosome profiling ( Figure 7 ) , providing a high-confidence collection of small alternative proteins for further studies . An important goal of this study is to associate potential functions to alternative proteins , which we can do through annotations . Because the sequence similarities and the presence of particular signatures ( families , domains , motifs , sites ) are a good indicator of a protein's function , we analyzed the sequence of the predicted alternative proteins in several organisms with InterProScan , an analysis and classification tool for characterizing unknown protein sequences by predicting the presence of combined protein signatures from most main domain databases ( Mitchell et al . , 2015 ) ( Figure 8; Figure 8—figure supplement 1 ) . We found 41 , 511 ( 23% ) human alternative proteins with at least one InterPro signature ( Figure 8b ) . Of these , 37 , 739 ( or 20 . 6% ) are classified as small proteins . Interestingly , the reference proteome has a smaller proportion ( 840 or 1 . 68% ) of small proteins with at least one InterPro signature , supporting a biological activity for alternative proteins . Similar to reference proteins , signatures linked to membrane proteins are abundant in the alternative proteome and represent more than 15 , 000 proteins ( Figure 8c–e; Figure 8—figure supplement 1 ) . With respect to the targeting of proteins to the secretory pathway or to cellular membranes , the main difference between the alternative and the reference proteomes lies in the very low number of proteins with both signal peptides and transmembrane domains . Most of the alternative proteins with a signal peptide do not have a transmembrane segment and are predicted to be secreted ( Figure 8c , d ) , supporting the presence of large numbers of alternative proteins in plasma ( Vanderperre et al . , 2013 ) . The majority of predicted alternative proteins with transmembrane domains have a single membrane spanning domain but some display up to 27 transmembrane regions , which is still within the range of reference proteins that show a maximum of 33 ( Figure 8e ) . We extended the functional annotation using the Gene Ontology . A total of 585 alternative proteins were assigned 419 different InterPro entries , and 343 of them were tentatively assigned 192 gene ontology terms ( Figure 9 ) . 15 . 5% ( 91/585 ) of alternative proteins with an InterPro entry were detected by MS or/and ribosome profiling , compared to 13 . 7% ( 22 , 055/161 , 110 ) for alternative proteins without an InterPro entry ( p-value=1 . 13e-05 , Fisher's exact test and chi-square test ) . Thus , predicted alternative proteins with InterPro entries are more likely to be detected , supporting their functional role . The most abundant class of predicted alternative proteins with at least one InterPro entry are C2H2 zinc finger proteins with 110 alternative proteins containing 187 C2H2-type/integrase DNA-binding domains , 91 C2H2 domains and 23 C2H2-like domains ( Figure 10a ) . Eighteen of these ( 17 . 8% ) were detected in public proteomic and ribosome profiling datasets ( Table 2 ) , a percentage that is similar to reference zinc finger proteins ( 20 . 1% ) . Alternative proteins have between 1 and 23 zinc finger domains ( Figure 10b ) . Zinc fingers mediate protein-DNA , protein-RNA and protein-protein interactions ( Wolfe et al . , 2000 ) . The linker sequence separating adjacent finger motifs matches or resembles the consensus TGEK sequence in nearly half the annotated zinc finger proteins ( Laity et al . , 2001 ) . This linker confers high-affinity DNA binding and switches from a flexible to a rigid conformation to stabilize DNA binding . The consensus TGEK linker is present 46 times in 31 alternative zinc finger proteins ( Supplementary file 4 ) . These analyses show that a number of alternative proteins can be classified into families and will help deciphering their functions . We compared the functional annotations of the 585 alternative proteins with an InterPro entry with the reference proteins expressed from the same genes . Strikingly , 89 of 110 altORFs coding for zinc finger proteins ( Figure 10 ) are present in transcripts in which the CDS also codes for a zinc finger protein . Overall , 138 alternative/reference protein pairs have at least one identical InterPro entry and many pairs have more than one identical entry ( Figure 11a ) . The number of identical entries was much higher than expected by chance ( Figure 11b , p<0 . 0001 ) . The correspondence between InterPro domains of alternative proteins and their corresponding reference proteins coded by the same genes also indicates that even when entries are not identical , the InterPro terms are functionally related ( Figure 11c; Figure 11—figure supplement 1 ) . The presence of identical domains remains significant ( p<0 . 001 ) even when the most frequent domains , zinc fingers , are not considered ( Figure 11—figure supplement 2 ) . The presence of identical domains within alternative/reference protein pairs encoded in the same genes may result from alternative splicing events which connect an altORF in a different reading frame than the CDS with a coding exon in the CDS reading frame . Such alternative proteins would likely be unannotated isoforms of the corresponding reference proteins . Thus , we examined whether there is any association between the % of overlap or identity and functional similarity within alternative/reference protein pairs . We performed blast searches of the 183 , 191 predicted alternative proteins against 54 , 498 reference proteins using BlastP . All altORFs with more than 80% identity and overlap had already been removed to generate our database ( as indicated in the Materials and methods ) . We found 100 ( 0 . 055% ) alternative proteins with 25% to 100% identity and 10% to 100% overlap with their reference protein pairs . Among them , 20 ( 0 . 00054% ) alternative proteins have identical InterPro signatures with their respective reference proteins ( Supplementary file 5 ) . The distribution of the percentage of sequence identity and overlap between alternative-reference protein pairs with ( w/ , n = 20 ) or without ( w/o , n = 80 ) identical Interpro signature is shown in Figure 11—figure supplement 3 . We observed no significant differences between the two groups ( p-value=0 . 6272; Kolmogorov Smirnov test ) . We conclude that there is no significant association between identity/overlap and the presence of identical domains in alternative/reference protein pairs . Recently , the interactome of 118 human zinc finger proteins was determined by affinity purification followed by mass spectrometry ( Schmitges et al . , 2016 ) . This study provides a unique opportunity to test if , in addition to possessing zinc finger domains , some pairs of reference and alternative proteins coded by the same gene may functionally interact . We re-analyzed the MS data using our alternative protein sequence database to detect alternative proteins in this interactome ( Supplementary file 6 ) . Five alternative proteins ( IP_168460 . 1 , IP_168527 . 1 , IP_270697 . 1 , IP_273983 . 1 , IP_279784 . 1 ) were identified within the interactome of their reference zinc finger proteins . This number was higher than expected by chance ( p<10−6 ) based on 1 million binomial simulations of randomized interactomes . These physical interactions within zinc finger alternative/reference protein pairs suggest that there are examples of functional relationships between large and small proteins coded by the same genes . Finally , we integrated the expression analyses and the conservation analyses of alternative/reference protein pairs to produce a high-confidence list of alternative proteins predicted to have a function and found 2715 alternative proteins in mammals ( H . sapiens to B . taurus ) , and 44 in vertebrates ( H . sapiens to D . rerio ) ( Supplementary file 7 ) . From this list , we focused on alternative proteins detected with at least two peptide spectrum matches or with high TIS reads and selected altMiD51 ( IP_294711 . 1 ) among the top 2% of alternative proteins detected with the highest number of unique peptides in proteomics studies , and altDDIT3 ( IP_211724 . 1 ) among the top 2% of altORFs with the most cumulative reads in translation initiation ribosome profiling studies . AltMiD51 is a 70 amino acid alternative protein conserved in vertebrates ( Andreev et al . , 2015 ) and conserved with its reference protein MiD51 from humans to zebrafish ( Supplementary file 7 ) . Its coding sequence is present in exon 2 of the MiD51/MIEF1/SMCR7L gene . This exon forms part of the 5’UTR for the canonical mRNA and is annotated as non-coding in current gene databases ( Figure 12a ) . Yet , altMiD51 is robustly detected by MS in several cell lines ( Supplementary file 2: HEK293 , HeLa Kyoto , HeLa S3 , THP1 cells and gut tissue ) , and we validated some spectra using synthetic peptides ( Figure 12—figure supplement 1 ) , and it is also detected by ribosome profiling ( Supplementary file 1 ) ( Vanderperre et al . , 2013; Andreev et al . , 2015; Kim et al . , 2014 ) . We confirmed co-expression of altMiD51 and MiD51 from the same transcript ( Figure 12b ) . Importantly , the tripeptide LYR motif predicted with InterProScan and located in the N-terminal domain of altMiD51 ( Figure 12a ) is a signature of mitochondrial proteins localized in the mitochondrial matrix ( Angerer , 2015 ) . Since MiD51/MIEF1/SMCR7L encodes the mitochondrial protein MiD51 , which promotes mitochondrial fission by recruiting cytosolic Drp1 , a member of the dynamin family of large GTPases , to mitochondria ( Losón et al . , 2013 ) , we tested for a possible functional connection between these two proteins expressed from the same mRNA . We first confirmed that MiD51 induces mitochondrial fission ( Figure 12—figure supplement 2 ) . Remarkably , we found that altMiD51 also localizes at the mitochondria ( Figure 12c; Figure 12—figure supplement 3 ) and that its overexpression results in mitochondrial fission ( Figure 12d ) . This activity is unlikely to be through perturbation of oxidative phosphorylation since the overexpression of altMiD51 did not change oxygen consumption nor ATP and reactive oxygen species production ( Figure 12—figure supplement 4 ) . The decrease in spare respiratory capacity in altMiD51-expressing cells ( Figure 12—figure supplement 4a ) likely resulted from mitochondrial fission ( Motori et al . , 2013 ) . The LYR domain is essential for altMiD51-induced mitochondrial fission since a mutant of the LYR domain , altMiD51 ( LYR→AAA ) was unable to convert the mitochondrial morphology from tubular to fragmented ( Figure 12d ) . Drp1 ( K38A ) , a dominant negative mutant of Drp1 ( Smirnova et al . , 1998 ) , largely prevented the ability of altMiD51 to induce mitochondrial fragmentation ( Figure 12d; Figure 12—figure supplement 5a ) . In a control experiment , co-expression of wild-type Drp1 and altMiD51 proteins resulted in mitochondrial fragmentation ( Figure 12—figure supplement 5b ) . Expression of the different constructs used in these experiments was verified by western blot ( Figure 12—figure supplement 6 ) . Drp1 knockdown interfered with altMiD51-induced mitochondrial fragmentation ( Figure 13 ) , confirming the proposition that Drp1 mediates altMiD51-induced mitochondrial fragmentation . It remains possible that altMiD51 promotes mitochondrial fission independently of Drp1 and is able to reverse the hyperfusion induced by Drp1 inactivation . However , Drp1 is the key player mediating mitochondrial fission and most likely mediates altMiD51-induced mitochondrial fragmentation , as indicated by our results . AltDDIT3 is a 31 amino acid alternative protein conserved in vertebrates and conserved with its reference protein DDIT3 from human to bovine ( Supplementary file 7 ) . Its coding sequence overlaps the end of exon 1 and the beginning of exon 2 of the DDIT3/CHOP/GADD153 gene . These exons form part of the 5’UTR for the canonical mRNA ( Figure 14a ) . To determine the cellular localization of altDDIT3 and its possible relationship with DDIT3 , confocal microscopy analyses were performed on HeLa cells co-transfected with altDDIT3GFP and DDIT3mCherry . Expression of these constructs was verified by western blot ( Figure 14—figure supplement 1 ) . Interestingly , both proteins were mainly localized in the nucleus and partially localized in the cytoplasm ( Figure 14b ) . This distribution for DDIT3 confirms previous studies ( Cui et al . , 2000; Chiribau et al . , 2010 ) . Both proteins seemed to co-localize in these two compartments ( Pearson correlation coefficient 0 . 92 , Figure 14c ) . We further confirmed the statistical significance of this colocalization by applying Costes’ automatic threshold and Costes’ randomization colocalization analysis and Manders Correlation Coefficient ( Figure 14d; Figure 14—figure supplement 2 ) ( Bolte and Cordelières , 2006 ) . Finally , in lysates from cells co-expressing altDDIT3GFP and DDIT3mCherry , DDIT3mCherry was immunoprecipitated with GFP-trap agarose , confirming an interaction between the small altDDTI3 and the large DDIT3 proteins encoded in the same gene ( Figure 14e ) . We have provided the first functional annotation of altORFs with a minimum size of 30 codons in different genomes . The comprehensive annotation of H . sapiens altORFs is freely available to download at https://www . roucoulab . com/p/downloads ( Homo sapiens functional annotation of alternative proteins based on RefSeq GRCh38 ( hg38 ) predictions ) . In light of the increasing evidence from approaches such as ribosome profiling and MS-based proteomics that the one mRNA-one canonical CDS assumption is untenable , our findings provide the first clear functional insight into a new layer of regulation in genome function . While many observed altORFs may be evolutionary accidents with no functional role , several independent lines of evidence support translation and a functional role for thousands of alternative proteins: ( 1 ) overrepresentation of altORFs relative to shuffled sequences; ( 2 ) overrepresentation of altORF Kozak sequences; ( 3 ) active altORF translation detected via ribosomal profiling; ( 4 ) detection of thousands of alternative proteins in multiple existing proteomic datasets; ( 5 ) correlated altORF-CDS conservation , but with overrepresentation of highly conserved and fast-evolving altORFs; ( 6 ) overrepresentation of identical InterPro signatures between alternative and reference proteins encoded in the same mRNAs; and ( 7 ) presence of clear , striking examples in altMiD51 , altDDIT3 and 5 alternative proteins interacting with their reference zinc finger proteins . While far from proven in our study , three of these lines of evidence ( 5 , 6 , and 7 ) would support the intriguing hypothesis that many altORFs code for proteins that cooperate functionally with the proteins coded by their CDSs . This hypothesis would also agree with recently increasing evidence that small proteins often regulate the function of larger proteins ( Couso and Patraquim , 2017 ) . Further experimental examples and more detailed co-conservation studies will be needed to address this hypothesis . The presence of two or more coding sequences in the same gene provides a mechanism for coordinated transcriptional regulation . Consequently , the transcription of these coding sequences can be turned on or off together , similar to prokaryotic operons . We speculate that having genes composed of a CDS and one or more altORFs gives rise to fewer , denser transcription units and allows cells to adapt more quickly with optimized energy expenditure to environmental changes . Upstream ORFs , here labeled altORFs5’ , are important translational regulators of canonical CDSs in vertebrates ( Johnstone et al . , 2016 ) . Interestingly , the altORF5’ encoding altDDIT3 was characterized as an inhibitory upstream ORF ( Jousse et al . , 2001; Young et al . , 2016 ) , but evidence of endogenous expression of the corresponding small protein was not sought . The detection of altMiD51 and altDDIT3 suggests that a fraction of altORFs5’ may have dual functions as translation regulators and functional proteins . Our results raise the question of the evolutionary origins of these altORFs . A first possible mechanism involves the polymorphism of initiation and stop codons during evolution ( Lee and Reinhardt , 2012; Andreatta et al . , 2015 ) . For instance , the generation of an early stop codon in the 5’end of a CDS could be followed by the evolution of another translation initiation site downstream , creating a new independent ORF in the 3’UTR of the canonical gene . This mechanism of altORF origin , reminiscent of gene fission , would at the same time produce a new altORF that has identical protein domains with the annotated CDS , as we observed for a substantial fraction ( 24% ) of the 585 alternative proteins with an InterPro entry . A second mechanism would be de novo origin of ORFs , which would follow the well-established models of gene evolution de novo ( Schlötterer , 2015; Knowles and McLysaght , 2009; Neme and Tautz , 2013 ) in which new ORFs are transcribed and translated and have new functions or await the evolution of new functions by mutations . The numerous altORFs with no detectable protein domains may have originated this way from previously non-coding regions or in regions that completely overlap with CDS in other reading frames . Detection is an important challenge in the study of small proteins . A TIS detected by ribosome profiling does not necessarily imply that the protein is expressed as a stable molecule , and proteomic analyses more readily detect large proteins that generate several peptides after enzymatic digestion . In addition , evolutionarily novel genes tend to be poorly expressed , again reducing the probability of detection ( Schlötterer , 2015 ) . Here , we used a combination of five search engines , thus increasing the confidence and sensitivity of hits compared to single-search-engine processing ( Vaudel et al . , 2015; Shteynberg et al . , 2011 ) . This strategy led to the detection of several thousand alternative proteins . However , ribosome profiling and MS have technical caveats and the comprehensive contribution of small proteins to the proteome will require more efforts , including the development of new tools such as specific antibodies . Only a relatively small percentage of alternative proteins ( 22 . 6% ) are functionally annotated with Interpro signatures , compared to reference proteins ( 96 . 9% ) . An obvious explanation is the small size of alternative proteins with a median size of 45 amino acids , which may not be able to accommodate large domains . It has been proposed that small proteins may be precursors of new proteins but require an elongation of their coding sequence before they display a useful cellular activity ( Carvunis et al . , 2012; Couso and Patraquim , 2017 ) . According to this hypothesis , it is possible that protein domains appear only after elongation of the coding sequence . Alternatively , InterPro domains were identified by investigating the reference proteome , and alternative proteins may have new domains and motifs that remain to be characterized . Finally , an unknown fraction of predicted altORFs may not be translated or may code for non-functional peptides . In conclusion , our deep annotation of the transcriptome reveals that a large number of small eukaryotic proteins , possibly even the majority , are still not officially annotated . Our results with altMiD51 , altDDIT3 , and some zinc-finger proteins also suggest that some small and large proteins coded by the same mRNA may cooperate by regulating each other’s function or by functioning in the same pathway , confirming the few examples in the literature of unrelated proteins encoded in the same genes and functionally cooperating ( Table 3 ) ( Quelle et al . , 1995; Abramowitz et al . , 2004; Bergeron et al . , 2013; Lee et al . , 2014; Yosten et al . , 2016 ) . To determine whether or not this functional cooperation is a general feature of small/large protein pairs encoded in the same gene will require more experimental evidence . Throughout this manuscript , annotated protein coding sequences and proteins in current databases are labeled annotated coding sequences or CDSs and reference proteins , respectively . For simplicity reasons , predicted alternative protein coding sequences are labeled alternative open-reading frames or altORFs . To generate MySQL databases containing the sequences of all predicted alternative proteins translated from reference annotation of different organisms , a computational pipeline of Perl scripts was developed as previously described with some modifications ( Vanderperre et al . , 2013 ) . Genome annotations for H . sapiens ( release hg38 , Assembly: GCF_000001405 . 26 ) , P . troglodytes ( Pan_troglodytes-2 . 1 . 4 , Assembly: GCF_000001515 . 6 ) , M . musculus ( GRCm38 . p2 , Assembly: GCF_000001635 . 22 ) , D . melanogaster ( release 6 , Assembly: GCA_000705575 . 1 ) , C . elegans ( WBcel235 , Assembly: GCF_000002985 . 6 ) and S . cerevisiae ( Sc_YJM993_v1 , Assembly: GCA_000662435 . 1 ) were downloaded from the NCBI website ( http://www . ncbi . nlm . nih . gov/genome ) . For B . taurus ( release UMD 3 . 1 . 86 ) , X . tropicalis ( release JGI_4 . 2 ) and D . rerio ( GRCz10 . 84 ) , genome annotations were downloaded from Ensembl ( http://www . ensembl . org/info/data/ftp/ ) . Each annotated transcript was translated in silico with Transeq ( Rice et al . , 2000 ) . All ORFs starting with an AUG and ending with a stop codon different from the CDS , with a minimum length of 30 codons ( including the stop codon ) and identified in a distinct reading frame compared to the annotated CDS when overlapping the CDS , were defined as altORFs . An additional quality control step was performed to remove initially predicted altORFs with a high level of identity with reference proteins . Such altORFs typically start in a different coding frame than the reference protein but through alternative splicing , end with the same amino acid sequence as their associated reference protein . Using BLAST , altORFs overlapping CDSs chromosomal coordinates and showing more than 80% identity and overlap with an annotated CDS were rejected . AltORF localization was assigned according to the position of the predicted translation initiation site ( TIS ) : altORFs5’ , altORFsCDS and altORFs3’ are altORFs with TISs located in 5’UTRs , CDSs and 3’UTRs , respectively . Non-coding RNAs ( ncRNAs ) have no annotated CDS and all ORFs located within ncRNAs are labeled altORFsnc . The presence of the simplified Kozak sequence ( A/GNNATGG ) known to be favorable for efficient translation initiation was also assessed for each predicted altORF ( Kozak , 2002 ) . The global aggregates of initiating ribosome profiles data were obtained from the initiating ribosome tracks in the GWIPS-viz genome browser ( Michel et al . , 2014 ) with ribosome profiling data collected from five large-scale studies ( Lee et al . , 2012; Ji et al . , 2015; Fritsch et al . , 2012; Stern-Ginossar et al . , 2012; Gao et al . , 2015 ) . Sites were mapped to hg38 using a chain file from the UCSC genome browser ( http://hgdownload . soe . ucsc . edu/goldenPath/hg19/liftOver/hg19ToHg38 . over . chain . gz ) and CrossMap v0 . 1 . 6 ( RRID:SCR_001173 ) . Similar to the methods used in these studies , an altORF is considered as having an active TIS if it is associated with at least ten reads at one of the seven nucleotide positions of the sequence NNNAUGN ( AUG is the predicted altORF TIS ) . An additional recent study was also included in our analysis ( Raj et al . , 2016 ) . In this study , a threshold of 5 reads was used . Raw sequencing data for ribosome protected fragments in harringtonine treated cells were aligned to the human genome ( GRCh38 ) using bowtie2 ( 2 . 2 . 8 ) ( Langmead and Salzberg , 2012 ) . Similar to the method used in this work , altORFs with at least five reads overlapping one position in the kozak region were considered as having an experimentally validated TIS . Each annotated transcript was shuffled using the Fisher-Yates shuffle algorithm . In CDS regions , all codons were shuffled except the initiation and stop codons . For mRNAs , we shuffled the 5’UTRs , CDSs and 3’UTRs independently to control for base composition . Non-coding regions were shuffled at the nucleotide level . The resulting shuffled transcriptome has the following features compared to hg38: same number of transcripts , same transcripts lengths , same nucleotide composition , and same amino-acid composition for the proteins translated from the CDSs . Shuffling was repeated 100 times and the results are presented with average values and standard deviations . The total number of altORFs is 539 , 134 for hg38 , and an average of 489 , 073 for shuffled hg38 . AltORFs and kozak motifs in the 100 shuffled transcriptomes were detected as described above for hg38 . Both alternative and reference proteomes were investigated . Pairwise ortholog and paralog relationships between the human proteomes and the proteomes from other species , were calculated using an InParanoid-like approach ( Sonnhammer and Östlund , 2015 ) , as described below ( RRID:SCR_006801 ) . The following BLAST ( RRID:SCR_001010 ) procedure was used . Comparisons using our datasets of altORFs/CDS protein sequences in multiple FASTA formats from Saccharomyces cerevisiae , Caenorhabditis elegans , Drosophila melanogaster , Danio rerio , Xenopus tropicalis Bos taurus , Mus musculus , Pan troglodytes , Homo sapiens were performed between each pair of species ( Homo sapiens against the other species ) , involving four whole proteome runs per species pair: pairwise comparisons ( organism A vs organism B , organism B vs organism A ) , plus two self-self runs ( organism A vs organism A , organism B vs organism B ) . BLAST homology inference was accepted when the length of the aligned region between the query and the match sequence equalled or exceeded 50% of the length of the sequence , and when the bitscore reached a minimum of 40 ( Remm et al . , 2001 ) . Orthologs were detected by finding the mutually best scoring pairwise hits ( reciprocal best hits ) between datasets A-B and B-A . The self-self runs were used to identify paralogy relationships as described ( Sonnhammer and Östlund , 2015 ) . Basewise conservation scores for the alignment of 100 vertebrate genomes including H . sapiens were obtained from UCSC genome browser ( http://hgdownload . soe . ucsc . edu/goldenPath/hg38/phyloP100way/ ) ( RRID:SCR_012479 ) . Conservation PhyloP scores relative to each nucleotide position within codons were extracted using a custom Perl script and the Bio-BigFile module version 1 . 07 ( see code file ) . The PhyloP conservation score for the wobble nucleotide of each codon within the CDS was extracted . For the 53 , 862 altORFs completely nested inside 20 , 814 CDSs , the average PhyloP score for wobble nucleotides within the altORF region was compared to the average score for the complete CDS . To generate controls , random regions in CDSs with a similar length distribution as altORFs were selected and PhyloP scores for wobble nucleotides were extracted . We compared the differences between altORF and CDS PhyloP scores ( altORF PhyloP – CDS PhyloP ) to those generated based on random regions . We identified expected quantiles of the differences ( ‘DQ’ column in the table ) , and compared these to the observed differences . Because there was greater conservation of wobble nucleotide PhyloP scores within altORFs regions located farther from the center of their respective genes ( r = 0 . 08 , p<0 . 0001 ) , observed differences were adjusted using an 8 knot cubic basis spline of percent distance from center . These observed differences were also adjusted for site-specific signals as detected in the controls .	Proteins are often referred to as the workhorses of the cell , and these molecules affect all aspects of human health and disease . Thus , deciphering the entire set of proteins made by an organism is often an important challenge for biologists . Genes contain the instructions to make a protein , but first they must be copied into a molecule called an mRNA . The part of the mRNA that actually codes for the protein is referred to as an open reading frame ( or ORF for short ) . For many years , most scientists assumed that , except for in bacteria , each mature mRNA in an organism has just a single functional ORF , and that this was generally the longest possible ORF within the mRNA . Many also assumed that RNAs copied from genes that had been labelled as “non-coding” or as “pseudogenes” did not contain functional ORFs . Yet , new ORFs encoding small proteins were recently discovered in RNAs ( or parts of RNA ) that had previously been annotated as non-coding . Working out what these small proteins actually do will require scientists being able to find more of these overlooked ORFs . The RNAs produced by many organisms – from humans and mice to fruit flies and yeast – have been catalogued and the data stored in publicly accessible databases . Samandi , Roy et al . have now taken a fresh look at the data for nine different organisms , and identified several thousand examples of possibly overlooked ORFs , which they refer to as “alternative ORFs” . This included more than 180 , 000 from humans . Further analysis of other datasets that captured details of the proteins actually produced in human cells uncovered thousands of small proteins encoded by the predicted alternative ORFs . Many of the so-called alternative proteins also resembled parts of other proteins that have a known activity or function . Lastly , Samandi , Roy et al . focused on two alternative proteins and showed that they both might affect the activity of the proteins coded within the main ORF in their respective genes . These findings reveal new details about the different proteins encoded within the genes of humans and other organisms , including that many mRNAs encode more that one protein . The implications and applications of this research could be far-reaching , and may help scientists to better understand how genes work in both health and disease .	[ "Abstract", "Introduction", "Results", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "computational", "and", "systems", "biology" ]	2017	Deep transcriptome annotation enables the discovery and functional characterization of cryptic small proteins
High-content phenotypic screening has become the approach of choice for drug discovery due to its ability to extract drug-specific multi-layered data . In the field of epigenetics , such screening methods have suffered from a lack of tools sensitive to selective epigenetic perturbations . Here we describe a novel approach , Microscopic Imaging of Epigenetic Landscapes ( MIEL ) , which captures the nuclear staining patterns of epigenetic marks and employs machine learning to accurately distinguish between such patterns . We validated the MIEL platform across multiple cells lines and using dose-response curves , to insure the fidelity and robustness of this approach for high content high throughput drug discovery . Focusing on noncytotoxic glioblastoma treatments , we demonstrated that MIEL can identify and classify epigenetically active drugs . Furthermore , we show MIEL was able to accurately rank candidate drugs by their ability to produce desired epigenetic alterations consistent with increased sensitivity to chemotherapeutic agents or with induction of glioblastoma differentiation . The epigenetic landscape of a cell is largely determined by the organization of its chromatin and the pattern of DNA and histone modifications . These confer differential accessibility to areas of the genome and through direct and in-direct regulation of all DNA-related processes , form the basis of the cellular phenotype ( Jenuwein and Allis , 2001; Lawrence et al . , 2016; Berger , 2007; Goldberg et al . , 2007 ) . By collecting global information about the epigenetic landscape , for example using ATAC- or histone ChIP-seq , we can derive multilayered information regarding cellular states ( Miyamoto et al . , 2018; Mikkelsen et al . , 2007 ) . These include stable cell phenotypes such as quiescence , senescence , or cell fate , as well as transient changes such as those induced by cytokines and chemical compounds . However , current methods for collecting such information are not adapted for high-content drug screening . Over the past decade the decreasing cost and remarkable scalability of high content screening have made it a particularly attractive alternative for drug discovery . More recently , novel image processing tools coupled with multiparametric analysis and machine learning have significantly impacted our ability to investigate and understand the output of phenotypic screens ( Kang et al . , 2016; Scheeder et al . , 2018 ) . Despite these advantages , such assays have not been adapted to extract and utilize information form the cellular epigenetic landscape . While malignant glioblastoma is the most common and lethal brain tumor , current therapeutic options offer little prognostic improvement , so the median survival time has remained virtually unchanged for decades ( Jhanwar-Uniyal et al . , 2015; Parvez , 2008; Burger and Green , 1987 ) . Tumor-propagating cells ( TPCs ) are a subpopulation of glioblastoma cells with increased tumorigenic capability ( Patel et al . , 2014 ) operationally defined as early-passaged ( <15 ) glioblastoma cells propagated in serum-free medium ( Nakano et al . , 2008 ) . Compared to the bulk of the tumor , TPCs are more resistant to drugs , such as temozolomide ( TMZ ) and radiation therapy ( Bao et al . , 2006; Safa et al . , 2015 ) . This resistance may explain the failure of traditional therapeutic strategies based on cytotoxic drugs targeting glioblastoma . Multiple approaches aimed at reducing or circumventing the resilience of TPCs have been proposed . These include targeting epigenetic enzymes ( i . e . , enzymes that write , remove , or read DNA and histone modifications ) to increase sensitivity to cytotoxic treatments ( Jones et al . , 2016; Strauss and Figg , 2016; Lee et al . , 2017; Romani et al . , 2018 ) ; and differentiating TPCs to reduce their tumorigenic potential ( von Wangenheim and Peterson , 1998; Von Wangenheim and Peterson , 2001; von Wangenheim and Peterson , 2008; Lee et al . , 2015; Song et al . , 2016; Garros-Regulez et al . , 2016 ) . Here , we introduce Microscopic Imaging of the Epigenetic Landscape ( MIEL ) , a novel high-content screening platform that profiles chromatin organization using the endogenous patterns of histone modifications present in all eukaryotic cells . We validate the platform across multiple cell lines and drug concentrations demonstrating its ability to classify epigenetically active compounds by molecular function , and its utility in identifying off-target drug effects . We show MIEL can accurately rank candidate drugs by their ability to produce a set of desired epigenetic alterations such as glioblastoma differentiation . We have developed a novel phenotypic screening platform , Microscopic Imaging of Epigenetic Landscape ( MIEL ) , which interrogates the epigenetic landscape at both population and single cell levels using image derived features and machine learning . MIEL takes advantage of epigenetic marks such as histone methylation and acetylation , which are always present in eukaryotic nuclei and can be revealed by immunostaining . MIEL analyzes the immunolabeling patterns of epigenetic marks using conventional image analysis methods for nuclei segmentation , feature extraction , and previously described machine-learning algorithms ( Collins et al . , 2015 ) ( Figure 1a and Materials and methods ) . Primarily , we utilized four histone modifications: H3K27me3 and H3K9me3 , which are associated with condensed ( closed ) facultative and constitutive heterochromatin , respectively; H3K27ac , associated with transcriptionally active ( open ) areas of chromatin , especially at promoter and enhancer regions; and H3K4me1 , associated with enhancers and other chromatin regions ( Figure 1a; Creyghton et al . , 2010; Shlyueva et al . , 2014 ) . To focus on the intrinsic pattern of epigenetic marks , we use only texture-associated features ( e . g . , Haralick's texture features [Haralick et al . , 1973] , threshold adjacency statistics , and radial features [Hamilton et al . , 2007] ) for multivariate analysis . Previous studies have successfully employed similar features for cell painting techniques combined with multiparametric analyses to accurately classify subcellular localization of proteins ( Hamilton et al . , 2007 ) , cellular subpopulations ( Loo et al . , 2009 ) , and drug mechanisms of action ( Collins et al . , 2015; Caie et al . , 2010; Gustafsdottir et al . , 2013; Loo et al . , 2007 ) . We employed three main methods of data visualization and analysis: ( 1 ) To visualize similarity between multiple cell populations across all acquired features we conducted multidimensional scaling ( MDS ) using the Euclidean distance between the multivariate centroids of all populations being compared and displayed the results as a 2D scatter plot ( termed distance map; Materials and methods and Figure 1a ) . ( 2 ) To classify multiple cell populations , we employed quadratic discriminant analysis of multivariate centroids ( DA; Materials and methods and Figure 1a ) . ( 3 ) Single cells within each cell populations were classified using a Support Vector Machine ( SVM; Materials and methods and Figure 1a ) . The most commonly used cellular assays for epigenetic drug discovery are lysis and ELISA based assays , such as AlphaLISA ( PerkinElmer ) . Imaging-based alternatives rely on staining for relevant histone modification and monitoring changes in average fluorescent intensity ( Sayegh et al . , 2013; Luense et al . , 2015 ) . Using MIEL , we screened a library of 222 epigenetically active compounds , many with known targets among epigenetic writers , erasers , or readers ( SBP epigenetic library , Figure 1—figure supplement 1a , b ) . Our analysis focused on MIEL’s ability to ( 1 ) detect active compounds; ( 2 ) group drugs by function and identify off-target effects; ( 3 ) be robust across cell lines and drug concentrations; ( 4 ) rank active drugs , and derive information regarding drug mechanism of action . To test the ability of the MIEL approach to detect active compounds and compare it to intensity-based methods , primary-derived TPCs ( GBM2 cell line ) were treated with epigenetically active drugs for 24 hr ( 10 µM , triplicates ) . Treated cells were immuno-labeled for multiple histone modifications expected to exhibit alterations following drug treatment ( H3K9me3 , H3K27me3 , H3K27ac and H3K4me1 ) . Image analysis including nuclei segmentation and features extraction was conducted , as previously described ( Collins et al . , 2015 ) on Acapella 2 . 6 ( PerkinElmer ) . Phenotypic profiles were generated for each compound or control ( DMSO ) treated wells . These are vectors composed of 1048 ( 262 features per modification X four modifications ) texture features derived from the staining of each histone modification and representing the average value for each feature across all stained cells in each cell population ( drug or DMSO ) . When treatment reduced cell count to under 50 imaged nuclei per well , the compound was deemed toxic and excluded from analysis . Following feature normalization by z-score , we calculated the Euclidean distance between vectors of the compounds and DMSO- treated cells . These distances were then normalized ( z-score ) to the average distance between DMSO replicates and the standard deviation of these distances . Compounds with a distance z-score of greater than three were defined as active ( see Materials and methods section ) . This analysis identified 122 compounds that induced significant epigenetic changes . Active compounds were not uniformly distributed across all functional drug categories . Rather , we identified 10 categories in which 50% of the drugs were categorized as active and nontoxic and 13 categories in which 25% or fewer of the drugs induced detectable epigenetic alterations following a 24 hr treatment ( Figure 1b ) . To compare MIEL with current thresholding methods , we repeated the calculation using mean fluorescence intensity for all histone modifications by normalizing ( z-score ) each drug against DMSO; active compounds were defined as compounds for which z-scored intensity for at least one of the histone modifications was greater than three or less than −3 . As a result , we identified 94 active compounds , which were distributed across functional categories similarly to MIEL-identified compounds ( Figure 1b ) . For each functional category , the number of compounds identified as active using thresholding was smaller than the number identified using MIEL ( Figure 1b ) demonstrating MIEL’s increased detection sensitivity over standard thresholding . To determine the contribution of individual histone modifications , we repeated both MIEL and thresholding analyses individually for each of the four modifications . Using MIEL-based analysis , a single modification yielded similar detection rates to the combination of modifications across most functional categories ( Figure 1—figure supplement 2a ) . Using intensity-based analysis , individual modifications yielded lower detection rates compared to the combination of modifications and displayed equal or reduced detection rates when compared to MIEL in all categories and modifications ( Figure 1—figure supplement 2a ) . Of note , 3 of the four modifications used for MIEL analysis showed similar detection rates across most of the functional categories . However , detection rates using H3K27me3 were consistently reduced across most active categories ( Figure 1—figure supplement 2a ) except for EZH1/2 inhibitors , possibly due to the role these enzymes play in regulating this posttranslational modification . To further compare MIEL and thresholding , we estimated the magnitude of epigenetic alterations induced by active compounds . We calculated the fold increase in distance from DMSO ( normalized to average distance between DMSO replicates ) as well as the fold change in fluorescence intensity for active compounds in each category . In all categories , MIEL showed an increased effect ( Figure 1—figure supplement 2b ) . These results demonstrate that , across all tested epigenetic marks , detecting epigenetically active compounds was markedly improved using the MIEL method compared to current image-based thresholding methods . One key advantage of phenotypic profiling methods like MIEL is the ability to classify compounds by function and identify nonspecific effects through comparison with profiles of well-defined controls . To assess whether MIEL could correctly group compounds by function , we applied discriminant analysis ( DA ) to all active , nontoxic compounds from categories that had at least three such compounds ( 85 compounds; seven categories and DMSO ) . Two replicates from each drug and 38 DMSO replicates were used as a training set for a quadratic DA , using all texture features derived from images of the four histone modifications . The third replicate for each compound , as well as 10 DMSO replicates , was used as a test set to validate the model . Results showed that MIEL separated multiple categories of epigenetically active drugs with an average accuracy of 91 . 4% ( Figure 1c , d ) . Although many of the epigenetically active compounds induced alterations in average fluorescence ( Figure 1—figure supplement 2b ) , a DA utilizing intensity measurements from all four channels was ineffective at separating the various categories and yielded only 51 . 6% average accuracy ( Figure 1—figure supplement 3a ) . To test whether individual histone modification textures contain sufficient information to distinguish between the various drug classes , we performed DA using features derived from each modification . Although this degraded MIEL’s ability to separate compound subclasses , which affected similar changes in histone modification such as Class I and Pan HDAC inhibitors , MIEL was still able to separate major categories , such as histone phosphorylation and deacetylation ( Figure 1—figure supplement 3b ) . DNA labeling dyes such as DAPI and Hoechst can partially recapitulate the staining pattern of H3K9me3 , which labels constitutive heterochromatin . To test the ability of DNA labeling dyes to capture information regarding chromatin organization and their usefulness for function based classification , we used texture features derived from the DAPI channel to repeat the functional classification . This yielded an overall classification accuracy of 65 . 6% ( Figure 1—figure supplement 3b ) compared with 86 . 4% provided by H3K9me3 ( Figure 1—figure supplement 3b ) . Despite reduced overall accuracy , it is evident that DAPI and other DNA dyes may be an informative and cheap alternative to histone staining in at least some applications when the analyzed epigenetic landscape are very distinct . Of note , the compound library used in this study included Pan HDACi , Class I HDACi , and Class I HDACi , known to also target HDAC6 . HDAC inhibitors targeting both Class I and HDAC six displayed the same profile as Pan HDAC , and DA showed the two categories to be undistinguishable . This was likely due to the high expression of HDAC Class I and HDAC six and low expression of other HDACs in the GBM2 glioblastoma line ( Figure 1—figure supplement 4a , b , c ) . Of the 85 compounds tested , 7 ( 8 . 2% ) were identified as active but were misclassified by MIEL . One of these was valproic acid , a commonly used anticonvulsant ( Peterson and Naunton , 2005 ) which also functions as a Pan HDAC inhibitor at high concentrations ( Phiel et al . , 2001 ) . Though valproic acid is expected to inhibit HDACs only at high concentrations ( >1 . 2 mM ) , a short 24 hr treatment induced detectable epigenetic changes even at low concentrations ( <30 µM ) . However , when we quantified H3K27ac and H3K27me3 immunofluorescence intensity at these concentrations , no increase in histone acetylation or decrease in histone methylation similar to other Pan HDAC inhibitors ( TSA , SAHA; Figure 1—figure supplement 5a ) was seen . To test , whether observed epigenetic changes resulted in corresponding transcriptomic alterations , we sequenced RNA from GBM2 cells treated with either DMSO , TSA , SAHA or valproic acid ( 15 µM ) for 24 hr and identified all genes altered by at least one of the drugs ( as compared to DMSO; 118 genes ) . This analysis indicated that the Pan HDAC inhibitors induced similar transcriptomic changes that were not apparent in the transcriptomic profile of valproic acid-treated cells ( Figure 1—figure supplement 5b ) . To test whether MIEL profiles reflected global drug-induced transcriptomic profiles , FPKM values for all expressed genes ( FPKM >1 in at least one cell population ) were used to calculate the Euclidean distance between all 4 cell populations . FPKM-based distances were then correlated to image texture feature-based distances , which yielded a high and significant correlation between these metrics ( R = 0 . 91 , pv <0 . 05; Figure 1—figure supplement 5c ) . Taken together , these demonstrate a unique ability of the MIEL approach to identify epigenetically active compounds , to accurately categorize them according to their molecular mechanism of action , and to detect off-target effects of compounds with known mechanism of action . As drugs vary in potency , predicting the function of unknown drugs relies on generating functional category-specific profiles that remain valid over a range of activity levels . To determine whether MIEL could correctly identify the functional category of drugs with different potencies , we treated GBM2 cells with drugs from several active categories at a range of concentrations ( 0 . 1 , 0 . 3 , 1 , 3 , 10 µM ) and conducted DA aimed at separating the different concentrations in each class . We found that for most drug categories tested ( inhibitors of: Aurora , JAK , SIRT and EZH1/2 ) , DA yielded low average classification accuracies ( Figure 2a - Aurora kinase: 43 . 3%; Figure 2—figure supplement 1a - EZH1/2: 62 . 5% , SIRT:4 6 . 2% , JAK: 37 . 2 ) , indicating similar MIEL profiles across tested drug concentrations . However , Pan HDAC and HDAC Class I inhibitors displayed progressive profile changes , allowing DA to separate the different concentrations at higher accuracy ( Figure 2a – HDAC Pan: 80 . 9%; Figure 2—figure supplement 1a - HDAC Class I: 82 . 2% ) . In addition to their on-target effect , drugs may induce epigenetic alterations through toxicity and stress . To estimate the impact of toxicity on drug induced profile changes and its contribution to drug misclassification across a range of concentrations , we plotted z-scored distance from DMSO ( effect size ) against z-scored nuclei count ( a proxy for cytotoxicity ) for GBM2 cells treated at a range of drug concentrations ( 0 . 1 , 0 . 3 , 1 , 3 , 10 µM ) . This demonstrated that some compound classes , such as Aurora and JAK inhibitors , induce epigenetic alterations only in concentrations at which cell count is significantly reduced , whether through toxicity or direct effect on proliferation ( Figure 2b – dark blue and pink respectively ) , while other compounds , such as HDAC inhibitors , characteristically have a concentration range where epigenetic alterations are not accompanied by reduced cell counts ( Figure 2b – green and yellow ) . Interestingly , both SIRT and EZH1/2 ( Figure 2b – light-blue and red , respectively ) inhibitors affected significant epigenetic changes without inducing significant changes in cell count . These results indicated the MIEL platform is ideally positioned to analyze dose-dependent effects from drug treatment . In particular , our data suggest that low ( 0 . 1 µM ) and high ( 10 µM ) concentration of HDAC inhibitors resulted in distinct and separable epigenetic landscapes , suggesting potentially distinct chromatin/gene expression profiles and divergent biological outcomes when using a low vs high concentration of such compounds . Testing candidate drugs in multiple cell lines can help gauge their inclusivity and identify tumor subtypes that do not respond to a specific drug or drug class . To test whether MIEL readouts were coherent across multiple glioblastoma TPCs , we treated 4 cell lines with a subset of drugs from the epigenetic library ( 57 drugs ) , derived phenotypic profiles , and calculated their effect size ( z-scored Euclidean distance from DMSO replicates ) . This revealed a significant positive correlation between all 4 cell lines pointing to similarities in their drug sensitivity profiles and demonstrating the robustness of the MIEL read out ( Figure 2c , d ) . To assess the ability of MIEL to group compounds by function across multiple cell lines we employed DA to classify DMSO and drug treated TPCs across these 4 GBM lines . This analysis enabled accurate separation of cells treated with drugs modulating distinct functions , such as EZH1/2 or SIRT inhibitors ( 5 and 3 compounds respectively; mean 100% accuracy; Figure 2e ) . However , we were unable to separate drug subclasses with similar functions , such as class I and pan HDACs inhibitors ( 6 and 17 compounds respectively; mean accuracy 76 . 8%; Figure 2e ) . These results demonstrate the ability of MIEL to correctly categorize by function drugs with varying degrees of potency across multiple cells lines . Finally , although individual drug activity correlated well across cells lines , the magnitude of the effect for some drug classes was highly correlated to the expression levels of the target gene . For example , SIRT inhibition was significantly more effective in lines showing reduced Sirt1 expression ( the main SIRT to deacetylate histone 3; n = 4 compounds , p<0 . 02; Figure 2—figure supplement 1b , c ) , and there was a significant inverse correlation between Sirt1 expression and the effect size ( R = −0 . 87; Figure 2—figure supplement 1c ) . These results further highlight the sensitivity of MIEL and its ability to reflect internal transcriptomic differences between cell populations . MIEL analysis indicated that the magnitude of drug induced profile changes , as measured by distance from DMSO replicates , varies between individual drugs within each drug class ( Figure 3—figure supplement 1a ) . To test whether these differences are biologically meaningful , we correlated MIEL-based activity readouts with the ability of epigenetic drugs to synergize with other treatments as these are often designed to work as part of a combination therapy ( Lee et al . , 2017; Romani et al . , 2018 ) . One common approach is to use epigenetic drugs to sensitize tumor cells to standard of care cytotoxic treatments ( Strauss and Figg , 2016; Zhou et al . , 2015; Li et al . , 2017; Entin-Meer et al . , 2007 ) , such as radiation and temozolomide ( TMZ ) , which are used to treat glioblastoma . To identify drug classes that sensitize glioblastoma TPCs to cytotoxic therapy , GBM2 cells were treated with epigenetic drugs for 2 days prior to radiation or TMZ . Cytotoxic treatment was carried out for 4 days at levels that induced a 50% reduction in cell numbers ( 1Gy radiation or 200 µM TMZ; Figure 3a ) . At the end of day six treatment , cells were counted , and a combined drug index ( CDI ) was calculated ( see Materials and methods ) . Though we did not identify any drugs that synergized ( CDI < 0 . 7 ) with the radiation therapy ( Figure 3b , right panel ) , multiple PARP and BET inhibitors ( PARPi and BETi ) sensitized cells to TMZ ( Figure 3b , left panel ) . PARPi have been extensively studied in this context and have been shown to function through multiple non-epigenetic mechanisms such as PARP trapping ( Murai et al . , 2012; Lord and Ashworth , 2017; Kedar et al . , 2012 ) . Consistent with this , most PARPi did not induce detectable epigenetic changes using MIEL ( Figure 3d , Figure 3—figure supplement 1b ) , and we found no correlation between the magnitude of epigenetic changes as measured by MIEL and CDI ( Figure 3d – bottom panel ) . To date , only a single report utilizing the BETi OTX015 ( Berenguer-Daizé et al . , 2016 ) has pointed to synergy with TMZ , prompting us to validate this finding in five additional glioblastoma lines . In three lines , BETi increased the TMZ effectiveness ( average CDI: 454M 0 . 76 ± 0 . 28 , PBT24 0 . 78 ± 0 . 12 and GBM2 0 . 51 ± 0 . 2; Mean ± SD; n = 11 BETi; Figure 3c ) . In the other three lines , the drugs did not synergize and , in many cases , were found to be protective against ( CDI > 1 ) TMZ ( average CDI: SK262 1 . 4 ± 0 . 26 , 101A 1 . 4 ± 0 . 22 and 217M 1 . 2 ± 0 . 21; Mean ± SD; n = 11 BETi; Figure 3c; p- values for all pairwise comparisons Figure 3c ) . We detected only few BETi-induced epigenetic changes in our initial screen conducted over 24 hr ( Figure 1b ) . However , following a 6 days treatment 6 out of 11 BETi induced significant ( average z-scored distance from DMSO replicates >3 ) epigenetic changes in all cell line tested ( Figure 3d , Figure 3—figure supplement 1b ) . In lines displaying TMZ and BETi synergy , the degree of BETi activity , as measured by MIEL , significantly correlated with the degree of synergism ( Figure 3d – top panel ) . This demonstrated that for individual compounds , MIEL can predict relative drug activity and suggests an epigenetic component for the mechanism of BETi-TMZ synergy . O6-alkylguanine DNA alkyltransferase ( MGMT ) , which provides the main line of defense against DNA alkylating agents such as TMZ , has been found to be epigenetically silenced through DNA methylation in a large fraction of glioblastoma tumors ( Karayan-Tapon et al . , 2010; Hegi et al . , 2005 ) . To gain a better understanding of the mechanism by which BETi sensitize glioblastoma TPCs to TMZ treatment , we quantified MGMT expression in the six lines tested using qPCR . This analysis showed that while all lines expressed similar BET transcription factors ( TFs ) levels , such as Brd2 ( Figure 3e ) , and were thus susceptible to BET inhibitors , only the three lines displaying BETi-TMZ synergy expressed MGMT ( Figure 3e ) . Treating those three lines with BETi , dramatically reduced MGMT expression ( Figure 3f ) . Finally , combining BET inhibitors with the MGMT inhibitor Lomeguatrib did not increase sensitivity to TMZ above the levels conferred by Lomeguatrib alone ( Figure 3g ) . In sum , we have discovered that several BETi synergized with TMZ treatment by reducing MGMT expression . We applied MIEL to rank BETi according to their magnitude of epigenetic effect and demonstrated that they ranks according their ability to synergize with TMZ suggesting that their mechanism of action involves epigenetic change . In contrast , the activity of PARP inhibitors didn’t correlate with magnitude of epigenetic effect , suggesting an alternative mechanism of action . Thus , we propose that the MIEL approach is well positioned to systematically analyze and identify epigenetically active compounds , then provide critical initial information for their mechanism of action . By altering histone and DNA modifications , epigenetic drugs have a direct effect on the MIEL readout . To test the ability of MIEL to identify and classify in-direct epigenetic changes we tested its utility for identifying drugs inducing GBM differentiation . Previous attempts to design screening strategies for this purpose have met with multiple difficulties . One critical problem is the lack of informative markers faithfully reporting GBM differentiation that could be used for high-throughput screening ( Patel et al . , 2014 ) . The lack of informative markers for GBM differentiation and the ability of MIEL to identify compounds producing desired epigenetic alterations prompted us to test the feasibility of using this approach to screen for drugs inducing GBM TPCs differentiation . For this , we first tested the ability of MIEL to discriminate between different cell fates . We analyzed 3 cell types: primary human fibroblasts , induced pluripotent stem cells ( iPSCs ) derived from these fibroblasts , and neural progenitor cells ( NPCs ) differentiated from the iPSCs . The fibroblasts were isolated from three unrelated donors ( WT-61 , WT-101 , WT-126 ) and used to obtain corresponding iPSC and NPC lines . Cellular identities of the 3 cell types were verified by immune-fluorescence for Sox2 and Oct4 ( Figure 4a ) , and MIEL analysis was carried out using data from either H3K4me1 and H3K9me3 or H3K27ac and H3K27me3 staining , with both combinations providing similar results . Multivariate centroids were calculated for each cell population and plotted on a distance map to visualize the relative Euclidean distance between various cell populations . The fibroblasts , iPSCs , and NPCs each segregate to form three visually distinct territories ( Figure 4—figure supplement 1c ) . We separated the nine lines by cell-fates using DA , which showed an accurate separation of the different cell-fates across all three donors ( average accuracy 100%; Figure 4b , Figure 4—figure supplement 1e ) . A similar analysis aimed at separate the different donors showed only low accuracy ( average accuracy 55 . 5%; Figure 4c , Figure 4—figure supplement 1f ) . To determine whether it was possible to discriminate between individual cells with different fates , a Support Vector Machine ( SVM ) classifier was trained on a subset of fibroblasts , iPSCs , and NPCS from the three donors . Classification of the test set indicated a high degree of separation between the different fates at a single cell level ( Figure 4—figure supplement 1b , d ) . Additionally , MIEL analysis ( using only H3K9me3 ) was able to discriminate between primary hematopoietic cell types freshly isolated from mouse bone marrow , namely lymphoid , myeloid , and stem/progenitors ( Figure 4—figure supplement 2 ) . However , closely related hematopoietic stem and progenitor cells were not readily separated ( Figure 4—figure supplement 2 ) . These results underscore MIEL’s ability to discriminate multiple different cell types and differentiation states uniquely based on their single-cell epigenetic landscapes both in cultured and primary cells of human and mouse origin . We tested MIEL’s ability to distinguish TPCs and differentiated glioma cells ( DGCs ) , derived from the same primary human GBMs ( Suvà et al . , 2014 ) . Three TPC/DGC pairs were derived in parallel from three genetically distinct glioblastoma tumor samples ( MGG4 , MGG6 , and MGG8 ) over a 3 month period using either serum-free FGF/EGF conditions for TPCs or 10% serum for DGCs ( Suvà et al . , 2014 ) . Visualization using distance maps demonstrated that TPCs and DGCs segregate to form two visually distinct territories ( Figure 4—figure supplement 1g ) and were separated with high accuracy using DA ( mean accuracy 100%; Figure 4d ) . SVM-based pairwise classification of single cells distinguished TPCs from their corresponding DGC lines with an average accuracy of 83% , using any of the four epigenetic marks tested ( H3K27me3 , H3K9me3 , H3K27ac , and H3K4me1; Figure 4e ) . An SVM classifier derived from images of the MGG4 TPC/DGC pair separated all 3 TPC/DGC pairs with 88% average accuracy , providing proof of principle for the derivation of a signature for non-tumorigenic cells obtained following serum differentiation of primary glioblastoma cells ( Figure 4f ) . These findings suggest that MIEL can readily distinguish undifferentiated TPCs from differentiated DGCs based on multiparametric signatures of these glioblastoma cells using only the patterns of universal epigenetic marks . For the purpose of establishing a screening protocol , we tested whether short serum or Bmp4 treatment is sufficient to induce a differentiation-like phenotype in TPCs . We treated several glioblastoma cell lines for 3 days with either serum or Bmp4 , then quantified expression of core transcription factors previously shown to determine the TPC transcriptomic program of TPCs ( Suvà et al . , 2014 ) . Immunostaining revealed that the four transcription factors Sox2 , Sall2 , Brn2 and Olig2 were downregulated by both serum and Bmp4 in a cell line-dependent manner ( Figure 4—figure supplement 3a ) . RNAseq analysis of serum- and Bmp4-treated GBM2 cells revealed that 3 days of treatment reduced ( vs untreated cells ) expression of most genes previously found to constitute the transcriptomic stemness signature ( Patel et al . , 2014 ) ( Figure 4—figure supplement 3b ) . Additionally , both serum and Bmp4 were found to attenuate TCP growth rate ( Figure 4—figure supplement 3c ) . To identify the cellular processes altered by these treatments , we conducted differential expression analysis . Expression of 4852 genes was significantly altered ( p<0 . 01 and −1 . 5 < Fold Change>1 . 5 ) by either serum or Bmp4 . Gene Ontology ( GO ) analysis of these altered genes indicated enrichment in multiple GO categories consistent with initiation of TPC differentiation; these include cell cycle , cellular morphogenesis associated with differentiation , differentiation in neuronal lineages , histone modification , and chromatin organization ( Figure 4—figure supplement 4 ) . These results demonstrate that a 3 day treatment with either serum or Bmp4 is sufficient to induce transcriptomic changes characteristic of TPC differentiation . Previous work indicated distinct features of glioblastoma differentiation induced with BMP compared to serum ( Carén et al . , 2015 ) . Indeed , we observed distinct expression changes , including differences in expression of genes regulating chromatin organization and histone modifications ( Figure 4—figure supplement 5a , b ) , between serum- and Bmp4-induced glioblastoma differentiation . To test the ability of MIEL to detect short term TPCs differentiation we treated four genetically distinct glioblastoma lines with serum or BMP4 , then conducted MIEL analysis using either H3K9me3 and H3K4me1 or H3K27ac and H3K27me3 . Discriminant analysis allowed high accuracy separation of these treatments across all cell lines using both histone modification combinations ( mean accuracy 100%; Figure 4h; Figure 4—figure supplement 5c ) . The global gene expression profile represents a gold standard for defining the cellular state ( Liang et al . , 2005 ) . Therefore , we correlated the relative distances between distinct cellular states , using MIEL-based and global gene expression-based metrics . We sequenced untreated and 3 days serum- or Bmp4-treated GBM2 TPCs ( three replicates each ) and used FPKM values of all expressed genes ( FPKM >1 in at least one cell population ) to calculate the Euclidean distance matrix between all cell populations . FPKM-based distances were then correlated to image texture feature-based distances . The resulting Pearson correlation coefficient of R = 0 . 93 ( p<0 . 001 ) suggests a high correlation between these two metrics ( Figure 4i , j ) , demonstrating that MIEL is capable of distinguishing closely related glioblastoma differentiation routes induced by serum or BMP and validating the robustness of the MIEL approach for analyzing glioblastoma differentiation . To test whether MIEL can identify compounds inducing GBM TPCs differentiation based on serum/Bmp4 signature , we screened the Prestwick compound library ( 1200 compounds ) . GBM2 TPCs were treated for 3 days with Prestwick compounds at 3 µM fixed , then immune-labeled for H3K27ac and H3K27me3 . GBM2 cells treated with DMSO , serum , BMP4 , or compound were compared within the same plate ( to avoid imaging artifacts and normalization issues ) . To identify epigenetically active compounds , we calculated the Euclidean distance to the DMSO center for each DMSO replicate and Prestwick compound . Distances were z-scored , and active compounds were defined as compounds for which z-scored distance was greater than 3 . Compounds with less than 50 cells imaged were considered toxic and excluded from analysis . This analysis detected 144 active compounds . To identify compounds inducing epigenetic changes reminiscent of serum- BMP4-induced differentiation , we used quadratic DA to build a model separating untreated , serum- and Bmp4-treated cells and classified all 144 active compounds to these categories ( Figure 5a , b ) . A total 31 compounds were classified as similar to either serum or Bmp4 ( i . e . , differentiated ) . Of these , 20 compounds belonged to 1 of the following four categories: Na/K-ATPase inhibitors of the digoxin family , molecules that disrupt microtubule formation or stability , topoisomerase inhibitors , or nucleotide analogues that disrupt DNA synthesis ( Figure 5b ) . To further narrow down the list of candidates , we conducted pairwise SVM classification of DMSO- and either serum- or BMP4-treated cells , then selected compounds that induced at least 50% of the cells to be classified as either serum- or BMP4-treated . We then calculated the Euclidean distance between candidate compounds and serum- or BMP4-treated cells and selected compounds where the distance to one or both treatments was less than the distance between DMSO and that treatment . This yielded 20 candidate compounds , of which 15 belonged to 1 of the four categories mentioned above; the top two compounds from each category were chosen for further analysis ( Figure 5—figure supplement 1a ) . GBM2 cells were treated for 3 days with DMSO , serum , Bmp4 or candidate compounds at 0 . 3 , 1 or 3 µM , fixed , and then immunostained for H3K27ac and H3K27me3 . Using pairwise SVM based classifications of untreated cells and either serum- or Bmp4-treated cells we identified for each of the eight compounds the lowest concentration at which 50% or more of the cells were classified as treated ( Figure 5—figure supplement 1b ) . These concentrations were used for all subsequent experiments ( Supplementary file 1 - Table S7 ) . Because most of these compounds are known for their cytotoxic effects , we verified the growth rates of drug-treated glioblastoma cells . With the exception of digoxin , which was cytostatic , drug treatment resulted in growth rates comparable with that induced by serum or BMP4 ( Figure 5—figure supplement 2a ) . We used immunofluorescence to test for expression of core TPC transcription factors ( Sox2 , Sall2 , Brn2 and Olig2 ) . Except for trifluridine , all compounds induced statistically significant reductions in Sox2; digoxin and digitoxigenin also induced a significant reduction of Sall2 and Brn2; Olig2 expression was unaltered by any treatment ( Figure 5—figure supplement 2b ) . Next , we investigated whether the compounds identified using MIEL can induce transcriptomic changes similar to serum and Bmp4 treatment and quantified the ability of MIEL to predict compounds best at mimicking these treatments . GBM2 cells were treated with DMSO , serum , Bmp4 , or each of the eight candidate compounds; after 3 days , RNA was extracted and sequenced . Transcriptomic profiles of the eight compounds were ranked according to average Euclidean distance ( based on FPKM values for all expressed genes ) from serum and BMP4-treated cells . To safeguard against potential artefacts of cytotoxicity , we compared gene expression-based ranking with measured cellular growth rates from drug treatments and found no positive correlation ( Figure 5—figure supplement 2c ) . Next , we compared Sox2 levels under all treatment conditions to determine whether expression of this transcription factor can identify drugs that best mimic serum or BMP4 . We found no positive correlation between Sox2 expression and the transcriptomic-based rankings ( Figure 5—figure supplement 2d ) , suggesting that Sox2 levels alone are insufficient to stratify the compounds . Finally , to compare MIEL-based signatures to the transcriptomic profile , we ranked MIEL readouts of cells treated with the eight drugs according to average Euclidean distance from serum- or Bmp4-treated cells ( calculated using texture features derived from images of H3K27ac , H3K27me3 , H3K9me3 and H3K4me1 ) . Comparison of the MIEL-based metric with the gene expression-based metric revealed a high degree of positive correlation between MIEL- and gene expression-based rankings ( Pearson correlation coefficient R = 0 . 92 , p<0 . 001; Figure 5c ) . To further visualize these results , we constructed heat maps depicting fold change in expression levels of genes associated with several GO terms enriched by serum and Bmp4 . Our top candidate , etoposide , altered expression of a large portion of genes in similar fashion to that of serum and BMP4; in contrast , the lowest-ranking candidate , digoxin , induced changes in gene expression , which were rather different from serum and BMP4 ( Figure 5d ) . Taken together , the above results reflect the unique ability of MIEL to identify molecules that shift epigenetic signature of glioblastoma TPCs towards DGCs . Critically , MIEL is capable of ranking such molecules according to their change-inducing potency and that ranking robustly correlate with global expression-based readouts of glioblastoma differentiation . Previous studies have demonstrated that image-based profiling can distinguish between classes of compounds with very distinct functions , such as Aurora and HADC inhibitors ( Kang et al . , 2016 ) . One objective of our study was to estimate the resolution of separation between categories of compounds with similar functions . We found that a single histone modification was sufficient to separate highly distinct classes ( Figure 1—figure supplement 3b ) . However , separating similar classes ( e . g . , Aurora and JAK inhibitors , which affect histone phosphorylation , or Pan and Class I HADCs , which affect histone acetylation ) required staining for at least one additional histone modification ( Figure 1d ) . Despite their many advantages , cellular assays , including high-content assays , are often used as secondary screens for epigenetic drugs due to multiplicity of enzyme family members and an inability to determine direct enzymatic activity ( Martinez and Simeonov , 2015 ) . Consequently , MIEL’s ability to separate closely related functional categories on top of other advantages make this profiling approach an attractive alternative for primary screens . Phenotypic profiling methods have been previously used to identify genotype‐specific drug responses by comparing profiles across multiple isogenic lines ( Breinig et al . , 2015 ) . Here we show that activity of biologics ( i . e . , serum and Bmp4 ) that induces glioblastoma differentiation , as well as that of 57 epigenetic compounds , was significantly correlated across four different primary glioblastoma lines ( Figure 2c , d , e; Figure 4h ) . We also showed that variation in activity levels correlated with target expression levels and that various categories can be distinguished across cell lines . Together , these suggest that MIEL could be used to identify cell lines showing an aberrant reaction to selected drugs and , therefore , aid in identifying optimal treatments for individual patients . Similar applications have previously been used to tailor specific kinase inhibitors to patients with chronic lymphocytic leukemia ( CLL ) who display venetoclax resistance ( Oppermann et al . , 2016 ) . Given the limited success of cytotoxic drugs in treating glioblastoma , we focused on alternative approaches: ( 1 ) epigenetic drugs aimed at sensitizing glioblastoma TPCs to such treatments , and ( 2 ) inducing glioblastoma differentiation . We have demonstrated MIEL’s ability to rank candidate drug activity to correctly predict the best candidates for achieving the desired effect . The importance of this is highlighted in large ( hundreds of thousands of compounds ) chemical library screens , which typically identify many possible hits needing to be reduced and confirmed in secondary screens ( Hughes et al . , 2011; Strovel et al . , 2004 ) . Our results show a significant correlation between BET inhibitor activity , as defined by MIEL ( Figure 3d ) , and their ability to synergize and increase TPC sensitivity to TMZ and reveal a previously unknown role for BET inhibitors in reducing MGMT expression ( Figure 3e , f , g ) . Previous studies have demonstrated upregulation of several BET transcription factors in glioblastomas ( Pastori et al . , 2014; Wadhwa and Nicolaides , 2016 ) and multiple pre-clinical studies have investigated the potential of BET inhibition as a single drug treatment for glioblastoma ( Xu et al . , 2018; Ishida et al . , 2017; Cheng et al . , 2013 ) . However , while clinical trials with the BET inhibitor OTX015 demonstrated low toxicity at doses achieving biologically active levels , no detectable clinical benefits were found ( Hottinger et al . , 2016 ) . This prompted approaches using drug combination treatments ( Ramadoss and Mahadevan , 2018 ) such as combining HDACi and BETi ( Heinemann et al . , 2015; Bhadury et al . , 2014 ) . The mechanism by which BETi induces increased TMZ sensitivity has not been described . Recently , a distal enhancer regulating MGMT expression was identified ( Chen et al . , 2018 ) . Activation of this enhancer by targeting a Cas9-p300 fusion to its genomic locus increased MGMT expression while deletion of this enhancer reduced MGMT expression ( Chen et al . , 2018 ) . As BET transcription factors bind elevated H3K27ac levels found in enhancers ( Sengupta et al . , 2015; Lovén et al . , 2013 ) , this may suggest a possible mechanism for BETi-induced reduction of MGMT expression , which in turn results in increased sensitivity to the DNA alkylating agent TMZ . Silencing the MGMT gene through promoter methylation has long been known to increase responsiveness to TMZ treatment and improve prognosis in patients with glioblastoma ( Karayan-Tapon et al . , 2010; Hegi et al . , 2005; Esteller et al . , 2000 ) . Yet Despite that , clinical trials that combine TMZ and MGMT inhibitors have not improved therapeutic outcomes in such patients , possibly due to the 50% reduction in dose of TMZ , which is required to avoid hematologic toxicity ( Quinn et al . , 2009a; Quinn et al . , 2009b; Quinn et al . , 2009c ) . Thus , BETi offers an attractive line of research , though further studies are needed to determine whether the elevated sensitivity of glioblastoma to BETi , and its ability to reduce MGMT expression could be exploited to improve patient outcome . We analyzed serum and BMP4 , two established biologicals known to induce glioblastoma differentiation in culture ( Lee et al . , 2006; Piccirillo et al . , 2006; Pollard et al . , 2009 ) and established signatures of the differentiated glioblastoma cells based on the pattern of epigenetic marks that could be applied across several genetic backgrounds . This is the first time that a signature for glioblastoma differentiation suitable for high-throughput drug screening has been obtained . Indeed , results of previous studies using bulk glioblastoma analysis ( Carén et al . , 2015 ) or single-cell sequencing ( Patel et al . , 2014 ) could not be readily applied for high-throughput screening . As a proof of principle , we analyzed the Prestwick chemical library ( 1200 compounds ) to validate MIEL’s ability to select and prioritize small molecules , which mimic the epigenetic and transcriptomic effects of serum and BMP4 . Surprisingly , we observed that the degree of reduction in endogenous SOX2 protein levels following drug treatment did not correlate with the degree of differentiation assessed by global gene expression ( Figure 5—figure supplement 2d ) ; in contrast , MIEL-based metrics did correlate . This result , taken together with MIEL’s ability to distinguish multiple cells types ( iPSCs , NPCs , fibroblasts , hematopoietic lineages; Figure 4b , c; Figure 4—figure supplement 2 ) across several genetic backgrounds , demonstate that the MIEL approach can readily identify compounds inducing desired changes in cell fate and that it can serve as a cost-effective tool for prioritizing compounds during the primary screenings . By tapping into the wealth of information contained within the cellular epigenetic landscape through modern high-content profiling and machine-learning techniques , the MIEL approach represents a valuable tool for high-throughput screening and drug discovery and is especially relevant when the desired cellular outcome cannot be readily defined using conventional approaches . Monolayer cultures of patient-derived GMB TPCs were propagated on Matrigel-coated plates in DMEM:F12 Neurobasal Medium ( 1:1; Gibco ) , 1% B27 supplement ( Gibco ) , 10% BIT 9500 ( StemCell Technologies ) , 1 mM glutamine , 20 ng/ml EGF ( Chemicon ) , 20 ng/ml bFGF , 5 µg/ml insulin ( Sigma ) , and 5 mM nicotinamide ( Sigma ) . The medium was replaced every other day and the cells were enzymatically dissociated using Accutase prior to splitting . Fibroblasts , iPSCs , and iPSC-derived NPCs were cultured as previously described ( Marchetto et al . , 2010; Kim et al . , 2011 ) . For TPC differentiation treatments cells were cultured in DMEM:F12 Neurobasal Medium ( 1:1 ) , 1% B27 supplement , 10% BIT 9500 , 1 mM glutamine supplemented with either Bmp4 ( 100 ng/ml; R and D Systems ) or FBS ( 10% ) . Cells were rinsed with PBS and fixed in 4% paraformaldehyde in PBS for 10 min at room temperature . After blocking with PBSAT ( 2% BSA and 0 . 5% Triton X-100 in PBS ) for 1 hr at room temperature , the cells were incubated overnight at 4°C with primary antibodies diluted in PBSAT . Primary antibodies are listed in Supplementary file 1 - Table S1 , and the appropriate fluorochrome-conjugated secondary antibodies were used at 1:500 dilution . Nuclear co-staining was performed by incubating cells with either Hoechst-33342 or DAPI nuclear dyes . For MIEL analysis , cells were imaged on either an Opera QEHS high-content screening system ( PerkinElmer ) using ×40 water immersion objectives or an IC200-KIC ( Vala Sciences ) using a × 20 objective . Images collected were analyzed using Acapella 2 . 6 ( PerkinElmer ) . At least 40 fields/well for Opera and five fields/well for IC200 were acquired and at least two wells per population were used . Features of nuclear morphology , fluorescence intensity inter-channel co-localization , and texture features ( Image moments , Haralick , Threshold Adjacency Statistics ) were calculated using custom algorithms ( Source Code File one and www . andrewslab . ca ) . A full list of the features used is available from the authors . Values for each cell were generated and exported to Microsoft Excel or MATLAB for further analysis . For Sall2 , Olig2 , Brn2 , Sox2 , Oct4 , and GFAP immunostaining , images were captured on an IC200-KIC ( Vala Sciences ) using a × 20 objective . Between 3 and 8 fields per well were acquired and analyzed using Acapella 2 . 6 ( PerkinElmer ) . For all nuclear markers , average intensities in nucleus or fold change compared to untreated cells are shown . Unless stated otherwise , at least three wells and a minimum of 300 cells for each condition were compared using the unpaired two-tailed t-test . The image features-based profile for each cell population ( e . g . , cell types , treatments , technical repetition ) was represented using a vector ( center of distribution vectors ) in which every element is the average value of all cells in that population for a particular feature . The vector’s length is given by the number of features chosen ( 262 per histone modification ) . Raw feature values were normalized by z-scoring to the average and standard deviation of all populations being compared . All cells in each population were used to calculate center vectors , and each population contained at least 50 cells . Activity level for each drug was determined by calculating the distance from DMSO . For this , feature values of all DMSO replicates center vectors were used to calculate the DMSO center vector . Euclidean distance of each compound and each DMSO replicate to the DMSO center vector was calculated . Distances were z-scored to the average distance and standard deviation of DMSO replicates from the DMSO center vector . Transcriptomic-based profile for each cell population was represented using a vector in which every element is the z-scored FPKM value for a single gene in that population . The length of the vector is given by the number of genes used to construct the profile . The Euclidean distance between all vectors ( either image features or transcriptomic based ) was calculated to assemble a dissimilarity matrix ( size N × N , where N is the number of populations being compared ) . For representation , the N × N matrix was reduced to a Nx2 matrix with MDS using the Excel add-on program Xlstat ( Base , v19 . 06 ) , and displayed as a 2D scatter plot . Quadratic discriminant analysis was conducted using the Excel add-on program xlstat ( Base , v19 . 06 ) . The model was generated in a stepwise ( forward ) approach using default parameters . All features derived from images of tested histone modification were used for analysis following normalization by z-score . Features displaying multicollinearity were reduced . Model training was done using multiple DMSO replicates and at least two replicates from each cell-line or drug treatment . The model was tested on at least 8 DMSO replicates and at least one replicate from each cell line or treatment . SVM classification was conducted as previously described ( Collins et al . , 2015 ) . Cell-level data in total populations ( minimum 400 cells per population ) were normalized to z-scores , and a subset of cells from each population being classified was randomly chosen as the training set ( subset size at least 100 × the population number bei ng classified ) . The training set was used for a SVM classifier ( MATLAB svmtrain function ) . The remaining cells ( test set ) were then classified using the SVM-derived classifier to assess the accuracy of classification ( MATLAB svmclassify function ) . Here , the accuracy of all pairwise classifications was given as the average accuracy calculated for each population . To classify the similarity of multiple cell populations , we classified known populations ( e . g . , different treatments or cell fates ) to generate known bins and then used the same classifiers on the unknown population to categorize each cell . GBM2 cells were plated at 4000 cells/well and exposed to epigenetic compounds ( Supplementary file 1 - Table S2 ) at 10 µM for 1 day in 384-well optical bottom assay plates ( PerkinElmer ) . Negative control was DMSO ( 0 . 1% ) , 48 DMSO replicates per plate , three technical replicates ( wells ) were treated per compound . Cells were fixed and stained with histone modification-specific antibodies ( H3K27ac and H3K27me3 , H3K9me3 , H3K4me1 ) and AlexaFluor-488- or AlexaFluor-555-conjugated secondary antibodies . DNA was stained with DAPI followed by imaging and feature extraction . To compare data from multiple plates , average feature values in each plate were normalized to DMSO . Here , feature values of all DMSO replicates center vectors in each plate , then were used to calculate the plate-wise DMSO vector . Raw feature values for all center vectors of all populations in each plate were normalized to the plate-wise DMSO vector; normalized feature values were z-scored as above . To identify active compounds , activity level for each compound was calculated as above , and active compounds were defined as compounds for which activity z-score was >3 . Compounds reducing the number of imaged cells per well below 50 were considered toxic and excluded from analysis . GBM2 cells were plated and stained as above . For each compound ( Supplementary file 1 - Table S3 ) , cells were treated at 0 . 1 , 0 . 3 , 1 . 0 , 3 . 0 , 10 . 0 uM . Activity levels were calculated as above . Average cell count was calculated across the replicates for each compound to compare epigenetic changes and toxicity . Cell counts were z-scored against the average and standard deviation of all DMSO replicates . Distances ( z-scored ) and cell counts ( z-scored ) were averaged for each functional class at each concentration . Total RNA was isolated from GBM2 cells using the RNeasy Kit ( Qiagen ) , 0 . 25 ug total RNA was used to isolate mRNAs and for library preparation . Library preparation and sequencing were conducted by the SBP genomics core ( Sanford-Burnham NCI Cancer Center Support Grant P30 CA030199 ) . PolyA RNA was isolated using the NEBNext Poly ( A ) mRNA Magnetic Isolation Module , and barcoded libraries were made using the NEBNext Ultra II Directional RNA Library Prep Kit for Illumina ( NEB , Ipswich MA ) . Libraries were pooled and single-end sequenced ( 1 × 75 ) on the Illumina NextSeq 500 using the High-Output V2 kit ( Illumina ) . Read data , processed in BaseSpace ( https://basespace . illumina . com ) , were aligned to Homo sapiens genome ( hg19 ) using STAR aligner ( https://code . google . com/p/rna-star/ ) with default settings . Differential transcript expression was determined using the Cufflinks Cuffdiff package ( https://github . com/cole-trapnell-lab/cufflinks ) . For heat maps showing fold change in expression , FPKM values in each HDACi-treated population were divided by the average FPKM values of DMSO-treated GBM2 and values shown as log2 of the ratio . Go enrichment analysis was conducted using PANTHER v11 ( Mi et al . , 2017 ) using all genes identified as differentially expressed following either serum or Bmp4 treatment . To highlight differences in expression levels between serum- and Bmp4-treated GBM2 cells , FPKM values in each sample were z-scored . Zscore= ( FPKMObservation-FPKMAverage ) /FPKMSD ( FPKMObservation- FPKM value obtain through sequencing; FPKMAverage – average of all FPKM values in all samples for a certain gene; FPKMSD – standard deviation of FPKM values for a certain gene ) . Heat maps were generated using Microsoft Excel conditional formatting . To compare drug-induced epigenetic changes across multiple glioblastoma cell lines , 101A , 217M , GBM2 and PBT24 cells were plated at 4000 cells/well and treated with compounds for 24 hr . Compounds and concentrations are shown in Supplementary file 1 - Table S4 . Activity level was calculated as above . Pearson coefficient and significance of correlation for activity levels in each pair of cell lines were calculated using the Excel add-on program xlstat ( Base , v19 . 06 ) . Euclidean distances were calculated using either transcriptomic data ( FPKM ) or texture features . Pearson’s correlation coefficient ( R ) was transformed to a t-value using the formula ( t = R × SQRT ( N-2 ) /SQRT ( 1-R2 ) where N is the number of samples , R is Pearson correlation coefficient; the p-value was calculated using Excel t . dist . 2t ( t ) function . For compound prioritization , Euclidean distance between the compound treated and serum- or Bmp4-treated GBM2 cells was calculated based on either FPKM ) or image features . The average distance for both serum and Bmp4 treatments was normalized to the average distance of untreated cells to serum and Bmp4 . Cells were plated at 1500 cells/well in 384-well optical bottom assay plates ( PerkinElmer ) . Two sets of the experiment were prepared; DMSO ( 0 . 1% ) was used for negative controls at 48 DMSO replicates per plate; three replicates ( wells ) were treated per compound . Compound concentrations used are shown in Supplementary file 1 - Table S5 . Cells in both sets were pre-treated with epigenetic compounds for 2 days prior to cytotoxic treatment . Cytotoxic treatment , either 200 µM temozolomide ( TMZ , Sigma ) or 1Gy x-ray radiation ( RS2000; RAD Source ) was carried out for 4 days on single set ( ‘treatment set’ ) ; for TMZ treatment , DMSO control was given to the second set . A single radiation dose was given at day 3; TMZ was given twice at days 3 and 5 of the experiment . Cells were fixed , stained with DAPI , and scored using an automated microscope ( Celigo; Nexcelom Bioscience ) . For each compound , fold change in cell number was calculated for both the ‘treatment set’ ( Drug+Cytotox ) and the ‘control set’ ( Drug ) , compared to DMSO-treated wells in the control set . The effect of radiation or TMZ alone was calculated as fold reduction of DMSO-treated wells in the treatment set compared to DMSO-treated wells in the control set ( Cytotox ) . The coefficient of drug interaction ( CDI ) was calculated as ( Drug+Cytotox ) / ( Drug ) X ( Cytotox ) . For conformation experiments , the same regiment and CDI calculations were carried out on SK262 , 101A , 217M , 454M , and PBT24 glioblastoma cell lines; PARPi and BETi were used at same concentration as the initial screen on GBM2 ( Table S5 ) . GBM2 cells were plated at 2000 cells/well and exposed to Prestwick compounds ( 3 µM; Supplementary file 1 - Table S6 ) for 3 days in 384-well optical bottom assay plates ( PerkinElmer ) . Cells were then fixed and stained with rabbit polyclonal anti-H3K27ac and mouse monoclonal anti-H3K27me3 antibodies followed by AlexaFluor-488- or AlexaFluor-555-conjugated secondary antibodies . Positive controls contained BMP4 ( 100 ng/ml ) and serum ( 10% ) ; negative controls contained DMSO ( 0 . 1% ) . DNA was counterstained with Hoechst . Images were acquired using Perkin Elmer Opera QEHS . MIEL analysis was conducted as described above .	Each cell contains a complete copy of a person’s genes coded in their DNA . However , for a cell to perform its specific role , it only needs a small fraction of this genetic information . The mechanisms that control which genes a cell is using fall under the umbrella of ‘epigenetics’ ( meaning beyond genetics ) . These mechanisms involve changes in the way that DNA is organized inside the cell nucleus and changes in how accessible different parts of the genome are to various cellular components . DNA is long and fragile so , to maintain its integrity , it is wrapped around protein complexes called histones . Adding chemical modifications to histones is one of the main epigenetic mechanisms that cells use to regulate which genes are turned on and off . Several methods allow researchers to read patterns of histone modification and use this information to derive what state a cell is in and how it might behave . Improving these methods is of particular interest in drug development , where this information could reveal the effects , and side-effects , of new treatments . Unfortunately , existing techniques are costly in both time and money , and they are not well suited to analyzing epigenetic changes caused by the large numbers of compounds tested during drug development . To overcome this barrier , Farhy et al . have developed a new system called ‘Microscopic Imaging of the Epigenetic Landscape’ ( MIEL for short ) . The system allows them to quickly analyze the epigenetic changes caused by each of a large number of different chemical compounds when they are used on cells . MIEL tags different histone modifications by staining each with a different color , and then uses automated microscopy to produce images . It then extracts information from these images using advanced image analysis tools . The changes induced by different drugs can then be compared and categorized using machine learning algorithms . To test the MIEL system , Farhy et al . grew brain cancer cells ( derived from human tumors ) in the lab , and treated them with compounds that target proteins involved in histone modifications . Using their newly created pipeline , Farhy et al . were able to identify the unique epigenetic changes caused by these compounds , and train the system to correctly predict which type of drug the cells had been treated with . In a different set of experiments Farhy et al . demonstrate the utility of their new pipeline in identifying drugs that induce a set of epigenetic changes associated with a reduced ability to regrow tumors . This new system could help screen thousands of compounds for their epigenetic effects , which could aid the design of new treatments for many diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "medicine", "cancer", "biology" ]	2019	Improving drug discovery using image-based multiparametric analysis of the epigenetic landscape
Dopamine is thought to regulate learning from appetitive and aversive events . Here we examined how optogenetically-identified dopamine neurons in the lateral ventral tegmental area of mice respond to aversive events in different conditions . In low reward contexts , most dopamine neurons were exclusively inhibited by aversive events , and expectation reduced dopamine neurons’ responses to reward and punishment . When a single odor predicted both reward and punishment , dopamine neurons’ responses to that odor reflected the integrated value of both outcomes . Thus , in low reward contexts , dopamine neurons signal value prediction errors ( VPEs ) integrating information about both reward and aversion in a common currency . In contrast , in high reward contexts , dopamine neurons acquired a short-latency excitation to aversive events that masked their VPE signaling . Our results demonstrate the importance of considering the contexts to examine the representation in dopamine neurons and uncover different modes of dopamine signaling , each of which may be adaptive for different environments . Dopamine is thought to be a key regulator of learning from appetitive as well as aversive events ( Schultz et al . , 1997; Wenzel et al . , 2015 ) . It has been proposed that dopamine neurons act as a teaching signal in the brain by signaling the discrepancy between the values of actual and predicted rewards , that is , reward prediction error ( RPE ) ( Bayer and Glimcher , 2005; Cohen et al . , 2012; Hart et al . , 2014; Roesch et al . , 2007; Schultz , 2010; Schultz et al . , 1997 ) . Although accumulating evidence supports this idea with respect to rewarding and reward-predicting events ( Bayer and Glimcher , 2005; Cohen et al . , 2012; Eshel et al . , 2015; Hart et al . , 2014; Roesch et al . , 2007; Schultz , 2010; Schultz et al . , 1997 ) , how dopamine neurons integrate information about aversive events remains highly controversial . Pioneering work by Wolfram Schultz and colleagues introduced the idea that dopamine neurons signal RPE . This work demonstrated that dopamine neurons in the midbrain of monkeys exhibit a highly specific set of responses to reward ( Mirenowicz and Schultz , 1994 ) . When the animal receives reward unexpectedly , dopamine neurons fire a burst of action potentials . If a sensory cue reliably predicts reward , however , dopamine neurons decrease their response to reward , and instead burst to the cue . Finally , if an expected reward is omitted , dopamine neurons pause their firing at the time they usually receive reward ( Hollerman and Schultz , 1998; Schultz et al . , 1997 ) . Subsequently , the idea of RPE coding by dopamine neurons has been substantiated by further experiments in a variety of species including monkeys ( Bayer and Glimcher , 2005; Hollerman and Schultz , 1998; Waelti et al . , 2001 ) , rats ( Flagel et al . , 2011; Pan et al . , 2005; Roesch et al . , 2007 ) , mice ( Cohen et al . , 2012; Eshel et al . , 2015 ) and humans ( D'Ardenne et al . , 2008 ) . This signal is proposed to underlie associative learning ( Rescorla and Wagner , 1972 ) , and bears a striking resemblance to machine learning algorithms ( Sutton and Barto , 1998 ) . Many of the previous studies that characterized dopamine responses used rewarded outcomes with varying degrees of predictability . Comparatively fewer studies have used aversive stimuli in the context of prediction errors . Among studies that have used aversive stimuli , these provide differing reports as to how dopamine neurons respond to aversive stimuli ( Fiorillo , 2013; Schultz , 2015; Wenzel et al . , 2015 ) . It is thought that the majority of dopamine neurons are inhibited by aversive stimuli ( Mileykovskiy and Morales , 2011; Mirenowicz and Schultz , 1996; Tan et al . , 2012; Ungless et al . , 2004 ) . However , a number of electrophysiological recording studies have reported that dopamine neurons are activated by aversive stimuli both in anesthetized ( Brischoux et al . , 2009; Coizet et al . , 2006; Schultz and Romo , 1987 ) and awake animals ( Guarraci and Kapp , 1999; Joshua et al . , 2008; Matsumoto and Hikosaka , 2009 ) , although the proportions and locations of aversion-activated neurons differed among these studies . The differences in the results between these studies could be due to the heterogeneity of dopamine neurons or to differences in experimental conditions ( e . g . type of aversive stimuli; type of anesthesia ) . Furthermore , another study using fast-scan cyclic voltammetry found that dopamine neurons are excited during successful avoidance of aversive stimuli ( Oleson et al . , 2012 ) , which could be 'rewarding' . Therefore , some of the excitatory responses to aversive stimuli may not be due to aversiveness alone . Some of these discrepancies could correspond to differences in dopamine signaling depending on the projection target . Roitman et al . ( 2008 ) monitored dopamine dynamics in the nucleus accumbens using cyclic voltammetry while the animal received intra-oral administrations of a sucrose or quinine solution ( Roitman et al . , 2008 ) . This study found that these stimuli caused opposite responses: dopamine release was increased by sucrose and decreased by quinine ( McCutcheon et al . , 2012 ) , suggesting that at least the majority of dopamine neurons projecting to the nucleus accumbens are inhibited by aversive stimuli . Matsumoto and Hikosaka ( 2009 ) examined the diversity of dopamine neurons in context of prediction error . They showed that dopamine neurons that are activated by the prediction of aversive stimuli are located in the lateral part of the substantia nigra pars compacta ( SNc ) , supporting the notion that dopamine subpopulations are spatially segregated ( Matsumoto and Hikosaka , 2009 ) . Consistent with this finding , Lerner et al . ( 2015 ) showed , using calcium imaging with fiber photometry , that SNc neurons projecting to the dorsolateral striatum are activated by aversive stimuli ( electric shock ) whereas those projecting to the dorsomedial striatum are inhibited ( Lerner et al . , 2015 ) . Lammel et al . ( 2011 ) provided further evidence for spatial heterogeneity by showing that dopamine neurons projecting to the medial prefrontal cortex , located in the medial ventral tegmental area ( VTA ) exhibited a form of synaptic plasticity ( AMPA/NMDA ratio ) in response to aversive stimuli ( formalin injection ) whereas dopamine neurons projecting to the dorsolateral striatum did not ( Lammel et al . , 2011 ) although how these neurons change their firing patterns in response to aversive stimuli remains unknown . In contrast to the above findings suggesting that dopamine neurons are heterogeneous with respect to signaling aversive events , Schultz , Fiorillo and colleagues have argued that dopamine neurons largely ignore aversiveness ( Fiorillo , 2013; Schultz , 2015; Stauffer et al . , 2016 ) . One argument is that the excitation of dopamine neurons caused by aversive stimuli may be due to a 'generalization' or 'spill-over' effect of rewarding stimuli . Specifically , Mirenowicz and Schultz ( 1996 ) showed that when rewarding and aversive stimuli are predicted by similar cues ( e . g . in a same sensory modality ) , aversion-predicting cues increase their tendency to activate dopamine neurons ( 'generalization' ) ( Mirenowicz and Schultz , 1996 ) . Kobayashi and Schultz ( 2014 ) showed that in a high-reward context , cues that predict a neutral outcome ( e . g . a salient picture ) increased their tendency to activate dopamine neurons compared to the neutral cues in a low reward context ( Kobayashi and Schultz , 2014 ) . Based on these and other observations ( Fiorillo et al . , 2013; Nomoto et al . , 2010 ) , they proposed that the early response reflects attributes such as stimulus generalization and intensity , and the later response reflects the subjective reward value and utility ( Schultz , 2016; Stauffer et al . , 2016 ) . One influential paper by Fiorillo ( 2013 ) concluded that dopamine neurons represent prediction errors with respect to reward but not aversiveness ( Fiorillo , 2013 ) . That is , dopamine neurons ignore aversive events . Recording from non-human primates , Fiorillo used three pieces of evidence to support this claim: First , dopamine neurons’ responses to aversive outcomes ( air puff ) were indistinguishable from their responses to neutral outcomes . Second , although most dopamine neurons reduced their reward responses when the reward was predicted , their response to aversive events was unaffected by prediction . Third , dopamine neurons did not integrate the value of aversive events when combined with rewarding events . From these results , the author proposed that the brain represents reward and aversiveness independently along two dimensions ( Fiorillo , 2013 ) . As a result , the author proposed that different molecules regulate different types of reinforcement learning: dopamine for reward and a different molecule for aversiveness . If proven true , these ideas are fundamental in understanding how the brain learns from reward and aversion . However , it remains to be clarified whether these observations can be generalized . The conclusions in many of the studies cited above relied upon indirect methods such as spike waveforms and firing properties ( Ungless and Grace , 2012 ) in order to identify dopamine neurons . These identification methods differed among studies and have recently been called into question ( Lammel et al . , 2008; Margolis et al . , 2006; Ungless and Grace , 2012 ) . The ambiguity of cell-type identification criteria across studies makes it difficult to consolidate data on dopamine signaling . For example , Ungless et al . showed that some neurons in the VTA that were excited by aversive events and identified as dopaminergic using standard electrophysiological criteria were revealed not to be dopaminergic when they were examined with juxtacellular labeling ( Ungless et al . , 2004 ) . Furthermore , Schultz has argued that some previous recording studies may not have targeted areas rich in dopamine neurons ( Schultz , 2016 ) . To circumvent this problem , we tagged dopamine neurons with a light-gated cation channel , channelrhodopsin-2 ( ChR2 ) and unambiguously identified dopamine neurons based on their responses to light ( Cohen et al . , 2012 ) . In the present study , we monitored the activity of identified dopamine neurons using a series of behavioral tasks designed to determine how dopamine neurons encode prediction of aversive events in addition to reward . Our results demonstrate that , in contrast to the proposal by Fiorillo ( 2013 ) , dopamine neurons in VTA indeed are able to encode complete VPE , integrating information about both appetitive and aversive events in a common currency . Importantly , the ability of dopamine neurons to encode VPE depends on both reward contexts and the animal’s trial-by-trial behavioral state . We recorded the spiking activity of total 176 neurons in the VTA using tetrodes while mice performed classical conditioning tasks ( Table 1 ) . To identify neurons as dopaminergic , we optogenetically tagged dopamine neurons ( Cohen et al . , 2012 ) . We then used a method developed previously ( Stimulus-Associated spike Latency Test [SALT] ) ( Eshel et al . , 2015; Kvitsiani et al . , 2013; Tian and Uchida , 2015 ) to determine whether light pulses significantly changed a neuron’s spike timing ( p<0 . 001 , Figure 1 ) . To ensure that spike sorting was not contaminated by light artifacts , we compared the waveforms between spontaneous and light-evoked spikes , as described previously ( Cohen et al . , 2012 ) . Dopamine neurons were mostly recorded from the central and posterior part of the lateral VTA including the parabrachial pigmented nucleus ( PBP ) , parainterfascicular nucleus ( PIF ) and paranigral nucleus ( PN ) ( Figure 1G , K , O ) . We obtained 72 optogenetically-identified dopamine neurons in total ( 5 ± 4 neurons per mouse; mean ± S . D . ; n = 14 mice ) . 10 . 7554/eLife . 17328 . 003Figure 1 . Optogenetic identification of dopamine neurons in the ventral tegmental area ( VTA ) . ( A ) Voltage trace from 10 pulses of 10 Hz light stimulation ( cyan bars , top ) of a representative dopamine neuron . A spontaneous spike and a light-triggered spike were magnified at the bottom . ( B ) Responses from this neuron to 10 Hz ( left ) and 50 Hz ( right ) stimulation . ( C ) Isolation of an identified dopamine neuron from noise and other units . ( D ) Histogram of p values testing whether light-activation induced significant changes in spike timing ( n = 62 neurons ) in the mixed prediction task . The p values were derived from SALT ( Stimulus-Associated spike Latency Test; see Materials and methods ) . Neurons with p values < 0 . 001 and waveform correlations > 0 . 9 were considered identified ( grey ) . P values and waveform correlations were calculated using light stimulation with all the frequencies ( 1–50 Hz ) . ( E ) Probability of light-evoked spike as a function of stimulation frequency for each dopamine neuron ( grey ) and the average across dopamine neurons ( blue circles and bars , median and interquartile range ) . ( F ) Histograms of the mean ( left ) and S . D . ( right ) spike latency to light stimulation with all the frequencies ( 1–50 Hz ) for 26 identified dopamine neurons . ( G ) Reconstruction of the positions of individual dopamine neurons recorded in the mixed prediction task . Each circle represents a lesion site from individual animals used in the mixed prediction task . Each horizontal line on the track ( indicated by a vertical line over the lesion site ) indicates estimated recording positions of individual dopamine neurons . Labeled structures: parabrachial pigmented nucleus of the VTA ( PBP ) , parainterfascicular nucleus of the VTA ( PIF ) , paranigral nucleus of the VTA ( PN ) , red nucleus ( RN ) , substantia nigra pars compacta ( SNc ) , and substantia nigra pars reticulata ( SNr ) . Scale bar , 1 mm . ( H–J ) Optogenetic identification of dopamine neurons recorded in high and low reward probability tasks ( 29 dopamine neurons identified out of 73 neurons ) . Conventions are the same as in D–F . ( K ) Reconstruction of the positions of individual dopamine neurons recorded in high ( red ) and low ( cyan ) reward probability tasks . Conventions are the same as in G . ( L–N ) Optogenetic identification of dopamine neurons recorded in high reward probability task 2 ( 17 dopamine neurons identified out of 41 neurons ) . ( O ) Reconstruction of the positions of individual dopamine neurons recorded in high reward probability task 2 ( magenta ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17328 . 00310 . 7554/eLife . 17328 . 004Table 1 . Summary of task conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 17328 . 004TaskCS ( % outcome ) Reward trials ( % ) Free reward ( % ) OutcomeOdor A ( Reward CS ) Odor B ( Nothing CS ) Odor C ( Air puff CS ) Odor D ( Reward and air puff CS ) Mixed prediction taskWater250025132Air puff007575Low reward probability taskWater200076Air puff0090High reward probability taskWater9000306Air puff0090High reward probability task 2Water9000307Air puff0080 We devised several different tasks to characterize dopamine activities in response to mild aversive air puff ( Table 1 ) . 'Mixed prediction task' ( low reward context ) was designed to examine interaction between the prediction of reward and the prediction of aversiveness . 'Low reward probability task' ( low reward context ) and 'high reward probability task' ( high reward context ) were specifically designed to test the effects of reward probability on dopamine responses: two task conditions differed only with respect to reward probabilities . 'High reward probability task 2' ( high reward context ) was originally conducted to replicate the diverse responses to aversive stimuli in dopamine neurons , which were reported in multiple previous studies . The effects of reward contexts were also examined with the mixed prediction task and the high reward probability task 2 . In the present study , we first focused on the characterization of dopamine activities in low reward contexts ( Figures 2–4 ) . Then , we compared dopamine activities between low and high reward contexts ( Figure 5 ) . Finally , we examined dopamine activities in relation to behaviors in different contexts ( Figure 6 ) . 10 . 7554/eLife . 17328 . 005Figure 2 . Dopamine neurons integrate values of both valences , reward and aversion . ( A ) Task design in the mixed prediction task . ( B ) Mean ± S . E . M . of firing rates of optogenetically-identified dopamine neurons during all four trial conditions; reward ( blue ) , nothing ( black ) , punishment ( red ) , and both reward and punishment ( magenta ) . ( C ) Scatter plot of the mean responses during the CS epoch ( 0–1 s , indicated by a solid black line in B ) for reward versus nothing . The baseline firing rate ( −1–0 s from odor onset ) was subtracted for each neuron . Black filled circles indicate neurons with significant difference between responses to the CS predicting reward and that predicting nothing ( unpaired t test , p<0 . 05 ) . ( D ) Scatter plot of the mean responses during the CS epoch for nothing versus punishment . Black filled circles indicate neurons with significant difference between responses to the CS predicting nothing and that predicting punishment . ( E ) Comparison of the responses of individual neurons ( n = 26 ) during CS ( 0–1 s ) predicting reward ( blue ) , nothing ( black ) and punishment ( red ) . For all box plots , the central mark is the median , the edges of the box are the 25th and 75th percentiles , and the whiskers extend to the most extreme data points not considered outliers ( points 1 . 5 × interquartile range away from the 25th or 75th percentile ) , and outliers are plotted individually as plus symbols . t ( 25 ) = 3 . 7 , p=0 . 001 , paired t test . One outlier >5 Hz in response to Odor A is not represented . ( F ) Scatter plot of the mean responses during the CS epoch for reward versus reward and punishment . Black filled circles indicate neurons with significant difference between responses to the CS predicting reward and that predicting reward and punishment . ( G ) Scatter plot of the mean responses during the CS epoch for reward and punishment versus punishment . Black filled circles indicate neurons with significant difference between responses to the CS predicting punishment and that predicting reward and punishment . ( H ) Comparison of the responses during CS predicting reward ( blue ) , both reward and punishment ( magenta ) , and punishment ( red ) . 1t ( 25 ) = 2 . 5 , p=0 . 02; t ( 25 ) = 4 . 4 , p=2 . 0 × 10−4 , paired t test . One outlier >5 Hz in response to Odor A is not represented . DOI: http://dx . doi . org/10 . 7554/eLife . 17328 . 00510 . 7554/eLife . 17328 . 006Figure 2—figure supplement 1 . Comparison of CS responses using dopamine neurons from two animals instead of three . ( A–C ) Comparison of the responses of individual dopamine neurons ( A , n = 18; B , n = 18; and C , n = 16 from two mice ) during CS ( 0–1 s ) predicting reward ( blue ) , nothing ( black ) and punishment ( red ) . For all box plots , the central mark is the median , the edges of the box are the 25th and 75th percentiles , and the whiskers extend to the most extreme data points not considered outliers ( points 1 . 5 × interquartile range away from the 25th or 75th percentile ) , and outliers are plotted individually as plus symbols . One outlier >5 Hz in response to Odor A is not represented in A and B . DOI: http://dx . doi . org/10 . 7554/eLife . 17328 . 00610 . 7554/eLife . 17328 . 007Figure 3 . Dopamine neurons signal aversive prediction error . ( A ) Mean ± S . E . M . of firing rate of dopamine neurons in response to predicted ( red ) and unpredicted air puff ( purple ) . ( B ) Scatter plot of the responses to predicted and unpredicted air puff ( 0–600 ms after air puff , indicated by a black solid line in A ) . Each data point represents an individual dopamine neuron . The baseline firing rate ( −1–0 s from odor onset ) was subtracted for each neuron . Black filled circles indicate neurons with significant difference between responses to unpredicted and predicted air puff ( unpaired t test , p<0 . 05 ) . ( C ) Comparison of the responses to predicted and unpredicted air puff ( n = 38 ) . For all box plots , central mark is the median , box edges are 25th and 75th percentiles , whiskers extend to the most extreme data points not considered outliers ( points 1 . 5 × interquartile range away from the 25th or 75th percentile ) , and outliers are plotted as plus symbols . t ( 37 ) = 2 . 4 , p=0 . 02 , paired t test . ( D ) Histogram of changes in firing rate during the US epoch ( 0–600 ms ) of predicted versus unpredicted air puff . The population average of the auROC curve was significantly different from 0 . 5 ( n = 38 , t ( 37 ) = 2 . 5 , p=0 . 015 , one-sample t test ) . Red arrow indicates mean auROC value . ( E ) Mean ± S . E . M . of firing rate around the outcome period in air puff omission ( red ) and nothing ( black ) trials . ( F ) A scatter plot of the firing rate during the outcome period in air puff omission trials and nothing trials ( 0–1000 ms after US onset , indicated by a black solid line in E ) subtracted by the baseline firing rate ( −1–0 s from odor onset ) for each neuron . Black filled circles indicate neurons with significant difference between firing rates during the outcome period in air puff omitted and nothing trials . ( G ) Comparison of the responses during air puff omission trial and nothing trial conditions . t ( 37 ) = 2 . 8 , p=0 . 008 , paired t test . ( H ) Histogram of changes in firing rate during the US epoch ( 0–1000 ms ) of air puff omission versus nothing trials . The population average of auROC curve was significantly different from 0 . 5 ( n = 38 , t ( 37 ) = 2 . 7 , p=0 . 011 , one-sample t test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17328 . 00710 . 7554/eLife . 17328 . 008Figure 3—figure supplement 1 . Reward prediction error coding by dopamine neurons in low reward contexts . ( A ) Mean ± S . E . M . of firing rate of dopamine neurons in response to predicted ( blue ) and unpredicted water ( cyan ) . Data were collected from two different low reward probability tasks—the mixed prediction task and the low reward probability task . ( B ) Scatter plot of the responses to predicted and unpredicted water ( 0–600 ms after water , indicated by a black solid line in A ) . Each data point represents an individual dopamine neuron ( n = 37 ) . The baseline firing rate ( −1–0 s from odor onset ) was subtracted for each neuron . Black filled circles indicate neurons with significant difference between response to unpredicted water and that to predicted water ( unpaired t test , p<0 . 05 ) . ( C ) Comparison of the responses to predicted and unpredicted water ( n = 37 dopamine neurons ) . Central mark is the median , box edges are 25th and 75th percentiles , whiskers extend to the most extreme data points not considered outliers ( points 1 . 5 × interquartile range away from the 25th or 75th percentile ) , and outliers are plotted as plus symbols . *t ( 36 ) = 4 . 5 , p=7 . 6 × 10−5 , paired t test . ( D ) Histogram of changes in firing rate during the US epoch ( 0–600 ms ) of predicted versus unpredicted water . The population average of auROC curve was significantly different from 0 . 5 ( n = 37 , t ( 36 ) = 4 . 4 , p=8 . 1 × 10−5 , one-sample t test ) . Red arrow indicates mean auROC value . DOI: http://dx . doi . org/10 . 7554/eLife . 17328 . 00810 . 7554/eLife . 17328 . 009Figure 4 . Correlation between responses related to aversive stimuli in dopamine neurons . ( A ) Scatter plot of the responses to unpredicted air puff ( 0–600 ms from air puff onset ) and the effects of prediction on the responses to air puff US ( subtraction of responses to unpredicted air puff from responses to predicted air puff , 0–600 ms from air puff onset ) in dopamine neurons . Each data point represents an individual dopamine neuron ( n = 38 ) . The baseline firing rate ( −1–0 s from odor onset ) was subtracted for each neuron . Solid line , best-fit linear regression . Pearson’s correlation , r = 0 . 69 , p=1 . 9 × 10−6 . ( B ) Scatter plot of the responses to CS predicting air puff ( 0–1000 ms from odor onset ) and the effects of prediction on the responses to air puff in dopamine neurons ( n = 38 ) . Pearson’s correlation , r = 0 . 39 , p=0 . 016 . ( C ) Scatter plot of the responses of individual dopamine neurons ( n = 37 ) to unpredicted water and air puff ( 0–600 ms from water and air puff onsets , respectively ) . No correlation between these two responses ( Pearson’s correlation , r = -0 . 04 , p=0 . 834 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17328 . 00910 . 7554/eLife . 17328 . 010Figure 4—figure supplement 1 . Correlation between responses related to air puff omission and the effects of prediction on the responses to air puff US . Scatter plot of the firing rate during the outcome period in air puff omission trials ( 0–1 s after air puff onset ) subtracted by the baseline firing rate ( −1–0 s from odor onset ) and the effects of prediction on the responses to air puff ( subtraction of responses to unpredicted air puff from responses to predicted air puff , 0–600 ms from air puff onset ) . Each data point represents an individual dopamine neuron ( n = 38 ) . Pearson’s correlation , r = -0 . 38 , p=0 . 02 . Solid line , best-fit linear regression . DOI: http://dx . doi . org/10 . 7554/eLife . 17328 . 01010 . 7554/eLife . 17328 . 011Figure 5 . Representation of negative value of aversive stimuli depends on reward context . ( A ) Task design in the high reward probability task . ( B ) Mean ± S . E . M . of firing rate of optogenetically-identified dopamine neurons during two trial conditions; nothing ( black ) and punishment ( red ) . ( C ) Scatter plot of the mean responses during the CS epoch ( 0–1 s , indicated by a solid black line in B ) for punishment versus nothing . The baseline firing rate ( −1–0 s from odor onset ) was subtracted for each neuron . Filled grey circles ( 11 out of 17 circles ) , dopamine neurons showing significant inhibition to punishment CS than the average baseline firing rate ( n = 80 trials , p<0 . 05 , one-sample t test ) . ( D ) Comparison of the responses of individual neurons during CS ( 0–1 s ) predicting nothing ( black ) and punishment ( red ) . For all box plots , central mark is the median , box edges are 25th and 75th percentiles , whiskers extend to the most extreme data points not considered outliers ( points 1 . 5 × interquartile range away from the 25th or 75th percentile ) , and outliers are plotted as plus symbols . t ( 16 ) = 2 . 1 , p>0 . 05 , paired t test . n . s . , not significant . ( E ) Histogram of changes in firing rate during the CS epoch ( 0–1 s ) of nothing versus punishment . The population average of the area under the receiver-operating characteristic ( auROC ) curve was not significantly different from 0 . 5 ( n = 17 , t ( 16 ) = 1 . 9 , p>0 . 05 , one-sample t test ) . Red arrow indicates mean auROC value . ( F ) Task design in the low reward probability task . ( G ) Mean ± S . E . M . of firing rate of optogenetically-identified dopamine neurons during two trial conditions; nothing ( black ) and punishment ( red ) . ( H ) Scatter plot of the mean responses during the CS epoch ( 0–1 s , indicated by a solid black line in G ) for punishment versus nothing . Filled grey circles ( 12 out of 12 circles ) , dopamine neurons showing significant inhibition to punishment CS than the average baseline firing rate ( n = 80 trials , p<0 . 05 , one-sample t test ) . ( I ) Comparison of the responses of individual neurons during CS ( 0–1 s ) predicting nothing ( black ) and punishment ( red ) . t ( 11 ) = 3 . 8 , p=0 . 003 , paired t test . ( J ) Histogram of changes in firing rate during the CS epoch ( 0–1 s ) of nothing versus punishment . The population average of the auROC curve was significantly different from 0 . 5 ( n = 12 , t ( 11 ) = 4 . 2 , p=0 . 002 , one-sample t test ) . ( K ) Comparison of the percentage of dopamine neurons showing significant inhibition to punishment CS than baseline ( Air puff CS < Baseline ) between high and low reward probability tasks . chi ( 1 ) = 5 . 34 , p=0 . 02 , chi-square test . Error bar , S . E . M . DOI: http://dx . doi . org/10 . 7554/eLife . 17328 . 01110 . 7554/eLife . 17328 . 012Figure 5—figure supplement 1 . The response to air puff-predicting CS was consistently smaller than that to nothing-predicting CS in low reward contexts . ( A–F ) Comparisons of responses of individual dopamine neurons during CS ( 0–1 s ) predicting nothing ( black ) and punishment ( red ) in high reward probability task 2 ( A , n = 17 , t ( 16 ) = 0 . 35 , p=0 . 73 , paired t test ) , high reward probability task ( B , n = 17 , t ( 16 ) = 2 . 1 , p=0 . 05 ) , high reward contexts ( data collected from two high reward probability tasks; C , n = 34 , t ( 33 ) = 1 . 5 , p=0 . 14 ) , mixed prediction task ( D , n = 26 , t ( 25 ) = 3 . 7 , p=1 . 2 × 10−3 ) , low reward probability task ( E , n = 12 , t ( 11 ) = 3 . 8 , p=2 . 8 × 10−3 ) , and low reward contexts ( data pooled from mixed prediction task and low reward probability task; F , n = 38 , *t ( 37 ) = 5 . 0 , p=1 . 5 × 10−5 ) . For all box plots , central mark is the median , box edges are 25th and 75th percentiles , whiskers extend to the most extreme data points not considered outliers ( points 1 . 5 × interquartile range away from the 25th or 75th percentile ) , and outliers are plotted as plus symbols . n . s . , not significant . ( G ) Comparison of the percentage of dopamine neurons showing significant inhibition to punishment CS than baseline ( Air puff CS < Baseline ) between high and low reward contexts ( n = 34 and 38 , respectively ) . chi ( 1 ) = 7 . 25 , p=7 . 1 × 10−3 , chi-square test . Error bar , S . E . M . ( H ) Comparison of the percentage of dopamine neurons showing significant inhibition to punishment CS than nothing CS ( Air puff CS < Nothing CS ) between high and low reward contexts ( n = 34 and 38 , respectively ) . chi ( 1 ) = 4 . 57 , p=0 . 03 , chi-square test . DOI: http://dx . doi . org/10 . 7554/eLife . 17328 . 01210 . 7554/eLife . 17328 . 013Figure 5—figure supplement 2 . The later response to air puff-predicting CS was smaller than that to nothing-predicting CS in the low reward probability task . ( A–B ) Comparison of the later responses ( 200–1000 ms after odor onset ) of individual neurons to CSs predicting nothing ( black ) and punishment ( red ) in different reward probability tasks ( A , n = 12 dopamine neurons in the low reward probability task; B , n = 17 in the high reward probability task ) . For all box plots , the central mark is the median , the edges of the box are the 25th and 75th percentiles , and the whiskers extend to the most extreme data points not considered outliers ( points 1 . 5 × interquartile range away from the 25th or 75th percentile ) , and outliers are plotted individually as plus symbols . t ( 11 ) = 4 . 0 , p=2 . 0 × 10−3; and t ( 16 ) = 2 . 0 , p=0 . 07 , paired t test . n . s . , not significant . ( C ) Comparison of the percentage of dopamine neurons showing significant inhibition to punishment CS than nothing CS ( Air puff CS < Nothing CS ) between high and low reward probability tasks . chi ( 1 ) = 4 . 44 , p=0 . 04 , chi-square test . Error bar , S . E . M . DOI: http://dx . doi . org/10 . 7554/eLife . 17328 . 01310 . 7554/eLife . 17328 . 014Figure 5—figure supplement 3 . Scatter plot of responses of individual dopamine neurons to air puff-predicting CS and unpredicted air puff in high reward contexts . Scatter plot of the responses to air puff-predicting CS ( 0–1 s from odor onset ) and unpredicted air puff US ( 0–600 ms from air puff onset ) . Each data point represents an individual dopamine neuron ( n = 34 ) . The baseline firing rate ( −1–0 s from odor onset ) was subtracted for each neuron . DOI: http://dx . doi . org/10 . 7554/eLife . 17328 . 01410 . 7554/eLife . 17328 . 015Figure 6 . Relation between CS response of dopamine neurons and behavior . ( A ) Eye blinking behavior during all three trial conditions in an example session in the high reward probability task . Red color indicates small eye area . ( B ) Average eye area during all three trial conditions in the example session; reward ( blue ) , nothing ( black ) and punishment ( air puff , red ) . ( C ) Comparison of the eye area during baseline ( −1–0 s from odor onset , black ) and delay period ( 1–2 s , red ) in punishment trial condition from an example animal ( n = 21 sessions ) . For all box plots , central mark is the median , box edges are 25th and 75th percentiles , whiskers extend to the most extreme data points not considered outliers ( points 1 . 5 × interquartile range away from the 25th or 75th percentile ) , and outliers are plotted as plus symbols . *t ( 20 ) = 14 . 7 , p=3 . 4 × 10−12 , paired t test . ( D ) Comparison of the responses of individual dopamine neurons ( n = 23 ) during punishment CS ( 0–1 s ) between blink ( dark red ) and no-blink trials ( magenta ) . The baseline firing rate ( −1–0 s from odor onset ) was subtracted for each neuron . t ( 22 ) = 3 . 0 , p=0 . 007 , paired t test . ( E ) Comparison of the responses of individual dopamine neurons during punishment CS ( 0–1 s ) in blink and no-blink trials in high and low reward probability tasks ( n = 13 and 10 neurons from these two tasks , respectively ) . The baseline firing rate ( −1–0 s from odor onset ) was subtracted for each neuron . **t ( 9 ) = 7 . 4 , p=4 . 3 × 10−5 , one-sample t test;t ( 9 ) = 3 . 0 , p=0 . 016; and t ( 12 ) = 1 . 3 , p=0 . 207 , paired t test . n . s . , not significant . ( F ) Comparison of the responses of individual dopamine neurons during reward-predicting CS ( 0–1 s ) between trials with anticipatory and no anticipatory licks ( ≥3 and <3 licks s−1 during delay period , respectively ) . In 90% water trials ( left ) , only 9 out of 34 dopamine neurons were collected , as the number of trials in which the animal did not show anticipatory licking was small . In 20–25% water trials ( right ) , 33 out of 38 dopamine neurons were collected . The baseline firing rate ( −1–0 s from odor onset ) was subtracted for each neuron . 1t ( 8 ) = 2 . 322 , p=0 . 0488 , paired t test; 2t ( 8 ) = 3 . 141 , p=0 . 0138 , one-sample t test . DOI: http://dx . doi . org/10 . 7554/eLife . 17328 . 01510 . 7554/eLife . 17328 . 016Figure 6—figure supplement 1 . Extraction of eye area from video frames . ( A ) A histogram showing the distribution of pixels with different intensity ( 0–255 levels , bottom ) in an example frame of the right eye region ( top ) . A threshold 234 pixel intensity was used to separate eye area from the background . ( B ) Pixels with intensity smaller than the eye threshold was set to 1 ( white ) and others were set to 0 ( black ) . ( C ) White small patches outside of the eye on the binary image in B were removed . ( D ) Smoothing the eye patch . ( E ) The averaged pixel intensities of example video frames were plotted . Troughs of the value indicate short infrared light source off ( <25 ms ) delivered 2 s after every US onset . DOI: http://dx . doi . org/10 . 7554/eLife . 17328 . 01610 . 7554/eLife . 17328 . 017Figure 6—figure supplement 2 . All the animals showed anticipatory eye-blinking to air puff in both low and high reward probability conditions . ( A–H ) Comparison of the eye area during baseline ( −1–0 s from odor onset , black ) and delay period ( 1–2 s from odor onset , red ) in punishment ( air puff ) trial condition . For all box plots , central mark is the median , box edges are 25th and 75th percentiles , whiskers extend to the most extreme data points not considered outliers ( points 1 . 5 × interquartile range away from the 25th or 75th percentile ) , and outliers are plotted as plus symbols . Each boxplot represents data from each animal ( 5 from the high reward probability task , and 3 from the low reward probability task ) . *t ( 21 ) = 3 . 9 , p=8 . 5 × 10−4 ( n = 22 sessions ) , paired t test in A; t ( 17 ) = 6 . 3 , p=8 . 4 × 10−6 ( n = 18 sessions ) , paired t test in B; U = 342 , p=0 . 02 ( n = 32 sessions ) , Mann-Whitney U test in C; **t ( 20 ) = 14 . 7 , p=3 . 4 × 10−12 ( n = 21 sessions ) , paired t test in D; t ( 16 ) = 2 . 9 , p=0 . 01 ( n = 17 sessions ) , paired t test in E; *t ( 15 ) = 4 . 2 , p=7 . 0 × 10−4 ( n = 16 sessions ) , paired t test in F; t ( 25 ) = 8 . 2 , p=1 . 4 × 10−8 ( n = 26 sessions ) , paired t test in G; and *t ( 18 ) = 5 . 2 , p=5 . 6 × 10−5 ( n = 19 sessions ) , paired t test in H . ( I ) Comparison of the eye area during delay period ( 1–2 s after odor onset ) in all three trial conditions; reward ( blue ) , nothing ( black ) and punishment ( red ) . Data were from 8 mice . 1t ( 7 ) = 3 . 6 , p=0 . 009; and 2t ( 7 ) = 3 . 2 , p=0 . 01 , paired t test . DOI: http://dx . doi . org/10 . 7554/eLife . 17328 . 01710 . 7554/eLife . 17328 . 018Figure 6—figure supplement 3 . Correlation between eye-blinking behavior and inhibitory responses to air puff-predicting CS during the later response window in low reward contexts . Comparison of the responses of individual dopamine neurons during late period of punishment CS ( 200–1000 ms ) in the high and low reward probability tasks ( n = 13 and 10 neurons from these two tasks , respectively ) . The baseline firing rate ( −1–0 s from odor onset ) was subtracted for each neuron . For all box plots , central mark is the median , box edges are 25th and 75th percentiles , whiskers extend to the most extreme data points not considered outliers ( points 1 . 5 × interquartile range away from the 25th or 75th percentile ) , and outliers are plotted as plus symbols . *t ( 9 ) = 7 . 0 , p=6 . 4 × 10−5 , one-sample t test; t ( 9 ) = 3 . 0 , p=0 . 016; and t ( 12 ) = 0 . 22 , p=0 . 83 , paired t test . n . s . , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 17328 . 01810 . 7554/eLife . 17328 . 019Figure 6—figure supplement 4 . Anticipated licking behavior during delay period . ( A ) Histograms of the lick rate during the delay period ( 1–2 s from odor onset ) during 90% water trials ( blue ) and nothing trials ( black ) from all the sessions in which dopamine neurons were identified . ( B ) Comparison of the lick rates during the delay period between 90% water trials and 20–25% water trials ( n = 34 and 38 sessions , respectively ) . Central mark is the median , box edges are 25th and 75th percentiles , whiskers extend to the most extreme data points not considered outliers ( points 1 . 5 × interquartile range away from the 25th or 75th percentile ) , and an outlier is plotted as plus symbols . ***t ( 70 ) = 8 . 8 , p=5 . 3 × 10−13 , unpaired t test . DOI: http://dx . doi . org/10 . 7554/eLife . 17328 . 019 A previous study reported that dopamine neurons do not integrate information about aversiveness along with reward-related information when rewarding liquid and an air puff are delivered to a monkey at the same time ( Fiorillo , 2013 ) . However , this method may produce complex interactions between the two different outcomes . To test how reward and aversion interact and affect dopamine responses , we devised a 'mixed prediction' paradigm ( Figure 2 ) in which a single odor ( Odor D in Figure 2A , conditioned stimulus , CS ) predicted both a rewarding and a mildly aversive event in a complementary and probabilistic manner: a reward ( water ) was delivered in 25% of the trials and an aversive event ( air puff ) was delivered in the remaining 75% of the trials . For comparison , we included the following trial types: Odor A predicted water in 25% of trials ( nothing in 75% ) , Odor C predicted air puff in 75% of trials ( nothing in 25% ) , and Odor B predicted no outcome . Each behavioral trial began with the odor CS ( 1 s ) , followed by a 1-s delay and an unconditioned stimulus ( US ) . We chose higher probability for air puff than water in order to balance the strength of positive and negative values in the task; we suspect that the magnitude of the negative value of mild air puff is much smaller than the magnitude of the positive value of water , which could cause us to overlook a small effect of predicted air puff on the CS response . We first asked whether the recorded dopamine neurons were inhibited or excited by odor cues ( CSs ) that predicted different outcomes . We found that the vast majority of the neurons were inhibited by the air puff-predicting CS while excited by the reward-predicting CS ( Figure 2B–D ) . On average , the firing rate during the CS period was significantly lower for the air puff-predicting CS than for the CS predicting nothing , while it was higher for the reward-predicting CS than for the CS that predicted nothing ( Figure 2E ) . A similar tendency was observed using data from two animals instead of three ( i . e . leaving one animal out of three ) ( Figure 2—figure supplement 1 ) . Among 26 identified dopamine neurons , 85% ( 22 neurons ) were significantly modulated by these three odors ( p<0 . 05 , one-way ANOVA ) , and 59% ( 13 of 22 significant neurons ) showed the monotonic CS value coding ( water > nothing > air puff ) . These results suggest that the firing of identified dopamine neurons was negatively modulated by the stimulus predicting aversive events . We next examined whether prediction of aversion in addition to reward changed the response of dopamine neurons . In contrast to the previous study ( Fiorillo , 2013 ) , we found that the majority of neurons showed an intermediate response to the CS predicting both water and air puff ( Odor D ) compared to the CSs predicting water only ( Odor A ) or air puff only ( Odor C ) ( Figure 2B , F , G ) . As a population , the net response to these CSs increased monotonically according to the values of both water and air puff , with the CS response to Odor D falling in between that of Odor A and Odor C ( Figure 2H ) . 89% ( 23 of 26 neurons ) of identified dopamine neurons were significantly modulated by these three odors ( p<0 . 05 , one-way ANOVA ) , and 65% ( 15 of 23 significant neurons ) showed the monotonic CS value coding ( water > water and air puff > air puff ) . These results indicate that VTA dopamine neurons combine values for both reward and punishment along a one-dimensional value axis . It has been shown that dopamine neurons’ responses to reward are greatly reduced when the reward is predicted , a signature of prediction error coding ( Schultz et al . , 1997 ) . We replicated these findings here even in low reward probability conditions ( 20–25% , Figure 3—figure supplement 1; see Materials and methods ) . We next examined whether these dopamine neurons show prediction error coding for aversive events . To address this question , we occasionally delivered air puff during inter-trial intervals without any predicting cues . These responses to unpredicted air puff were compared to the responses to air puff in trials when air puff was predicted by an odor cue . We found that the inhibitory response to an air puff was significantly reduced when the air puff was predicted by an odor cue ( Figure 3A–D ) . To further examine whether dopamine neurons showed prediction error coding for aversive events , we compared the firing rate during the outcome period in air puff omission trials with that in trials that predict nothing . We found that the omission of a predicted air puff slightly but significantly increased firing rates , compared to no change in nothing trials ( Figure 3E–H ) although we observed variability in air puff omission responses . Together , these results demonstrate that dopamine neurons signal prediction errors for aversive events in addition to rewarding events . These results indicate that dopamine neurons have the ability to signal VPEs for both appetitive and aversive events , supporting previous work by Matsumoto and Hikosaka ( Matsumoto and Hikosaka , 2009 ) and contrasting with previous work by Fiorillo ( Fiorillo , 2013 ) . Although we found that most dopamine neurons were inhibited by air puff ( mildly aversive event ) , there was a considerable variability in the extent to which individual dopamine neurons were inhibited . Does this diversity support a functional diversity across dopamine neurons in the lateral VTA ? In a previous study , dopamine neurons in the lateral VTA exhibited neuron-to-neuron variability in the magnitude of response to a given size of reward ( Eshel et al . , 2016 ) . Despite this variability in responsivity , the response functions of individual dopamine neurons were scaled versions of each other , indicating a remarkable homogeneity . One consequence of this scaled relationship is that neurons that responded strongly to a given size of reward were more greatly suppressed by reward expectation . In other words , reward expectation suppressed a neuron’s reward response in proportion to the size of its response to unexpected reward . Does the same relationship hold for inhibitory responses to air puff ? To address this question , we examined the correlation between aversion-related responses in dopamine neurons ( Figure 4 , Figure 4—figure supplement 1 ) . We indeed found a similar relationship: dopamine neurons that were strongly inhibited by air puff also exhibited a larger prediction-dependent reduction of their responses to air puff ( Figure 4A , Pearson’s r = 0 . 69 , p=1 . 9 × 10−6 ) . In other words , the ratio between individual dopamine neurons’ responses to unpredicted versus predicted air puff was preserved across neurons . In addition , similar to reward responses ( Eshel et al . , 2016 ) , inhibitory responses to the air puff-predicting CS were correlated with prediction-dependent reduction of responses to air puff US ( Figure 4B , Pearson’s r = 0 . 39 , p=0 . 016 ) . These results indicate that the response function was preserved across dopamine neurons in the case of aversive stimuli . We next examined the relationship between responses of dopamine neurons to reward and to aversion . We compared responses of dopamine neurons to unpredicted water and unpredicted air puff ( Figure 4C ) . We observed no obvious unique clusters across neurons , supporting the notion that there was no clear subpopulation of dopamine neurons specialized in signaling reward versus aversion in the lateral VTA . Rather , we found that most of dopamine neurons were inhibited by unpredicted aversive stimuli and excited by unpredicted rewarding stimuli . Interestingly , we did not find any negative or positive correlation of neurons’ responses to water and air puff; the proportion of the response magnitudes in response to reward versus aversion was diverse across neurons . These results indicate that although the response function either for reward prediction error or for aversion prediction error was homogeneous across dopamine neurons , these two functions were relatively independent , suggesting that different mechanisms may produce dopamine responses to reward and aversion . Although the above results suggested that most of the dopamine neurons that we recorded from the lateral VTA were inhibited by aversive events , contrasting results were obtained in some previous studies . In monkeys , it was found that , on average , the responses to aversive stimuli were indistinguishable from responses evoked by neutral stimuli ( Fiorillo , 2013 ) . In addition , previous studies in mice ( Cohen et al . , 2012; Tian and Uchida , 2015 ) mirrored these contrasting results in VTA dopamine neurons . These results suggest that the difference between studies is not due to a species difference , raising the possibility that our task parameters altered the dopamine response . Multiple studies found that even non-rewarding stimuli can excite dopamine neurons with short latency . A recent study reported that whether a neutral stimulus elicits these short-latency excitations depends on reward context , and that excitation is larger in a high versus a low reward context ( Kobayashi and Schultz , 2014 ) . We noted that the reward probability used in the task described above ( mixed prediction task ) was much lower ( 15% rewarded trials overall ) than in previous studies ( e . g . 50% rewarded trials overall in Cohen et al . , 2012 ) . This raises the possibility that in a high reward context , short-latency excitations to aversive stimuli masked inhibitory responses to aversive stimuli , and thus clear inhibition by aversive stimuli has not been observed in previous studies because these studies typically used relatively high reward probabilities . To directly test whether reward probability affected dopamine neurons’ responses to aversive events , we recorded the activity of dopamine neurons in two task conditions that differed only with respect to reward probabilities ( Figure 5 ) . In the high reward probability condition , the probability of water in Odor A trials was 90% ( 36% reward trials overall ) ( Figure 5A ) while in the low reward probability condition , the reward probability in Odor A trials was 20% ( 13% reward trials overall ) ( Figure 5F ) . Consistent with the previous study ( Kobayashi and Schultz , 2014 ) , nothing-predicting CS ( neutral cue ) elicited short-latency excitation more prominently in the high reward compared to the low reward probability condition ( Figure 5B , G ) . Next we examined the response to the air puff-predicting CS . In the low reward context , the difference between the responses to air puff- and nothing-predicting CSs remained significantly different ( Figure 5H–J ) , consistent with the above experiment ( Figure 2 ) . In contrast , we found that in the high reward probability condition , dopamine neurons exhibited biphasic responses to the air puff-predicting CS: short-latency excitation followed by later inhibition . Furthermore , dopamine neurons’ net responses to the air puff-predicting CS and nothing-predicting CS were no longer significantly different ( Figure 5C–E ) . This result , obtained in a high reward probability condition , is similar to those obtained in previous studies ( Cohen et al . , 2012; Fiorillo , 2013; Tian and Uchida , 2015 ) . In the low reward context , all the dopamine neurons ( 12 of 12 dopamine neurons; p<0 . 05 , one-sample t test; filled grey circles in Figure 5H ) were inhibited by the air puff-predicting CS whereas in the high reward context , a large fraction of dopamine neurons ( 6 of 17 dopamine neurons; p>0 . 05 , one-sample t test; filled white circles in Figure 5C ) did not show consistent inhibition by air puff CS . That is , in the high reward context , a little fraction of neurons showed a stronger inhibition to the air puff CS compared to the nothing CS ( 100% and 65% , low and high reward context , respectively; p=0 . 02 , chi-square test; Figure 5K ) . We obtained additional data using a task condition similar to the high reward context ( high reward probability task 2; see Materials and methods , and Table 1 ) ( n = 17 identified dopamine neurons ) . Furthermore , the mixed prediction task ( Figure 2 ) provides additional data for a low reward context ( n = 26 identified dopamine neurons ) . Similar results were obtained by using dopamine neurons in each of these experiments or by pooling neurons from these experiments separately for high and low reward contexts ( n = 34 and 38 identified dopamine neurons , respectively ) ( Figure 5—figure supplement 1 ) . The above analyses used a relatively large time window that contains the entire response period ( 0–1 , 000 ms ) . Because dopamine responses in high reward contexts exhibited biphasic responses ( early excitation followed by later inhibition ) , we further analyzed the data by separating these time windows into smaller bins . Because there is no known mechanism by which downstream neurons can read out these windows separately , analysis using a large window can be considered more conservative . However , previous studies have proposed that different information may be conveyed in these time windows ( Schultz , 2016; Stauffer et al . , 2016 ) . We obtained similar results even if we compared only later time bins ( 200–1 , 000 ms ) , excluding the early excitation phase ( Figure 5—figure supplement 2 ) . By excluding the early excitation period ( 0–200 ms ) , many dopamine neurons showed inhibition to air puff-predicting CS in both low and high reward contexts compared to the baseline firing ( 92% and 82% , respectively ) . However , during this inhibition phase , most dopamine neurons ( 65% ) did not distinguish the air puff CS from the nothing CS in the high reward context while most dopamine neurons ( 75% ) showed more inhibition to the air puff CS than to the nothing CS in the low reward context ( i . e . 35% and 75% of neurons distinguished air puff CS from nothing CS in high and low reward contexts , respectively; p=0 . 04 , chi-square test; Figure 5—figure supplement 2C ) . This suggests that although many dopamine neurons exhibit inhibitory responses during the later response window in high reward contexts , the information that this inhibition conveys may be different from that in low reward contexts; the inhibition in the high reward context largely reflected 'no reward' rather than the negative value of an air puff ( Fiorillo , 2013 ) . Neurons that showed a large excitation to the air puff CS were not necessarily the same group of neurons which showed excitation to the air puff itself , consistent with a previous study ( Matsumoto and Hikosaka , 2009 ) ( Figure 5—figure supplement 3 ) . These results demonstrate that dopamine neurons’ responses to aversion-predicting cues are greatly affected by reward contexts , and suggest that dopamine neurons’ ability to faithfully represent negative values of aversive cues are undermined in high reward contexts . In order to examine dopamine responses to aversive stimuli more carefully in relation to behavior , we characterized dopamine activities and anticipatory eye-blinking on a trial-by-trial basis . We quantified the area of the right eye ( including both sclera and pupil ) concurrently with neuronal recording in high and low reward probability contexts ( Figure 6; Figure 6—figure supplement 1; see also Materials and methods ) . We observed that the eye area became smaller after the onset of a CS predicting air puff and became larger after a CS predicting reward ( Figure 6A , B ) . In air puff trials , the eye area during the delay period was significantly smaller than before CS presentation ( −1–0 s from CS onset ) , indicating anticipatory eye-blinking ( Figure 6C , n = 21 sessions , p=3 . 4 × 10−12 , paired t test ) . We confirmed that in both low and high reward probability conditions , all of the animals showed significant anticipatory eye-blinking ( Figure 6—figure supplement 2 ) . The eye area during the delay period was significantly smaller in air puff trials than in nothing trials ( Figure 6—figure supplement 2I ) . These results indicate that our air puff conditions were aversive enough to cause anticipatory eye-blinking during the recording experiments , although we noticed that the amount of eye-blinking differed across trials . Because the level of anticipatory eye-blinking varied across trials , we next divided air puff trials in each session into two groups , 'blink' ( small eye size ) and 'no-blink' ( big eye size ) trials ( see Materials and methods ) , and then examined the correlation between blinking and the responses of dopamine neurons to the air puff CS . We found that the firing rates of dopamine neurons to air puff CS in blink trials were significantly smaller than that in no-blink trials ( Figure 6D ) . In other words , inhibition of dopamine neurons during the CS period , but not excitation , predicted aversion-related behavior . The correlation between trial-by-trial dopamine activity and anticipatory blinking was even clearer if we consider reward contexts ( Figure 6E ) . In the low reward probability condition , the inhibitory response of dopamine neurons in blink trials was significantly greater than in no-blink trials ( p=0 . 02 , paired t test ) . Of note , in the high reward probability condition , the inhibitory response of dopamine neurons was greatly reduced even when the animals showed anticipatory eye-blinking ( Figure 6E , p=0 . 009 , unpaired t test ) . This result suggests that dopamine responses may not directly trigger eye-blinking behavior . Rather , the results are consistent with the idea that dopamine neurons’ inhibitory responses to aversive cues signal negative values of the outcome , but not the action itself . Importantly , dopamine neurons showed significant inhibition only when animals showed anticipatory eye-blinking in the low reward context ( Figure 6E , p=4 . 3 × 10−5 , one-sample t test ) . The results do not change even when we only used a later window of dopamine CS responses , excluding the early excitation period ( 0–200 ms ) ( Figure 6—figure supplement 3 ) . Whereas mice showed anticipatory blinking in 48% of air puff trials ( 90% air puff trials ) , they showed more consistent anticipatory licking in water trials ( 98% in 90% water trials ) ( see Materials and methods ) . This could be due to the fact that we used only a mildly aversive air puff ( to prevent the animal from being discouraged to perform the task altogether ) , whereas water is highly rewarding . Although the number of trials in which the animal did not show anticipatory licking was small , we observed a similar relationship between dopamine responses and behavior: dopamine neurons were consistently excited by the 90% reward cue when they showed anticipatory licking , but not when they did not show anticipatory licking ( Figure 6F ) . These results indicate the importance of both reward contexts and behavioral outcomes to understand how dopamine neurons represent reward and aversion . Without monitoring behaviors , investigators may easily miss weak inhibitory responses to mildly aversive stimuli in dopamine neurons . Dopamine has long been thought to be a key regulator of reinforcement learning . One dominant theory posits that dopamine acts as a teaching signal that broadcasts an RPE signal to the rest of the brain . Recent studies using optogenetics have established that activation of dopamine neurons alone is sufficient for appetitive conditioning ( Steinberg et al . , 2013; Tsai et al . , 2009; Witten et al . , 2011 ) whereas suppression is sufficient for aversive conditioning ( Chang et al . , 2016; Danjo et al . , 2014; Ilango et al . , 2014; Tan et al . , 2012; van Zessen et al . , 2012 ) , although activation of dopamine neurons that project to the cortex or dopamine neurons in the dorsal raphe has potential to induce aversion and/or other functions ( Lammel et al . , 2014; Matthews et al . , 2016; Popescu et al . , 2016 ) . Furthermore , pharmacological studies have suggested that normal dopamine signaling is required for appetitive as well as aversive conditioning ( Cooper et al . , 1974; Flagel et al . , 2011; Wenzel et al . , 2015 ) . These results have provided convergent evidence supporting the role of dopamine in learning . However , whether dopamine neurons signal prediction errors with respect to aversive events remained controversial , and remained an obstacle towards establishing the role of dopamine as the teaching signal proposed in reinforcement learning theories . Comparatively fewer experiments have used combinations of aversive stimuli and rewarding stimuli to characterize the dopamine response . Fiorillo ( 2013 ) proposed that reward and aversion are processed separately in the brain , based on the observation that dopamine neurons signaled information about reward but largely ignored aversive events ( Fiorillo , 2013 ) . This result contradicts some previous studies that showed consistent inhibition of dopamine neurons by aversive stimuli or the cues that predict them ( Matsumoto and Hikosaka , 2009; McCutcheon et al . , 2012; Roitman et al . , 2008 ) . Our results suggest that there are different modes of dopamine signaling: in one mode , dopamine neurons indeed integrate the information about reward and aversion and signal VPE . This is an ideal teaching signal for reinforcement learning to maximize future values . Further , we found that the response function to aversive stimuli was preserved across dopamine neurons , suggesting that each dopamine neuron has the potential to contribute a prediction error of aversiveness , as well as of reward ( Eshel et al . , 2016 ) . However , in a high reward context , dopamine neurons largely lose their ability to signal integrated VPEs . Our results in high reward contexts are consistent with those observed previously ( Fiorillo , 2013 ) , and we also showed that similar results were obtained in previous recordings of optogenetically-identified dopamine neurons ( Cohen et al . , 2012; Tian and Uchida , 2015 ) . This raises the possibility that one of the apparent differences observed between previous electrophysiological studies is due to different experimental parameters with respect to reward contexts . It should be noted that many physiological experiments tend to include highly rewarding training sessions in order to motivate animals . In natural environments in which wild animals forage , rewards might not be as abundant as in these experimental conditions . Our results indicate that dopamine neurons signal VPE with high fidelity in low reward contexts . In examining the temporal dynamics of dopamine responses , we realized that on average , the peak of excitation for reward cues occurred earlier than the trough of inhibition for aversive cues . Interestingly , dopamine responses to the cue predicting both rewarding and aversive outcomes in our mixed prediction tasks first showed excitation and then inhibition , different from the flatter responses to nothing cues . This clear temporal difference raises the possibility that information about values from rewarding and aversive outcomes are not yet integrated in presynaptic neurons and arise from different sources of inputs to dopamine neurons . A recent electrophysiological recording study from monosynaptic inputs to dopamine neurons also suggested that different presynaptic neurons may convey values for rewarding versus aversive stimuli ( Tian et al . , 2016 ) . Consistent with this idea , we did not observe a correlation between the magnitude of single dopamine neurons’ responses to reward and aversiveness ( Figure 4C ) , in contrast to correlations within reward-related responses ( Eshel et al . , 2016 ) and within aversiveness-related responses ( Figure 4A , B ) . Of note , a previous study ( Eshel et al . , 2016 ) showed that , in high reward context , neurons that were highly responsive to unexpected rewards tended to also be highly responsive to aversive events: they showed greater levels of suppression below baseline . The results in the previous study are reminiscent of Fiorillo’s study , which found that the inhibitory phase in biphasic responses of dopamine neurons did not encode negative values of aversive stimulus , but rather encode 'no reward' ( Fiorillo , 2013 ) . The results in the previous study ( Eshel et al . , 2016 ) could be consistent with the present results if the inhibition represents 'no reward' but not the negative value of aversive stimulus , and this no-reward response is correlated with other reward-related responses of dopamine neurons . Our results indicate that dopamine neurons represent aversive information in a reward context dependent manner . Our results are consistent with a previous study which proposed that dopamine neurons change their response patterns depending on reward context ( Kobayashi and Schultz , 2014 ) . The authors found that neutral stimuli excited dopamine neurons more strongly in high- compared to low-reward contexts ( Kobayashi and Schultz , 2014 ) . The present study extends this finding to aversive stimuli . Short latency excitations of dopamine neurons have been observed in various experiments and have been attributed to generalization ( Kobayashi and Schultz , 2014; Mirenowicz and Schultz , 1996 ) , stimulus intensities ( Fiorillo et al . , 2013 ) , motivational salience ( Bromberg-Martin et al . , 2010; Matsumoto and Hikosaka , 2009 ) , trial starts ( Bromberg-Martin et al . , 2010 ) or stimulus detection ( Nomoto et al . , 2010 ) . Our data do not distinguish these possibilities and the short latency excitation in the high reward context is likely to comprise a combination of these . Importantly , however , our data in the high reward context showed that the short-latency excitations compromised the monotonic value coding of dopamine neurons , and the difference between responses to air puff-predicting CS and nothing-predicting CS was diminished . This means that dopamine neurons did not simply add a constant amount of spikes ( the same amount of excitation ) on top of the monotonic value coding . Thus , our observations suggest that the combination of these factors and/or additional factors distorted normal value coding in dopamine neurons in high reward context . In our experiments , high- and low-reward contexts differed with respect to the probability of rewarded trials . This suggests that dopamine responses to aversion depend on the frequency of reward , which may in turn change the animal’s state . It remains to be examined how the frequency of rewards changes dopamine responses and whether dopamine responses could be modulated by other manipulations of the environment such as the amount of reward or the strength or frequency of aversive events . In addition to overall reward contexts , we found that the inhibitory responses of dopamine neurons changed on a finer time-scale; the inhibition of dopamine neurons by air puff-predicting cues was correlated with trial-by-trial variability of aversion-related behaviors in low reward contexts . A similar correlation was observed between excitation of dopamine neurons by reward-predicting cues and reward-related behaviors . These results suggest that dopamine neurons track the predictions of values ( reward and aversiveness ) which may reflect animals’ states over various time-scales . A previous study also examined dopamine responses to aversive stimuli in relation to behaviors . Using cyclic voltammetry , the authors showed that , in response to electrical shock-predicting cues , the dopamine concentration in the ventral striatum increased when the rats exhibited an active avoidance behavior while it decreased when the rats showed freezing behavior ( Oleson et al . , 2012 ) . It is therefore proposed that dopamine responses depend on whether the animal exhibits active avoidance or passive reaction ( Oleson et al . , 2012; Wenzel et al . , 2015 ) . In the present study , we found that the degree of inhibition , not excitation , of dopamine neurons in response to the air puff-predicting CS was positively correlated with anticipatory eye-blinking behaviors ( Figure 6D ) . According to the above idea ( Oleson et al . , 2012; Wenzel et al . , 2015 ) , the anticipatory eye-blinking that we observed may be categorized as a passive avoidance behavior , which could be the reason as to why we observed inhibition , but not excitation of dopamine neurons correlated with anticipatory eye-blinking behaviors . Whereas dopamine neurons displayed a relatively uniform response function to aversion in low reward contexts , we observed diverse responses in high reward contexts , including some inhibitory and some excitatory responses to aversive events . What caused diverse responses to aversion in high reward contexts ? Increasing evidence supports the diversity of dopamine neurons depending on the location of the cell body and projection targets ( Lammel et al . , 2014; Roeper , 2013 ) . For example , it was reported that neurons in the lateral SNc signal salience ( Lerner et al . , 2015; Matsumoto and Hikosaka , 2009 ) or 'stable value' as opposed to 'flexible value' in the medial SNc ( Kim et al . , 2015 ) . Another study showed that dopamine neurons in the ventromedial VTA exhibited excitation to an aversive stimulus ( Brischoux et al . , 2009 ) . Previous studies showed that responses to aversive stimuli are diverse across dopamine neurons with different projection targets ( Lammel et al . , 2011; Lerner et al . , 2015 ) . Although the majority of dopamine neurons in the lateral VTA , our main recording site , project to the ventral and anterior dorsal striatum ( Lammel et al . , 2008; Menegas et al . , 2015 ) , our study did not distinguish the exact projection targets of dopamine neurons . It remains to be determined which subpopulations of dopamine neurons switch signaling modes depending on low versus high reward contexts . Our results demonstrated the importance of considering global contexts and behaviors and of unambiguously identifying dopamine neuron . It remains to be examined in future studies how reward frequency changes both the animal’s state and dopamine responses to punishment , and how these changes relate to our normal and abnormal behaviors . Further , there are complex temporal dynamics and diversity of dopamine activities . Considering these factors together is challenging but represents a firm step towards fully understanding the nature and function of dopamine signals . We used 15 adult male mice heterozygous for Cre recombinase under control of the <Slc6a3> gene that encodes the dopamine transporter ( DAT ) ( B6 . SJL-Slc6a3tm1 . 1 ( cre ) Bkmn/J , Jackson Laboratory; RRID:IMSR_JAX:006660 ) ( Bäckman et al . , 2006 ) . All mice were backcrossed with C57/BL6 mice . Eleven out of 15 mice were further crossed with tdTomato-reporter mice ( Gt ( ROSA ) 26Sortm9 ( CAG-tdTomato ) Hze , Jackson Laboratory ) to express tdTomato in dopamine neurons . Electrophysiological data were collected from 14 mice , and video data were from 8 mice . Animals were singly housed on a 12 hr dark/12 hr light cycle ( dark from 06:00 to 18:00 ) and each performed the conditioning task at the same time of day , between 08:00 and 16:00 . All procedures were approved by Harvard University Institutional Animal Care and Use Committee . Total 1 µl of adeno-associated virus ( AAV ) , serotype 5 , carrying an inverted ChR2 ( H134R ) -EYFP flanked by double loxP sites ( Atasoy et al . , 2008 ) [AAV5-DIO-ChR2-EYFP ( Tsai et al . , 2009 ) ] was injected stereotactically into the VTA ( 3 . 1 mm posterior to bregma , 0 . 5 mm lateral , 3 . 9 mm deep from dura and 3 . 5 mm posterior to bregma , 0 . 5 mm lateral , 4 . 2 mm deep from dura ) . We previously showed that expression of this virus in dopamine neurons is highly selective and efficient ( Cohen et al . , 2012 ) . After > 1 week from virus injection , a custom-made metal plate ( a head plate ) was implanted . A microdrive containing electrodes and an optical fiber was implanted in the VTA stereotactically in the same surgery . All the surgeries were performed under aseptic conditions under isoflurane inhalation anesthesia ( 1–2% at 0 . 8–1 . 0 L min−1 ) . The animals were given analgesics ( ketoprofen , 1 . 3 mg kg−1 intraperitoneal , and buprenorphine , 0 . 1 mg kg−1 intraperitoneal ) postoperatively . After >1 week of recovery , mice were water-deprived in their home cage . Animals were habituated for 1–2 days with the head restrained by a head plate before training on the task . Odors were delivered with a custom-designed olfactometer ( Uchida and Mainen , 2003 ) . Each odor was dissolved in mineral oil at 1:10 dilution . 30 µl of diluted odor was placed inside a filter-paper housing ( Thomas Scientific , Swedesboro , NJ ) . Odors were selected pseudorandomly from isoamyl acetate , eugenol , 1-hexanol , citral , and 4-heptanone for each animal . Odorized air was further diluted with filtered air by 1:8 to produce a 900 ml min−1 total flow rate . We delivered an odor for 1 s , followed by 1 s of delay and an outcome . Trials were pseudorandomly interleaved . In the mixed prediction task , during initial training period ( 1–3 days ) , an odor ( not used for Odors A–D ) preceded a drop of water ( 4 μl ) and Odor B preceded no outcome . After the initial training , Odors A–D were paired with water with 25% probability , no outcome ( 100% nothing ) , air puff with 75% probability , and water with 25% probability and air puff with remaining 75% probability . In high and low reward probability tasks , 2 odors ( not used for Odors A–C ) preceded a drop of water and no outcome , respectively , during initial training period . Later , Odor A was paired with water with 90% probability in the high reward probability task , and 20% probability in the low reward probability task . The probabilities of no outcome in Odor B trials and air puff in Odor C trials were 100% and 90% , respectively , in both reward probability tasks . In high reward probability task 2 , 2 odors ( Odor A and Odor B ) preceded a drop of water and no outcome , respectively , during initial training period . Later , Odor A was paired with water with 90% probability and Odor C was paired with air puff with 80% probability . Air puff was delivered to the animal's right eye . The strength of air puff was enough to cause anticipated eye-blinking behavior . In order to control any sounds caused by air puff accumulation , the air was accumulated at the offset of odor delivery in all the trial types and released at 2 . 3 s after the offset of odor delivery outside of a hemi-soundproof behavioral box except for air puff trials . Licks were detected by breaks of an infrared beam placed in front of the water tube . To quantify eye-blinking behavior trial-by-trial , animal’s face including right eye was recorded by a CCD camera ( Point Grey ) . The sampling rate was 60 Hz . To monitor animal’s eye under dark conditions , we put infrared light sources inside the behavior box . To synchronize the video frames with event time stamps for further analysis , the infrared light source was turned off briefly ( <25 ms ) 2 s after the onset of US . We also added no-odor trials ( 4% of the trials in the mixed prediction task , 12% in high and low reward probability tasks , and 13% in the high reward probability task 2 ) in which either water reward or air puff was presented unpredictably . Inter-trial intervals ( ITIs ) were drawn from an exponential distribution , resulting in a flat ITI hazard function truncated at 15 s such that expectation about the start of the next trial did not increase over time . Data in the mixed prediction task were obtained from 63 sessions ( 19–25 sessions per animal , 21 ± 3 sessions; mean ± S . D . , n = 3 mice ) ; data in the high reward probability task were obtained from 38 sessions ( 2–19 sessions per animal , 10 ± 8 sessions , n = 4 mice ) ; data in the low reward probability task were obtained from 20 sessions ( 2–10 sessions per animal , 7 ± 4 sessions , n = 3 mice ) ; data in the high reward probability task 2 were obtained from 39 sessions ( 1–22 sessions per animal , 10 ± 9 sessions , n = 4 mice ) . The animals performed between 208 and 476 trials per day ( 371 ± 81 trials; mean ± S . D . ) in the mixed prediction task , 272 trials per day in both high and low reward probability tasks , and between 182 and 454 trials per day ( 322 ± 42 trials; mean ± S . D . ) in the high reward probability task 2 . We recorded extracellularly from multiple neurons simultaneously using a custom-built 200-μm-fiberoptic-coupled screw-driven microdrive with six or eight implanted tetrodes ( four wires wound together ) . Tetrodes were glued to the fiber optic ( Thorlabs ) with epoxy ( Devcon ) . The ends of the tetrodes were 350–500 μm from the end of the fiber optic . Neural signals and time stamps for behavior were recorded using a DigiLynx recording system ( Neuralynx ) . Broadband signals from each wire filtered between 0 . 1 and 9000 Hz were recorded continuously at 32 kHz . To extract the timing of spikes , signals were band-pass-filtered between 300 and 6000 Hz . Spikes were sorted offline using MClust-3 . 5 software ( David Redish ) . At the end of each session , the fiber and tetrodes were lowered by 20–80 μm to record new neurons . Sessions of recordings were continued until the tetrodes reached the bottom of the brain where no units were recorded and large fluctuations of voltage traces were recorded from tetrodes . After the completion of the recording sessions , tetrodes were moved up to the depth where units were recorded or the depth where light-identified dopamine neurons were recorded to ensure that the following electrolytic lesions were in the brain . To verify that our recordings targeted dopamine neurons , we used ChR2 to observe stimulation-locked spikes ( Cohen et al . , 2012 ) . The optical fiber was coupled with a diode-pumped solid-state laser with analogue amplitude modulation ( Laserglow Technologies ) . Before and after each behavioral session , we delivered trains of 10 light pulses , each 5-ms long , at 1 , 2 , 5 , 10 , 20 and 50 Hz at 473 nm at 5–20 mW mm−2 . Spike shape was measured using a broadband signal ( 0 . 1–9000 Hz ) sampled at 30 kHz . This ensured that particular features of the spike waveform were not missed . We used two criteria to include a neuron in our data set . First , the neuron must have been well isolated [L-ratio < 0 . 05 ( Schmitzer-Torbert and Redish , 2004 ) , except for two units with L-ratio = 0 . 055 and 0 . 057] . Second , the neuron must have been recorded in or between the sessions when dopamine neurons were identified on the same tetrode to ensure that all neurons came from VTA . Recording sites were further verified histologically with electrolytic lesions using 5–20 s of 30 μA direct current and from the optical fiber track . Recording sites of individual dopamine neurons were reconstructed on the Franklin and Paxinos brain atlas ( Franklin and Paxinos , 2008 ) . The depths were estimated from the lesion site in each animal . To measure firing rates , peristimulus time histograms ( PSTHs ) were constructed using 1-ms bins . To calculate spike density functions , PSTHs were smoothed using a box filter ( 100 ms duration , t ± 50 ms ) . Average firing rates of responses to conditioned stimulus ( CS ) were calculated using a time window of 0–1000 or 200–1000 ms after odor onset . To obtain responses to the unconditioned stimulus ( US ) , we used a time window of 0–600 ms after the onset of US except that responses to air puff omission and nothing ( no outcome ) were calculated using a time window 0–1000 ms . Slightly different window sizes were also tested and gave qualitatively the same results . The baseline firing rates were obtained based on the activity in a time window during inter-trial-interval immediately preceding odor onset ( -1000 to 0 ms before odor onset ) . The baseline firing rates were computed by using data from all trial types . We calculated the area under the receiver-operating characteristic ( auROC ) value of each neuron using the trial-by-trial responses to CS , unpredicted and predicted outcomes in time windows previously described . The area of the right eye region was calculated as follows ( see also Figure 6—figure supplement 1 ) : ( 1 ) Eye threshold: Since in our recording settings , most of the face background area was saturated ( close to 255 pixel intensity ) , a threshold around 234 pixel intensity was used to separate eye area from the background . Pixels with intensity smaller than the threshold were set to 1 ( white ) and others were set to 0 ( black ) . ( 2 ) Remove dark patches outside of the eye: To remove the occasional dark patches outside of eye area in the raw image ( e . g . , top panel in Figure 6—figure supplement 1A ) , connected areas smaller than 500 pixels were deleted . ( 3 ) Smooth the eye patch: We performed morphological opening to remove spiky edges . Then we filled all black spots on the binary image ( e . g . , in Figure 6—figure supplement 1C ) smaller than 500 pixels to remove the bright spots inside of the eye area due to reflection . ( 4 ) Compute eye area: We found the largest connected regions on the binary image in Figure 6—figure supplement 1D and computed the area of this region in pixels and also computed the eccentricity by fitting the area to an eclipse ( MATLAB regionprops function ) . These codes for extracting eye areas from video files are available at https://github . com/hide-matsumoto/prog_hide_matsumoto_2016 . To analyze the eye-blinking behavior trial-by-trial , we synchronized video frames with Neuralynx timestamps as follows: ( 1 ) Detect frames when infrared light was off . When infrared light source was briefly turned off ( <25 ms ) , average pixel intensity of the frame steeply decreased . Thus , when the average pixel intensity of each frame in the session was plotted over time , the light-off frames were detected as troughs of the value ( Figure 6—figure supplement 1E ) . ( 2 ) We then matched these frames that have troughs of average pixel intensity with time stamps of infrared light source-off saved in Neuralynx . ( 3 ) Interpolate time of other frames: The timing of a video frame was interpolated using the time stamps of the two closest light-off frames . The eye areas extracted from video frames were further analyzed using event time stamps saved in Neuralynx . To compare eye areas across sessions , the computed eye areas were normalized by the maximum eye area ( 99th percentile of all the eye areas ) in every session . Trials were categorized into two groups , blink and no-blink trials , using the criteria that the averaged eye area during delay period in each air puff trial was larger ( no-blink trials ) or smaller ( blink trials ) than 0 . 5 . To check the percentage of trials that the animals showed anticipatory licking during the delay period ( 1–2 s from odor onset ) , trials were categorized into two groups , lick and no-lick trials , using the criteria that the lick rate during the delay period was larger ( lick trials ) or smaller ( no-lick trials ) than 3 . The threshold ( 3 licks s−1 ) was determined by comparing the distributions of the lick rates during the delay period in rewarded trials and those in nothing trials ( Figure 6—figure supplement 4A ) . For each statistical analysis provided in the manuscript , the Kolmogorov–Smirnov normality test was first performed on the data to determine whether parametric or non-parametric tests were required . Data were analyzed in MATLAB ( MathWorks ) and were shown as mean ± S . E . M . , unless otherwise stated . For unpaired t test , the equality of variance between two groups was first validated statistically . For paired and unpaired comparisons , two-sided tests were used . Bonferroni correction was applied for significance tests with multiple comparisons . To test monotonicity of CS responses , we chose neurons showing that ( 1 ) their CS responses were significantly modulated by odors ( examined by one-way ANOVA ) , ( 2 ) the response to reward-predicting CS was significantly larger than that to air puff-predicting CS ( p<0 . 05 , unpaired t test ) , and ( 3 ) the averaged response to CS predicting nothing ( or CS predicting both reward and air puff ) was intermediate between that to reward-predicting CS and that to air puff-predicting CS . Sample sizes in this study were based on previous literature in the field ( Cohen et al . , 2012; Eshel et al . , 2015 , 2016; Tian and Uchida , 2015 ) and were not pre-determined by a sample size calculation . Randomization and blinding were not employed . After recording , mice were transcardially perfused with saline and then with 4% paraformaldehyde under anesthesia . Brains were cut in 100 μm coronal sections . Brain sections from DAT-cre mice were immunostained with antibodies to tyrosine hydroxylase ( AB152 , 1:400 , Millipore; RRID:AB_390204 ) and secondary antibodies labeled with Alexa594 ( 1:200 , Invitrogen ) to visualize dopamine neurons . Sections were further stained with 4′ , 6-diamidino-2-phenylindole ( DAPI , Vector Laboratories ) to visualize nuclei . Recording sites were further verified to be amid EYFP- and tdTomato-positive or tyrosine hydroxylase-positive neurons in VTA .	There are many types of learning; one type of learning means that rewards and punishments can shape future behavior . Dopamine is a molecule that allows neurons in the brain to communicate with one another , and it is released in response to unexpected rewards . Most neuroscientists believe that dopamine is important to learn from the reward; however , there are different opinions about whether dopamine is important to learn from punishments or not . Previous studies that tried to examine how dopamine activities change in response to punishment have reported different results . One of the likely reasons for the controversy is that it was difficult to measure only the activity of dopamine-releasing neurons . To overcome this issue , Matsumoto et al . used genetically engineered mice in which shining a blue light into their brain would activate their dopamine neurons but not any other neurons . Tiny electrodes were inserted into the brains of these mice , and a blue light was used to confirm that these electrodes were recording from the dopamine-producing neurons . Specifically if the electrode detected an electrical impulse when blue light was beamed into the brain , then the recorded neuron was confirmed to be a dopamine-producing neuron . Measuring the activities of these dopamine neurons revealed that they were indeed activated by reward but inhibited by punishment . In other words , dopamine neurons indeed can signal punishments as negative and rewards as positive on a single axis . Further experiments showed that , if the mice predicted both a reward and a punishment , the dopamine neurons could integrate information from both to direct learning . Matsumoto et al . also saw that when mice received rewards too often , their dopamine neurons did not signal punishment correctly . These results suggest that how we feel about punishment may depend on how often we experience rewards . In addition to learning , dopamine has also been linked to many psychiatric symptoms such as addiction and depression . The next challenge will be to examine how the frequency of rewards changes an animal’s state and responses to punishment in more detail , and how this relates to normal and abnormal behaviors .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Midbrain dopamine neurons signal aversion in a reward-context-dependent manner
Stem cells are maintained in a specialized microenvironment called niche but the nature of stem cell niche remains poorly defined in many systems . Here we demonstrate that intestinal epithelium-derived BMP serves as a niche signal for intestinal stem cell ( ISC ) self-renewal in Drosophila adult midgut . We find that BMP signaling is asymmetric between ISC and its differentiated daughter cell . Two BMP ligands , Dpp and Gbb , are produced by enterocytes and act in conjunction to promote ISC self-renewal by antagonizing Notch signaling . Furthermore , the basement membrane-associated type IV collagens regulate ISC self-renewal by confining higher BMP signaling to ISCs . The employment of gut epithelia as a niche for stem cell self-renewal may provide a mechanism for direct communication between the niche and the environment , allowing niche signal production and stem cell number to be fine-tuned in response to various physiological and pathological stimuli . In adult life , many organs rely on stem cells to maintain their integrity by replenishing lost cells during tissue homeostasis and regeneration , yet the regulatory mechanisms that control stem cell proliferation , self-renewal , and differentiation are still not fully understood . Stem cells are thought to reside in specialized microenvironments called niches that produce signals required for stem cell self-renewal and maintenance ( Jones and Wagers , 2008; Morrison and Spradling , 2008; Losick et al . , 2011 ) ; however , the cellular basis of stem cell niches and the molecular nature of the niche signals have not been well defined in many systems . Drosophila adult midgut has emerged as an attractive system to study stem cell biology in adult tissue homeostasis and regeneration not only because the cell lineage of this tissue is relatively simple and well defined but also because it bears similarities to the mammalian intestine ( Casali and Batlle , 2009; Biteau et al . , 2011; Jiang and Edgar , 2012 ) . Drosophila posterior midgut contains self-renewing stem cells located adjacent to the basement membrane ( BM ) of the midgut epithelium ( Figure 1A; Micchelli and Perrimon , 2006; Ohlstein and Spradling , 2006 ) . These intestine stem cells ( ISCs ) undergo cell division and asymmetric fate determination to produce a renewed ISC and an enteroblast ( EB ) . The EB exits cell cycle and differentiates into either an absorptive enterocyte ( EC ) or a secretory enteroendocrine cell ( EE ) depending on Notch ( N ) pathway activity ( Figure 1A; Ohlstein and Spradling , 2007 ) . Fate determination between the two ISC daughter cells is regulated by N signaling ( Micchelli and Perrimon , 2006; Ohlstein and Spradling , 2006 , 2007; Bardin et al . , 2010 ) . Immediately after an ISC division , a high level of active Delta ( Dl ) is retained in the basally localized daughter cell that remains as ISC while the more apically localized daughter cell activates N signaling to become EB ( Ohlstein and Spradling , 2007 ) . How asymmetric N signaling between two ISC daughter cells is established has remained poorly understood . A recent study suggested that asymmetric segregation of aPKC could play a role ( Goulas et al . , 2012 ) , but additional mechanisms may exist . A previous study suggested that visceral muscle ( VM ) -derived Wingless ( Wg ) serves as a niche signal for ISC self-renewal ( Lin et al . , 2008 ) . However , other studies suggested that Wg does not regulate ISC self-renewal but instead regulates its proliferation ( Lee et al . , 2009; Cordero et al . , 2012 ) . Hence , it is still unclear whether ISC fate is influenced by an environmental signal ( s ) . 10 . 7554/eLife . 01857 . 003Figure 1 . BMP signaling is required for midgut regeneration . ( A ) Left: an ISC lineage in Drosophila adult midguts . ISC: intestinal stem cell; EB: enteroblast; EC: enterocyte; EE: enteroendocrine cell . ISC and EB are collectively called precursor cells . Dl and Su ( H ) -lacZ mark ISC and EB , respectively , whereas Pdm1 and Pros are the markers for EC and EE , respectively . esg-Gal4 and Myo1A-Gal4 are precursor and EC-specific Gal4 drivers , respectively . Right: sagittal view of Drosophila midgut epithelium immunostained with an anti-GFP antibody ( green ) , Phalloidin ( red ) and a nuclear dye ( DRAQ5 , blue ) . Arrows and asterisks indicate precursor cells and ECs , respectively . ( B ) Quantification of PH3+ cells in midguts from adults of the indicated genotypes ( mean ± SD , n = 20 for each genotype ) . Tkv and Put RNAi in precursor cells blocked damage-induced mitotic index . ( C–T ) 3- to 5-day-old adult females of esgF/O without ( C–E and L–N ) or with UAS-Tkv-RNAi105834 ( F–H and O–Q ) or UAS-Put-RNAi ( I–K and R–T ) were shifted to 29°C for 8 days and treated with sucrose , DSS and bleomycin for 2 days , followed by immunostaining for GFP and PH3 ( C–K ) , or GFP , Pdm1 and Pros ( L–T ) . Top views of midguts are shown in these panels and in panels of all other figures unless indicated otherwise . Scale bars in this and other figures ( except for Figure 6A–C ) are 10 μm . esgts: esg-Gal4 tub-Gal80ts . esgF/O: esg-Gal4 tub-Gal80ts UAS-GFP; UAS-flp Act>CD2>Gal4 . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 003 Drosophila midguts constantly undergo turnover and can regenerate after tissue damage ( Amcheslavsky et al . , 2009; Jiang et al . , 2009 ) . Several evolutionarily conserved signaling pathways , including Insulin , JNK , JAK-STAT , EGFR , Wg/Wnt , and Hpo pathways , have been implicated in the regulation of ISC proliferation during midgut homeostasis and regeneration ( Amcheslavsky et al . , 2009; Buchon et al . , 2009; Jiang et al . , 2009; Lee et al . , 2009; Karpowicz et al . , 2010; Ren et al . , 2010; Shaw et al . , 2010; Staley and Irvine , 2010; Amcheslavsky et al . , 2011; Biteau and Jasper , 2011; Jiang et al . , 2011; Xu et al . , 2011; Cordero et al . , 2012 ) . It is very likely that additional pathways are involved in the regulation of midgut homeostasis and regeneration . By carrying out in vivo RNAi screen , we identified components in the BMP pathway as essential regulators of midgut regeneration . Clonal analysis and lineage tracing experiments suggest that BMP signaling regulates ISC self-renewal as well as ISC proliferation and lineage differentiation . We showed that EC-derived Dpp and Gbb act in concert to promote ISC self-renewal by antagonizing N signaling-mediated differentiation . We provided evidence that BMP exists in an apical-basal activity gradient and that BM regulates ISC self-renewal by confining high BMP signaling to ISCs . To identify additional genes and pathways that regulate injury-induced ISC proliferation , we carried out in vivo RNAi screen in which candidate genes were knocked down in midgut precursor cells using the esg-Gal4 tub-Gal80ts ( esgts ) system , in which Gal4 is under the control of a temperature sensitive Gal80 ( McGuire et al . , 2004 ) . 3–5-day-old adult females expressing individual UAS-RNAi transgenes under the control of esgts were shifted to 29°C for 8 days and fed with tissue-damaging reagents such as DSS or bleomycin for 2 days , followed by immunostaining to examine ISC proliferation ( Ren et al . , 2010; Amcheslavsky et al . , 2011; Ren et al . , 2013 ) . The TGFβ/BMP signaling pathway has been implicated as an important regulator of stem cell biology in many systems ( Zhang and Li , 2005; Oshimori and Fuchs , 2012 ) . In Drosophila , BMP signal is transduced via two type-I receptors Thickvein ( Tkv ) and Saxophone ( Sax ) , and a type-II receptor Punt ( Put ) ( Nellen et al . , 1994; Moustakas and Heldin , 2009 ) . We found that inactivation of BMP signaling in adult midgut precursor cells by knocking down either type I ( esgts>Tkv-RNAi; VDRC#105834 ) or type II ( esgts>Put-RNAi; VDRC #107071 ) receptor blocked DSS- or bleomycin-induced ISC proliferation , as indicated by the diminished mitotic cells recognized by staining with an anti-phospho-histone 3 ( PH3 ) antibody ( Figure 1B ) . This is somewhat surprising given that BMP signaling restricts stem cell/progenitor cell proliferation in mammalian intestines ( Haramis et al . , 2004; He et al . , 2004 ) . To examine the role of BMP signaling in midgut regeneration , we employed the esgF/O ( esg-Gal4 tub-Gal80ts UAS-GFP; UAS-flp Act>CD2>Gal4 ) system in which all cells in the ISC lineage are labeled by GFP after shifting temperature to 29°C ( Jiang et al . , 2009 ) . Feeding adult flies with DSS or bleomycin induced a rapid turnover of midgut epithelia , as evident by the newly formed ECs ( marked by GFP+ Pdm1+ ) and EEs ( GFP+ , Pros+ ) 2–3 days after treatment ( Figure 1M , N ) . These guts also contained many dividing ISCs marked by PH3 staining ( Figure 1D , E ) . By contrast , mock treated guts only contained GFP+ precursor cells at this stage ( Figure 1C , L ) . Damage-induced ISC proliferation and epithelial turnover were blocked by inactivation of either type I or type II BMP receptor because GFP+ ECs or EEs , or PH3+ cells were rarely found in midguts expressing UAS-Tkv-RNAi105834 or UAS-Put-RNAi with esgF/O ( Figure 1F–K , O–T ) . Instead , these guts only contained GFP+ precursor cells ( Figure 1F–K , O–T ) , suggesting that BMP signaling is also required for intestinal epithelium differentiation . The observed reduction of mitotic index in BMP receptor knockdown midguts could be due to reduced stem cell activity or reduced stem cell number . To distinguish these possibilities , we examined the expression of Dl and Su ( H ) -lacZ , which mark ISC and EB , respectively ( Figure 1A; Micchelli and Perrimon , 2006; Ohlstein and Spradling , 2006 , 2007 ) , in adult midguts expressing either esgts>Tkv-RNAi105834 or esgts>Put-RNAi . 8 days after adult flies were cultured at 29°C , the number of Dl+ cells dropped significantly in Tkv or Put RNAi guts: less than 10% of esg>GFP+ precursor cells were DI+ compared with ∼50% in control guts , whereas the number of Su ( H ) -lacZ+ cells increased concomitantly ( Figure 2A–G ) . In control guts , most pairs of precursor cells contained one Dl+ cell and one Su ( H ) -lacZ+ cell ( Figure 2A , D–D″ ) . By contrast , most pairs of precursor cells deficient for BMP signaling contained two Su ( H ) -lacZ+ and no Dl+ cells ( Figure 2B–C , E–F″ ) . The stem cell loss phenotype caused by inaction of BMP signaling cannot be rescued by blocking apoptosis ( Figure 2—figure supplement 1 ) , suggesting that BMP signaling does not simply regulate ISC survival . In fact , an increase in the number of Su ( H ) -lacZ+/Su ( H ) -lacZ+ precursor pairs accompanied by a decrease in the number of Dl+/Su ( H ) -lacZ+ precursor pairs strongly suggest that ISC loss in Tkv or Put RNAi guts is due to precocious differentiation of ISC daughter cells into EBs . 10 . 7554/eLife . 01857 . 004Figure 2 . BMP signaling is required for ISC self-renewal . ( A–F″ ) 3- to 5-day-old adult females expressing esgts>GFP ( A , D–D″ ) or expressing esgts>GFP together with UAS-Tkv-RNAi ( B , E–E″ ) or UAS-Put-RNAi ( C , F–F″ ) were shifted to 29°C for 8 days , followed by immunostaining for DI ( red in A–C ) or Su ( H ) -lacZ ( red in D–F″ ) , GFP and DRAQ5 ( a nuclear marker ) . In the control guts , most pairs of precursor cells contain one Dl+ ISC and one Su ( H ) -lacZ+ EB ( A , D–D″ ) ; however , in Tkv or Put RNAi guts , most pairs of precursor cells contain two Su ( H ) -lacZ+ cells without Dl staining ( B–C , E–F″ ) . ( G ) Percentage of Dl+ or Su ( H ) -Z+ cells out of GFP+ precursor cells ( mean ± SD , n = 10 for each genotype ) . ( H ) Schematic drawing of an ISC division that produces differentially labeled twin-spot ( RFP+ GFP− and RFP− GFP+ ) through FRT-mediated mitotic recombination . The expression of GFP and RFP is under the control of the ubiquitin ( ubi ) promoter . Transgenic overexpression or RNAi through esgts allows determining the effect of gain- or loss-of-function of genes of interest on the outcome of an ISC division . ( I ) Schematic drawings of differentially labeled twin-spot clones generated by FLP/FRT-mediated mitotic recombination of dividing ISCs . ( J ) Scheme for twin-spot experiments involving Put RNAi . 3–5-day-old control or esgts>Put-RNAi adult flies were grown at 29°C for 7 days before heat shock ( hs ) to induce clones . After one-day recovery at 29°C , the flies were raised at 18°C for 4 days prior to analysis . ( K–L‴ ) Representative twin-spot clones from control and Put RNAi guts . ( M ) Quantification of twin spots of different classes from control and Put-RNAi guts: Con ( n = 110 , ISC/EB: 82% , ISC/ISC: 10% , EB/EB: 8% ) , Put-RNAi ( n = 110 , ISC/EB: 20% , ISC/ISC: 0% , EB/EB: 80% ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 00410 . 7554/eLife . 01857 . 005Figure 2—figure supplement 1 . Blocking apoptosis does not rescue ISC loss caused by inactivation of BMP signaling . ( A–D ) 3–5-day-old adult females expressing esgts>GFP ( Con; A ) , esgts>Diap1 ( B ) , esgts>Put-RNAi ( C ) , or esgts>Put-RNAi + Diap1 ( D ) were shifted to 29°C for 8 days , followed by immunostaining for DI and GFP . Overexpression of the apoptosis inhibitor Diap1 did not rescue the loss of Dl+ cells caused by Put RNAi . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 00510 . 7554/eLife . 01857 . 006Figure 2—figure supplement 2 . RFP/GFP two-color twin spot clonal analysis . ( A ) Confocal images of a posterior midgut containing FRT82 ubi-GFP ( red ) and FRT82 ubi-RFP ( red ) and immunostained with the nuclear marker DRAQ5 ( blue ) prior to clonal induction . Both ubi-GFP and ubi-RFP were expressed quite uniformly in the posterior region of adult midguts . ubi: ubiquitin promoter . B–D‴ , Examples of three indicated classes of twin spots generated by heat-shock induced FRT/FLP-mediated mitotic recombination in control midguts under normal homeostasis . 82% ( 90/110 ) of twin spots contained one multi-cellular clone and one single-cell clone derived from ISC/EB pairs whereas 10% ( 11/110 ) and 8% ( 9/110 ) contained two multicellular clones ( ISC/ISC class ) or two single-cell clones ( EB/EB class ) , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 006 To determine whether BMP signaling controls ISC/EB fate choice more definitively , we carried out lineage tracing experiments in which the two ISC daughter cells and their descendants were labeled by RFP+ ( red ) and GFP+ ( green ) , respectively , following FRT-mediated mitotic recombination ( Figure 2H ) . In this system , transgenic RNAi lines or UAS transgenes can be expressed in ISCs using esg-gal4 to assess the effect of inactivation or overexpression of genes of interest on the fate of the two ISC daughters that will generate distinctively labeled twin-spot clones ( Figure 2H , I ) . Furthermore , incorporation of Gal80ts allows temporal control of transgenic gene expression before and after clonal induction by temperature shift ( Figure 2J ) . We named this method ‘RGT’ for RFP and GFP labeled Twin-spot clonal analysis . The scheme for twin-spot experiments involving Put RNAi is shown in Figure 2J . 3–5-day-old control or esgts>Put-RNAi adult flies containing FRT82 ubi-RFP/FRT82 ubi-GFP were grown at 29°C for 7 days prior to clone induction by heat shock . After 1-day recovery at 29°C , the flies were raised at 18°C for 4 days prior to analysis . Temperature downshift restores normal BMP signaling , allowing normal lineage differentiation after clonal induction . Consistent with previous reports ( de Navascues et al . , 2012; O’Brien et al . , 2011 ) , the majority of twin spots ( 82%; 90/110 ) from the control guts contained one multi-cellular clone and one single-cell clone that are derived from asymmetric ISC/EB pairs ( Figure 2I , K–K‴ , M , Figure 2—figure supplement 2 ) , and only a small fraction of twin spots contained either two multi-cellular clones derived from symmetric ISC/ISC pairs ( 10%; 11/110 ) or two single-cell clones derived from symmetric EB/EB pairs ( 8%; 9/110 ) ( Figure 2I , M , Figure 2—figure supplement 2 ) . By contrast , the majority of twin spots ( 80%; 88/110 ) from esgts>Put-RNAi expressing guts falls into the symmetric EB/EB class , and the remaining twin spots ( 20%; 22/110 ) belongs to the asymmetric ISC/EB class ( Figure 2L–L‴ , M ) . Thus , loss of BMP signaling alters the outcome of ISC divisions from mostly asymmetric to predominantly symmetric non-self renewing . These results support the notion that BMP signaling regulates ISC/EB fate choice . To confirm the results obtained by RNAi , we generated and analyzed MARCM clones deficient for BMP signaling . putP ( also known as put10460 ) behaves like a genetic null allele ( Ruberte et al . , 1995 ) . 3–5-day-old adult flies were heat shocked to induce GFP+ marked putP mutant clones and then raised at 25°C for different periods of time prior to analysis . Because of the regional difference of ISC activity in the Drosophila midguts ( Buchon et al . , 2013; Marianes and Spradling , 2013 ) , we analyzed the ISC lineage clones only in the posterior midguts . At 12 days after clone induction ( ACI ) , control ISC lineage clones in the posterior region of midguts contained an average of 8 cells with single Dl+ positive cell and multiple ECs indicated by their large nuclei and Pdm1 staining ( Figure 3A , C , I , J ) . By contrast , the majority of putP ISC lineage clones contained two cells of small nuclei with no Dl staining but both exhibiting Su ( H ) -lacZ expression ( Figure 3B , I , J ) . Even at an early time point ( 5 days ACI ) whereby most of the ISC lineage clones contain only two cells , 91% ( 109/120 ) of the control clones contained one Dl+ cell and one Su ( H ) -lacZ+ cell whereas 82% ( 98/120 ) of the putP clones contained two Su ( H ) -lacZ+ cells ( Figure 3G , H ) , suggesting that BMP-signaling-deficient ISC daughters failed to self-renew but instead underwent precocious differentiation into two EBs . Of note , at 12 days ACI , the majority of control clones contained Pdm1+ ECs whereas none of the putP clones ( 0/150 ) exhibited Pdm1 staining ( Figure 3C , D , K ) ; however , at 18 days ACI , a large fraction of putP clones ( ∼70% , n = 95 ) exhibited Pdm1 staining ( Figure 3E , F , K ) . The delayed occurrence of Pdm1+ cells in putP clones further argues that BMP signaling is also required for proper ISC lineage differentiation into mature cells . 10 . 7554/eLife . 01857 . 007Figure 3 . Characterization of midgut phenotypes caused by differential inactivation of BMP pathway components . ( A–H ) Midguts containing the control clones ( A , C , E , G ) or putP clones ( B , D , F , H ) were immunostained for DI ( red in A , B , G , H ) or Pdm1 ( red in C–F ) , GFP ( green ) , and Su ( H ) -lacZ ( blue in A , B , G , H ) at 5 ( G , H ) , 12 ( A–D ) , or 18 ( E , F ) days after clone induction ( ACI ) . Control and mutant clones are marked by GFP expression . Control ISC lineage clones usually contain one Dl+ cell , one or more Su ( H ) -lacZ+ cells , and many cells with large nuclei at 12 days ACI . By contrast , most put mutant ISC lineage clones contain two cells that are Dl− but Su ( H ) -lacZ+ . At 5 days ACI , 91% of control ISC lineage clones contained one Dl+ cell and one Su ( H ) -lacZ+ cell while 82% of put mutant ISC lineage clones contained two Su ( H ) -lacZ+ cells . ( I ) Quantification of ISC lineage clones containing Dl+ cells 12 days ACI ( mean ± SD , n = 125 for each genotype ) . ( J ) Quantification of clone size for control ( Con ) or putP ISC lineage clones 12 days ACI ( n = 150 for each genotype ) . ( K ) Quantification of Pdm1+ clone frequency for control ( Con ) and putP ISC lineage clones at 12 or 18 days ACI ( mean ± SD , n = 150 for each genotype ) . ( L–S ) Adult midguts carrying MARCM clones of the indicated genotype were immunostained for Dl and GFP at 8 days ACI . Arrows indicate Dl+ cells and asterisks in H indicate clones without Dl+ cells . ( T ) Quantification of Dl+ clone frequency for the indicated genotypes at 8 days ACI ( mean ± SD , n = 130 for each genotype ) . ( U ) Quantification of clone size for the indicated genotypes at 8 days ACI ( n = 170 for each genotype ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 00710 . 7554/eLife . 01857 . 008Figure 3—figure supplement 1 . Tkv and Sax act redundantly in the regulation of ISC self-renewal . ( A ) Schematic drawing of Tkv and Sax coding regions with numbers indicating the nucleotide positions . The red lines indicate the regions targeted by individual RNAi lines . The red bars indicate the conserved region between Tkv and Sax coding sequences . ( B–E′ ) Control midguts ( esgts>GFP ) or midguts expressing the indicated RNAi lines with esgts>GFP at 29°C for 12 days were immunostained to show the expression of Dl ( red ) and esg>GFP ( green ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 00810 . 7554/eLife . 01857 . 009Figure 3—figure supplement 2 . Both tkv8 and mad1–2 mutant clones caused non-cell autonomous ISC overproliferation . Posterior midguts carrying control ( A ) , tkv8 ( B ) , mad1–2 ( C ) , or tkv8 mad1–2 ( D ) clones and immunostained for PH3 , GFP , and DRAQ5 at 8 days ACI . Arrows and asterisks indicate PH3+ cells inside and outside the clones , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 00910 . 7554/eLife . 01857 . 010Figure 3—figure supplement 3 . Inactivation of BMP signaling in ECs caused ISC proliferation . ( A–D ) Control midguts ( Myo1Ats ) or midguts expressing the indicated RNAi lines at 29°C for 8 days were immunostained for PH3 and DRAQ5 . Inactivation of BMP signaling in ECs resulted in elevated ISC proliferation . ( E ) Quantification of PH3+ cells for the indicated genotypes ( n = 10 for each genotype ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 010 tkv8 is a null allele that encodes a truncated protein containing only part of the extracellular domain ( Nellen et al . , 1994 ) . Surprisingly , compared to the control clones , tkv8 clones exhibited increased clone size and contained one or more Dl+ cells ( Figure 3L , M , T , U ) suggesting that tkv8 clones over-proliferated . Similar results have been observed by Guo et al ( Guo et al . , 2013 ) . BMP can transduce signal through two type I receptors Tkv and Sax ( Brummel et al . , 1994; Nellen et al . , 1994 ) , raising a possibility that Sax may support ISC self-renewal in the absence of Tkv . Indeed , we found that tkv8 clones expressing Sax-RNAi behaved similarly to putP clones even though Sax-RNAi alone did not exhibit a stem cell loss phenotype ( Figure 3N , O , T ) . We noticed that the sequence targeted by Tkv-RNAi105834 contained a region that is conserved between Tkv and Sax ( Figure 3—figure supplement 1A ) , suggesting that Tkv-RNAi105834 could inactivate both Tkv and Sax , which may explain why Tkv-RNAi105834 caused ISC loss while tkv8 did not . Indeed , Tkv-RNAi40937 , which targets a unique region in Tkv did not cause ISC loss but when combined with Sax-RNAi , resulted in stem cell loss ( Figure 3—figure supplement 1B–E′ ) . These results underscore the functional redundancy between Tkv and Sax in the control of ISC self-renewal . Furthermore , they suggest that different degrees of BMP pathway inactivation may result in distinct phenotypes with partial loss of BMP pathway activity causing ISC overproliferation whereas more complete loss of BMP signaling leading to ISC loss . To further test the idea that different degrees of BMP inactivation have distinct effects on ISC behavior , we generated MARCM clones for a hypomorphic allele of the BMP signal transducer Mad , mad1–2 ( Flybase ) . As expected , mad1–2 clones over-proliferated and behaved like tkv8 clones ( Figure 3P , T , U ) ; however , mad1–2 clones expressing Mad-RNAi exhibited stem cell loss phenotype ( Figure 3R , T , U ) . Strikingly , even though tkv8 or mad1–2 single mutant clones overproliferated , tkv8 mad1–2 double mutant clones failed to proliferate ( Figure 3U , Figure 3—figure supplement 2D ) and exhibited stem cell loss phenotype similar to put null mutant clones in the posterior midguts ( Figure 3S , T ) . Interestingly , PH3 staining of midguts containing either mad1–2 or tkv8 clones revealed increased mitotic index both outside and inside the mutant clones ( Figure 3—figure supplement 2A–C ) , suggesting that mad1–2 and tkv8 clones can exert a non-cell-autonomous effect on the proliferation of neighboring wild type ISCs . In contrast , tkv8 mad1–2 double mutant clones did not exhibit any PH3+ signal , nor did they stimulate the proliferation of neighboring wild type ISCs because no ectopic PH3+ cells were associated with mad1–2 tkv8 double mutant clones in the posterior midguts ( Figure 3U , Figure 3—figure supplement 2D ) . Similarly , we did not observe any PH3+ cells within or outside of put null clones ( data not shown ) . Because tkv8 and mad1–2 single mutant clones contained many ECs whereas put null or tkv8 mad1–2 double mutant clones contained little if any ECs at the time we did PH3 staining , we suspected that the non-cell-autonomous effect of tkv8 and mad1–2 single mutant clones was due to BMP signaling defects in ECs . In support of this notion , RNAi of Put , Tkv , or Mad in ECs resulted in elevated ISC proliferation ( Figure 3—figure supplement 3; Li et al . , 2013b ) . In Drosophila , the BMP signal transducer Mad is phosphorylated upon receptor activation; therefore , the levels of pMad signal are indicative of the levels of BMP pathway activity . By immunostaining with an anti-pMad antibody ( Persson et al . , 1998 ) , we observed high levels of pMad signal in Dl-lacZ+ cells and low levels in Su ( H ) >GFP+ cells ( Figure 4A–B′ ) , suggesting that BMP pathway is asymmetrically activated in a pair of ISC/EB cells with ISC transducing higher levels of BMP signaling activity than EB . Consistent with this notion , we found that dad-lacZ , which is induced by BMP signaling , exhibits high levels of expression in ISCs and low levels of expression in EBs ( Figure 4C ) . However , pMad signals were evenly distributed into two future daughter cells of a dividing ISC marked by PH3+ ( Figure 4D ) , suggesting that asymmetric BMP signaling is unlikely due to asymmetric inheritance of activated pathway components but rather due to asymmetric induction after ISC division . A small fraction ( 8/105 ) of ISC division resulted in the production of two Dl+ ISCs that contained equally high levels of pMad staining ( Figure 4E , E′ ) and both ISCs tend to lie in close proximity to the BM ( Figure 4F , F′ ) , suggesting that they were derived from symmetric cell division ( Goulas et al . , 2012 ) . We also observed strong pMad staining and dad-lacZ expression in ECs ( indicated by asterisks in Figure 4B–D , F , F′ ) , suggesting that BMP signaling is active in differentiated cells . 10 . 7554/eLife . 01857 . 011Figure 4 . Asymmetric BMP signaling regulates ISC self-renewal . ( A–A′ , C–E′ ) High magnification views of wild type adult midguts immunostained for pMad ( red in A–A′ , D–E′ ) , dad-lacZ ( red in C ) , Dl-lacZ ( blue in A′ , E′ ) , Su ( H ) -GFP ( green in A′ , C , E′ ) or PH3 ( green in D ) . ( B–B′ , F–F′ ) Sagittal views of wild type adult midguts immunostained for pMad ( red ) , Su ( H ) -GFP ( green in B′ ) , Dl-lacZ ( green in F′ ) , and Phalliodin ( blue ) . Arrows and arrowheads indicate ISCs and EBs , respectively . Asterisks indicate the pMad signals in ECs . ( G–O ) Adult midguts expressing esgts>GFP ( G , J , M ) , esgts>GFP + TkvQ235D ( H , K , N ) , or esgts>GFP + Dpp + Gbb ( I , L , O ) at 29°C for the indicated time periods were immunostained for Dl ( red in G–L ) , PH3 ( red in M–O ) , GFP ( green ) , and Su ( H ) -lacZ or DRAQ5 ( blue ) . ( P ) Scheme for twin-spot experiments involving TkvQ235D overexpression . 3–5-day-old control or esgts>TkvQ235D adult flies were grown at 29°C for 3 days before heat shock ( hs ) to induce clones . After 1-day recovery at 29°C , the flies were raised at 18°C for 4 days prior to analysis . ( Q–R ) Representative twin-spot clones from control and esgts>TkvQ235D guts . ( S ) Quantification of twin spots of different classes from control and esgts>TkvQ235D guts: Con ( n = 160 , ISC/EB: 83% , ISC/ISC: 9% , EB/EB: 8% ) , TkvQ235D ( n = 190 , ISC/EB: 23% , ISC/ISC: 77% , EB/EB: 0% ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 01110 . 7554/eLife . 01857 . 012Figure 4—figure supplement 1 . Effect of misexpressing Dpp or Gbb alone in precursor cells on ISC self-renewal . ( A–B′ ) Adult midguts expressing esgts>Dpp + GFP ( A , A′ ) or esgts>Gbb + GFP ( B , B′ ) at 29°C for 12 days were immunostained for Dl ( red ) , GFP ( green ) and DRAQ5 ( blue ) . Misexpression of Dpp or Gbb alone resulted in ectopic Dl+ cells that formed small clusters ( arrows ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 012 To determine whether asymmetric BMP signaling plays an instructive role in the regulation of ISC self-renewal , we ectopically activated the pathway in precursor cells by expressing a constitutively active form of Tkv ( TkvQ235D ) ( Nellen et al . , 1996 ) . Immunostaining with a pMad antibody confirmed that TkvQ235D induced high levels of BMP pathway activation in precursor cells ( data not shown ) . In control guts , ISCs ( Dl+ esg>GFP+ ) existed in isolation and many of them were accompanied by Su ( H ) -lacZ+ esg>GFP+ EBs ( arrows in Figure 4G ) . 4 days after shifting to 29°C , esgts>TkvQ235D guts contained many pairs of Dl+ Su ( H ) -lacZ- precursor cells ( arrows in Figure 4H ) . Expression of TkvQ235D for a longer period of time ( 8 days ) resulted in the formation of large clusters of ISC-like cells that contained diving cells ( Figure 4K , N ) , suggesting that ectopic BMP signaling promotes ISC self-renewal . Next , we conducted RGT experiments to confirm that BMP signaling promote ISC fate . The scheme for twin-spot experiments involving TkvQ235D overexpression is shown in Figure 4P . 3–5-day-old control or esgts>TkvQ235D adult flies were grown at 29°C for 3 days prior to clone induction by heat shock . After 1-day recovery at 29°C , the flies were raised at 18°C for 4 days to allow lineage differentiation after clonal induction . As shown in Figure 4Q–S , midguts expressing esgts>TkvQ235D generated symmetric twin clones of the ISC/ISC class at much higher frequency ( 77%; 146/190 ) than the control guts ( 9%; 14/160 ) , suggesting that ectopic BMP signaling activity promotes the symmetric self-renewing outcome of an ISC division . These results demonstrate that excessive BMP signaling favors ISC fate choice . Dpp and Gbb are the two major BMP ligands in Drosophila ( Moustakas and Heldin , 2009 ) . Coexpression of Dpp and Gbb using esgts also resulted in the formation of ISC-like cell clusters similar to ectopic expression of TkvQ235D ( Figure 4I , L , O ) ; however , expression of either Dpp or Gbb alone only produced smaller ISC-like cell clusters ( Figure 4—figure supplement 1; compared with Figure 4L ) . These results suggest that Dpp and Gbb act in concert to promote ISC self-renewal likely by forming a heterodimer ( see below ) ( Ray and Wharton , 2001 ) . N signaling plays a critical role in balancing ISC self-renewal and differentiation in Drosophila midguts . Gain-of-N signaling blocks ISC self-renewal and promotes differentiation whereas loss-of-N signaling leads to excessive ISCs at the expense of EBs ( Figure 5A–A′ , F–F′; Micchelli and Perrimon , 2006; Ohlstein and Spradling , 2006 ) . BMP signaling could inhibit N pathway activity , thereby blocking ISC differentiation . Alternatively , N signaling could promote EB fate by inhibiting BMP pathway activity . To distinguish these two possibilities , we determined the epistatic relationship between BMP and N signaling . In one set of experiments , we expressed a constitutively active form of N ( NICD ) ( Struhl et al . , 1993 ) in precursor cells that also transduced high levels of BMP signaling activity due to the ectopic expression of TkvQ235D or combined expression of Dpp and Gbb . We found that N activation completely suppressed the formation of ISC-like cell clusters induced by ectopic BMP signaling ( Figure 5D–E′ compared with Figure 5B–C′ ) . We also found that there were no GFP+ cells in many areas due to the differentiation of precursor cells into ECs induced by NICD in esgts>NICD or esgts>NICD + TkvQ235D or esgts>NICD + Dpp + Gbb ( data not shown ) , and that the remaining GFP+ precursor cells were all Dl− ( Figure 5D–E′ ) . In a reciprocal set of experiments , we inactivated N signaling by expressing esgts>N-RNAi in precursor cells in which the BMP signaling was blocked due to the expression of esgts>Tkv-RNAi105834 or esgts>Put-RNAi . We found that loss of N signaling restored Dl+ cells in BMP signaling deficient precursor cells ( Figure 5I–J′ compared with Figure 5G–H′ ) , suggesting that ISC can form in the absence of BMP signaling as long as N signaling is blocked . Furthermore , pMad staining indicates that most of the ISC-like cells in N RNAi guts exhibited low levels of BMP pathway activity ( Figure 5L–L′ compared with Figure 5K–K′; Figure 5—figure supplement 1 ) , consistent with N acting downstream of or in parallel with BMP . Taken together , these results suggest that BMP signaling promotes ISC self-renewal by antagonizing N-mediated differentiation . 10 . 7554/eLife . 01857 . 013Figure 5 . BMP signaling promotes ISC self-renewal by antagonizing N . ( A–E′ ) Adult midguts expressing NICD ( 8d ) ( A–A′ ) , TkvQ235D ( 8d ) ( B–B′ ) , Dpp + Gbb ( 12d ) ( C–C′ ) , or the indicated combinations of transgenes ( D–E′ ) under the control of esgts were immunostained for Dl ( red ) , GFP ( green ) and DRAQ5 ( blue ) . Coexpression of NICD suppressed the excessive Dl+ cells caused by TkvQ235D or Dpp/Gbb misexpression , leading to loss of Dl+ cells similar to expression of NICD alone . ( F–J′ ) Adult midguts expressing the indicated RNAi lines under the control of esgts for 8 days were immunostained for Dl ( red ) , GFP ( green ) , and DRAQ5 ( blue ) . N RNAi rescued Dl+ cells in midguts expressing Tkv-RNAi105834 or Put-RNAi . ( K–M′ ) High magnification views of adult midguts expressing esgts>GFP ( K–K′ ) , esgts>GFP + N-RNAi ( L–L′ ) , or esgts>GFP + NICD ( M–M′ ) and immunostained for pMad ( red ) , GFP ( green ) , and DRAQ5 ( blue ) . Arrows and arrowhead indicate ISC and EB , respectively . Asterisks indicate the pMad signals in ECs . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 01310 . 7554/eLife . 01857 . 014Figure 5—figure supplement 1 . Integrated pMad levels in control or N knock down precursor cells . Quantification of pMad signals in pairs of ISC/EB in the posterior control midguts ( Con; esgts>GFP ) or in clusters of precursor cells expressing esgts>N-RNAi ( mean ± SD: n = 20 for control ISC/EB pairs; N > 100 for N RNAi precursor cells ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 014 We noticed that midguts expressing esgts>Tkv-RNAi105834 + N-RNAi or esgts>Put-RNAi + N-RNAi contained smaller clusters of Dl+ cells compared with esgts>N-RNAi guts ( Figure 5F–F′ , Figure 5I–J′ ) , suggesting that BMP pathway activity is required for the proliferation of N-signaling-deficient ISCs . We also noticed that expression of NICD in precursor cells resulted in diminished pMad staining ( Figure 5M–M′ compared with Figure 5K–K′ ) , suggesting that excessive N signaling can inhibit BMP pathway activity . It is possible that elevated N activity in the presumptive EB can downregulate BMP signaling as a feedback mechanism ( ‘Discussion’ ) . We next sought to determine the source of BMP signals . To our surprise , GFP under the control of a dpp-Gal4 ( dpp>GFP ) ( Teleman and Cohen , 2000; Roy et al . , 2011 ) was detected in ECs along the anterior-posterior ( A-P ) axis ( Figure 6A , B , D , D′ ) but not in precursor cells or VM ( data not shown ) . However , dpp>GFP signals were not uniform along the A-P axis of the midguts but instead , exhibited discrete domains of high-level expression in the posterior ( p ) , middle ( m ) and anterior ( a ) regions ( Figure 6A , B ) . A similar observation was made by Li et al . ( 2013a ) . To confirm dpp expression pattern as well as to determine the source of Gbb in midguts , we employed a sensitive RNA in situ hybridization method that allows detection of individual mRNA ( Raj et al . , 2008 ) . Both dpp and gbb probes could detect endogenous as well as ectopically expressed gene products in wing imaginal discs ( Figure 6—figure supplement 1A–D ) . In situ hybridization confirmed that the expression pattern of dpp>GFP correlated with that of endogenous dpp mRNA ( Figure 6E , E′ , Figure 6—figure supplement 1E–H ) . Furthermore , dpp mRNA was not detected in precursor cells or VM ( Figure 6—figure supplement 2A–F′ ) . Similarly , gbb mRNA was detected in ECs but not in precursor cells or VM ( Figure 6F–F′ , Figure 6—figure supplement 2G–L′ ) , and high levels of gbb were detected in regions that expressed low levels of dpp ( Figure 6—figure supplement 1I–L ) . This complementary pattern may allow a broad and relatively even distribution of BMP activity along the A/P axis of midguts , as indicated by the dad-lacZ expression ( Figure 6C ) . Taken together , these results establish ECs as a major source of Dpp and Gbb in adult midguts . Increasing the production of Dpp and Gbb in ECs using the Myo1A-Gal4/tub-Gal80ts ( Myo1Ats ) system to express UAS-Dpp and UAS-Gbb resulted in increased number of Dl+ cells ( Figure 7B ) , indicating that paracrine BMP signaling initiated from ECs can promote ISC self-renewal . 10 . 7554/eLife . 01857 . 015Figure 6 . Both Dpp and Gbb are expressed in ECs . ( A and B ) Low magnification views of adult midguts expressing one ( a ) or two ( b ) copies of UAS-GFP transgene under the control of dpp-Gal4 were immunostained for GFP and DRAQ5 . dpp>GFP is expressed in most of the midgut epithelia with strong expression in the posterior ( p ) , middle ( m ) , and anterior ( a ) regions . ( C ) Low magnification view of a midgut expressing dad-lacZ . ( D–D′ ) High magnification view of the posterior region of a dpp>GFP expressing midgut immunostained for GFP , Pdm1 , and DRAQ5 . ( E–E′ ) RNAi in situ hybridization of a dpp>GFP expressing midgut ( posterior region ) shows the coincidence of dpp mRNA and dpp>GFP signals . dpp mRNA signal is detected in the ECs ( outlined by dashed line as examples ) . ( F–F′ ) RNA in situ hybridization of midguts expressing Myo1A>GFP shows that gbb mRNA is detected in ECs . Two ECs are marked by dashed line as examples . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 01510 . 7554/eLife . 01857 . 016Figure 6—figure supplement 1 . Characterization of dpp and gbb expression in Drosophila by RNA in situ hybridization . ( A–D ) RNA in situ hybridization with dpp ( A and B ) or gbb ( C and D ) probe for wild type wing discs ( A , C ) or wing discs expressing UAS-Dpp ( B ) or UAS-Gbb ( D ) with wing disc specific Gal4 driver MS1096 . ( E–L ) High magnification views of the indicated regions of midguts expressing dpp>GFP and hybridized with dpp ( E–H ) or gbb ( I–L ) probe . dpp mRNA expression correlates with that of Dpp>GFP ( E–H ) whereas the high gbb mRNA expression domain corresponds to the low expression region of Dpp>GFP ( I–L ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 01610 . 7554/eLife . 01857 . 017Figure 6—figure supplement 2 . dpp and gbb mRNAs are not detected in precursors or VM . High magnification views of the indicated regions of adult midguts expressing esg>GFP ( A–C′ , G–I′ ) or how>GFP ( D–F′’ , J–L′ ) and probed for dpp ( A–F′ ) or gbb ( G–L′ ) expression by RNA in situ hybridization . esg>GFP and how>GFP mark the precursor cells and VM , respectively . Neither dpp mRNA nor gbb mRNA was detected in precursor cells ( outlined by dashed circles in A , A′ , B , B′ , C , C′ , H , H′ ) or VM . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 01710 . 7554/eLife . 01857 . 018Figure 7 . EC-derived Dpp and Gbb regulate ISC self-renewal . ( A and B ) Control ( A ) or midguts coexpressing both Dpp and Gbb with Myo1Ats ( B ) were immunostained for Dl and Su ( H ) -lacZ . ( C–I ) Control guts ( C , G ) , guts expressing strong Dpp-RNAi line ( D , H ) , strong Gbb-RNAi line ( E , I ) , or a combination of weak Dpp- and Gbb-RNAi lines ( F ) were immunostained for Dl ( red ) , Su ( H ) -lacZ ( green ) , and DRAQ5 ( blue ) . ( J ) Quantification of Dl+ or Su ( H ) -Z+ cell number ( mean ± SD , n = 15 for each genotype ) . Scale bars in A–C are 100 μm . Of note , to ensure sufficient knockdown of Dpp or Gbb , two copies of individual RNAi lines were expressed in midguts for 25 days . See ‘Materials and methods’ the genotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 01810 . 7554/eLife . 01857 . 019Figure 7—figure supplement 1 . Characterization of Dpp and Gbb RNAi lines . ( A–E ) Adult wing phenotypes associated with Dpp ( B and C ) or Gbb ( D and E ) knockdown using a wing specific Gal4 driver , MS1096 , to express the indicated RNAi lines . Arrows in D and E indicate defects in the posterior cross vein . ( F–H ) Control guts ( F ) or midguts expressing the indicated RNAi lines ( G and H ) with Myo1Ats at 29°C for 25 days were immunostained for Dl ( red ) and DRAQ5 ( blue ) . ( I ) Number of Dl+ cells per field in midguts of the indicated genotypes ( mean ± SD: n = 15 for each genotype ) . ( J ) Knockdown efficiency measured by RT-qPCR after 25-day expression of the indicated RNAi lines using Myo1Ats ( mean ± SD: triplicates ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 01910 . 7554/eLife . 01857 . 020Figure 7—figure supplement 2 . Characterization of Dpp and Gbb knockdown in ECs . Adult midguts expressing Myo1Ats>GFP ( A and A′ ) , Myo1Ats>GFP + Dpp-RNAiS ( B and B′ ) , or Myo1Ats>GFP + Gbb-RNAiS ( C and C′ ) at 29°C for 25 days followed by immunostaining for pMad ( red ) , GFP ( green ) , and DRAQ5 ( blue ) . Control guts exhibited asymmetric pMad staining in precursor pairs ( outlined by dashed circles in A and A′ ) and high levels of pMad in ECs ( indicated by asterisks in A and A′ ) . pMad was diminished in both precursor cells and ECs when Dpp or Gbb was knocked down in ECs ( B–C′ ) . ( D–I ) Adult midguts expressing Myo1Ats>GFP ( D and G ) , Myo1Ats>GFP + Dpp-RNAiS ( E and H ) , or Myo1Ats>GFP + Gbb-RNAiS ( F and I ) at 29°C for 25 days followed by Sucrose ( D–F ) or DSS ( G–I ) treatment and immunostaining for PH3 ( red ) and DRAQ5 ( blue ) . Knockdown of Dpp or Gbb in ECs blocked DSS-induced ISC proliferation . ( J ) Quantification of PH3+ cells in midguts from adults of the indicated genotypes ( mean ± SD: n = 20 for each genotype ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 02010 . 7554/eLife . 01857 . 021Figure 7—figure supplement 3 . Partial loss of BMP in ECs stimulates ISC proliferation . ( A and B ) Control guts ( A ) and guts expressing strong Dpp-RNAi line ( B ) for 10 days were immunostained for PH3 ( red ) and DRAQ5 ( blue ) . ( C ) Quantification of PH3+ cells in midguts from adults of the indicated genotypes ( mean ± SD: n = 20 for each genotype ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 02110 . 7554/eLife . 01857 . 022Figure 7—figure supplement 4 . Characterization of Dpp and Gbb in trachea and VM . ( A ) RNA in situ hybridization of midguts expressing Btl>GFP with a dpp probe . dpp mRNA was detected in Btl>GFP+ tracheal cells . ( B–D ) Adult midguts expressing Btlts-Gal4 ( B ) , Btlts>Dpp ( C ) or Btlts>Dpp + Gbb ( D ) at 29°C for 12 days were immunostained for Dl ( red ) and DRAQ5 ( blue ) . Misexpression of either Dpp alone or both Dpp and Gbb in tracheal cells failed to induce ectopic ISCs . ( E–G ) Control midguts ( Con ) ( E ) or midguts expressing Dpp RNAi in trachea ( Btlts>Dpp-RNAiS ) ( F ) or VM ( Howts>Dpp-RNAiS ) ( G ) at 29°C for 25 days were immunostained for Dl and DRAQ5 . ( H ) Quantification of Dl+ cells for the indicated genotypes ( mean ± SD: n = 8 for each genotype ) . Knockdown of Dpp in either tracheal cells or VM did not significantly affect the number of Dl+ cells in the midguts . ( I ) Knockdown efficiency measured by RT-qPCR after 25-day expression of the indicated RNAi lines using Howts or Btlts ( mean ± SD: triplicates ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 022 If EC-derived Dpp and Gbb serve as niche signals for ISC self-renewal , one would expect that loss of Dpp or Gbb in ECs should lead to stem cell loss . To test this hypothesis , we inactivated Dpp or Gbb in ECs using RNAi . Two Dpp RNAi lines: UAS-Dpp-RNAiS and UAS-Dpp-RNAiW , and two Gbb RNAi lines: UAS-Gbb-RNAiS and UAS-Gbb-RNAiW , were employed . When expressed in wing discs , these RNAi lines caused wing phenotypes indicative of Dpp or Gbb inactivation , and the severity of the wing phenotypes indicated that UAS-Dpp-RNAiS and UAS-Gbb-RNAiS are strong lines whereas UAS-Dpp-RNAiW and UAS-Gbb-RNAiW are weak lines ( Figure 7—figure supplement 1B–E ) . This notion was further confirmed by examining the knockdown efficiency using RT-qPCR ( Figure 7—figure supplement 1J ) . Myo1Ats>Dpp-RNAi or Myo1Ats>Gbb-RNAi guts were examined for stem cell maintenance 25 days after shifting to 29°C . Of note , two copies of each UAS-RNAi line as well as UAS-Dicer2 were coexpressed to increase knockdown efficiency ( see ‘Materials and methods’ for full genotypes ) . Knockdown of Dpp or Gbb by strong RNAi lines ( Myo1Ats>Dpp-RNAiS or Myo1Ats>Gbb-RNAiS ) resulted in loss of Dl+ cells ( Figure 7D , E , H–J; Figure 7—figure supplement 1I ) . pMad staining confirmed that BMP signaling activity was diminished both in precursor cells and ECs in these guts ( Figure 7—figure supplement 2A–C′ ) . Consistent with the stem cell loss phenotype , Myo1Ats>Dpp-RNAiS or Myo1Ats>Gbb-RNAiS guts exhibited greatly reduced mitotic index in response to injury as compared with controlled guts ( Figure 7—figure supplement 2D–J ) . Although knockdown of Dpp or Gbb by weak RNAi lines ( Myo1Ats>Dpp-RNAiW or Myo1Ats>Gbb-RNAiW ) did not significantly affect Dl+ cell number ( Figure 7—figure supplement 1G–I ) , their combined expression ( Myo1Ats>Dpp-RNAiW + Gbb-RNAiW ) resulted in loss of Dl+ cells ( Figure 7F; Figure 7—figure supplement 1I ) . Of note , Dpp or Gbb RNAi for shorter period of time ( e . g . , 10 days ) did not cause stem cell loss but instead , resulted in ISC overproliferation due to less complete knockdown ( Figure 7—figure supplement 3; data not shown ) . Recent studies suggested that Dpp is expressed in tracheal cells that contact adult midgut epithelium ( Li et al . , 2013b ) or in VM ( Guo et al . , 2013 ) . We confirmed that Dpp is expressed in tracheal cells by RNA in situ hybridization ( Figure 7—figure supplement 4A ) ; however , overexpression of either Dpp alone or in conjunction with Gbb in tracheal cells using Btl-Gal4 Gal80ts ( Btlts ) did not increase Dl+ cell number ( Figure 7—figure supplement 4B–D ) . Furthermore , neither Dpp RNAi in trachea ( Btlts>Dpp-RNAiS ) nor in VM ( Howts>Dpp-RNAiS ) significantly affected Dl+ cell number ( Figure 7—figure supplement 4E–H ) , which is in contrast to Dpp or Gbb RNAi in ECs . RT-qPCR analysis revealed that dpp mRNA was reduced by only ∼30% when dpp was knocked down in these tissues ( Figure 7—figure supplement 4I ) . Furthermore , we did not observed a significant change in pMad staining in Btlts>Dpp-RNAiS or Howts>Dpp-RNAiS midguts ( data not shown ) . Taken together , these results suggest that EC-derived Dpp and Gbb are the major source of BMP that regulates ISC self-renewal . The finding that BMP ligands are produced and required in ECs for ISC self-renewal is counterintuitive because both ISCs and EBs are surrounded by ECs , raising an important question of how the two ISC daughter cells activate different levels of BMP signaling . Interestingly , when expressed in ECs , a GFP-tagged Dpp ( Dpp-GFP ) was enriched on the basal side of midgut epithelia ( Figure 8B , B′ ) whereas a control GFP was uniformly distributed along the apical basal axis ( Figure 8A , A′ ) . Co-staining with a Golgi marker revealed that Dpp-GFP was enriched on the basal side of the secretary pathway ( Figure 8C ) , raising the possibility that Dpp-GFP is preferentially secreted from the basal/basolateral side of ECs . Although a GFP-tagged Gbb ( Gbb-GFP ) was uniformly distributed along the apical/basal axis when expressed in ECs ( Figure 8—figure supplement 1A , A′ ) , coexpression of Dpp redistributed Gbb-GFP toward the basal side ( Figure 8—figure supplement 1B–C′ ) , suggesting that Dpp and Gbb may physically interact , a notion confirmed by coimmunoprecipitation experiments ( Figure 8—figure supplement 2 ) . Hence , Dpp and Gbb heterodimers may form an apical/basal activity gradient that allows basally situated ISC daughter cells to transduce higher levels of BMP signal than more apically localized daughter cells . 10 . 7554/eLife . 01857 . 023Figure 8 . Regulation of ISC self-renewal by Vkg and BMP activity gradient . ( A–C ) Low ( A and B ) and high ( A′ , B′ , C ) magnification sagittal views of adult midguts expressing Myo1Ats>GFP ( A–A′ ) or Myo1Ats>Dpp-GFP ( B–C ) and immunostained with the indicated antibodies or dyes . Golgi is marked by an anti-Lava lamp antibody . Adult flies expressing Dpp-GFP ( or GFP ) with Myo1Ats were raised at 29°C for 5 days , followed by immunostaining . ( D ) Sagittal view of a Vkg-GFP expressing gut immunostained with the indicated antibodies or dyes . Arrows and arrowhead indicate ISC and EB , respectively . ( E–G , I , J ) Wild type ( E and I ) or vkg mutant guts ( F , G , J ) were immunostained for Dl ( red in E–G ) , pMad ( red in I and J ) , Su ( H ) -lacZ and DRAQ5 . In control guts ( I ) , precursor cells exhibit high ( arrows ) and low ( asterisks ) levels of pMad staining in pairs . vkg mutant guts from 10–12-day-old females contained clusters of precursor cells with high levels of pMad signal ( outlined in J ) . ( H ) Quantification of Dl+ or Su ( H ) -Z+ cells in wild type and vkg mutant guts ( mean ± SD , n = 20 for each genotype ) . ( K–L ) The ectopic Dl+ phenotype in vkg mutant was rescued by dpp heterozygosity ( dpphr56/+ at 29°C ) . ( M ) Quantification of Dl+ cells in midguts of the indicated genotype ( mean ± SD , n = 20 for each genotype ) . vkg: vkgK00236/vkg01209 ; dpp−/+ vkg: dpphr56 vkgK00236/vkg01209 . ( N–Q ) Low and high magnification views of wild type ( N–O ) or vkg mutant ( P–Q ) midguts expressing Dpp-GFP and immunostained for Integrin/Mys , GFP and DRAQ5 . Arrows point to the BM in all panels . In wild type guts , Dpp-GFP is enriched at BM as indicated by colocalization of Dpp-GFP and Integrin/Mys signals ( yellow in O; asterisks ) , but the colocalization is greatly reduced in vkg mutant guts ( Q ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 02310 . 7554/eLife . 01857 . 024Figure 8—figure supplement 1 . Dpp regulates Gbb subcellular localization . ( A–C′’ ) Sagittal views of adult midguts expressing Myo1Ats>Gbb-GFP ( A and A′ ) , Myo1Ats>Gbb-GFP + Dpp ( B and B′ ) , or Myo1Ats>Gbb-GFP + Dpp-HA ( C and C′ ) were immunostained with antibodies against GFP ( green ) , Integrin/Mys ( red in A′ and B′ ) or HA ( red in C′ ) , and DRAQ5 . When expressed alone , Gbb-GFP was uniformly distributed along the apical/basal axis . Coexpression of either the non-tagged or HA-tagged Dpp redistributed Gbb-GFP , resulting its enrichment on the basal side of EC . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 02410 . 7554/eLife . 01857 . 025Figure 8—figure supplement 2 . Dpp physically interacts with Gbb . Adult midguts expressing Dpp-HA and Gbb-GFP individually or in combination with Myo1Ats at 29°C for 5 days were subjected to immunoprecipitation and western blot analysis with the indicated antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 02510 . 7554/eLife . 01857 . 026Figure 8—figure supplement 3 . Integrated pMad levels in precursor cells of control or vkg mutant midguts . Quantification of pMad signals in pairs of wild type ISC/EB ( Con; esgts>GFP ) or in clusters of vkg mutant ( vkg00236/01 , 029 ) precursor cells in the posterior midguts ( mean ± SD: n = 20 for control ISC/EB pairs , n > 40 for precursor cells in vkg mutant guts ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 02610 . 7554/eLife . 01857 . 027Figure 8—figure supplement 4 . Dpp localization at the BM is diminished in vkg mutant guts . ( A–A″ ) Adult midguts were immunostained for the Drosophila integrin βPS subunit Integrin/Mys ( red ) , Vkg-GFP ( green ) , Phalloidin and DRAQ5 ( blue ) . Integrin/Mys is enriched at the basal membrane ( BM ) marked by Vkg-GFP . ( B–E ) , Low ( B–B′ , D–D′ ) and high ( C and E ) magnification views of wild type ( B and C ) or vkg mutant ( D–E ) midguts expressing Dpp-GFP and immunostained for Integrin/Mys ( red ) , GFP ( green ) , Phalloidin and DRAQ5 ( blue ) . Dpp-GFP is enriched at the BM in wild type midguts as suggested by the overlap of Dpp-GFP and Integrin/Mys signals ( yellow; indicated by asterisks in C ) . Dpp-GFP is no longer enriched at BM in vkg mutant guts as shown by diminished colocalization between Dpp-GFP and Integrin/Mys ( E ) . Arrows point to the BM in all panels . ( F ) Quantification of the mean intensity of extracellular Dpp-GFP signals that colocalize with integrin/Mys signals in control or vkg mutant guts ( mean ± SD: n = 15 for each genotype ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 027 After an ISC division , the daughter cell that is in close contact with BM usually remains as a stem cell while the more apically localized daughter cell becomes an EB ( Figure 8D; Micchelli and Perrimon , 2006; Ohlstein and Spradling , 2006; Goulas et al . , 2012 ) , raising an interesting possibility that the BM may regulate ISC self-renewal . The type IV collagen encoded by viking ( vkg ) is localized at the BM in adult midguts ( Figure 8D; Amcheslavsky et al . , 2009 ) . A previous study demonstrated that type IV collagens could physically interact with Dpp to restrict the range of BMP signaling in Drosophila ovaries ( Wang et al . , 2008 ) . Interestingly , we found that vkg trans-heterozygous midguts ( vkgk00236/vkg01209 and vkgk00236/vkg197 ) contained an excessive number of Dl+ cells and greatly reduced number of Su ( H ) -lacZ+ cells ( Figure 8F–H ) , suggesting that compromised Vkg activity promotes ISC self-renewal . pMad staining revealed that vkg mutant guts contained clusters of precursor cells exhibiting high levels of BMP signaling activity ( compare Figure 8J with Figure 8I , Figure 8—figure supplement 3 ) , suggesting that Vkg regulates ISC self-renewal through restricting BMP signaling . As a further support , we found that reducing the dose of Dpp suppressed ectopic ISC formation in vkg mutant guts ( Figure 8K–M ) . In addition , we found that vkg mutant guts retained less Dpp-GFP at the BM compared with control guts ( Figure 8N–Q , Figure 8—figure supplement 4 ) . Hence , the BM in Drosophila midguts serves as an integral part of the ISC niche by regulating the extracellular distribution of the niche signal . Several recent studies revealed that BMP signaling plays several roles in Drosophila adult midgut homeostasis ( Guo et al . , 2013; Li et al . , 2013a , 2013b ) . First , peak levels of BMP signaling in the middle region of midguts specify copper cell differentiation ( Guo et al . , 2013; Li et al . , 2013a ) . Second , BMP signaling also regulates ISC proliferation because reduction in BMP pathway activity results in ISC overproliferation ( Guo et al . , 2013; Li et al . , 2013b ) , which is in line with a growth inhibitory role of BMP signaling in mammalian intestines ( Haramis et al . , 2004; He et al . , 2004 ) . However , it is controversial whether BMP signaling regulates ISC proliferation cell autonomously or non-cell autonomously ( Guo et al . , 2013; Li et al . , 2013b ) . While Li et al . suggested that BMP signaling protects EC integrity and therefore indirectly regulates ISC proliferation; Guo et al . argued that BMP signaling regulates ISC proliferation in a strictly cell-autonomous fashion . The observations that both tkv8 and mad1–2 mutant clones caused excessive proliferation of neighboring wild type ISCs ( Figure 3—figure supplement 2B , C ) and that inactivation of BMP signaling in ECs stimulated ISC proliferation ( Figure 3—figure supplement 3B–E ) clearly support a non-cell autonomous role of BMP signaling . Indeed , inactivation of BMP signaling in ECs stimulated the production of JAK-STAT and EGFR pathway ligands that fuel ISC proliferation ( Li et al . , 2013b ) ( our own unpublished observation ) . Nevertheless , it remains possible that BMP signaling could regulate ISC proliferation through both cell autonomous and non cell-autonomous mechanisms , as appear to be the case for Hpo signaling ( Karpowicz et al . , 2010; Ren et al . , 2010; Shaw et al . , 2010; Staley and Irvine , 2010; Ren et al . , 2013 ) . Uncertainty also exists regarding how BMP signaling regulates the growth and proliferation of mammalian intestines . While an early work suggested that BMP signaling acts directly on stem/progenitor cells by antagonizing Wnt signaling ( He et al . , 2004 ) , later studies argued that BMP signaling acts on stromal cells to indirectly regulate stem/progenitor cell proliferation ( Kim et al . , 2006; Auclair et al . , 2007 ) . Future studies are needed to clarify the exact role of BMP signaling in the regulation of ISC proliferation . In this study , we uncover novel functions of BMP in the regulation of Drosophila adult midgut homeostasis , that is , BMP serves as a niche signal to promote ISC self-renewal . In addition , we find that BMP signaling is also required for appropriate lineage differentiation into mature EC and EE . Several lines of evidence suggest that BMP regulates ISC/EB fate choice rather than simply serving as a growth/survival factor for ISC maintenance . First , loss of BMP signaling in precursor cells resulted in a rapid loss of ISC accompanied by an increase in the number of EB . For example , loss of BMP signaling by either put RNAi in precursor cells or put null ISC lineage clones produced mostly EB/EB pairs as indicated by cell specific markers as well as by two-color lineage tracing experiments ( Figures 2 and 3 ) . By contrast , blockage of cell growth/proliferation by inhibiting EGFR pathway or by inactivating dMyc only resulted in a gradual decline in the ISC number; and Ras or dMyc mutant clones retained the stem cell marker for a long period of time ( 2–3 weeks ) even though they failed to divide ( Xu et al . , 2011; Ren et al . , 2013 ) . Second , BMP signaling is asymmetric in ISC/EB pairs with the basally located ISCs exhibiting higher levels of BMP signaling activity than the more apically localized EBs . The differential BMP signaling is consistent with BMP pathway activity promoting ISC fate . Third and perhaps the most compelling evidence is that ectopic BMP signaling can promote stem cell fate at the expense of EB fate . For example , gain of BMP signaling either by overexpressing a constitutively active form of Tkv in precursor cells or by misexpressing Dpp and Gbb resulted in the formation of large ISC-like cell clusters ( Figure 4K , L ) . Furthermore , the twin-spot lineage tracing experiments confirmed that ectopic BMP signaling favors the symmetric self-renewing ( ISC/ISC ) outcome of an ISC division . In seeking for the source of BMP ligands , we were surprised to find that both dpp and gbb are largely expressed in ECs and their expression patterns are complementary , that is , higher levels of gbb mRNA were detected in regions where dpp expression is low and vice versa . Low levels of Dpp>GFP expression were detected in ECs along the entire A/P axis ( Figure 6B ) although our RNA in situ hybridization clearly missed low levels of dpp mRNA in certain regions . Similarly , our gbb RNA in situ probe may have missed low levels of gbb expression in the anterior and posterior regions of the midguts . Therefore , it is very likely that Dpp and Gbb are coexpressed in most if not all ECs albeit at different levels in different regions . A recent study confirmed that dpp is expressed in ECs ( Li et al . , 2013a ) . Although dpp expression was also detected in trachea cells ( Figure 7—figure supplement 4A; Li et al . , 2013b ) as well as in VM ( Guo et al . , 2013 ) , inactivation of Dpp from these tissues did not affect ISC maintenance ( Figure 7—figure supplement 4F–H ) . By contrast , inactivation of either Dpp or Gbb in ECs resulted in stem cell loss phenotype whereas increasing the production of Dpp and Gbb in ECs increased stem cell number ( Figure 7 , Figure 7—figure supplement 1I ) . These observations suggest that EC-derived BMP serves as the niche signal for ISC self-renewal although we cannot exclude the possibility that trachea- or VM-derived Dpp could play a minor role . Hence , our study establishes a new paradigm for studying stem cell niche and its regulation because in most other systems , stem cell niches are derived from lineages distinct from the stem cells they support . What then is the advantage of utilizing epithelia as a niche to control ISC self-renewal ? We speculate that the employment of midgut epithelia as stem cell niche may provide a mechanism for direct communication between the niche and the environment , allowing the production of niche signal and stem cell number to be regulated in response to various physiological and pathological stimuli . Hence , it would be interesting to explore in the future whether BMP production in ECs is dynamically regulated under various stress conditions where change in the stem cell number has also been observed ( Amcheslavsky et al . , 2009; Biteau et al . , 2008; Jiang et al . , 2009; McLeod et al . , 2010; O’Brien et al . , 2011 ) . Our MARCM clone analysis for Tkv shows that loss of Tkv can support the ISC self-renewal , but loss of Sax and Tkv at the same time causes the loss of ISCs . This result suggests that low levels of BMP pathway activity conferred by Sax-Put receptor complex appear to be enough to support ISC self-renewal ( although not enough to prevent ISC from overproliferation ) . On the other hand , our results showed that removal of either Dpp or Gbb resulted in ISC loss , implying that Dpp or Gbb homodimers failed to produce enough BMP activity to support ISC self-renewal . It is not clear why Dpp and Gbb homodimers fail to elicit low levels of BMP pathway activity similar to those transduced by Sax . One possibility is that Dpp and Gbb are produced at much lower levels in the midguts compared to early embryos and imaginal discs so that Dpp and Gbb homodimer concentration may not reach a critical threshold for effective signaling in midgut precursor cells . Second , competition between ECs and ISCs for limited amount of BMP may also restrict the availability of BMP ligands to ISCs . A third possibility is that extracellular Dpp or Gbb homodimers may not be stable in the midguts so that depletion of one ligand might cause concomitant reduction in the levels of the other . The finding that ECs are the major source of niche signal also raises an important question of how the apical/basal BMP activity gradient is established . Interestingly , we find that Dpp appears to be secreted preferentially from the basal side of ECs , and that Dpp can form a dimer with Gbb to influence its apical/basal distribution . Furthermore , we obtain evidence that BMP signaling range is restricted by the BM-enriched type IV collagen/Vkg . Because type IV collagen/Vkg can physically interact with Dpp ( Wang et al . , 2008 ) , we propose that basal secretion coupled with BM trapping may establish a steep BMP activity gradient so that basally localized ISC daughter cells receive higher levels of BMP than their more apically localized siblings , resulting in differential BMP signaling ( Figure 9 ) . 10 . 7554/eLife . 01857 . 028Figure 9 . A working model for how BMP regulates ISC self-renewal . Basal/basolateral secretion coupled with basement membrane ( BM ) trapping sets up an apical-basal BMP activity gradient consisting of Dpp-Gbb heterodimers . Basally localized ISC daughter cells activate higher levels of BMP signaling that promotes ISC self-renewal by antagonizing N . See text for details . DOI: http://dx . doi . org/10 . 7554/eLife . 01857 . 028 Our genetic epistasis experiments suggested that BMP signaling promotes ISC self-renewal by antagonizing N ( Figure 5 ) . Although it is possible that BMP and N pathways could act in parallel and exert opposing influence on ISC/EB fate choice , we favor a model in which BMP promotes ISC fate by inhibiting N pathway activity because loss of BMP signaling in precursor cells resulted in the ectopic expression of Su ( H ) -lacZ , which is a N pathway activity reporter , whereas gain of BMP signaling suppressed the expression of Su ( H ) -lacZ in precursor cells ( Figures 3 and 4 ) . Therefore , we propose that differential BMP signaling sets up a difference in the levels of N signaling activity between the two daughter cells after an ISC division . The initial small difference in the N pathway activity between the apical and basal ISC daughter cells is amplified by N feedback regulation ( Figure 9; Axelrod , 2010 ) : N signaling in the apical cell inhibits Dl expression and Dl accumulation in the basal cell further strengthens N signaling in its apical sibling . Intriguingly , we also found that excessive N signaling could block BMP pathway activity , as indicated by the blockage of pMad staining in precursor cells expressing NICD ( Figure 5M ) . We speculate that elevated N pathway activity in the apical ISC daughter cell may attenuate BMP signaling in this cell , forming another feedback mechanism to reinforce EB fate choice . Our model can explain the observations that the absolute levels of BMP pathway activity is not critical for ISC self-renewal since partial loss of BMP signaling did not lead to stem cell loss ( Figure 3 ) . As long as two ISC daughter cells transduce different levels of BMP signal , the N and BMP signaling feedback loops can amplify the initial small difference , leading to a bistable cell fate choice . A similar mechanism has been postulated to account for the bistable R3/R4 fate determination in the Drosophila compound eye , which is regulated by the interplay between Wg/Wnt and N signaling ( Cooper and Bray , 1999; Fanto and Mlodzik , 1999 ) . Differential BMP signaling might not be the only mechanism responsible for ISC self-renewal . A recent study revealed that aPKC is asymmetrically inherited by apically localized ISC daughter cells and that aPKC promotes N pathway activity ( Goulas et al . , 2012 ) . Therefore , asymmetric segregation of aPKC may dampen BMP signaling response in the apically localized ISC daughter cells , which could contribute to the differential BMP signaling . Future study will determine how BMP signaling inhibits N and how the extrinsic and intrinsic mechanisms are integrated to coordinate ISC self-renewal and differentiation . Mutant clones for tkv , put or mad were generated using the MARCM system ( Lee and Luo , 2001 ) . Fly stocks were crossed and cultured at 18°C . 5-day-old F1 adults with the appropriate genotypes were subjected to heat shock at 37°C for 1 hr . After clone induction , flies were raised at room temperature for 5 , 8 , 12 , or 18 days before dissection . For experiments involving tubGal80ts , crosses were set up and cultured at 18°C to restrict Gal4 activity . 2 to 3-day-old F1 adult flies were then shifted to 29°C to inactivate Gal80ts , allowing Gal4 to activate UAS transgenes . For twin-spot clonal analysis , 3–5-day-old adult flies were grown at 29°C for 7 days ( for Put RNAi experiments ) or 3 days ( for TkvQ235D overexpression experiments ) before heat shock at 37°C for 30 min to induce clones . After 1-day recovery at 29°C , the flies were raised at 18°C for 3–4 days . The guts were dissected out and analyzed by confocal microscopy . In general , 2–3-day-old F1 adult flies were then shifted to 29°C to inactivate Gal80ts for 8 days , then these adult flies were used for feeding experiments . Flies were cultured in an empty vial containing a piece of 2 . 5 × 3 . 75-cm chromatography paper ( Fisher , Pittsburgh , PA ) wet with 5% sucrose solution as feeding medium . Flies were fed with 5% of DSS ( MP Biomedicals , Santa Ana , CA ) or 25 μg/ml bleomycin ( Sigma , St . Louis , MO ) dissolved in 5% sucrose ( mock treatment ) for 2 days at 29°C . Female flies were used for gut immunostaining in all experiments . The entire gastrointestinal tracts were dissected out and fixed in 1 X PBS plus 8% EM grade paraformaldehyde ( Polysciences ) for 2 hr . Samples were washed and incubated with primary and secondary antibodies in a solution containing 1 X PBS , 0 . 5% BSA , and 0 . 1% Triton X-100 . The following primary antibodies were used: mouse anti-Delta ( DSHB ) , 1:100; mouse anti-Pros ( DSHB ) , 1:100; mouse anti-Arm ( DSHB ) , 1:100; mouse anti-integrin βPS/Mys ( DSHB ) , 1:100; rabbit anti-LacZ ( MP Biomedicals ) , 1:1000; rabbit and mouse anti-PH3 ( Millipore , Billerica , MA ) , 1:1000; goat anti-GFP ( Abcam , Cambridge , MA ) , 1:1000; mouse anti-pMad antibody ( Persson et al . , 1998 ) , 1:300; rabbit anti-Pdm1 ( gift from Dr Xiaohang Yang , Institute of Molecular and Cell Biology , Singapore ) , rabbit anti-Lava lamp , 1:300; 1:500; DRAQ5 ( Cell Signaling Technology , Danvers , MA ) , 1:5000; Phalloidin , 1:100 . Quantification of immunostaining was performed using ImageJ software . RNA fluorescent in situ hybridization ( FISH ) in the midgut was performed as described ( Raj et al . , 2008 ) . Forty eight 20-mer DNA oligos complementing the coding region of the target genes ( dpp and gbb ) were designed and labeled with a fluorophore ( http://www . biosearchtech . com/ , Petaluma , CA ) . For RNA in situ hybridization , the midguts were first dissected and fixed in 8% paraformaldehyde at 4°C for overnight , followed by washing with PBS and Triton X-100 ( 0 . 1% ) for three times ( 15 min each ) . The samples were further permeabilized in 70% ethanol at 4°C for overnight . The hybridization was performed according to the online protocol ( http://www . biosearchtech . com/stellarisprotocols ) . Total RNA was extracted from 10 female guts using RNeasy Plus Mini Kit ( #74134; Qiagen , Valencia , CA ) , and cDNA was synthesized using the iScript cDNA synthesis kit ( Bio-Rad , Hercules , CA ) . RT-qPCR was performed using iQ SYBR Green System ( Bio-Rad ) . Primer sequences used are: 5′-gtgcgaagttttacacacaaaga-3′ and 5′-cgccttcagcttctcgtc-3′ ( for dpp ) , and 5′-cgctggaactctcgaaataaa-3′ and 5′-ccacttgcgatagcttcaga-3′ ( for gbb ) . RpL11 was used as a normalization control . Relative quantification of mRNA levels was calculated using the comparative CT method . For each genotype , 30 Midguts were dissected and mashed in 400 µl lysis buffer: 50 mMTris-HCl ( pH 8 . 0 ) , 100 mM NaCl , 10 mM NaF , 1 mM Na3VO4 , 1% NP40 , 10% glycerol , 1 . 5 mM EDTA ( pH 8 . 0 ) , protease inhibitor tablets ( Roche , IN ) . 40 µl supernatants were taken out and placed into another tube as the whole cell lysates ( WCL ) . The remaining supernatants were used for IP . Anti-rabbit GFP antibodies and protein A beads were incubated with the lysate for overnight at 4°C , and the beads was washed for three time with the lysis buffer . The immunoprecipitates and WCLs were separated on SDS-PAGE , followed by western blot using anti-GFP and anti-HA antibodies . Figure 1: B: esgts: w; esg-Gal4 tub-Gal80ts UAS-GFP/+ , esgts>Tkv-RNAi105834: w; esg-Gal4 tub-Gal80ts UAS-GFP/UAS-Tkv-RNAi105834 , esgts>Put-RNAi: w; esg-Gal4 tub-Gal80ts UAS-GFP/UAS-Put-RNAi . C–E , L–N: w; esg-Gal4 tub-Gal80ts UAS-GFP/+; UAS-flp , act>CD2>gal4/+ , F–H , O–Q: w; esg-Gal4 tub-Gal80ts UAS-GFP/UAS-Tkv-RNAi105834; UAS-flp , act>CD2>gal4/+ , I–K , R–T: w; esg-Gal4 tub-Gal80ts UAS-GFP/Put-RNAi; UAS-flp , act>CD2>gal4/+ . Figure 2: A: w; esg-Gal4 tub-Gal80ts UAS-GFP/+ , B: w; esg-Gal4 tub-Gal80ts UAS-GFP/UAS-Tkv-RNAi105834 , C: w; esg-Gal4 tub-Gal80ts UAS-GFP/UAS-Put-RNAi , D: Su ( H ) -LacZ; esg-Gal4 tub-Gal80ts UAS-GFP/+ , E: Su ( H ) -LacZ; esg-Gal4 tub-Gal80ts UAS-GFP/UAS-Tkv-RNAi105834 , F: Su ( H ) -LacZ; esg-Gal4 tub-Gal80ts UAS-GFP/UAS-Put-RNAi , K: yw hsflp; esgGal4 Tub-Gal80ts; FRT82B ubi-GFP/FRT82B ubi-RFP , L: yw hsflp; esgGal4 Tub-Gal80ts/UAS-Put-RNAi; FRT82B ubi-GFP/FRT82B ubi-RFP . Figure 3: A , G: yw UAS-GFP hsflp/Su ( H ) -LacZ; tub-Gal4/+; FRT82B tub-Gal80/FRT82B , B , H: yw UAS-GFP hsflp/Su ( H ) -LacZ; tub-Gal4/+; FRT82B tub-Gal80/FRT82B putP , C , E: yw UAS-GFP hsflp/+; tub-Gal4/+; FRT82B tub-Gal80/FRT82B , D , F: yw UAS-GFP hsflp/+; tub-Gal4/+; FRT82B tub-Gal80/FRT82B putP , L: yw UAS-GFP hsflp; tub-Gal80 FRT40/FRT40; tub-Gal4/+ , M: yw UAS-GFP hsflp; tub-Gal80 FRT40/tkv8 FRT40; tub-Gal4/+ , N: yw UAS-GFP hsflp; tub-Gal80 FRT40/FRT40; tub-Gal4/UAS-Sax-RNAi . O: yw UAS-GFP hsflp; tub-Gal80 FRT40/tkv8 FRT40; tub-Gal4/UAS-Sax-RNAi , P: yw UAS-GFP hsflp; tub-Gal80 FRT40/mad1–2 FRT40; tub-Gal4/+ , Q: yw UAS-GFP hsflp; tub-Gal80 FRT40/FRT40; tub-Gal4/UAS-Mad-RNAi , R: yw UAS-GFP hsflp; tub-Gal80 FRT40/mad1–2 FRT40; tub-Gal4/UAS-mad-RNAi , S: yw UAS-GFP hsflp; tub-Gal80 FRT40/tkv8 mad1–2 FRT40; tub-Gal4 . Figure 4: A–B′ , E–F′: w; Su ( H ) Gal4 UAS-GFP/+; Dl-LacZ/+ , C: Su ( H ) Gal4 UAS-GFP/+; dad-LacZ/+ , G: Su ( H ) -LacZ; esg-Gal4 tub-Gal80ts UAS-GFP/+ , H: Su ( H ) -LacZ; esg-Gal4 tub-Gal80ts UAS-GFP/+; UAS-TkvQ235D/+ , I: Su ( H ) -LacZ; esg-Gal4 tub-Gal80ts UAS-GFP/+; UAS-Dpp/UAS-Gbb , J , M: esg-Gal4 tub-Gal80ts UAS-GFP/+ , K , N: esg-Gal4 tub-Gal80ts UAS-GFP/+; UAS-TkvQ235D/+ , L , O: esg-Gal4 tub-Gal80ts UAS-GFP/+; UAS-Dpp/UAS-Gbb . Q: yw hsflp; esgGal4 Tub-Gal80ts; FRT82B ubi-GFP/FRT82B ubi-RFP , R: yw hsflp/UAS-TkvQ235D; esgGal4 Tub-Gal80ts/+; FRT82B ubi-GFP/FRT82B ubi-RFP . Figure 5: A: esg-Gal4 tub-Gal80ts UAS-GFP/+; UAS-N ICD/+ , B: esg-Gal4 tub-Gal80ts UAS-GFP/+; UAS-TkvQ235D/+ , C: esg-Gal4 tub-Gal80ts UAS-GFP/+; UAS-Dpp UAS-Gbb , D: esg-Gal4 tub-Gal80ts UAS-GFP/+; UAS-N ICD/UAS-TkvQ235D , E: esg-Gal4 tub-Gal80ts UAS-GFP/+; UAS-N ICD/UAS-Dpp UAS-Gbb , F: esg-Gal4 tub-Gal80ts UAS-GFP/+; UAS-N-RNAi/UAS-N-RNAi , G: esg-Gal4 tub-Gal80ts UAS-GFP/UAS-Tkv-RNAi105834 , H: esg-Gal4 tub-Gal80ts UAS-GFP/UAS-Put-RNAi , I: esg-Gal4 tub-Gal80ts UAS-GFP/UAS-Tkv-RNAi105834; UAS-N-RNAi/UAS-N-RNAi , J: esg-Gal4 tub-Gal80ts UAS-GFP/UAS-Put-RNAi; UAS-N-RNAi/UAS-N-RNAi , K: esg-Gal4 tub-Gal80ts UAS-GFP , L: esg-Gal4 tub-Gal80ts UAS-GFP/+ , UAS-N-RNAi/UAS-N-RNAi . M: esg-Gal4 tub-Gal80ts UAS-GFP/+; UAS-N ICD/+ . Figure 6: A , D , E: W; UAS-GFP/+; dppGal4/+ , B: UAS-GFP/UAS-GFP; dppGal4/+ , F: w; UAS-GFP/+; Myo1AGal4/+ . Figure 7: A: Su ( H ) -LacZ; Myo1AGal4 tub-Gal80tsUAS-GFP/+ , B: Su ( H ) -LacZ; Myo1AGal4 tub-Gal80tsUAS-GFP/+; UAS-Dpp UAS-Gbb/+ , C: Myo1AGal4 tub-Gal80tsUAS-GFP/UAS-Dicers , D: Myo1AGal4 tub-Gal80tsUAS-GFP/UAS-Dicer2; UAS-Dpp-RNAiS/UAS-Dpp-RNAiS , E: Myo1AGal4 tub-Gal80tsUAS-GFP/UAS-Dicer2; UAS-Gbb-RNAiS/UAS-Gbb-RNAiS , F: Myo1AGal4 tub-Gal80tsUAS-GFP/UAS-Dicer2; UAS-Dpp-RNAiW UAS-Gbb-RNAiW/UAS-Dpp-RNAiW UAS-Gbb-RNAiW , G: Su ( H ) -LacZ; Myo1AGal4 tub-Gal80tsUAS-GFP/UAS-Dicer2 , H: Su ( H ) -LacZ; Myo1AGal4 tub-Gal80tsUAS-GFP/UAS-Dicer2; UAS-Dpp-RNAiS/UAS-Dpp-RNAiS , I: Su ( H ) -LacZ; Myo1AGal4 tub-Gal80tsUAS-GFP/UAS-Dicer2; UAS-Gbb-RNAiS/UAS-Gbb-RNAiS . Figure 8: A , Myo1AGal4 tub-Gal80ts/UAS-GFP , B , C: Myo1AGal4 tub-Gal80ts/+; UAS-Dpp-GFP/+ , D: Su ( H ) -LacZ; vkg-GFP/+ , E: Su ( H ) -LacZ , F: Su ( H ) -LacZ; vkgk00236/vkg01209 , G: Su ( H ) -LacZ; vkgk00236/vkg197 , J , K: vkgk00236/vkg01209 , L: vkgk00236dpphr56/vkg01209 , N–O: Myo1AGal4/+; tub-Gal80ts/UAS-Dpp-GFP , P–Q: vkg00236Myo1AGal4/vkg01209; tub-Gal80ts/UAS-Dpp-GFP . Figure 2—figure supplement 1: A: w; esg-Gal4 tub-Gal80ts UAS-GFP/+ , B: w; esg-Gal4 tub-Gal80ts UAS-GFP; UAS-Diap1 , C: w; esg-Gal4 tub-Gal80ts UAS-GFP/UAS-Put-RNAi , D: w; esg-Gal4 tub-Gal80ts UAS-GFP/UAS-Put-RNAi; UAS-Diap1 . Figure 2—figure supplement 2: A–D: yw hsflp; esgGal4 Tub-Gal80ts; FRT82B ubi-GFP/FRT82B ubi-RFP . Figure 3—figure supplement 1: B: w; esg-Gal4 tub-Gal80ts UAS-GFP/+ , C: w; esg-Gal4 tub-Gal80ts UAS-GFP/UAS-Tkv-RNAi40937 , D: w; esg-Gal4 tub-Gal80ts UAS-GFP/UAS-Sax-RNAi , E: w; esg-Gal4 tub-Gal80ts UAS-GFP/UAS-Tkv-RNAi40937; UAS-Sax-RNAi/+ . Figure 3—figure supplement 2: A: yw UAS-GFP hsflp; tub-Gal80 FRT40/FRT40; tub-Gal4/+ , B: yw UAS-GFP hsflp; tub-Gal80 FRT40/tkv8 FRT40; tub-Gal4/+ , C: yw UAS-GFP hsflp; tub-Gal80 FRT40/mad1–2 FRT40; tub-Gal4/+ , D: yw UAS-GFP hsflp; tub-Gal80 FRT40/tkv8 mad1–2 FRT40; tub-Gal4 . Figure 3—figure supplement 3: A: w; myo1A-Gal4 tub-Gal80ts UAS-GFP , B: w; myo1A-Gal4 tub-Gal80ts UAS-GFP/UAS-Dicer2; UAS-Tkv-RNAi105834 , C: w; myo1A-Gal4 tub-Gal80ts UAS-GFP/UAS-Put-RNAi . D: w; myo1A-Gal4 tub-Gal80ts UAS-GFP/UAS-Dicer2; UAS-mad-RNAi . Figure 4—figure supplement 1: A , A′: esg-Gal4 tub-Gal80ts UAS-GFP/+; UAS-Dpp/+ , B , B′: esg-Gal4 tub-Gal80ts UAS-GFP/+; UAS-Gbb/+ . Figure 6—figure supplement 1: B: MS1096; UAS-dpp , D: MS1096; UAS-Gbb , E–L: UAS-GFP/+; dpp-Gal4/+ . Figure 6—figure supplement 2: A–C′ , G–I′: esg-Gal4/+; UAS-GFP/+ , D–F′ , J–L′: UAS-GFP/+; how-Gal4/+ . Figure 7—figure supplement 1: F: Myo1AGal4 tub-Gal80tsUAS-GFP/+ , G: Myo1AGal4 tub-Gal80tsUAS-GFP/UAS-Dicer2; UAS-Dpp-RNAiw/UAS-Dpp-RNAiw , H: Myo1AGal4 tub-Gal80tsUAS-GFP/UAS-Dicer2; UAS-Gbb-RNAiw/UAS-Gbb-RNAiw . Figure 7—figure supplement 2: A , A′ , D , G: Myo1AGal4 tub-Gal80tsUAS-GFP/+ , B , B′ , E , E′: Myo1AGal4 tub-Gal80tsUAS-GFP/UAS-Dicer2; UAS-Dpp-RNAiS/UAS-Dpp-RNAiS , C , C′ , F , I: Myo1AGal4 tub-Gal80tsUAS-GFP/UAS-Dicer2; UAS-Gbb-RNAiS/UAS-Gbb-RNAiS . Figure 7—figure supplement 3: A: Myo1AGal4 tub-Gal80tsUAS-GFP/+ , B: Myo1AGal4 tub-Gal80tsUAS-GFP/UAS-Dicer2; UAS-Dpp-RNAiS/UAS-Dpp-RNAiS . Figure 7—figure supplement 4: A: btlGal4/+; UAS-GFP , B: btlGal4/+; tub-Gal80tsUAS-GFP , C: btlGal4/+; tub-Gal80tsUAS-GFP/UAS-Dpp , D: btlGal4/+; tub-Gal80tsUAS-GFP/+; UAS-Dpp UAS-Gbb/+ , E: BtlGal4; tub-Gal80ts , F: BtlGal4/UAS-Dicer2; UAS-Dpp-RNAiS tub-Gal80ts/UAS-Dpp-RNAiS , G: tub-Gal80tsUAS-GFP/UAS-Dicer2; How-Gal4 UAS-Dpp-RNAiS/UAS-Dpp-RNAiS . Figure 8—figure supplement 1: A: Myo1AGal4 tub-Gal80ts; UAS-Gbb-GFP , B: Myo1AGal4 tub-Gal80ts; UAS-Gbb-GFP/UAS-Dpp , C: Myo1AGal4 tub-Gal80ts; UAS-Gbb-GFP/UAS-Dpp-HA . Figure 8—figure supplement 4: A–A″ , vkg-GFP , B–C , Myo1AGal4 tub-Gal80ts/+; UAS-Dpp-GFP/+ , D–E , vkgk00236Myo1AGal4/vkg01209; tub-Gal80ts/UAS-Dpp-GFP .	Keeping an organ in top condition requires a steady supply of fresh cells to replace those that are dead or damaged . This is particularly critical for the epithelial cells lining the intestines , which only live for a few days , but are necessary for digesting food . These cells cannot simply reproduce by cell division , so they must be replenished by adult stem cells—adaptable cells that can produce any of the cell types found in a given organ . When an adult stem cell divides , two daughter cells are produced . Normally , one of these remains in the stem state , and the other becomes a particular type of cell for use in the organ . Exactly how each daughter cell knows what to become is unclear . However , it is known that in addition to communicating with each other , stem cells also communicate with their immediate surroundings , which is known as a niche . For many processes , the molecules and mechanisms used in niche signaling remain to be discovered . The midgut of fruit flies presents a relatively simple stem cell system for study , and has the added advantage that its cells behave in ways that are similar to the cells that make up the intestines of mammals . By developing a method of tracking the two daughter cells of a single stem cell simultaneously , Tian and Jiang have been able to uncover new details about how this niche operates . Epithelial cells in the gut produce molecules called bone morphogenetic proteins ( BMPs ) that influence how bone and many other types of body tissues form . Tian and Jiang have found that two types of BMP are the signals responsible for keeping daughter cells in the stem state . When released from the base of the epithelial cells , BMPs can only travel a very short distance before other proteins trap them . As a result , one of a pair of daughter cells receives a higher level of the signal and remains as a stem cell . This cell then sends a signal to the other daughter cell , telling it to form a specialized cell .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine" ]	2014	Intestinal epithelium-derived BMP controls stem cell self-renewal in Drosophila adult midgut
HNF4α has been implicated in colitis and colon cancer in humans but the role of the different HNF4α isoforms expressed from the two different promoters ( P1 and P2 ) active in the colon is not clear . Here , we show that P1-HNF4α is expressed primarily in the differentiated compartment of the mouse colonic crypt and P2-HNF4α in the proliferative compartment . Exon swap mice that express only P1- or only P2-HNF4α have different colonic gene expression profiles , interacting proteins , cellular migration , ion transport and epithelial barrier function . The mice also exhibit altered susceptibilities to experimental colitis ( DSS ) and colitis-associated colon cancer ( AOM+DSS ) . When P2-HNF4α-only mice ( which have elevated levels of the cytokine resistin-like β , RELMβ , and are extremely sensitive to DSS ) are crossed with Retnlb-/- mice , they are rescued from mortality . Furthermore , P2-HNF4α binds and preferentially activates the RELMβ promoter . In summary , HNF4α isoforms perform non-redundant functions in the colon under conditions of stress , underscoring the importance of tracking them both in colitis and colon cancer . Hepatocyte nuclear factor 4alpha ( HNF4α ) ( NR2Α1 ) is a highly conserved member of nuclear receptor superfamily of ligand-dependent transcription factors that is expressed in liver , kidney , pancreas , stomach and intestine ( Sladek et al . , 1990 ) . HNF4α is best known for its role in the liver where it is a master regulator of liver-specific gene expression and essential for adult and fetal liver function ( Hayhurst et al . , 2001; Kaestner , 2010; Bolotin et al . , 2010; Odom , 2004 ) . HNF4α is also known for its role in the pancreas where it regulates insulin secretion from beta cells ( Gupta et al . , 2007; 2005; Miura et al . , 2006 ) . Mutations in the HNF4Α gene and promoter regions are associated with Maturity Onset Diabetes of the Young 1 ( MODY1 ) ( Ellard and Colclough , 2006 ) . In contrast , the role of HNF4α in the intestine has only recently been investigated . Knockout of the Hnf4a gene in the embryonic mouse colon results in disrupted crypt topology , and a decreased number of epithelial and mature goblet cells ( Garrison et al . , 2006 ) , while the adult intestinal knockout shows defects in the balance between proliferation and differentiation as well as immune function , ion transport , epithelial barrier function and oxidative stress ( Ahn et al . , 2008; Cattin et al . , 2009; Darsigny et al . , 2009; 2010; Chahar et al . , 2014 ) . Dysregulation of the HNF4A gene is linked to several gastrointestinal disorders including colitis and colon cancer and a single nucleotide polymorphism in the HNF4A gene region is associated with ulcerative colitis ( Ahn et al . , 2008; Chellappa et al . , 2012; Tanaka et al . , 2006; Oshima et al . , 2007; Barrett et al . , 2009 ) . While it is clear that HNF4α is critical for normal colon function , it is not known which transcript variant is the most relevant . There are two different promoters ( proximal P1 and distal P2 ) in the HNF4α gene that are both active in the colon . The promoters are conserved from frog to human and , along with alternative splicing , give rise to nine different transcript variants of HNF4α ( Huang et al . , 2009 ) ( Figure 1A ) . The major isoforms of the P1 promoter are HNF4α1/α2 while the P2 promoter gives rise to HNF4α7/α8: distinct first exons result in an altered A/B domain which harbors the activation function 1 ( AF-1 ) while the DNA and ligand binding domains are identical . The two promoters are expressed under unique temporal and spatial conditions , with the large and small intestine being the only adult tissues that express both P1- and P2-HNF4α ( Tanaka et al . , 2006; Nakhei et al . , 1998 ) . While a loss of P1- but not P2-HNF4α has been noted in colon cancer ( Chellappa et al . , 2012; Tanaka et al . , 2006 ) , the specific roles of the HNF4α isoforms remain obscure . For example , P1-driven HNF4α acts as a tumor suppressor in mouse liver ( Hatziapostolou et al . , 2011; Walesky et al . , 2013a ) . In contrast , the HNF4A gene and protein are amplified in human colon cancer ( Cancer Genome Atlas Network , 2012; Zhang et al . , 2014 ) although the different isoforms were not distinguished in those studies . We recently showed that ectopic expression of P1- but not P2-HNF4α decreased the tumorigenic potential of the human colon cancer cell line HCT116 in a mouse xenograft model ( Vuong et al . , 2015 ) , suggesting that the different HNF4α isoforms may indeed play distinct roles in the colon . Here , we investigate the role of P1- and P2-HNF4α isoforms in the mouse colon using genetically engineered mice that express either the P1- or the P2-HNF4α isoforms ( Briançon and Weiss , 2006 ) . We show that in wildtype ( WT ) mice P1- and P2-HNF4α are expressed in different compartments in the colonic epithelium , interact with distinct sets of proteins , regulate the expression of unique sets of target genes , and play distinct roles during pathological conditions such as colitis and colitis-associated colon cancer ( CAC ) . We also provide genetic and biochemical evidence indicating that RELMβ , a member of the RELM/FIZZ family of cytokines , plays a critical role in the response of HNF4α to colitis and appears to be both directly and indirectly regulated by HNF4α . In the distal colon , the bottom two-thirds of the crypt and the top one-third , including surface epithelium , are functionally categorized as proliferative and differentiated compartments , respectively ( Potten et al . , 1997 ) . We used monoclonal antibodies specific to the different HNF4α isoforms ( Chellappa et al . , 2012; Tanaka et al . , 2006 ) ( Figure 1A ) to examine the distribution of P1- and P2-HNF4α along the crypt-surface axis . The P1/P2 antibody , which recognizes both P1- and P2-HNF4α , shows HNF4α expression in both crypt and surface epithelial cells ( Figure 1B ) , as reported previously ( Ahn et al . , 2008; Darsigny et al . , 2009; Chahar et al . , 2014 ) . In contrast , the isoform-specific antibodies reveal that P1-HNF4α is expressed mainly in the differentiated compartment , not in the proliferative compartment as defined by NKCC1 staining ( Figure 1C ) . P2-HNF4α was observed primarily in the bottom half of the crypt ( Figure 1B ) and co-localized with the proliferation marker Ki67 in isolated colonic crypts ( Figure 1D ) . While there was some expression of P2-HNF4α in the differentiated compartment ( i . e . , non Ki67 expressing cells ) , it was notably absent from the surface epithelium ( Figure 1B ) . 10 . 7554/eLife . 10903 . 003Figure 1 . Differential localization of HNF4α isoforms in mouse colonic crypts . ( A ) Schematic of the mouse ( and human ) Hnf4a gene showing the two promoters ( P1 and P2 ) ( top ) and the P1- and P2-driven HNF4α isoforms that they express ( bottom ) . The differential N-terminal A/B domain ( indicated in blue and orange ) as well as epitopes to isoform-specific ( αP1 and αP2 ) and common ( αP1/P2 ) antibodies are indicated . DBD , DNA binding domain; LBD , ligand binding domain; F , F domain . ( B–D ) IF and immunohistochemistry of distal colon ( B , C ) or isolated colonic crypts ( D ) stained for the indicated proteins using the antibodies in ( A ) ( B: 40X magnification; C , D: 25X magnification with digital zoom ) . NKCC1 ( Slc12a2 ) ( C ) and Ki67 ( D ) mark the proliferative compartment of the crypt . Representative images from two independent experiments ( n=2–4 mice per genotype ) are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 10903 . 003 Previous studies showed that HNF4A expression is decreased in human inflammatory bowel disease ( IBD ) patients and intestine-specific deletion of the mouse Hnf4a gene increases susceptibility to dextran sodium sulfate ( DSS ) -induced colitis ( Ahn et al . , 2008 ) and can lead to chronic inflammation even in the absence of DSS ( Darsigny et al . , 2009 ) . However , these studies do not address the role of the individual HNF4α isoforms . We treated young adult male mice ( WT ) with 2 . 5% DSS and found a statistically significant decrease in total HNF4α following 5 days of DSS treatment , as others have observed ( Ahn et al . , 2008; Chahar et al . , 2014 ) , and an increase in HNF4α during the recovery phase , especially P1-HNF4α ( Figure 2A , B ) . Contrary to the restricted expression of P1-HNF4α in the differentiated compartment in untreated mice , P1-HNF4α was also expressed near the bottom of the crypt after DSS treatment ( Figure 2C ) , consistent with substantial loss of proliferating cells following DSS treatment ( Tessner et al . , 1998 ) . 10 . 7554/eLife . 10903 . 004Figure 2 . Dysregulation of P1- and P2-HNF4α in mouse models of colitis and colon cancer . ( A ) IB of WCE from the distal colon of WT mice treated with 2 . 5% DSS for 5 days followed by 0 or 3 days recovery , and an analogous region of untreated ( Control ) mice . Each lane is from a different mouse . The position of the molecular weight marker ( 52 kD ) is shown . ( B ) Quantification of the HNF4α signal in ( A ) normalized to total protein , as determined by Coomassie staining of the same blot . For the purposes of quantification the outlier in lane 8 was omitted . P<0 . 05 , P<0 . 005 . ( C ) Representative IF of distal colon from untreated and DSS-treated WT mice ( n=3–4 per condition ) stained with P1-HNF4α antibody ( 40X magnification ) . Arrow indicates P1-HNF4α expressing cells near the bottom of the crypt in the DSS-treated mice . ( D ) IB as in ( A ) but from the tumor area of WT mice treated with 10 mg/kg AOM and three cycles of a 7-day DSS treatment and harvested at ~95 days . Three gels were run in parallel with the same extracts; one representative β-actin stain is shown . ( E ) IB analysis as in ( D ) but from mice injected six times with 10 mg/kg AOM and harvested at ~150 days . Shown is one representative of the three Coomassie stains performed for loading verification . DOI: http://dx . doi . org/10 . 7554/eLife . 10903 . 004 In a mouse model of colitis-associated colon cancer ( CAC ) in which a single injection of azoxymethane ( AOM ) is followed by multiple treatments of DSS in the drinking water , we found that P1-HNF4α is greatly reduced in tumors compared to untreated controls but that total HNF4α protein was only marginally reduced ( Figure 2D ) , suggesting that P2-HNF4α was not affected . The P1-HNF4α decrease correlated with an increase in active Src ( pSrc ) , consistent with our earlier finding that Src specifically phosphorylates and causes the degradation of human P1- but not P2-HNF4α ( Chellappa et al . , 2012 ) . We also observed a specific loss of P1-HNF4α protein in a mouse model of sporadic , non-colitis colon cancer ( Figure 2E ) , as we observed previously in humans ( Chellappa et al . , 2012 ) . To decipher the function of the HNF4α isoforms in the colon , we utilized HNF4α isoform-specific mice generated by an exon swap strategy ( Figure 3A top left ) ( Briançon and Weiss , 2006 ) . These mice express exclusively either P1-HNF4α ( α1HMZ ) or P2-HNF4α ( α7HMZ ) wherever HNF4α is endogenously expressed . Immunoblot analysis confirmed that the HNF4α protein level in the distal colon of the exon swap mice is equivalent to that of WT littermates , and that P2-HNF4α is the major isoform in the distal colon ( Figure 3A top right and Figure 3—figure supplement 1E ) . In α1HMZ mice , P1-HNF4α was detected in all epithelial cells in both the bottom of the crypt and the surface epithelium; a similar ubiquitous expression was observed for P2-HNF4α in α7HMZ mice ( Figure 3A bottom ) . 10 . 7554/eLife . 10903 . 005Figure 3 . Differential susceptibility of HNF4α isoform-specific mice to colitis-associated colon cancer . ( A ) Top left , Schematic of Hnf4a exon-swap ( i . e . , isoform-specific ) mice . Top right , IB as in Figure 2A of WCE from the distal colon of the exon-swap mice and their WT controls , probed with the common αP1/P2 antibody . See Figure 3—figure supplement 1E for verification of protein loading . Bottom , representative IF of distal colons from untreated α1HMZ and α7HMZ mice stained with either P1- or P2-driven HNF4α specific antibodies ( 40X magnification ) . N=3–4 mice per genotype examined . Scale for P1-HNF4α α1HMZ is 0 . 22 x 0 . 22 microns; all others are 0 . 36 x 0 . 36 microns . ( B ) Tumor growth in WTα1 ( n = 5 ) and α1HMZ ( n = 6 ) mice treated with 10 mg/kg AOM and two cycles of DSS ( 5 days per cycle ) and harvested at ~53 days . Right , number of tumors per mouse colon . Left , tumor load ( sum of the width or length of all macroscopic lesions in a given mouse ) . Each symbol represents results from one mouse . ( C ) Left , average length of crypt in WTα1 and α1HMZ mice , untreated ( Control ) or treated as in ( B ) N = 2–3 mice per condition; 26–56 crypts per mouse were measured . P<0 . 0005 between treated and control within a genotype and across genotypes in the treated condition . Right , Representative H&E stain ( 10X magnification ) of mice treated as in ( B ) Scale bar is 100 microns . ( D ) Tumor number in WTα1 ( n = 15 ) and α1HMZ ( n = 17 ) male mice treated as in ( B ) but with three cycles of DSS ( two cycles of 5 days and one cycle of 4-days ) and harvested at ~85 days . Top , total number of tumors per mouse . Bottom , number of tumors per mouse based on the tumor width . n . s . , non-significant . ( E ) As in ( B ) but for WTα7 ( n = 21 ) and α7HMZ ( n = 23 ) mice treated with 10 mg/kg AOM and 2–3 cycles of DSS ( 4–5 days per cycle ) and harvested at ~53–64 days . P-values between α7HMZ and WTα7 mice are indicated . Tumor data were pooled from three independent experiments . ( F ) Tumor number and load in WTα7 ( n = 20 ) and α7HMZ ( n = 14 ) mice as in ( E ) but harvested at ~85 days . The following figure supplement is available for Figure 3:DOI: http://dx . doi . org/10 . 7554/eLife . 10903 . 00510 . 7554/eLife . 10903 . 006Figure 3—figure supplement 1 . HNF4α isoform-specific mice subjected to AOM+DSS to induce colitis-associated colon cancer . ( A ) Spleen-to-body weight ratio of WTα1 ( n=5 ) and α1HMZ ( n=6 ) mice treated with 10 mg/kg AOM and two cycles of DSS ( 5 days per cycle ) and harvested at ~54 days . ( B ) Tumor load in WTα7 and α7HMZ male mice treated with 10 mg/kg AOM and 2–3 cycles of DSS ( 4–5 days per cycle ) and harvested after different time points as indicated . Each symbol represents results from one mouse . ( C ) IF of Ki67 ( green ) and nuclear ( red ) staining in the distal colon of treated ( 10 mg/kg AOM and 2 cycles of DSS , 5 days ) WTα7 and α7HMZ male mice harvested at 53–64 days . ( D ) Average percent of Ki67-positive cells counted in two to three fields in ( C ) from mice ( n=2-4 ) per condition ( ~2300 to 3000 total cells per genotype scored ) . n . s . , not statistically significant . ( E ) IB from Figure 3A in main text with Coomassie stain to show equal loading . DOI: http://dx . doi . org/10 . 7554/eLife . 10903 . 006 After 53 days of AOM+DSS treatment , the α1HMZ mice had significantly fewer and smaller tumors compared to WT controls ( Figure 3B ) . In addition , despite similar crypt length in untreated WT and α1HMZ mice , the α1HMZ mice did not exhibit the characteristic increase in crypt length associated with mutagen exposure observed in WT mice ( Richards , 1977 ) ( Figure 3C ) . In fact , the crypt length decreased compared to both treated WT and untreated α1HMZ . We also observed fewer infiltrating immune cells ( Figure 3C right ) as well as decreased spleen-to-body weight ratio in the treated α1HMZ mice ( Figure 3—figure supplement 1A ) . After 85 days of treatment , the difference in tumor number was less pronounced: a significant decrease in tumor number was observed in α1HMZ mice only in the smallest tumors ( 0–2 mm ) ( Figure 3D ) . In contrast to α1HMZ mice , the α7HMZ mice exhibited a greater tumor load and tumor number than their WT controls after 53–64 days of treatment ( Figure 3E ) . However , at the later time point ( 85 days ) , the effect was lost mainly due to increased tumor burden in the WT mice ( Figure 3F and Figure 3—figure supplement 1B ) . Interestingly , there was no difference in the percent of Ki67-staining cells between WT and α7HMZ mice ( 53–64 day treatment ) ( Figure 3—figure supplement 1C , D ) . More striking than tumor induction by AOM+DSS in the α1HMZ and α7HMZ mice was their response to an acute DSS treatment to induce colitis --2 . 5% DSS in drinking water for 5 days . There was a ~73% mortality rate for α7HMZ mice that occurred starting after three days of recovery when the mice were switched to normal tap water ( Figure 4A ) . During the recovery phase , α7HMZ mice exhibited a significant decrease in body weight and colon length ( Figure 4B and Figure 4—figure supplement 1A ) , and a worse histological score ( due to more severe crypt damage , inflammation and ulceration ) compared to their WT littermates ( Figure 4C and Figure 4—figure supplement 1B ) . There was also an increased spleen-to-body weight ratio ( Figure 4—figure supplement 1C ) when the mice were maintained and treated in an open access vivarium . IB analysis revealed that , in contrast to the WT mice that lost expression of both HNF4α isoforms after five days of DSS and then had an increase in P1-HNF4α expression at 3-day recovery ( Figure 2A ) , in the α7HMZ mice P2-HNF4α protein amount is notably increased upon DSS treatment and then decreased after a 3-day recovery , as observed by both IB and IF ( Figure 4D ) . At 12 days of recovery , we observed a massive infiltration of immune cells and a continued striking loss in crypt structure in α7HMZ mice compared to WT mice ( Figure 4—figure supplement 1D ) . 10 . 7554/eLife . 10903 . 007Figure 4 . Differential susceptibility of HNF4α isoform-specific mice to DSS-induced colitis . ( A ) Percent mortality of WTα7 ( n = 28 ) and α7HMZ ( n = 16 ) mice treated with 2 . 5% DSS for 5 days . α7HMZ mice typically died during day 3 to 12 of recovery following DSS treatment . Data pooled from two independent experiments . Not shown is a third experiment with older mice ( 21–23 weeks ) with similar results ( WT: 1 of 5 mice died; α7HMZ: 3 of 6 mice died ) . ( B ) Change in bodyweight ( represented as% initial body weight ) ( left ) and colon length ( right ) of WT ( n = 4 ) and α7HMZ ( n = 4 ) mice treated as indicated . Significant comparisons are indicated with a P-value . ( C ) Left , representative H&E stain of WT and α7HMZ mice treated with 2 . 5% DSS for 5 days followed by 0 or 3 days of recovery . Right , histological scores of colitis in WTα7 ( n = 4 ) and α7HMZ ( n = 4 ) mice . ( D ) Left , IB for HNF4α ( P1/P2 antibody ) of WCE from the distal colon of α7HMZ mice treated as indicated . Right , representative IF of distal colon from α7HMZ mice treated with 2 . 5% DSS for 5 days -/+ recovery as indicated and stained with P1/P2-HNF4α antibody ( green ) and TO-PRO3 ( red ) for nuclei ( 40X magnification ) . Extracts from four mice per genotype ( out of n = 5–7 ) were randomly chosen for IB analysis on a single gel/blot; sections from 3 mice per genotype were examined . ( E ) Colon length of WT ( n = 8 ) and α1HMZ ( n = 10 ) male mice treated with 2 . 5% DSS for 5 days followed by 3 days of recovery . Results from two independent experiments were pooled . ( F ) Representative H&E stain ( left ) and histological scores ( right ) of colitis in WTα1 ( n = 8 ) and α1HMZ ( n = 10 ) mice treated as in ( E ) . The following figure supplements are available for Figure 4:DOI: http://dx . doi . org/10 . 7554/eLife . 10903 . 00710 . 7554/eLife . 10903 . 008Figure 4—figure supplement 1 . Increased susceptibility of α7HMZ mice to DSS-induced colitis . Results from an independent DSS experiment from that shown in Figure 4 A-C . Mice from this experiment were analyzed by IB and IF in Figure 4D . ( A ) Colon length of WTα7 and α7HMZ mice treated with 2 . 5% DSS for 5 days followed by 0 or 3 days of recovery . P<0 . 001 , between WTα7 and α7HMZ at 3 days of recovery . N=4–7 per genotype per treatment . ( B ) Representative H&E stain of distal colon of WTα7 and 7HMZ mice treated in ( A ) . ( C ) Spleen-to-body weight ratio of WT7 ( n=7 ) andα7HMZ ( n=5 ) mice treated with 2 . 5% DSS for 5 days and sacrificed the following day . ( D ) Representative H&E stain of distal colon from WTα7 and α7HMZ mice treated with 2 . 5% DSS for 5 days followed by 12 days of recovery . N=5 per genotype treated , 2–3 mice sectioned . DOI: http://dx . doi . org/10 . 7554/eLife . 10903 . 00810 . 7554/eLife . 10903 . 009Figure 4—figure supplement 2 . Increased inflammation in α7HMZ mice in DSS-induced colitis . ( A ) Spleen-to-body weight ratio of α7HMZ ( n=7 ) and α1HMZ ( n=7 ) mice treated for 4 days with 2 . 5% DSS followed by 18 days of recovery . ( B ) Crypt length of α7HMZ ( n=3 ) and α1HMZ ( n=3 ) mice treated as in ( A ) : 26-37 crypts analyzed per genotype . ( C ) Representative photographs of colons from α7HMZ and α1HMZ mice treated with 2 . 5% DSS for 4 days and allowed to recover for 18 days . ( D ) Representative H&E stain ( 10X magnification ) of α7HMZ and α1HMZ male mice treated as in ( A ) along with histological score from three mice per genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 10903 . 009 In contrast to the extreme sensitivity of the α7HMZ mice to DSS-induced colitis , α1HMZ mice were less susceptible than their WT controls as indicated by increased colon length ( Figure 4E ) and well-preserved crypt structure and decreased histological score ( Figure 4F ) . There was no difference in spleen-to-body weight ratio between the α1HMZ mice and WT controls ( data not shown ) . Clinical and histological changes occurring a few weeks after DSS treatment are referred to as chronic or advanced changes ( Perše and Cerar , 2012 ) . To examine chronic effects , we allowed the mice to recover for 18 days after a somewhat milder DSS treatment ( 4 days ) to reduce the mortality of α7HMZ mice . Despite the shorter DSS treatment , after 18 days , the α7HMZ mice still exhibited elevated spleen-to-body weight ratio , increased crypt length , more visibly inflamed colons and immune cell infiltration and overall higher histological scores compared to α1HMZ mice ( Figure 4—figure supplement 2A–D ) . Expression profiling of the distal colon revealed a significant change in a substantial number of genes in the untreated isoform-specific mice compared to their WT controls ( Figure 5—figure supplement 1A ) . There was an overall greater effect in α1HMZ than α7HMZ mice in terms of the number of dysregulated genes with a large fold change , which could be due to the fact that P1-HNF4α typically has a more potent transactivation function than P2-HNF4α ( Eeckhoute et al . , 2003 ) . On the other hand the number of genes altered at lower fold change was higher in α7HMZ compared to α1HMZ mice , consistent with more P2-HNF4α protein in the distal colon of WT mice than P1-HNF4α ( Figure 3A ) . Gene Ontology ( GO ) analysis of the differentially regulated genes showed that in α1HMZ mice there is a marked upregulation of genes involved in wound healing and immune response , as well as a variety of metabolic processes typically associated with differentiation ( Figure 5A ) . In contrast , in α7HMZ mice there is a significant upregulation of genes involved in cell cycle and DNA repair and a decrease in genes involved in cell adhesion , motility and ion transport ( Figure 5B ) . ( See Figure 5—source data 1A-1G for the fold change in the top 100 dysregulated genes and the genes in the aforementioned GO categories , respectively ) . 10 . 7554/eLife . 10903 . 010Figure 5 . Altered gene expression , interacting proteins , migration and ion transport in HNF4α isoform-specific mice . ( A , B ) Comparative Gene Ontology ( GO ) of genes differentially regulated ( ≥two-fold ) in the distal colon of untreated α1HMZ ( A ) and α7HMZ ( B ) mice . ( C ) Top , Venn diagram of total number of HNF4α-interacting proteins from RIME analysis found in α7HMZ only , α1HMZ only or both α7HMZ and α1HMZ colons , as described in Material and methods . Indicated are nuclear proteins that have been implicated in regulating gene expression and associated with human or mouse colon cancer , IBD , Crohn’s disease and/or ulcerative colitis , as well as other pro-proliferative proteins found only in α7HMZ colons . Shown also are transcription factors that interact with HNF4α in both genotypes . Bold , proteins mentioned in text . Bottom , Total number of proteins in the indicated categories that show a significant interaction with HNF4α in the exon swap mice . TF , transcription factor; RNA binding proteins; kinase and phosphatase categories include only protein kinases and phosphatases , as well as relevant scaffolding proteins b . ( D ) Untreated HNF4α isoform-specific mice and their WT littermates ( n = 3–4 per genotype ) were injected with BrdU ( 75 mg/kg ) and sacrificed at 2 hr or 48–50 hr . The distance migrated by the BrdU+ cells from the bottom of the crypt between 2 hr and 48–50 hr is plotted as% crypt length; 5–38 crypts per mouse were scored . ( E ) Intestinal chloride secretion in response to 10 µM forskolin and 100 µM carbachol represented as change in short-circuit current ( ΔIsc ) . Left , WTα1 ( n = 6 ) and α1HMZ ( n = 5–8 ) mice . P<0 . 02 between α1HMZ and WTα1 . Right , WTα7 ( n = 4 ) and α7HMZ ( n = 3 ) mice . P<0 . 05 versus WTα7 . Results from one experiment per genotype are shown: a second independent experiment for α7HMZ yielded similar results ( not shown ) . ( F ) Left , RELMβ mRNA expression in the distal colon of untreated α1HMZ , α7HMZ and their WT controls from microarrays in ( A , B ) , represented as an average of the three Retnlb probes . P<0 . 008 versus WTα7 . Right , RELMβ protein level quantified by ELISA in the mid colon homogenate of untreated WTα7 ( n = 5 ) and α7HMZ ( n = 3 ) mice . Shown are means of technical triplicates . The following supplementary figure and source data are available for Figure 5:DOI: http://dx . doi . org/10 . 7554/eLife . 10903 . 01010 . 7554/eLife . 10903 . 011Figure 5—source data 1 . Transcriptomic analysis of HNF4α isoform-specific mice . ( A ) Top 100 genes DOWN in distal colon of α1HMZ male mice compared to WT controls . ( B ) Top 100 genes UP in distal colon of α1HMZ male mice compared to WT controls . ( C ) Top 100 genes DOWN in distal colon of α7HMZ male mice compared to WT controls . ( D ) Top 100 genes UP in distal colon of α7HMZ male mice compared to WT controls . ( E ) Up-regulated genes involved in wound healing and immune function enriched in α1HMZ mice . ( F ) Up-regulated genes involved in cell cycle and DNA repair in α7HMZ mice . ( G ) Down-regulated genes involved in cell adhesion and ion transport in α7HMZ mice . DOI: http://dx . doi . org/10 . 7554/eLife . 10903 . 01110 . 7554/eLife . 10903 . 012Figure 5—source data 2 . Proteomic analysis of HNF4α isoform-specific mice . ( A ) List of proteins that interact with HNF4α in α1HMZ and α7HMZ colons from RIME analysis meeting the criteria described in Figure 5C . ( B ) Select proteins that interact with HNF4α in α1HMZ and α7HMZ colons from RIME analysis used to prepare the graph in Figure 5C . ( C ) All peptides that interact with HNF4α in both α7HMZ and α1HMZ colons from RIME analysis in which there are at least 2 positives for each genotype . ( D ) All peptides that interact with HNF4α preferentially in α7HMZ colons in which there are 2 or more positives for α7HMZ compared to α1HMZ . ( E ) All peptides that interact with HNF4α preferentially in α1HMZ colons in which there are 2 or more positives for α1HMZ compared to α7HMZ . Figure 5C . DOI: http://dx . doi . org/10 . 7554/eLife . 10903 . 01210 . 7554/eLife . 10903 . 013Figure 5—figure supplement 1 . Transcriptomic and BrdU analysis of HNF4α isoform-specific mice . ( A ) Total number of genes up- or down-regulated ( non log fold change ) in α1HMZ and α7HMZ untreated young adult male mice compared to their respective WT controls as determined by Affymetrix Exon arrays . ( B ) Position of BrdU+ cells in the distal colon of WTα7 and α7HMZ mice at 2 hr and 48–50 hr after injection . Left , distance in microns . Right , distance as percent of crypt length from the crypt bottom . P<0 . 02 for α7HMZ ( n=4 mice ) versus WTα7 ( n=3 mice ) at 48 hr . Data pooled from two independent experiments: 5–37 crypts counted per mouse . ( C ) Distance migrated in microns by Brdu+ cells between 2 hr and 48 hr from mice treated in ( B ) . ( D ) As in ( B ) but for WTα1 ( n=4 ) and α1HMZ ( n=4 ) mice . P<0 . 005 between genotypes at 48–50 hr . Data are from one experiment . ( E ) As in ( C ) but for WTα1 and α1HMZ mice . ( F ) Total number of BrdU+ cells per crypt ( 5–38 crypts counted per mouse ) in the distal colon of WTα7 and α7HMZ mice at 3 hr ( n=7 per genotype ) and 25 hr ( n=3–4 per genotype ) after BrdU injection . DOI: http://dx . doi . org/10 . 7554/eLife . 10903 . 013 The DNA binding domains of P1- and P2-HNF4α are 100% identical and the isoforms have similar in vitro DNA binding specificity and chromatin immunoprecipitation ( ChIP ) -seq profiles in human colon cancer cells ( Vuong et al . , 2015 ) . Therefore , to elucidate the mechanism responsible for differential gene expression in mouse colon , we performed RIME ( Rapid Immunoprecipitation Mass spectrometry of Endogenous proteins ) on HNF4α in the colons of α1HMZ and α7HMZ mice ( Figure 5C ) . The isoforms share 76 interacting proteins , including previously reported HNF4γ ( Daigo et al . , 2011 ) , a well known co-regulator for nuclear receptors ( NRIP1 , RIP140 ) and DPF2 , a BRG1-associated factor ( BAF45 ) . However , there were more proteins uniquely binding HNF4α in α7HMZ and α1HMZ colons -- 138 and 99 , respectively ( Figure 5C top and Figure 5C—source data 2A–E ) . Src tyrosine kinase , for example , bound uniquely in α1HMZ colons , consistent with our previous report that Src preferentially phosphorylates and interacts with HNF4α1 in cell-based and in vitro systems ( Chellappa et al . , 2012 ) and validating RIME for identification of differential interacting proteins in vivo . In contrast , CUL4A , a core component of a cullin-based E3 ubiquitin ligase complex and overexpressed in cancer ( Kopanja et al . , 2009 ) , and PCM1 , a centrosome binding protein translocated to the JAK2 locus in certain leukemias ( Reiter et al . , 2005 ) , both bound uniquely in α7HMZ colons . Both CUL4A and PCM1 are required for efficient cell proliferation , genome stability and/or proper centrosome function ( Erger and Casale , 1998; Farina et al . , 2016 ) , consistent with the upregulation of genes involved in cell cycle and DNA repair in α7HMZ colons ( Figure 5B ) , and accelerated tumorigenesis in α7HMZ mice ( Figure 3E ) . Cross-referencing the interacting proteins to those in the literature associated with colon cancer and inflammatory bowel disease ( IBD ) revealed several additional relevant proteins for each genotype , the vast majority of which ( 62 . 9% ) are known transcription regulators , protein kinases or phosphatases and associated proteins ( Figure 5C ) . For example , NDRG2 , a kinase downstream of the mTOR/SGK pathway and a tumor suppressor that mediates apoptosis ( Deuschle et al . , 2012 ) , and EMD , a nuclear membrane protein phosphorylated by Src ( Tifft et al . , 2009 ) , both preferentially bind HNF4α1 and have been negatively associated with colon cancer . Likewise , HNF4α1 was preferentially bound by catalytic subunits of AMPK ( PRKAA1/2 ) and is known to be phosphorylated by AMPK , which decreases its protein stability ( Hong et al . , 2003 ) . However , AMPK suppresses cell proliferation via inhibition of mTOR and activation of p53 pathways ( Motoshima et al . , 2006 ) and low levels of AMPK activity are correlated with poor survival in metastatic colon cancer patients ( Zulato et al . , 2014 ) , indicating that additional studies are required to elucidate the impact of AMPK signaling on HNF4α in colitis and colon cancer . In contrast , protein kinase C beta 2 ( PRKCB ) preferentially interacts with HNF4α7 and is known to be both necessary and sufficient to confer susceptibility to AOM-induced colon carcinogenesis in the colonic epithelium ( Liu , 2004 ) . All told , there were hundreds of proteins that preferentially interacted with the HNF4α isoforms , including many signaling molecules as well as RNA binding proteins and transcription factors , providing multiple potential mechanisms for differential gene expression . While the isoform-specific mice did not exhibit any overt morphological abnormalities in their intestines under normal conditions , the gene expression analysis ( and AOM/DSS and DSS colitis results ) suggested potential functional differences . Since there was a decrease in cell motility genes in α7HMZ mice , we examined migration of BrdU-labeled cells up the crypt and found that 48 hr after injection the average position of the BrdU-labeled cells ( both in absolute terms and relative to the bottom of the crypt ) was lower in α7HMZ mice compared to WT: this resulted in a statistically significant decreased migration of the BrdU+ cells during the 45 hr period ( Figure 5D and Figure 5—figure supplement 1B , C ) . Conversely , the position of the BrdU+ cells , and hence migration , was significantly higher in α1HMZ mice ( Figure 5D and Figure 5—figure supplement 1D , E ) . Despite these differences , there was a similar number of total BrdU+ cells in WT and α7HMZ mice ( Figure 5—figure supplement 1F ) . Since the ion transport genes were also decreased in α7HMZ , we examined electrogenic chloride secretion in isolated colonic mucosa . The distal colon of WT and α7HMZ mice exhibited a similar transmucosal electrical resistance and basal Isc ( data not shown ) . However , the α7HMZ distal colon was refractory to stimulation with calcium-dependent ( carbachol ) and cAMP-dependent chloride secretagogues ( forskolin ) , while the α1HMZ distal colon showed a markedly increased Isc response to forskolin ( Figure 5E ) . Since impaired chloride secretion is observed in colitis ( Hirota and McKay , 2009 ) , this differential response to forskolin as well as cell migration could explain , at least in part , the differential sensitivity of the α1HMZ and α7HMZ mice to DSS . During experimental colitis the cytokine RELMβ is known to activate the innate immune system in response to loss of epithelial barrier function and increased exposure to gut microbiota: hence , mice lacking the Retnlb gene are known to be resistant to experimental colitis ( Hogan et al . , 2006; McVay et al . , 2006 ) . Interestingly , one of the genes most significantly upregulated in the untreated α7HMZ colon was Retnlb ( 5 . 3-fold increase versus WT controls ) ; RELMβ protein levels were also increased ( Figure 5F ) . In contrast , there was no significant difference in RELMβ gene expression between α1HMZ mice and their WT littermates ( Figure 5F left ) . To determine whether RELMβ plays a causal role in the susceptibility of α7HMZ mice to colitis , we crossed α7HMZ mice with a RELMβ knock out ( Retnlb-/- ) to generate RbKO/α7HMZ mice ( Figure 6—figure supplement 1A ) . We confirmed that RELMβ expression is lost in these mice , that HNF4α protein levels are unchanged by the RELMβ knock out and that the α7HMZ allele has the same effect in the 'Rb line' ( designated C57BL/6N+J , see Materials and methods for details ) in terms of body weight loss and increased RELMβ expression after DSS treatment ( Figure 6A and Figure 6—figure supplement 1B–D ) . 10 . 7554/eLife . 10903 . 014Figure 6 . RELMβ knockout decreases susceptibility of α7HMZ mice to colitis . ( A ) RELMβ protein level quantified by ELISA in the midcolon homogenate of mice with the indicated genotype treated with 2 . 5% DSS for 6 days . Genotypes are indicated as Retnlb/Hnf4a . N = 3–5 mice per genotype as indicated by each dot . ( B ) Histological scores of colitis in WT/WT ( n = 6 ) , WT/α7HMZ ( n = 9 ) and Rb KO/ α7HMZ ( n = 11 ) male mice treated with 2 . 5% DSS for 5 days followed by 3 days of recovery . Multiple sections per mouse were scored . ( C ) Percent change in body weight during and following DSS treatment ( 2 . 5% for 6 days ) . Day 0 is the start of treatment . Left , graph over time from one experiment . N = 3–5 mice per genotype . # Indicates P<0 . 05 on day 10 and 11 between WT/WT and α7HMZ/WT; indicates P<0 . 01 on day 10 and 11 between α7HMZ/WT and α7HMZ/Rb KO . Right , meta-analysis of percent weight loss at 3 days of recovery after 6 days of treatment with 2 . 5% DSS from nine independent experiments . N = 12–47 mice per genotype . The WT/α7HMZ data include both the α7HMZ C57BL/6N parent as well as the α7HMZ C57BL/6N+J generated from the Retnlb-/- cross . See Figure 6—figure supplement 1 for a comparison of the two α7HMZ lines . ( Data from one experiment in which all mice , including the WT/WT controls , had lower than normal body weight loss were excluded from the analysis . ) ( D ) Colon length from mice treated with 2 . 5% DSS for 6 days followed by different recovery periods . N = 3–14 mice per genotype per treatment . P<0 . 05 versus WT/WT at different time points . #P<0 . 01 or ##P<0 . 002 versus RbKO/α7HMZ at different time points . Data are pooled from 12 independent experiments . ( E ) Kaplan-Meier survival curve of WT/α7HMZ ( n = 4 ) and KO/α7HMZ ( n = 9 ) mice after 6 days 2 . 5% DSS in one experiment . Meta-analysis of several independent experiments also showed that out of a total of 24 KO/α7HMZ mice allowed to go past 3 days of recovery , only one mouse died ( 3 . 6% mortality ) . In contrast , 13 out of 29 WT/α7HMZ mice ( 44 . 8% ) either died or had to be sacrificed due to severe distress . Data for WT/α7HMZ mice in both the C57BL/6N and C57BL/6N+J lines were combined: no difference in mortality was noted between the lines . ( F ) WT , α7HMZ and α1HMZ mice ( all in C57BL/6N background , n = 3–63-6 per genotype per treatment ) were treated with 2 . 5% DSS for 6 days alone or followed by 3 days of recovery . Left , RELMβ protein quantified by ELISA in the midcolon homogenate: shown are means of technical triplicates from one experiment . Right , colon length . RELMβ ELISA: P<0 . 03 versus WT; #P<0 . 01 versus α7HMZ . Colon length: P<0 . 01 versus WT; #P<0 . 0002 versus α7HMZ . The following figure supplement is available for Figure 6::DOI: http://dx . doi . org/10 . 7554/eLife . 10903 . 01410 . 7554/eLife . 10903 . 015Figure 6—figure supplement 1 . Verification of RELMβKO/α7HMZ mice . ( A ) Verification of genotype of RELMβ KO mice crossed into α7HMZ mice . Shown is a representative DNA agarose gel of PCR products generated with primers for the Retnlb and Hnf4a loci for 2 mice per genotype out of hundreds analyzed . Genotypes are written as Retnlb/Hnf4a loci . ( B ) IB of distal colon WCE using the HNF4α P1/P2 antibody showing roughly equivalent amounts of HNF4α protein in WT/WT and RbKO/α7HMZ mice . Extracts from one mouse are loaded per lane . ( C ) Percent change in body weight of α7HMZ mice from the two different lines -- the parental α7HMZ line in C57BL/6N and the progeny from the cross with the RELMβ KO ( C57BL/6N+J ) -- after 6 days of 2 . 5% DSS followed by 3 days of recovery from 4 ( C57BL/6N ) to 6 ( C57BL/6N+J ) independent experiments . N=15-26 mice per genetic background . One experiment was excluded in which the WT/WT mice serving as controls lost less than the usual amount of body weight , suggesting insufficient DSS treatment . ( D ) Representative IF from one of two mice stained per genotype for RELMβ in the distal colon of the indicated genotypes treated for 6 days with 2 . 5% DSS followed by 3 days recovery . Red , RELMβ . Blue , DAPI . ( E ) Percent change in body weight during and following DSS treatment ( 2 . 5% for 6 days ) . Day 0 is the start of treatment . Both α1HMZ ( n=13 up to Day 6; n=8 up to Day 9 ) and α7HMZ ( n=9 up to Day 6; n=4 up to Day 9 ) are in the C57BL/6N background . Data are pooled from two independent experiments except that α7HMZ Days 7–9 is from one experiment . P<0 . 05 , P<0 . 005 , P<0 . 0005 . DOI: http://dx . doi . org/10 . 7554/eLife . 10903 . 015 Interestingly , while the histological score of the RbKO/α7HMZ mice was somewhat improved compared to WT/α7HMZ at three days of recovery , the difference was not significant ( P=0 . 13 ) ( Figure 6B ) . In contrast , the weight loss of the RbKO/α7HMZ mice was completely restored to WT/WT levels ( Figure 6C ) , as was the colon length ( Figure 6D ) and overall survival ( Figure 6E ) . Notably , the RELMβ protein levels in the α1HMZ mice were significantly reduced at three days of recovery ( although elevated right at the end of the DSS treatment ) and inversely correlated with colon length ( Figure 6F ) . These results suggest that elevated RELMβ expression in untreated and DSS-treated α7HMZ mice and decreased expression during recovery in α1HMZ mice might contribute to their increased and decreased susceptibility to DSS-induced colitis , respectively . Interestingly , body weight loss is attenuated in α1HMZ during DSS treatment , however this protective effect is lost during recovery ( Figure 6—figure supplement 1E ) . All together our results suggest that both P1- and P2-HNF4α are critical for full recovery following DSS treatment . To address the mechanism responsible for increased RELMβ expression in α7HMZ mice we performed ChIP in CaCo2 cells , which express predominantly P2-HNF4α ( Chellappa et al . , 2012 ) . We used the Support Vector Machine ( SVM ) learning tool in the HNF4 Binding Site Scanner to predict three potential HNF4α binding sites within 1 . 5 kb of the transcription start site ( +1 ) of human RETNLB ( Figure 7A , left and Figure 7—figure supplement 1A-B ) , two of which are in the vicinity of NFκB and CDX2 binding sites ( Wang , 2005; He et al . , 2003 ) . We found that endogenous HNF4α binds the two regions that encompass the SVM sites ( Region 1 and 2 ) ( Figure 7A , right ) . The mouse Retnlb promoter also contains predicted HNF4α binding sites close to +1 , one of which is highly conserved in human ( Figure 7—figure supplement 1B-C , indicating that RELMβ expression may be directly regulated by P2-HNF4α in both mouse and human . Luciferase assays in LS174T goblet-like cells with RELMβ reporter constructs containing HNF4α binding sites in ChIP region 2 ( Figure 7—figure supplement 1D ) confirmed that P2-HNF4α activates the RELMβ promoter significantly more than P1-HNF4α ( Figure 7B ) . siRNA knockdown of endogenous P1- and P2-HNF4α in LS174T cells also showed a greater effect of loss of endogenous P2-HNF4α on RELMβ expression than P1-HNF4α ( Figure 7—figure supplement 1E ) . In contrast , on a known HNF4α-responsive promoter , ApoB , P1-HNF4α activates better than P2-HNF4α ( Figure 7—figure supplement 1F ) . 10 . 7554/eLife . 10903 . 016Figure 7 . Direct and indirect mechanisms of regulation of RELMβ expression by HNF4α isoforms: impact on DSS sensitivity and recovery . ( A ) P2-HNF4α binds the promoter of the RETNLB gene in colonic epithelial cells . Left , schematic of the human RETNLB promoter showing predicted SVM binding sites for HNF4α , as well as sites for NFκB , KLF4 , STAT6 and CDX2 ( He et al . , 2003 ) . Right , ChIP results for endogenous HNF4α in Caco2 cells at RELMβ Region 1 and Region 2 , as well as an HMOX1 control . In , input ( 1/10 dilution ) ; Ig , IgG; H4 , anti-HNF4α . ( B ) Left , uciferase activity of pGL2 . basic and RELMβ reporter constructs in LS174T cells cotransfected with vector ( pcDNA3 . 1 ) , human HNF4α2 or HNF4α8 . Shown is the RLU normalized to protein concentration . Data are represented as mean of triplicates + SD of one independent experiment . P<0 . 05 , P<0 . 005 between vector control and HNF4α2 or HNF4α8 . $P<0 . 05 , $$P<0 . 005 HNF4α2 versus HNF4α8 . Right , IB of extracts from LS174T cell line cotransfected with -870 . hRELMβ reporter and HNF4α isoforms . ( C ) Gut permeability measured by appearance of FITC dextran ( 4 kDa ) in serum of WT , α7HMZ and α1HMZ mice either untreated , treated with 2 . 5% DSS for 6 days alone or followed by 3 days of recovery . ( n = 7–10 for all groups except α7HMZ 6d DSS + 3d recovery where n = 4 ) . P-values were determined by Student’s T-test . ( D ) List of genes related to barrier function altered in the distal colon of α7HMZ and α1HMZ mice compared to their WT controls . Shown is nonlog fold change from the microarray experiment in Figure 5 . ( E ) Summary of various phenotypes of HNFα isoform-specific mice ( α7HMZ and α1HMZ ) relative to WT mice in untreated , DSS and AOM+DSS treated animals as indicated . n . d . , not done . = , no change . ( F ) Distribution of HNF4α isoforms in colonic crypts and their effect on crypt structure in DSS-induced colitis . See text for details . The following figure supplement is available for Figure 7:DOI: http://dx . doi . org/10 . 7554/eLife . 10903 . 01610 . 7554/eLife . 10903 . 017Figure 7—figure supplement 1 . Predicted HNF4α binding sites in the human and mouse RELMβ gene and RELMβ reporter assays . ( A–C ) . Screenshots of UCSC Genome Browser shows putative HNF4α binding sites in human ( AB ) and mouse ( C ) RELMβ genes as well as the fragments amplified by the primers for Region 1 ( A ) and Region 2 ( B ) . The SVM score from HNF4 Binding Site Scanner is given for the predicted sites in the RELMβ gene as well as the canonical HNF4α motif , AGGTCAaAGGTCA ( A ) . Sites with scores above 1 . 0 are potential binders; sites with scores >1 . 5 are predicted to be excellent binders . ( D ) Schematic of human RELMβ luciferase constructs used in ( E ) and Figure 7B with transcription factor binding sites and Region 2 bound by HNF4α in ChIP indicated . ( E ) Left , luciferase activity of pGL2 . basic and RELMβ reporter constructs in LS174T cells transfected with siControl , siP1-HNF4α or siP2-HNF4α . Shown is the RLU normalized to β-gal activity . Data are mean of triplicates + SD of one independent experiment . P<0 . 005 between siControl and siP1-HNF4α or siP2-HNF4α . $$P<0 . 005 siP1-HNF4α versus siP2-HNF4α . Right , IB of WCE from LS174T cells transfected with the indicated siRNAs performed in parallel to the luciferase experiment . ( F ) Left , luciferase activity of ApoB ( -85–47 . E4 ) promoter construct co-transfected with HNF4α2 or HNF4α8 expression vector ( 500 ng ) in COS-7 cells . Bar graphs represent mean ± SD of triplicate samples from one independent experiment . *HNF4α2/8 vector compared to empty vector , $HNF4α2 vector compared to HNF4α8 vector , P<0 . 05 . Right , IB analysis of HNF4α protein level in COS-7 cells . COS-7 cells do not express endogenous HNF4α protein . DOI: http://dx . doi . org/10 . 7554/eLife . 10903 . 017 Both P1 and P2-HNF4α are decreased after six days of DSS treatment in WT mice when RELMβ expression is increased ( Figures 2A and 6F ) , suggesting that additional mechanisms are at play in the upregulation of the Retnlb gene . Therefore , since RELMβ expression is known to be activated by decreased epithelial barrier function ( McVay et al . , 2006 ) , we conducted in vivo epithelial permeability assays using Fluorescein isothiocyanate–dextran 4 kDa ( FITC-dextran ) and found that α7HMZ mice have moderately decreased barrier function as shown by increased FITC-dextran in the serum of both untreated and DSS-treated mice ( Figure 7C ) . Furthermore , we found decreased FITC-dextran in α1HMZ mice compared to α7HMZ at 3 days of recovery ( Figure 7C ) , suggesting improved barrier function , consistent with the lower levels of RELMβ and longer colon length in α1HMZ mice ( Figure 6F ) . Barrier function and colon length are both indicators of colon health . Analysis of the expression profiling data in the untreated mice revealed dysregulation of several genes related to barrier function ( Figure 7D and Figure 5—source data 1G ) . For example , Il4i1 and Il13ra2 , known signaling pathways critical for RELMβ expression ( Artis et al . , 2004 ) , are both increased in α7HMZ mice . There was also a concerted decrease in the expression of genes involved in cell adhesion and paracellular permeability in α7HMZ ( Cdh1 , Cldn15 , Cldn16 and Sh3bp1 ) , which would contribute to decreased barrier function and hence increased DSS sensitivity and RELMβ expression in α7HMZ . In a mouse model of CAC , α1HMZ mice exhibited decreased tumor load , suggesting that expression of HNF4α1 from the P2 promoter in the proliferative compartment may protect against tumorigenesis , consistent with studies showing a loss of P1-HNF4α in human colon cancer ( Chellappa et al . , 2012; Tanaka et al . , 2006; Oshima et al . , 2007 ) , a tumor suppressive role for P1-HNF4α in mouse liver ( Hatziapostolou et al . , 2011; Walesky et al . , 2013b ) and our recent colon cancer xenograft studies showing that ectopic expression of P1-HNF4α reduces tumor growth ( Vuong et al . , 2015 ) . In contrast , α7HMZ mice showed an initial increase in tumorigenesis , which could reflect the absence of P1-HNF4α . Since there was no increase in the number of Ki67- or BrdU-positive cells in α7HMZ colons , no visible tumors after a chronic colitis regimen ( three cycles of DSS treatment ) ( data not shown ) nor acceleration of tumor growth in the presence of ectopic expression of P2-HNF4α in the xenograft model ( Vuong et al . , 2015 ) , there is no indication that P2-HNF4α actively promotes proliferation . Rather , P2-HNF4α appears to be merely permissive of cell proliferation , consistent with its expression in the proliferative compartment and its retention in human colon cancer ( Chellappa et al . , 2012; Tanaka et al . , 2006 ) . . Interestingly , HNF4α has been shown to act as an oncogene in gastric cancer and only P2-HNF4α is expressed in the stomach ( Chang et al . , 2014; Dean et al . , 2010 ) . An additional and/or alternative explanation for the differences in CAC-induced tumors in the isoform-specific mice could be their remarkable differences in DSS sensitivity , which in turn could be due to opposing chloride secretory responses and epithelial migration ( Figure 7E ) . Since the chloride secretory pathway is required for maintaining proper luminal hydration , which helps protect the epithelium from physical damage ( Barrett and Keely , 2000 ) , this suggests that the isoform-specific mice have different barrier functions , which we confirmed with FITC-dextran assays . Decreased expression of cell adhesion genes in α7HMZ colons -- such as E-cadherin ( Cdh1 ) , an established HNF4α target ( Battle et al . , 2006; Elbediwy et al . , 2012 ) critical for both migration of cells along crypt-villi axis and epithelial barrier function ( Grill et al . , 2015; Schneider et al . , 2010 ) -- would contribute to decreased barrier function . In contrast , downstream effectors of IL-18 signaling ( Il18r1 and Il18rap ) , which are implicated in intestinal epithelial barrier function ( Nowarski et al . , 2015 ) , were increased in α7HMZ mice , while Il18bp , a decoy receptor for IL-18 which attenuates signaling , is decreased . All told , there are several key cell adhesion and cytokine signaling genes that are dysregulated in α7HMZ colons that could contribute to decreased barrier function and subsequently a pro-inflammatory state ( Hogan et al . , 2006; McVay et al . , 2006 ) , which could in turn contribute to the enhanced colitis and tumorigenesis observed in α7HMZ mice . One such cytokine is RELMβ , which we show is a direct target of HNF4α and preferentially activated by P2-HNF4α . DSS also causes epithelial injury and a need for rapid proliferation , expansion , migration and differentiation of intestinal epithelial cells to promote wound healing and regeneration ( Sturm , 2008 ) . Hence , the inability of α7HMZ mice to effectively recover from DSS could be attributed to defective migration , chloride secretion ( ion transport ) and/or differentiation ( Figure 7F ) . BrdU-labeled cells exhibited greater migration in α1HMZ and lower migration in α7HMZ mice: α7HMZ mice also had reduced expression of genes involved in cell motility . Cldn15 is downregulated in α7HMZ and upregulated in α1HMZ colons and a known target of HNF4α ( Darsigny et al . , 2009 ) . Cldn15 dysregulation could explain the decrease in secretory capacity in α7HMZ mice and hence their inability to recover after DSS as a basal level of secretion is important for proper gut formation ( Anderson and Van Itallie , 2009; Tamura et al . , 2008 ) . Cdx2 , another established target of HNF4α ( Saandi et al . , 2013 ) and a major player in intestinal differentiation ( Suh and Traber , 1996; Lorentz et al . , 1997 ) , has a similar expression profile . Finally , the involvement of these processes in recovery from DSS could explain why the crypt structure in the α7HMZ mice is not completely ameliorated by the RELMβ knockout even though body weight and colon length loss and lethality are: it has been noted previously that RELMβ expression per se does not alter colonic epithelial proliferation McVay et al . , 2006 and its role in affecting the barrier function is still debated ( Hogan et al . , 2006; McVay et al . , 2006 ) . In summary , the results presented here indicate that while P1- and P2-HNF4α isoforms can substitute for each other during normal development and homeostasis , under conditions of stress they play notably different roles . Those roles seem to be driven by unique interacting partners leading to differential expression of target genes . The results also show that any factor that disrupts the balance between the HNF4α isoforms in the colon could have serious functional consequences . Those factors include Src tyrosine kinase ( Chellappa et al . , 2012 ) , as well as any one of a number of other signaling molecules that interact preferentially with the isoforms . Future studies will be required to elucidate all the underlying mechanisms but it is anticipated that several will be important in diagnosing and treating gastrointestinal diseases involving HNF4α . Care and treatment of animals were in strict accordance with guidelines from the University of California Riverside Institutional Animal Care and Use Committee ( Protocol number A200140014 ) . Mice were maintained in isolator cages under 12 hr light/dark cycles at ~21°C on bedding ( Andersons bed OCOB Lab 1/8 1 . 25CF ) from Newco ( Rancho Cucamonga , CA ) and either fed a standard lab chow ( LabDiet , #5001 , St . Louis , MO ) and maintained in an open access vivarium or fed an irradiated chow ( LabDiet , #5053 ) and housed in a specific pathogen-free ( SPF ) vivarium ( α7HMZ and Retnlb-/- matings ) . All experiments were performed in an open access vivarium except those in Figure 4CEF where the mice were born in an open-access vivarium and then moved to an SPF facility before treatment ( due to a required institutional change ) . Subsequently , mice born in the SPF facility were brought to an open access facility at least two weeks prior to DSS treatment . Transgenic mice on a mixed 129/Sv plus C57BL/6 background carrying exon 1A or exon 1D in both the P1 and P2 promoter ( α1HMZ or α7HMZ , respectively ) have been described previously ( Briançon and Weiss , 2006 ) . Both lines were maintained as heterozygotes ( HTZ ) ; wildtype ( WT ) and homozygous ( HMZ ) were mated for a single generation to generate mice for experiments . Appropriate , age-matched WT controls for both the α1HMZ and α7HMZ lines were used ( designated WTα1 and WTα7 , respectively , Figures 1–5 ) . The α7HMZ and α1HMZ mice were backcrossed to C57BL/6N for 10+ generations and used with C57BL/6N WT controls ( Figure 6 ) . The backcrossed α7HMZ mice were crossed with RELMβ knockout ( Retnlb-/- ) mice which were generated as previously described ( Hogan et al . , 2006 ) using VelociGene technology . The Retnlb-/- mice were backcrossed 6+ generations in C57BL/6J to generate RbKO/α7HMZ mice in a C57BL/6N+J background . WT/α7HMZ mice from the RELMβ cross showed essentially identical susceptibility to DSS as the α7HMZ parent in the C57BL/6N background ( as well as the original exon swap mice in the mixed background ) ( Figure 6—figure supplement 1C and data not shown ) . All experiments with RbKO/α7HMZ mice were compared to RbWT/α7HMZ from the RELMβ cross except for the meta analysis in Figure 6C which included data from the parental α7HMZ line in C57BL/6N . Mice of the same genotype were housed three to five per cage , randomly assigned to treatment groups at the beginning of the experiment and subjected to a single experimental regime in their cages . Mice were euthanized by CO2 asphyxiation and tissues harvested in the mid morning to mid afternoon . Male mice ( 10 to 16 weeks old ) were treated with 2 . 5% DSS ( MW 36 , 000–50 , 000 Da , MP Biomedicals , #160110 , Santa Ana , CA ) in water given ad libitum for four to six days and sacrificed immediately or allowed to recover up to 18 days with tap water . WT mice were treated in parallel in each experiment as controls for the DSS: the same lot number of DSS was used for a given group of experiments whenever possible to avoid lot-to-lot variation . Mice in severe distress ( weighing 13 grams or less , or excessively hunched and lethargic ) were euthanized prior to the termination of the experiment except in experiments measuring mortality . To avoid confounding effects due to unrelated illnesses , when an animal became unexpectedly ill , all mice in the cage were excluded from the analysis . CAC was established as described ( Neufert et al . , 2007 ) . Briefly , we intraperitoneally ( i . p . ) injected male mice ( 6 to 10 weeks old ) with 10 mg/kg AOM ( National Cancer Institute , Bethesda , MD ) on Day 1 in the morning . On Day 2 mice were given 2 . 5% DSS in water for four to seven days , followed by 16 days of untreated water; the cycle was repeated one or two additional times . Mice were sacrificed at day 46 to 95; tumor number counted by visual inspection and tumor size measured with digital calipers were determined in a blind fashion . Sporadic colon cancer was induced in male mice ( 6 to 8 weeks old ) by i . p . injection of 10 mg/kg AOM once a week for six consecutive weeks . Mice were sacrificed 28 weeks after the first injection . Distal colons were fixed in 10% phosphate buffered formalin and stained with hematoxylin and eosin ( H&E ) or for immunofluorescence ( IF ) as described previously ( Tanaka et al . , 2006; Lytle et al . , 2005 ) . For antigen retrieval , tissue sections were soaked in 1% SDS in phosphate buffered saline ( PBS ) and microwaved for 2 min for all antibodies except for P2-HNF4α which was autoclaved at 121°C for 20 min in 10 mM citrate buffer . Images were captured with a Zeiss 510 confocal microscope . Mouse monoclonal antibodies to P1/P2-driven HNF4α ( #PP-H1415-00 ) , P1-driven HNF4α ( #PP-K9218-00 ) and P2-driven HNF4α ( #PP-H6939-00 ) were from R&D Systems ( Minneapolis , MN ) . Antibodies to Ki67 were from Abcam ( #ab1667 , Cambridge MA ) . Rabbit NKCC1 antibody ( TEFS2 ) has been described previously ( McDaniel and Lytle , 1999 ) . Alexa fluor anti-mouse and anti-rabbit secondary antibodies and TO-PRO-3 nuclear stain ( red ) were from Life Technologies ( Carlsbad , CA ) . For RELMβ IF staining , antigen retrieval was performed by immersion of slides in 95–100°C pre-heated sodium citrate buffer ( 10 mM ) . Following cooling to room temperature , slides were rinsed twice with PBS/0 . 1% Tween 20 for 5 min and then blocked with 5% normal donkey serum ( Jackson Immuno Research Labs , Westgrove , PA ) in StartingBlock ( Thermo Scientific , Carlsbad , CA ) . Sections were stained with rabbit anti-RELMβ antibody ( #500-P215 , Peprotech , Rocky Hill , NJ ) , followed by fluorochrome-conjugated anti-rabbit antibody ( Abcam ) , and counterstaining with DAPI ( Cell Signaling Technology , Danvers , MA ) . Whole cell extracts ( WCE ) were prepared from either snap-frozen or fresh tissue using ice-cold Triton lysis buffer by motorized ( Wheaton , Millville , NJ ) or manual homogenization . Triton lysis buffer was 20 mM Tris pH 7 . 5 , 150 mM NaCl , 10% glycerol , 1% NP40 , 1% Triton-X-100 , 1 mM EDTA , 2 mM EGTA plus inhibitors ( 1 μg/ml of aprotonin , leupeptin and pepstatin , 1 mM of sodium orthovanadate , sodium fluoride and ( PMSF ) , phosphatase inhibitor cocktail I & II ( 1:100 ) , protease inhibitor cocktail ( 1:10 – 1:100 ) , Sigma-Aldrich , St . Louis , MO ) . Protein extracts ( ~20–100 μg ) were analyzed by 10% SDS-PAGE followed by transfer to Immobilon ( EMD Millipore , Billerica , MA ) before staining with antibodies or Coomassie for protein loading . Blinded histology scoring of H&E stained sections was performed according to three criteria . Crypt damage: 0 = intact crypts , 1–2 = loss of basal area , 3–4 = entire crypt loss with erosion , 5 = confluent erosion . Leukocyte inflammation: 0 = no inflammatory infiltrate , 1 = leukocyte infiltration in the lamina propria , 2 = leukocyte infiltration extending into the submucosa , 3 = transmural and confluent extension on inflammatory cells . Ulceration: 0 = no ulcers , 1–2 = presence of ulcers , 3 = confluent and extensive ulceration . Young adult male mice were injected i . p . with 75 mg/kg BrdU ( BD Biosciences , #550891 , San Jose , CA ) and sacrificed after 2 to 3 hr , 25 hr or 48–50 hr . Distal colons fixed in formalin were sectioned and immunostained with BrdU antibody as per manufacturer instruction ( BD Biosciences , #550803 ) . Images were captured at 40X ( Zeiss Axioplan , Jen Germany ) and crypt dimensions were measured using SPOT Imaging software ( Sterling Heights , MI ) . Distal colons from 19-week-old male WT mice from the mixed background ( WTα7 ) were rinsed in PBS , placed in a 4% bleach solution for 20 min , washed three times in PBS and then incubated with 3 mM EDTA , 0 . 5 mM ( DTT ) in PBS for 90 min at 4°C followed by a PBS wash . Colonic crypts were isolated by vigorous shaking; they were fixed and immunostained as described ( Chellappa et al . , 2012 ) . The short circuit current ( Isc ) and electrical resistance across the mucosal layer of mouse distal colon was measured using an Ussing chamber as described ( Bajwa et al . , 2009 ) . Electrogenic Cl− secretion was recorded as the Isc evoked by sequential addition of 100 µM carbachol and 20 µM forskolin ( Sigma-Aldrich ) to the serosal bath . Mouse Exon 1 . 0 ST Arrays ( Affymetrix , Santa Clara , CA ) were hybridized at the University of California Riverside Genomics Core using polyA+ RNA extracted from the distal colon of young adult male mice ( 12 to 16 weeks old ) fed a standard lab chow ad libitum in an open access vivarium . RNA was pooled from two to three mice per genotype and applied to one array; a second array was processed in a similar fashion for a total of four to six mice assayed per genotype . Results from the two arrays were averaged . Isoform-specific mice were compared to their appropriate , age-matched WT controls . Data were analyzed using Robust Multi-array Average ( RMA ) background adjustment and quantile normalization on probe-level data sets with Bioconductor packages , Exonmap , and Affy software . To determine the differentially expressed transcripts only the probes with P<0 . 05 ( Student’s t-test ) and more than two-fold change were considered . Gene Ontology ( GO ) overrepresentation analysis was conducted using DAVID . Microarray data have been deposited in the Gene Expression Omnibus MIAME-compliant database ( Accession number GSE47731 ) ( http://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE47731 ) . Colon tissue ( ~1 cm ) was weighed and homogenized in 0 . 5 mL PBS , followed by ELISA with capture and detection biotinylated antibodies for anti-RELMβ ( Cat #500-P215Bt , Peprotech ) according to the manufacturer’s instructions . Samples were compared to a serial-fold dilution of recombinant mouse RELMβ protein ( #450-26B , Peprotech ) and calculated as ng per gram tissue . All ELISAs were performed in technical triplicates . Human colonic epithelial cells Caco2 cells ( ATCC HTB-37 ) were grown in DMEM ( Dulbecco’s modified Eagle’s medium with 4 . 5 g/liter glucose , L-glutamine , and pyruvate ) supplemented with 20% fetal bovine serum ( FBS ) ( BenchMark; cat#100–106 ) and 100 U/mL penicillin-streptomycin ( 1% P/S ) at 37°C and 5% CO2 . At ~95% confluency the cells were crosslinked with formaldehyde and subjected to ChIP analysis using the affinity purified anti-HNF4α antisera ( α445 ) , which recognizes the very C-terminus of both P1- and P2-HNF4α , as described previously ( Vuong et al . , 2015 ) . The following primers in the RETNLB promoter were used in the PCR for 40 cycles: Region 1 forward 5’-CTCCTCCACCTCTTCCTCCT-3’ and reverse 5’-CATCCTAATCCCCCTTCTCC-3’ ( 301 bp ) ; Region 2 forward 5’-CCTTTGCTCTGGATCTCTGC-3’ and reverse 5’- ATGAGCCCCCAAAAGAACTC-3’ ( 405 bp ) . Primers in the HMOX1 promoter were used as a positive control , forward 5’-CCTCTCCACCCCACACTGGC-3’ and reserve 5’-GCGCTGAGGACGCTCGAGAG-3’ ( 179 bp ) . Primers were designed using the UCSC genome browser ( https://genome . ucsc . edu/ ) and Primer3 ( v . 0 . 4 . 0 ) ( http://bioinfo . ut . ee/primer3-0 . 4 . 0/ ) . Predicted HNF4α binding sites ( Support Vector Machine , SVM , algorithm ) were identified using the HNF4α Binding Site Scanner ( http://nrmotif . ucr . edu/ ) ( Bolotin et al . , 2010 ) . Intestinal epithelial permeability was assessed by measuring the appearance of FITC-dextran ( FD-4 , Sigma ) in mouse serum as described previously ( Brandl et al . , 2009 ) . Untreated or treated ( 2 . 5% DSS for six days or 2 . 5% DSS for six days , followed by days of recovery with tap water ) WT , α7HMZ or α1HMZ mice were fasted overnight and then gavaged with FITC-dextran ( 60 mg/100 g body weight ) 4 hr before harvesting . Blood was collected either from the inferior vena cava or by cardiac puncture and allowed to sit on ice for 30 min . Serum was collected after centrifuging the blood for 15 min at 2000 xg at 4°C . FITC-dextran measurements were performed in duplicate or triplicate by fluorometry at 490 nm . Data were analyzed using Prism 6 software ( GraphPad Prism version 6 for Mac , GraphPad Software , La Jolla , CA ) ; outliers were identified by the ROUT method and removed . Human HNF4α2 ( NM_000457 ) and HNF4α8 ( NM_175914 . 3 ) constructs in pcDNA3 . 1 ( + ) vector were gifts from Dr . Christophe Rachez ( Pasteur Institute , Paris , France ) as described previously ( Chellappa et al . , 2012 ) . pGL2 . basic , human RELMβ reporter constructs and LS174T cells were gifts from Dr . Gary Wu ( Wang et al . , 2005 ) . ON-TARGET siRNA targeting P1- and P2-HNF4α were custom synthesized from Dharmacon . si P1-HNF4α: Sense , 5'-U U G A G A A U G U G C A G G U G U U U U -3'; Antisense 3'-U U A A C U C U U A C A C G U C C A C A A - ( 5'-P ) 5' . siP2-HNF4α: Sense , 5'-G U G G A G A G U U C U U A C G A C A U U-3'; Antisense , 3'-U U C A C C U C U C A A G A A U G C U G U- ( 5'-P ) 5' . ON-TARGETplus Non-targeting siRNA #1 ( D-001810-01-20 ) was used as siControl . LS174T cell lines were grown in DMEM supplemented with 10% FBS and penicillin-streptomycin ( 1% P/S ) and maintained at 37°C and 5% CO2 . For siRNA experiments , 8X106 LS174T cells were plated in 60-mm plates and transfected 24 hr after plating with 100 nM siRNA using RNAi Max ( Invitrogen ) according to the manufacturer’s protocol . Forty-eight hours after transfection , cells were split into a 24-well plate ( 8X106 cells per well ) , and 24 hr later transfected with Lipofectamine 3000 according to the manufacturer’s protocol ( Invitrogen ) with CMV . βgal ( 50 ng ) and pGL2 . basic or human RELMβ reporter constructs ( 1 μg ) . For HNF4α transfections , 8X106 LS174T cells were plated in 24-well plates and 24 hr later transfected with human HNF4α2 or HNF4α8 ( 100 ng ) , CMV . βgal ( 50 ng ) and pGL2 . basic or human RELMβ reporter constructs ( 1 μg ) . Cells were harvested 24 hr after transfection using passive lysis buffer ( Promega ) . Luciferase and β-gal activity were measured as described previously ( Chellappa et al . , 2012 ) . RIME was carried out as previously described ( Mohammed et al . , 2013 ) , with the following modifications . Whole colon from α1HMZ and α7HMZ untreated male mice ( n = 3 per genotype , ~16 weeks of age , backcrossed into C57/BL6N and maintained in an SPF vivarium ) were fixed in 1 . 1% methanol-free formaldehyde ( 5 mL ) [in 1 x phosphate buffered saline ( PBS plus 1 mM PMSF and DTT , 2 μg/mL leupeptin and aprotinin , and 1:200 protein phosphatase inhibitors I ( 2&3 ) ( Sigma ) ] for 10 min at RT; crosslinking was then stopped with 0 . 125 M glycine for 5 min at RT . Fixed colon was further processed into single cells at 4°C . Colon was lightly minced with a razor blade and disaggregated in PBS plus inhibitors ( as above ) by passing through a motorized homogenizer . The cells were then drained through a cell strainer and dounced using a hand-held homogenizer . Cells were swelled in 1 . 0 mL Hypotonic Buffer ( 10 mM HEPES-KOH pH 7 . 9 , 10 mM KCl , 1 . 5 mM MgCl2 ) plus inhibitors for 10 min at 4°C and centrifuged to collect the nuclei . The pellet was washed with Nuclei Buffer ( 1% SDS , 50 mM Tris-Cl pH 8 . 0 , 10 mM EDTA ) plus inhibitors and D3 Buffer ( 0 . 1% SDS , 10 mM Tris-Cl , 1 mM EDTA ) . The pellet was resuspended in fresh 0 . 5 mL D3 Buffer plus inhibitors , transferred to a 1 . 0 mL AFA millitube with a plastic stirring rod and more D3 Buffer was added to fill the tube ( total volume ~780 μL ) . Samples were sonicated for 9 min ( 4 min per 200 cycles/bursts , 5 . 0 duty force , and 140 peak power; one min delay ) in a Covaris S220 , Bioruptor and pre-cleared for 30 min in 10 μL of magnetic beads ( Pierce Thermo Scientific , cat#88802 ) . Prior to use , magnetic beads were washed 3 x 1 . 0 mL in cold PBS . Samples were split in half and diluted 1:1 with Immunoprecipitation ( IP ) Dilution Buffer ( 0 . 01% SDS , 20 mM Tris-Cl pH8 . 0 , 1 . 1% Triton X-100 , 167 mM NaCl , 1 . 2 mM EDTA ) and placed in non-stick tubes . Each half sample was incubated with 40 μL of magnetic beads that were pre-incubated with 21 μg of anti-HNF4α ( α-445 ) ( Sladek et al . , 1990 ) or rabbit IgG in 0 . 05% Tween PBS at 4°C overnight . The following day , IPs were washed 3 x 1 . 0 mL with ice cold Radioimmunoprecipitation assay ( RIPA ) Buffer ( 15 mM Tris-Cl , 150 mM NaCl , 1% NP-40 , 0 . 7% deoxycholate ) , washed 1 x 400 μL with DNaseI Buffer ( 40 mM Tris-Cl , 1 mM CaCl2 , 10 mM NaCl , 6 mM MgCl2 ) and incubated in 100 μL DNaseI Buffer and 8 μL of DNaseI enzyme ( 4 μg/μL ) for 20 min at 30°C . Additional washes were done ( 3 x 0 . 5 mL with RT RIPA Buffer and 2 x 1 . 0 mL with cold RIPA Buffer ) . IP’d material was washed twice with cold 50 mM NH4CO3 . At the last wash ( 0 . 5 mL ) , samples were transferred to new tubes . Wash buffer was removed and IP beads subjected to mass spectrometry as follows: Sample beads were washed with trypsin digestion buffer , digested with trypsin overnight and subjected to 2D-nanoLC/MS/MS analysis at the UCR Institute of Integrated Genome Biology Proteomics Core as described previously ( Drakakaki et al . , 2012 ) . Briefly , a MudPIT approach employing a two-dimension nanoAcquity UPLC ( Waters , Milford , MA ) and an Orbitrap Fusion method ( Thermo Scientific , San Jose , CA ) was used to analyze all sample . Data were acquired using Orbitrap fusion method ( Hebert et al . , 2014 ) with acquisition time set from 1–70 min . For MS2 scanning only precursor ions with intensity of 50 , 000 or higher were selected and scanned from most intense to least intense precursor ions within 4 s . A 5-s exclusion window was applied to all abundant ions to avoid repetitive MS2 scanning on the same precursor ions using 10 ppm error tolerance . All raw MS data were processed and analyzed using Proteome Discoverer version 2 . 1 ( Thermo Scientific , San Jose , CA ) . Only proteins with 1% FDR cut-off ( q<0 . 01 ) were considered for subsequent analysis . Proteins had to be present in at least two out of the three replicates with the HNF4α antibody and not in any IgG control in both α1HMZ and α7HMZ samples to be considered for the 'both' category . To be considered in the 'α7HMZ only' category , the protein had to appear either in two of three α7HMZ samples but in none of the three α1HMZ samples , or in three of the α7HMZ samples and only one of the α1HMZ samples . A similar strategy was used for the 'α1HMZ only' proteins . Proteins were converted to gene symbols and cross-referenced with human and mouse genes associated with colon cancer , IBD , Crohn’s disease and ulcerative colitis found in the literature ( Franke et al . , 2010; Jostins et al . , 2012 ) and a Pubmed-Gene search conducted in April 2016 , followed by manual curation . Gene Ontology using Panther ( www . pantherdb . org ) as well as manual curation resulted in the TF , RNA binding and protein kinase and phosphatase categories in Figure 5C bottom . The Human Protein Atlas ( http://www . proteinatlas . org/ ) was used to confirm expression in the colon and nucleus . Sample sizes for DSS and AOM+DSS regimes were determined on the basis of mouse-to-mouse variation in body weight loss and tumor number/load ( respectively ) observed in pilot experiments . Each mouse was considered to be a biological replicate; technical replicates refer to multiple analyses of the same tissue from a given animal . All results are expressed as the mean ± s . e . m , of sample size n . Significance was tested by analysis of variance or Student's t-test . Probabilities less than 5% ( P<0 . 05 ) were considered to be significant . For RIME , a cut off of q<0 . 01 ( 1% FDR ) was used .	The digestive system in animals consists of a network of organs – including the liver , stomach , pancreas and intestines – that work together to break down food and deliver energy to the rest of the body . Many proteins called transcription factors help to guide the development of these organs and keep them healthy throughout life . Among these is a protein called HNF4α . In various diseases of the digestive system , such as gastric cancer or inflammatory bowel disease , the production of HNF4α is not properly regulated . Gene expression can be activated by transcription factors binding to regions of DNA called promoters . The gene that encodes HNF4α has two promoters called P1 and P2 , and each produce several different versions of the HNF4α protein . The colon contains intestinal glands ( also known as colonic crypts ) that contain a lower part in which cells actively divide and an upper part of non-dividing cells that help with digestion . Previous studies have shown that if the mouse colon is unable to produce HNF4α , the structure of the crypts is disrupted . By studying crypts taken from the colon of mice , Chellappa et al . have now found that P1-HNF4α proteins are mainly produced at the top of the crypts , whereas P2-HNF4α proteins are found mainly at the bottom . Chellappa et al . then used two sets of genetically engineered mice: one that can only produce P1-HNFα proteins , and one that only has P2-HNFα proteins . Under normal conditions both sets of mice appeared healthy . However , differences became apparent if the mice were subjected to treatments that cause colitis or colitis-associated colon cancer . Mice that could only produce P1-HNF4α proteins were less susceptible to colitis and got fewer and smaller tumors than normal mice . By contrast , mice that could only produce P2-HNF4α experienced more colitis and developed more tumors than normal mice . Comparing the genes expressed in the colon cells of these two types of mice revealed several differences . In particular , much more of a pro-inflammatory protein called RELMβ was produced in P2-only mice . Chellappa et al . then proceeded to show that RELMβ is essential for the susceptibility of P2-mice to coliltis . Overall , the experiments show that P1-HNF4α and P2-HNF4α perform different tasks both in the healthy and the diseased mouse colon . In future it will be important to work out how the balance between the two sets of proteins is disrupted in diseases of the colon .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "cell", "biology" ]	2016	Opposing roles of nuclear receptor HNF4α isoforms in colitis and colitis-associated colon cancer
Genomic amplification of the androgen receptor ( AR ) is an established mechanism of antiandrogen resistance in prostate cancer . Here , we show that the magnitude of AR signaling output , independent of AR genomic alteration or expression level , also contributes to antiandrogen resistance , through upregulation of the coactivator GREB1 . We demonstrate 100-fold heterogeneity in AR output within human prostate cancer cell lines and show that cells with high AR output have reduced sensitivity to enzalutamide . Through transcriptomic and shRNA knockdown studies , together with analysis of clinical datasets , we identify GREB1 as a gene responsible for high AR output . We show that GREB1 is an AR target gene that amplifies AR output by enhancing AR DNA binding and promoting EP300 recruitment . GREB1 knockdown in high AR output cells restores enzalutamide sensitivity in vivo . Thus , GREB1 is a candidate driver of enzalutamide resistance through a novel feed forward mechanism . Androgen receptor ( AR ) targeted therapy is highly effective in advanced prostate cancer but is complicated by the emergence of drug resistance , called castration-resistant prostate cancer ( CRPC ) ( Shen and Abate-Shen , 2010; Watson et al . , 2015 ) . The most common mechanism of CRPC is restored AR signaling , primarily through amplification of AR ( Chen et al . , 2004; Robinson et al . , 2015 ) . The importance of AR amplification as a clinically important drug resistance mechanism is underscored by recent data showing that AR amplification , detected in circulating tumor DNA or in circulating tumor cells ( CTCs ) , is correlated with reduced clinical benefit from the next generation AR inhibitors abiraterone or enzalutamide ( Annala et al . , 2018; Podolak et al . , 2017 ) . Genomic landscape studies of prostate cancer have revealed several molecular subtypes defined by distinct genomic drivers ( Berger et al . , 2011; Cancer Genome Atlas Research Network , 2015; Taylor et al . , 2010 ) . In addition to this genomic heterogeneity , primary prostate cancers also display heterogeneity in AR transcriptional output , measured by an AR activity score ( Hieronymus et al . , 2006 ) . Notably , these differences in transcriptional output occur in the absence of genomic alterations in AR , which are generally found only in CRPC ( Cancer Genome Atlas Research Network , 2015 ) . One potential explanation for this heterogeneity in AR transcriptional output is through coactivators and other AR regulatory proteins such as FOXA1 , SPOP , FOXP1 and TRIM24 ( Cancer Genome Atlas Research Network , 2015; Geng et al . , 2013; Groner et al . , 2016; Pomerantz et al . , 2015; Takayama et al . , 2014 ) . Much of the work to date has focused on inter-tumoral heterogeneity . Here , we address the topic of intra-tumoral heterogeneity in AR transcriptional output , for which we find substantial evidence in prostate cancer cell lines and in primary prostate tumors . Using a sensitive reporter of AR transcriptional activity to isolate cells with low versus high AR output , we show that high AR output cells have an enhanced response to low doses of androgen and reduced sensitivity to enzalutamide , in the absence of changes in AR mRNA and protein expression . To understand the molecular basis for these differences , we performed transcriptome and shRNA knockdown studies and identified three genes ( GREB1 , KLF8 and GHRHR ) upregulated in high AR output cells , all of which promote AR transcriptional activity through a feed-forward mechanism . Of these , we prioritized GREB1 for further characterization because GREB1 mRNA levels are increased in primary prostate tumors that have high AR activity . GREB1 amplifies AR transcriptional activity through a two-part mechanism: by promoting EP300 recruitment and by enhancing AR binding to chromatin . Importantly , GREB1 knockdown converted high AR output cells to a low AR output state and restored enzalutamide sensitivity in vivo . Collectively , these data implicate GREB1 as an AR signal amplifier that contributes to prostate cancer disease progression and antiandrogen resistance . Previous work using a KLK3 promoter/GFP reporter ( PSAP-eGFP ) showed that LNCaP prostate cancer cells display varying levels of eGFP expression . Characterization of low GFP cells in this analysis revealed reduced AR levels and increased expression of stem cell and developmental gene sets ( Qin et al . , 2012 ) . We explored this question in the context of the contemporary data on heterogeneity in AR transcriptional output using a different AR-responsive reporter , ARR3tk-eGFP , where eGFP expression is driven by the probasin promoter modified to contain three AR responsive elements ( Snoek et al . , 1998 ) . LNCaP ( Figure 1 ) and CWR22PC-EP ( Figure 1—figure supplement 1 ) prostate cancer cells containing a single copy of the reporter construct were derived by infection with lentivirus containing the reporter at a low multiplicity of infection ( MOI ) ( Figure 1A ) . Remarkably , we observed >100 fold range in eGFP expression , as measured by flow cytometry , despite similar levels of AR by immunofluorescence microscopy ( Figure 1B , C , Figure 1—figure supplement 1A ) . We then used flow cytometry to isolate eGFP-positive cells from both ends of the spectrum of AR transcriptional output , which we refer to as ARsig-hi ( high AR output ) and ARsig-lo ( low AR output ) cells , respectively ( Figure 1C , Figure 1—figure supplement 1A ) . ARsig-hi cells also express higher levels of endogenous AR target genes ( FKBP5 , KLK3 , TRPM8 ) ( Figure 1D , E , Figure 1—figure supplement 1B , C ) , and have an overall increase in AR transcriptional activity based on RNA-sequencing analysis ( Figure 1F ) . In addition , the ARsig-lo and ARsig-hi transcriptional phenotypes remain stable for over 30 days post-sorting ( Figure 1G , Figure 1—figure supplement 1D ) . Interestingly , ARsig-lo cells showed upregulation of gene sets related to proliferation and cell cycle ( Figure 1—source data 1 ) . Of note , Qin et al . ( 2012 ) reported downregulation of these gene sets in their low/absent KLK3 cells , suggesting that the two reporters read out different transcriptional activities . Importantly , the difference in AR output between ARsig-lo and ARsig-hi cells is not explained by different levels of AR expression or nuclear translocation , since both were comparable in each subpopulation ( Figure 1D , E , Figure 1—figure supplement 1B , C , Figure 1—figure supplement 2 ) . We next asked if isolated ARsig-lo and ARsig-hi populations have different responses to ligands such as dihydrotestosterone ( DHT ) or antagonists such as enzalutamide . ARsig-hi cells showed enhanced sensitivity to DHT in a dose-dependent manner ( Figure 1H; Figure 1—figure supplement 1E ) . This result is similar to the effect of increased AR expression in conferring sensitivity to low doses of androgen ( Chen et al . , 2004 ) , but now without a change in AR level . To address sensitivity to enzalutamide , we used LNCaP/AR xenografts ( derived from LNCaP cells ) because this model has a track record of revealing clinically relevant mechanisms of enzalutamide resistance ( Arora et al . , 2013; Balbas et al . , 2013 ) . As we did with LNCaP and CWR22PC-EP cells , we derived ARsig-lo and ARsig-hi subpopulations by flow cytometry and also observed differential AR output despite similar levels of AR expression ( Figure 1—figure supplement 3A–C ) . Remarkably , ARsig-hi cells developed enzalutamide resistance significantly faster that ARsig-lo or parental cells when injected into castrated mice treated with enzalutamide ( Figure 1I ) . Having demonstrated heterogeneous AR output within prostate cancer cell lines , we asked if similar , intra-tumoral heterogeneity is observed clinically by immunohistochemical analysis of KLK3 and AR expression in several primary cancers . Consistent with previous reports ( Qin et al . , 2012; Ruizeveld de Winter et al . , 1994 ) , we observed heterogeneous KLK3 staining that is not strictly correlated with AR level . For example , we found variable intensity of KLK3 staining in tumor cells with comparable levels of AR staining ( lined boxes; Figure 1—figure supplement 4 ) and , conversely , variable intensity of AR staining in tumor cells with similar KLK3 staining ( dotted circles; Figure 1—figure supplement 4 ) . Although this is a small dataset , the results indicate that the AR transcriptional heterogeneity we observe in prostate cancer cell lines is present in patient samples . Emerging technologies for conducting single cell RNA and protein analysis in clinical material will enable deeper investigation of this question . To elucidate the molecular basis underlying the differences in ARsig-lo and ARsig-hi cells , we performed RNA-sequencing and found 69 genes upregulated in ARsig-lo cells and 191 genes upregulated in ARsig-hi cells ( fold change ≥1 . 5 , p<0 . 05 , Figure 2—source data 1 ) . In addition to enrichment of gene sets regulated by androgen ( Figure 1F ) , human prostate luminal and basal cell gene sets were enriched in ARsig-hi and ARsig-lo cells , respectively ( Figure 2A ) . Based on these results , we postulated that high AR output could be a consequence of upregulation of transcriptional co-activators and/or of genes involved in luminal differentiation . We therefore filtered the list of 191 genes upregulated in ARsig-hi cells and identified 33 genes annotated as co-activators or luminal genes ( Figure 2—source data 2 ) , then measured the consequence of shRNA knockdown of each one on AR output in ARsig-hi cells ( Figure 2B ) . Three of the 33 candidate genes ( GREB1 , GHRHR , KLF8 ) inhibited AR activity when knocked down in ARsig-hi cells , with successful knockdown confirmed by qRT-PCR ( Figure 2C , D ) . AR knockdown served as a positive control , and ACPP ( one of the 30 genes that did not score ) served as a negative control . Interestingly , all three hits are transcriptional upregulated by DHT simulation ( Figure 2E ) , which likely explains their increased expression in ARsig-hi cells . Among the three , GREB1 emerged as the most compelling candidate for further investigation based on interrogation of clinical datasets . Specifically , we found a statistically significant positive correlation ( r ) between GREB1 RNA level and AR output score ( Cancer Genome Atlas Research Network , 2015; Hieronymus et al . , 2006 ) across the primary prostate tumors from the TCGA dataset , but not GHRHR or KLF8 ( Figure 2F ) . Consistent with this , increased expression of GREB1 , but not GHRHR or KLF8 , was observed in TCGA cases with high AR scores ( top 5% ) versus low AR scores ( bottom 5% ) ( Figure 2F , Figure 2—source data 3 ) . To be sure that GREB1 is relevant in other model systems , we confirmed GREB1 upregulation in CWR22PC-EP ARsig-hi cells ( Figure 2—figure supplement 1A ) and reduced AR output after GREB1 knockdown ( Figure 2—figure supplement 1B ) . We further validated the knockdown data using CRISPR/Cas9 , which also showed inhibition of AR output ( by flow cytometry ) and highly reduced KLK3 expression in LNCaP ARsig-hi sublines expressing different sgRNAs targeting GREB1 , without detectable changes in AR protein level ( Figure 2G , H ) . GREB1 was first reported as an estrogen-regulated gene in breast cancer ( Rae et al . , 2005 ) then shown to bind directly to ER , presumably through its LxxLL motif , and function as an ER coactivator by promoting interaction with cofactors ( Mohammed et al . , 2013 ) . To determine if GREB1 also functions as an AR coactivator , we introduced exogenous GREB1 ( HA-GREB1 ) into ARsig-lo LNCaP and CWR22PC-EP cells and derived stably expressing sublines ( Figure 3A , Figure 3—figure supplement 1A ) . GREB1 overexpression enhanced DHT-induced AR target gene expression in a dose-dependent manner ( Figure 3B , C , Figure 3—figure supplement 1B ) , indicating that GREB1 also promotes AR activity . In breast cancer , GREB1 functions as a coactivator through binding to ER and recruitment of the EP300/CBP complex to ER target genes ( Mohammed et al . , 2013 ) . We find that GREB1 functions similarly in prostate cells , as shown by co-immunoprecipitation documenting AR-GREB1 interaction ( Figure 3D ) and ChIP experiments showing recruitment of GREB1 to KLK3 and FKBP5 enhancer regions ( Figure 3E ) . Furthermore , ARsig-hi cells showed a GREB1-dependent increase in EP300 binding ( Figure 3F , G ) and GREB1 overexpression increased EP300 recruitment to AR target genes in ARsig-lo cells ( Figure 3—figure supplement 2A ) . Knockdown of EP300 suppressed the effect of GREB1 overexpression on DHT-induced AR target gene upregulation in ARsig-lo cells ( Figure 3—figure supplement 2B , refer also to Figure 3B ) , suggesting that EP300 is required for the function of GREB1 as an AR co-factor . In addition to this canonical coactivator function of promoting assembly of an active transcription complex , we found that GREB1 also impacts AR DNA binding . For example , knockdown or CRISPR deletion of GREB1 in ARsig-hi cells significantly reduced binding of AR to the KLK3 enhancer and , conversely , GREB1 overexpression promoted AR recruitment in ARsig-lo cells ( Figure 3H , Figure 3—figure supplement 2C ) . AR ChIP-sequencing revealed that this effect is genome-wide , with a significant reduction in the mean height of AR peaks in GREB1-depleted cells ( Figure 3I–K ) . Importantly , the location of AR peaks ( enhancer , promoter ) was identical in intact versus GREB1 knockdown cells and there were no differences in consensus binding sites ( Figure 3—figure supplement 2D , E ) . Therefore , GREB1 enhances AR DNA efficiency but not alter DNA-binding site specificity . As seen previously in our analysis of ARsig-hi cells , total and nuclear AR levels were not changed by GREB1 knockdown or overexpression ( Figure 3C , Figure 3—figure supplement 2F , G ) . Of note , earlier studies of GREB1 in breast cancer did not report any effect on ER DNA binding ( Mohammed et al . , 2013 ) , which we confirmed by GREB1 knockdown in MCF7 breast cancer cells ( Figure 3—figure supplement 3A , B ) . Thus , GREB1 functions as a coactivator of both ER and AR but through somewhat different mechanisms . To address the possibility that other hormone receptor coactivators might also function differently in prostate cells , we asked if NCOA1 and NCOA2 , previously shown to recruit the EP300/CBP complex to AR ( Leo and Chen , 2000 ) , also influence AR DNA binding . To do so , we knocked down both genes in ARsig-hi cells based on prior work showing redundancy between NCOA1 and NCOA2 ( Leo and Chen , 2000; Wang et al . , 2005 ) . AR reporter activity and target gene expression was inhibited in NCOA1/2-depleted cells , as expected , but AR occupancy of AR binding sites was unchanged ( Figure 3—figure supplement 3C–E ) . Thus , in addition to a role in EP300/CBP recruitment , GREB1 has unique effects on AR DNA binding that distinguish it from other coactivators . Having demonstrated that GREB1 is overexpressed in ARsig-hi cells and functions as an AR coactivator , we asked if GREB1 is required for maintenance of the ARsig-hi state . First we evaluated the consequences of GREB1 knockdown on transcription . Consistent with experiments in ARsig-lo cells showing that GREB1 overexpression enhanced AR transcriptional activity ( Figure 3B , C , Figure 3—figure supplement 1B ) , GREB1 knockdown inhibited baseline and DHT-induced AR target gene expression in ARsig-hi cells ( Figure 4A–C , Figure 4—figure supplement 1A , B ) . RNA-sequencing confirmed enrichment of androgen down-regulated gene sets in GREB1-depleted cells ( Figure 4D ) as well as downregulation of the 20 AR target genes used to calculate the AR activity score in TCGA tumors ( Figure 4—figure supplement 1C ) . GREB1 knockdown cells also showed enrichment of the same prostate basal gene set that was enriched in ARsig-lo cells ( Figure 4D , refer also to Figure 2A ) . Additional analysis of RNA-seq data suggests that GREB1 is a major molecular determinant of the ARsig-hi state: specifically , ( i ) GREB1 knockdown impaired the induction of >70% of all DHT-induced genes ( Figure 4E , Figure 4—source datas 1 and 2 ) and ( ii ) the top 100 gene sets enriched in GREB1-depleted ARsig-hi cells and ARsig-lo cells show significant overlap ( Figure 4F , Figure 4—source data 3 ) . Earlier we showed that ARsig-hi cells rapidly acquire resistance to enzalutamide ( refer to Figure 1I ) . To determine the role of GREB1 in this drug resistant phenotype , we performed knockdown experiments using the LNCaP/AR xenograft . After confirming that AR activity was inhibited in ARsig-hi cells ( Figure 4—figure supplement 1D , E ) , we injected LNCaP/AR ARsig-hi xenografts with GREB1 shRNAs into castrated mice treated with enzalutamide and found a significant delay in the development of enzalutamide resistance after 10 weeks ( Figure 4G ) . Clinical data from CRPC patients also supports for a role of GREB1 in enzalutamide resistance . Although the samples are not matched pre- and post-treatment , we observed an overall increase in GREB1 expression in those who progressed on enzalutamide treatment ( Figure 4H ) . When we analyzed tumor purity content and stromal signature score as described previously ( Carter et al . , 2012; Shah et al . , 2017; Yoshihara et al . , 2013 ) , no significant difference was observed between samples collected pre- vs . post-treatment ( Figure 4—figure supplement 1F ) . There is abundant evidence from tumor sequencing studies that genomic alterations in AR ( amplification and/or mutation ) are present in over 50% of CRPC patients ( Cancer Genome Atlas Research Network , 2015; Robinson et al . , 2015 ) and that AR amplification is associated with a less favorable clinical response to abiraterone or enzalutamide treatment ( Annala et al . , 2018 ) . Therefore , high levels of AR transcriptional output can promote castration-resistant disease progression . Here we show that prostate cancers can amplify AR output through increased expression of the dual AR/ER coactivator GREB1 , in the absence of genomic AR alterations . As with genomic AR amplification , increased AR output driven by high GREB1 expression is also associated with enzalutamide resistance . In addition to demonstrating the importance of transcriptional heterogeneity in drug resistance , we also show that GREB1 amplifies AR activity by a novel two-part mechanism . Similar to canonical coactivators such as NCOA1/2 , GREB1 binds AR and promotes the assembly of an active transcription complex by recruitment of histone acetyl transferases such as EP300/CBP ( Lee and Lee Kraus , 2001 ) . However , GREB1 has the additional property of improving the efficiency of AR binding to DNA , which further enhances AR transcriptional output . Although conceptually distinct from canonical coactivators , this dual mechanism of AR activation is may not be unique to GREB1 . For example , TRIM24 has been shown to function as an oncogenic AR cofactor and , similar to GREB1 , knockdown of TRIM24 impairs recruitment of AR to target genes ( Groner et al . , 2016 ) . Curiously , the effect of GREB1 on AR DNA binding is not seen with ER , suggesting different conformational consequences of GREB1 binding on AR and ER , respectively , then influence DNA binding . One curious observation is the fact that prostate cancers can maintain transcriptional heterogeneity as a stable phenotype , despite the fact that GREB1 expression drives a feed forward loop which , in principle , should result in an increased fraction of high AR output cells over time . One potential explanation for the ability of these populations to maintain stable proportions of high versus low AR output cells at steady state is the fact that androgen has growth inhibitory effects at higher concentrations ( Culig et al . , 1999 ) . Because GREB1 amplifies the magnitude of AR output in response to normal ( growth stimulatory ) androgen concentrations , the biologic consequence of high GREB1 levels could be the same growth suppression seen with high androgen concentrations . This model predicts that high AR output cells would gain a fitness advantage under conditions of androgen deprivation or pharmacologic AR inhibition , as demonstrated by the enzalutamide resistance observed in xenograft models . Further work is required to understand the clinical implications of our work , particularly whether GREB1 levels in CRPC patients are predictive of response to next generation AR therapy . While we show that GREB1 levels are elevated in the tumors of CRPC patients who have progressed on enzalutamide , it will be important to address this question prospectively , prior to next generation AR therapy . It is also important to note that the positive correlation of GREB1 levels with high AR activity is largely based on the hormone-naïve TCGA cohort . It is also possible that the LNCaP cell line used for functional studies has an AR point mutation could potentially influence response to GREB1 expression , but we obtained similar results in 22PC cells that lack this mutation ( Veldscholte et al . , 1992 ) . In terms of therapeutic implications , GREB1 knockdown experiments provide genetic evidence that GREB1 is required for in vivo enzalutamide resistance in xenograft models . Although pharmacologic strategies to inhibit GREB1 function are not currently available , a small molecule inhibitor that blocks protein-protein interactions between the AR N-terminal domain and CBP/EP300 is currently in clinical development ( Andersen et al . , 2010 ) ( NCT02606123 ) . This work provides precedent that similar strategies to disrupt GREB1/AR interaction may be possible . LNCaP and MCF7 cell lines were obtained from American Type Culture Collection ( ATCC , Manassas , VA ) and maintained in RPMI ( LNCaP ) or DMEM ( MCF7 ) +10% FBS ( Omega Scientific , Tarzana , CA ) . LNCaP/AR cell line was generated and maintained as previously described ( Chen et al . , 2004 ) . CWR22Pc was a gift from Marja T . Nevalainen ( Thomas Jefferson University , Philadelphia , PA ) and CWR22Pc-EP was generated and maintained as previously described ( Mu et al . , 2017 ) . Cell lines were authenticated by exome sequencing methods , and were negative for mycoplasma contamination testing . Rapidly cycling eGFP AR reporter cells were collected using Accumax dissociation solution ( Innovative Cell Technologies , San Diego , CA ) , and dead cells were counterstained with DAPI ( Invitrogen , Grand Island , NY ) . For FACS-sorting of ARsig-lo and ARsig-hi cells , 5% of the entire population with lowest and highest eGFP expression was sorted out using BD FACSAria cell sorter . The 5% cutoff was used because it generates at least a 100-fold difference in median AR-GFP reporter signal between ARsig-lo and ARsig-hi cells and also allows us to have sufficient numbers of sorted cells to conduct various assays . For flow cytometric analysis of reporter activity , eGFP expression was measured using the BD-LDRII flow cytometer and analysis was done using FlowJo software . The lentiviral eGFP AR reporter ( ARR3tk-eGFP/SV40-mCherry ) was generated by switching 7xTcf promoter of 7xTcf-eGFP/SV40-mCherry ( Addgene , Cambridge , MA , 24304 ) with probasin promoter containing 3xARE ( ARR3tk ) ( Snoek et al . , 1998 ) . For shRNA knockdown experiments , SCEP vector was generated by substituting GFP cassette of SGEP ( pRRL-GFP-miRE- PGK-PuroR , gift from Johannes Zuber ) ( Fellmann et al . , 2013 ) with mCherry cassette . The following guide sequences were used for knockdown: shAR . 177: TAGTGCAATCATTTCTGCTGGC shGREB1-1: TTGTCAGGAACAGACACTGGTT shGREB1-2: TTTCAGATTTATATGATTGGAG shGREB1-3: TTGACAAGATACCTAAAGCCGA shKLF8 . 3467: TTGAGTTCTAAAGTTTTCCTGA shKLF8 . 2180: TATTTGTCCAAATTTAACCTAA shKLF8 . 2684: TTATAAAACAATCTGATTGGGC shGHRHR . 544: TAAAAGTGGTGAACAGCTGGGT shGHRHR . 1571: TTTATTGGCTCCTCTGAGCCTT shGHRHR . 1583: TTCATTTACAGGTTTATTGGCT shEP300-1: TCCAGAAAGAACTAGAAGAAAA shEP300-2: TTAATCTATCTTCAGTAGCTTG shNCOA1-1: TTCTTCTTGGAACTTGTCGTTT shNCOA2-1: TTGCTGAACTTGCTGTTGCTGA shNCOA2-2: TTAACTTTGCTCTTCTCCTTGC shRenilla was previously described as Ren . 713 targeting Renilla luciferase ( Fellmann et al . , 2013 ) . Pools of 3 shRNAs were used to knockdown GREB1 , KLF8 and GHRHR in a small-scale shRNA screen , and shGREB1-1 was used for further studies . For CRISPR/Cas9 experiments , lentiCRISPRv2 vector gifted by F . Zhang ( Addgene , 52961 ) was used with the following guide sequences designed using http://crispr . mit . edu/ website: SgGREB1-7: AGGCATGTCCTGCGTGCCGC SgGREB1-8: TCACGGGCATACGAGCAGTA sgNT was previously described ( Wang et al . , 2014 ) . pCMV6-GREB1 plasmid was a gift from J . Carroll ( Cancer Research UK Cambridge Institute , Cambridge , UK ) . The lentiviral GREB1 cDNA plasmid was constructed by cloning GREB1 cDNA from pCMV6-GREB1 into Tet-inducible pLV-based lentiviral expression vector with HA-tag . Lentiviral transduction of cells was performed as described previously ( Mu et al . , 2017 ) . To make AR reporter cell line , cells were infected with ARR3tk-eGFP/SV40-mCherry at low multiplicity of infection ( MOI ) to enable each cell has one copy of reporter construct , and the transduced cells were sorted by mCherry flow cytometry . To inactivate GREB1 gene , we single-cell cloned the cells infected with lentiCRISPRv2 vector containing SgGREB1-7 or SgGREB1-8 , and isolated a clone that had genomic alteration at target sequence . Three clones were generated by using SgGREB1-7 ( SgGREB1-7-2 , 7–11 and 7–12 ) and one clone was generated by using SgGREB1-8 ( SgGREB1-8-2 ) . FACS-based small-scale shRNA screen with 33 selected genes was performed as follows: FACS-sorted ARsig-hi cells were plated in 12 well plate ( 1 . 5 × 105 cells per well , Corning , 353043 ) and each well was infected with pool of 3 SEPC shRNAs against each gene on the following day . Cells with stable integration of hairpins were selected with 2 μg/ml puromycin . 9 days after infection , half of the cells in each well was used to analyze eGFP AR reporter activity using flow cytometry , and the other half was subjected to qRT-PCR to determine knockdown level of the gene . We performed the screen in duplicate and each replicate included wells infected with shRenilla or shAR as controls . The median fluorescence intensity ( MFI ) of eGFP was measured using FlowJo software . The shRNAs decreased eGFP MFI more than 1 . 5 fold compared to shRenilla ( normalized value lower than 0 . 667 ) in both duplicate were considered as hits . The list of 33 genes used in the screen and the summary of median eGFP intensity can be found at Figure 2—source data 2 . To compare time to acquire enzalutamide resistance in vivo , FACS-sorted bulk , ARsig-lo and ARsig-hi populations derived from LNCaP/AR were cultured for 6 days after sorting to obtain enough number of cells for xenograft assay . 2 × 106 cells were injected subcutaneously into the flank of physically castrated CB17 SCID mice in a 50:50 mix of matrigel ( BD Biosciences , San Jose , CA ) and regular culture medium ( five mice , 10 tumors per group ) , and enzalutamide treatment was initiated on the day of injection . To test the effect of GREB1 knockdown on development of enzalutamide resistance , FACS-sorted ARsig-hi population derived from LNCaP/AR was infected with control or three different shGREB1 constructs 2 days after sorting . Cells with stable integration of hairpin were selected with 2 μg/ml puromycin . 5 days after infection , 2 × 106 cells were injected subcutaneously into the flank of castrated CB17 SCID mice ( five mice , 10 tumors per group ) , and enzalutamide treatment was initiated on the day of injection . The same cell populations used for injection were also used to test eGFP AR reporter activity using flow cytometry , and qRT-PCR to test knockdown level of GREB1 . Measurements were obtained weekly using Peira TM900 system ( Peira bvba , Belgium ) . All animal experiments were performed in compliance with the approved institutional animal care and use committee ( IACUC ) protocols ( #06-07-012 ) of the Research Animal Resource Center of Memorial Sloan Kettering Cancer Center . Protein was extracted from cells using Triton lysis buffer and quantified by BCA Protein Assay ( ThermoFisher Scientific , Waltham , MA , 23225 ) . Nuclear/cytoplasmic fractionation was achieved with Subcellular Protein Fractionation Kit ( ThermoFisher Scientific , 78840 ) . Protein lysates were subjected to SDS-PAGE and immunoblotted with the following antibodies against: AR ( Abcam , Cambridge , United Kingdom , ab108341 ) , KLK3 ( Cell Signaling Technology , Danvers , MA , 5365 ) , FKBP5 ( Cell Signaling , 8245 ) TRPM8 ( Epitomics , Burlingame , CA , 3466–1 ) , tubulin ( Santa Cruz Biotechnology , Dallas , TX , sc-9104 ) , Cyclophilin B ( Abcam , ab178397 ) , BRD4 ( Cell Signaling , 13440 ) , TOP2B ( Abcam , ab58442 ) , HA ( Cell Signaling , 3724 ) . For AR immunoprecipitation , at least 1 . 5 mg of total protein was incubated with AR antibody ( Abcam , ab108341 ) overnight at 4°C followed by the addition of Protein A/G agarose beads ( Santa Cruz , sc-2003 ) for 2 hr . Immune complexes were extensively washed with Triton buffer and solubilized using Laemmli sample buffer ( BioRad , Hercules , CA ) . For immunofluorescence staining , cells were fixed with 4% formaldehyde , permeabilized with 0 . 2% Triton-X , blocked with 5% normal goat and 5% normal horse serum , stained with anti-AR ( Santa Cruz , sc-816 ) primary and Alexa Fluor 647 ( Invitrogen ) secondary antibodies , and mounted with DAPI mounting solution ( Vector Lab , Burlingame , CA ) . For Immunohistochemistry , tumor sections were stained with anti-AR ( Agilent , Santa Clara , CA , 441 ) and KLK3 ( Biogenex , Fremont , CA ) antibodies using Leica Bond RX ( Leica Biosystems , Wetzlar , Germany ) . Total RNA was isolated using the QiaShredder kit ( Qiagen , Germantown , MD ) for cell lysis and the RNeasy kit ( Qiagen ) for RNA purification . For quantitative PCR with reverse transcription ( RT–qPCR ) , we used the High-Capacity cDNA Reverse Transcription Kit ( Applied Biosystems , Grand Island , NY ) to synthesize cDNA according to the manufacturer's protocol . Real-time PCR was performed using gene-specific primers and 2X SYBR green quantfast PCR Mix ( Qiagen , 1044154 ) . Data were analyzed by the DDCT method using GAPDH as a control gene and normalized to control samples , which were arbitrarily set to 1 . To test DHT-induced AR target gene upregulation , cells were hormone-deprived in 10% charcoal-stripped dextran-treated fetal bovine serum ( Omega Scientific ) media for 2 days and then treated with indicated concentration of DHT for 24 hr . Triplicate measurements were made on at least three biological replicates . The primer sequences used for q-PCR are listed at Supplementary file 1 . For RNA-seq , library preparation , sequencing and expression analysis were performed by the New York Genome Center . Libraries were prepared using TruSeq Stranded mRNA Library Preparation Kit in accordance with the manufacturer’s instructions and sequenced on an Illumina HiSeq2500 sequencer ( rapid run v2 chemistry ) with 50 base pair ( bp ) reads . Partek Genomics Suite software ( Partek Inc , St . Louis , MO ) was used to analyze differentially expressed genes between ARsig-lo vs . ARsig-hi ( Fold change ≥1 . 5 , p<0 . 05 ) . To analyze RNA-seq data from ARsig-hi cells with shRenilla vs . shGREB1 , reads were aligned to the NCBI GRCh37 human reference using STAR aligner ( Dobin et al . , 2013 ) . Quantification of genes annotated in Gencode vM2 were performed using featureCounts and quantification of transcripts using Kalisto ( Bray et al . , 2016 ) . QC were collected with Picard and RSeQC ( Wang et al . , 2012 ) ( http://broadinstitute . github . io/picard/ ) . Normalization of feature counts was done using the DESeq2 package ( http://www-huber . embl . de/users/anders/DESeq/ ) . Differentially expressed genes were defined as a 1 . 5 fold difference , p<0 . 05 of DESeq-normalized expression . For GSEA , statistical analysis was performed with publicly available software from the Broad Institute ( http://www . broadinstitute . org/gsea/index . jsp ) . The basal and luminal gene signatures used for GSEA ( Supplementary file 2 ) were generated by conducting RNA-sequencing with normal human basal vs . luminal prostate cells isolated as previously described ( Karthaus et al . , 2014 ) . Full description of this study will be reported separately . ChIP experiments were performed as previously described ( Arora et al . , 2013 ) , using SDS-based buffers . Antibodies were used at a concentration of 5 ug per 1 mL of IP buffer , which encompassed approximately 8 million cells per IP . Antibodies used were: AR ( Santa Cruz , sc-816 ) , EP300 ( Santa Cruz , sc-585 ) , HA ( Abcam , ab9110 ) , ER ( Santa Cruz , sc-8002 ) . The primer sequences used for ChIP-qPCR are listed at Supplementary file 1 . For ChIP–seq , library preparation and RNA-seq were performed by the NYU Genome Technology Center . Libraries were made using the KAPA Biosystems Hyper Library Prep Kit ( Kapa Biosystems , Woburn , MA , KK8504 ) , using 10 ng of DNA as input and 10 PCR cycles for library amplification . The libraries were sequenced on a HiSeq 2500 , as rapid run v2 chemistry , paired-end mode of 51 bp read length . The ChIP-seq reads were aligned to the human genome ( hg19 , build 37 ) using the program BWA ( VN: 0 . 7 . 12; default parameters ) within the PEMapper . Duplicated reads were marked by the software Picard ( VN: 1 . 124; http://broadinstitute . github . io/picard/index . html ) and removed . The software MACS2 ( Feng et al . , 2012 ) ( -q 0 . 1 ) was used for peak identification with data from ChIP input DNAs as controls . Peaks of sizes > 100 bp and with at least one base pair covered by >18 reads were selected as the final high confident peaks . Peaks from shGREB1/control conditions were all merged to obtain non-overlapping genomic regions , which were then used to determine conditional specific AR binding . Overlapped peaks were defined as those sharing at least one base pair . To generate graphs depicting AR ChIP–seq read density in ±2 kilobase regions of the AR peak summits , the same number of ChIP–seq reads from different conditions were loaded into the software ChAsE ( Younesy et al . , 2016 ) , and the resulting read density matrices were sorted by the read densities in the shRenilla control , before coloring . The read density was also used to select peaks with significant signal difference between shGREB1 and controls . The criteria for assigning peaks to genes have been described previously ( Rockowitz and Zheng , 2015 ) . The MEME-ChIP software ( Machanick and Bailey , 2011 ) was applied to 300 bp sequences around the peak summits for motif discovery , and the comparison of sequence motifs was also analyzed with HOMER ( http://homer . ucsd . edu/homer/ ) . All analysis of human prostate cancer data was conducted using previously published datasets of The Cancer Genome Atlas ( TCGA ) ( Cancer Genome Atlas Research Network , 2015 ) and PCF/SU2C ( Robinson et al . , 2015 ) , which can be explored in the cBioPortal for Cancer Genomics ( http://www . cbioportal . org ) . Tumor purity content was estimated computationally using the ABSOLUTE method ( Carter et al . , 2012 ) , based on mutant allele variant fractions and zygosity shifts . Stromal signature score was applied to the normalized RNA-seq expression dataset ( Yoshihara et al . , 2013 ) . For comparison of pooled data between two different groups , unpaired t tests were used to determine significance . For comparison of data among three groups , one-way ANOVA was used to determine significance . In vitro assays represent three independent experiments from biological replicates , unless otherwise indicated . In all figures , p<0 . 05 , p<0 . 01 , p<0 . 001 and **p<0 . 0001 . For GSEA , statistical analysis was performed with publicly available software from the Broad Institute ( http://www . broadinstitute . org/gsea/index . jsp ) . The sample size estimate was based on our experience with previous experiments ( Balbas et al . , 2013; Bose et al . , 2017; Chen et al . , 2013 ) . No formal randomization process was used to assign mice to a given xenograft assay , and experimenters were not blinded .	The prostate is a roughly walnut-sized gland that makes up part of the reproductive system in men . The normal development of this gland depends on a protein known as the androgen receptor . This protein also plays an important role in driving the growth of prostate cancers , and doctors routinely treat such cancers with drugs that block the androgen receptor . While these treatments often shrink the tumors at first , the prostate cancer cells commonly become resistant to the existing “antiandrogen” drugs , highlighting the need to find new drugs for this cancer . The main way that prostate cancers become resistant to antiandrogen drugs is by making more of the androgen receptor . As such , a better understanding of this protein’s activity may prove vital to discovering new treatments . Together with other proteins called co-factors , the androgen receptor binds to DNA and switches on a set of target genes that drive the growth of prostate cancers . The activity of these genes , referred to as “androgen receptor output” , varies between different patients with prostate cancer and even between different cells from a single patient’s tumor . This variation may occur even when the level of the androgen receptor is constant , which suggests that other factors affect the output of the androgen receptor . Lee et al . set out to discover if cells with different androgen receptor outputs , but constant androgen receptor levels , respond differently to antiandrogen drugs . First , human prostate cancer cells were separated according to their androgen receptor output . Lee et al . then treated all the cells with an antiandrogen drug known as enzalutamide: tumors grown from cells with a high output became resistant to the drug faster than cells with low output . Next , a large-scale experiment revealed the differences in gene activity between cells with high and low outputs . On average , the cells with a high androgen receptor output had more of an androgen receptor co-factor called GREB1 than the cells with a low output . Biochemical experiments showed that the GREB1 protein interacts with the androgen receptor and amplifies the expression of the receptor’s target genes . When the levels of the GREB1 protein were experimentally decreased in prostate cancer cells with a high androgen receptor output , the cells became less resistant to the antiandrogen drug . Future work will be needed to know if GREB1 levels are a good proxy for patients with high androgen receptor output . The current work predicts that those patients will respond less well to current antiandrogen drugs . A better understanding of how GREB1 and androgen receptor cooperate may also be useful for developing new drugs to treat prostate cancer .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cancer", "biology" ]	2019	GREB1 amplifies androgen receptor output in human prostate cancer and contributes to antiandrogen resistance
Pediatric neural tumors are often initiated during early development and can undergo very rapid transformation . However , the molecular basis of this early malignant susceptibility remains unknown . During Drosophila development , neural stem cells ( NSCs ) divide asymmetrically and generate intermediate progenitors that rapidly differentiate in neurons . Upon gene inactivation , these progeny can dedifferentiate and generate malignant tumors . Here , we find that intermediate progenitors are prone to malignancy only when born during an early window of development while expressing the transcription factor Chinmo , and the mRNA-binding proteins Imp/IGF2BP and Lin-28 . These genes compose an oncogenic module that is coopted upon dedifferentiation of early-born intermediate progenitors to drive unlimited tumor growth . In late larvae , temporal transcription factor progression in NSCs silences the module , thereby limiting mitotic potential and terminating the window of malignant susceptibility . Thus , this study identifies the gene regulatory network that confers malignant potential to neural tumors with early developmental origins . Many pediatric tumors are thought to initiate during prenatal stages and are able to rapidly progress towards malignancy , sometimes within a few months ( Marshall et al . , 2014 ) . Yet , they contain very few genetic alterations ( Huether et al . , 2014; Parsons et al . , 2011; Pugh et al . , 2013; Vogelstein et al . , 2013; Wu et al . , 2014 ) suggesting that transformation in infancy is not driven by the gradual accumulation of genetic lesions over many years , as for most adult cancers . Instead , cells born during early development appear predisposed to malignant transformation . However , the developmental programs and gene networks that govern this early malignant susceptibility remain to be deciphered ( Chen et al . , 2015b ) . Drosophila is a well-established animal model to investigate basic principles of tumorigenesis in the developing or ageing organism ( Gonzalez , 2013; Siudeja et al . , 2015 ) . In particular , it has been used to demonstrate that single gene inactivation perturbing the asymmetric divisions of neural stem cells ( NSCs ) , called neuroblasts ( NBs ) in Drosophila , during development can rapidly cause NB amplification and aggressive malignant tumors in transplantation assays ( Caussinus and Gonzalez , 2005; Knoblich , 2010 ) . However , the underlying mechanisms of transformation are still unknown ( Caussinus and Gonzalez , 2005; Knoblich , 2010 ) . Normal NBs are active from embryogenesis to pupal stages and generate the neurons and glial cells that constitute the Drosophila central nervous system ( CNS ) . Two main types of NBs have been identified . Upon asymmetric division , most NBs ( type-I ) self-renew while giving rise to an intermediate progenitor , called the ganglion mother cell ( GMC ) , which usually divides once to generate two post-mitotic neurons or glia . In contrast , a small number of NBs ( type-II ) located in the central brain region of the CNS , generates intermediate neural progenitors ( INPs ) that can produce a few GMCs allowing for an amplification of post-mitotic progeny in the lineage ( Homem and Knoblich , 2012 ) ( Figure 1—figure supplement 1A ) . NBs undergo a limited number of divisions during development and invariably stop dividing before adulthood ( Truman and Bate , 1988 ) . For NBs located in the ventral nerve cord ( VNC ) of the CNS , this limited mitotic potential is governed by a NB-intrinsic clock that schedules their terminal differentiation during metamorphosis ( Maurange et al . , 2008 ) . This timing mechanism is set in NBs by the sequential expression of a series of 'temporal' transcription factors that has the ability to endow each progeny with a different neuronal identity according to their birth order ( Kohwi and Doe , 2013; Maurange , 2012 ) . In addition , NBs in the VNC need to progress up to a late temporal factor in the series to become competent to respond to the hormonal cues promoting cell cycle exit and terminal differentiation during metamorphosis ( Homem et al . , 2014; Maurange et al . , 2008 ) . In VNC NBs , there are four known temporal transcription factors ( Hunchback ( Hb ) -> Kruppel ( Kr ) -> Pdm -> Castor ( Cas ) ) mainly expressed during embryogenesis ( Baumgardt et al . , 2009; Grosskortenhaus et al . , 2005; Isshiki et al . , 2001; Kambadur et al . , 1998 ) . Cas is re-expressed in early larval NBs presumably followed by other , yet unknown , temporal factors required to set up a late global transition of neuronal identity during larval development and to schedule NB termination during metamorphosis ( Maurange et al . , 2008 ) . Progression throughout the sequence is governed by cross-regulatory transcriptional interactions between the temporal transcription factors , and can be blocked by continuous mis-expression of a temporal factor or by its inactivation ( Figure 1—figure supplement 1B ) ( Isshiki et al . , 2001 ) . Transitions between temporal transcription factors can also be promoted by Seven-up ( Svp ) , an orphan nuclear receptor orthologous to mammalian COUP-TF transcription factors . In particular , Svp is transiently expressed in embryonic NBs , to promote the early Hb->Kr transition , and in larval NBs to trigger a global temporal transition allowing NBs to switch from generating an early subpopulation of neurons expressing the BTB transcription factor Chinmo to a later sub-population expressing other markers ( Benito-Sipos et al . , 2011; Kanai et al . , 2005; Maurange et al . , 2008; Mettler et al . , 2006 ) . Inactivation of Svp during early larval stages blocks NBs in an early temporal identity . Consequently , late svp-/- NBs continuously generate Chinmo+ neurons , fail to undergo terminal differentiation during metamorphosis , and continue to divide in adults ( Maurange et al . , 2008 ) . Multiple series of temporal transcription factors have been uncovered in the different regions of the CNS , and recent data suggests that this temporal patterning system is evolutionary conserved and operating in mammalian NSCs ( Brand and Livesey , 2011; Konstantinides et al . , 2015; Li et al . , 2013; Mattar et al . , 2015 ) . Remarkably , inactivation of genes involved in the differentiation of INPs or GMCs can cause their reversion to a NB-like progenitor that , unlike normal NBs , possesses an unrestrained mitotic potential causing malignant tumors . This highly penetrant phenotype has , for example , been observed in the case of mutations inactivating the transcription factor Prospero ( Pros ) in GMCs ( Betschinger et al . , 2006; Choksi et al . , 2006 ) , or inactivating the NHL translational repressor Brat , the transcription factor Earmuff/dFezf , or components of the SWI/SNF complex in INPs ( Figure 1—figure supplement 1A ) ( Bello et al . , 2006; Betschinger et al . , 2006; Eroglu et al . , 2014; Koe et al . , 2014; Lee et al . , 2006; Weng et al . , 2010 ) . More recently , it has been described that inactivation of the transcription factors Nerfin1/INSM1 or Lola in post-mitotic neurons is sufficient to induce their progressive dedifferentiation into GMC- and NB-like states , and to cause unlimited proliferation ( Froldi et al . , 2015; Southall et al . , 2014 ) ( Figure 1—figure supplement 1A ) . While the mechanisms by which these factors induce or maintain differentiation has been thoroughly investigated and explain the observed amplification of NB-like cells upon loss-of-function , the reasons why dedifferentiated NBs ( dNBs ) acquire an unlimited proliferation potential remain unknown . Here , we test the hypothesis that the unlimited mitotic potential underlying the malignant properties of dNBs is caused by the deregulation of the temporal specification system . We find that neural tumors only undergo malignant transformation if dedifferentiation is induced during an early developmental window . This early malignant susceptibility of neural cells is governed by the temporal patterning system that regulates whether or not a pre-existing early oncogenic module , that controls NB mitotic activity during development , can be co-opted to trigger malignant growth . This work therefore uncovers how the temporal transcription factor series regulates NSC mitotic potential during development and governs the malignant susceptibility of neural cells according to their birth-order . Our study provides a model that may help understand the ontogeny of human tumors with early developmental origins . In order to precisely track the growth of single tumors throughout developmental and adult stages , we used the poxn-GAL4 driver to express GFP ( UAS-GFP ) and an RNAi construct against pros ( UAS-prosRNAi ) in a targeted subset of six VNC NBs from the beginning of larval development ( Figure 1A , B and Figure 1—figure supplement 2 ) . Consequently , many dNBs ( Mira+ ) are generated at the expense of neurons ( Elav+ ) , forming six tumor-like structures of proliferating progenitors in late larval VNCs ( Figure 1B and Figure 1—figure supplement 2 ) . While wild-type ( wt ) NBs undergo terminal differentiation during metamorphosis and are absent in adults ( Figure 1A ) , tumors of dNBs persist in adults and continue growing , forming in 6 day-old adults , large tumors that have fused and invaded the whole VNC ( Figure 1B–D ) . At this stage , tracking of poxn>prosRNAi tumors generated in the VNCs shows that they invade adjacent tissues such as the central brain or halters ( Figure 1C ) . Therefore , NB tumors induced by the loss of Pros during early larval stages and maintained in their natural environment resist differentiation cues operating during metamorphosis , and invariably acquire an unlimited growth potential as well as invasive properties . As such , we define them as malignant tumors . 10 . 7554/eLife . 13463 . 003Figure 1 . A subset of dNBs induced by Pros knock-down propagates malignant tumors in adults . The scale bar in all images represents 30 μm . NBs and dNBs are always labeled using an anti-Mira antibody . Neurons are labelled using anti-Elav . ( A ) Schematic drawing representing a ventral view of the late larval ( L3 ) and adult Drosophila CNS . Ventral nerve cord ( VNC ) . NBs are represented as red circles . The poxn-GAL4 driver is active in six lateral NBs of the larval VNC ( marked in green on the scheme ) . In poxn-GAL4 , UAS-GFP larvae ( poxn>GFP ) , GFP labels the six NBs ( white arrows ) and their recently generated progeny due to transient GAL4 and GFP perdurance . All NBs are absent in the adult VNC . ( B ) In poxn-GAL4 , UAS-prosRNAi , UAS-GFP , UAS-dcr2 larvae ( poxn>prosRNAi , GFP ) , six tumors of dNBs are generated . dNBs are represented on the scheme as green circles filled in red . A subset of dNBs persist and form small tumors in 1 day-old adult VNCs . ( C ) In 6 day-old adults , poxn>prosRNAi , GFP tumors cover the whole VNC and invade adjacent tissues such as the brain , and halteres ( D ) Mean tumor volumes quantified in wt poxn> GFP adult VNCs and in poxn>prosRNAi , GFP 1 and 6 day-old adult VNCs . No tumor is observed in wt adults . 1 day-old poxn>prosRNAi , GFP VNCs ( n= 5 VNCs , m = 1 . 4x105 , SEM = 6 . 3x104 ) and 6 day-old poxn>prosRNAi , GFP VNCs ( n = 7 VNCs , m = 1 . 5x106 , SEM = 1 . 8x105 ) . p-value is 2 . 5x10-3 . ( E ) poxn>prosRNAi , GFP tumors are almost exclusively composed of dNBs in late L3 , and devoid of neurons ( Elav ) . At around 20 hr after pupa formation , a brief pulse of neuronal differentiation in poxn>prosRNAi , GFP tumors is seen . GFP briefly labels recently differentiated Elav+ neurons due to transient GAL4 perdurance . In adults , persisting dNBs reconstitute malignant tumors . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 00310 . 7554/eLife . 13463 . 004Figure 1—figure supplement 1 . Progeny-to-NSC dedifferentiation and temporal progression in the developingcentral nervous system of Drosophila . ( A ) Type-I and Type-II NBs in the larval central nervous system . Type-I NBs ( large red circle ) divide asymmetrically to self-renew and generate GMCs ( medium red circle with nuclear Pros ) , which divide once to generate two neurons ( small gray circles with nuclear Nerfin ) or glia . Type-II NBs divide asymmetrically to self-renew and generate immature Intermediate Neural Precursors ( imINPs , expressing Brat ) . imINPs maturate into INPs and undergo several rounds of asymmetric divisions to self-renew and generate GMCs ( expressing nuclear Pros+ ) . Each GMC divides once , giving birth to two differentiating neurons or glial cells . ( B ) During embryonic and larval development , each VNC NB sequentially expresses a series of transcription factors ( Hunchback ( Hb ) , Kruppel ( Kr ) , Pdm1/2 ( Pdm ) , Castor ( Cas ) and Seven-up ( Svp ) ) inherited by the GMC . The various GMCs subsequently generate different types of neurons ( differently colored circles ) , which identity is determined by their birth-order . NBs that are mutant for temporal factors , such as cas or svp , fail to progress further in the series and continuously generate the same type of neurons . Such NBs fail to exit the cell cycle during pupal stages and continue dividing during adulthood . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 00410 . 7554/eLife . 13463 . 005Figure 1—figure supplement 2 . poxn>prosRNAi larvae possess tumors in the VNC but not in the brain . poxn-Gal4 expression ( GFP staining ) in the larval CNS is limited to six lateral NBs and their closed progeny in the VNC ( 1 ) . Thus , removing pros by RNAi using poxn-Gal4 leads to the formation of six tumors in the VNC . poxn-GAL4 is not active in brain NBs , although being expressed in a small subset of neurons ( 2 and 3 ) . Consequently , no tumors are generated in the brain . NBs are labeled with Mira in red , and the neurons are labeled with Elav in blue . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 00510 . 7554/eLife . 13463 . 006Figure 1—figure supplement 3 . L1/L2-induced MARCM pros-/- clones generate malignant tumors in adult . ( A ) wt late L3 NBs express Mira and generate neurons labeled with anti-Elav . NBs are absent in adult VNCs . GFP+ pros-/- MARCM clones induced during early larval ( L1/L2 ) development generate dNBs that form tumors that persist in adults and cover the VNC . ( B ) In late L3 , GFP+ pros-/- MARCM clones induced during L1/L2 are almost exclusively composed of dNBs ( Mira+ ) and are devoid of neurons ( Elav- ) . At around 20 hr after pupa formation , a large number of dNBs differentiates into Elav+ neurons . In adults , persisting dNBs reconstitute malignant tumors . Yellow asterisk indicates axonal bundles induced by earlier neuronal differentiation of dNBs during metamorphosis . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 006 Tracking poxn>prosRNAi tumors throughout development revealed that a large number of dNBs undergo neuronal differentiation at around 20 hr after pupa formation ( APF ) . At this time , a large population of GFP+ neurons ( up to 79% of the tumor cell population ( ± 3%; n = 3416; 3 tumors ) ) could be transiently observed due to the transitory persistence of GFP from dNBs ( Figure 1E ) . Similar figures of neural differentiation were not observed in tumors when examined during larval or adult stages . To ensure that this burst of differentiation was not inherent to the use of RNAi , nor specific to the poxn lineages , we used the MARCM technology ( Lee and Luo , 1999 ) to generate random pros-/- clones during early larval development ( L1/L2 ) . While such clones lead to malignant tumor growth in adults ( Figure 1—figure supplement 3A ) , a similar burst of differentiation was observed in most lineages at around 20 hr of metamorphosis ( Figure 1—figure supplement 3B ) . Interestingly , this event coincides with the timing of wt NB terminal differentiation that occurs in response to the production of the steroid hormone ( Homem et al . , 2014; Maurange et al . , 2008 ) . Thus , most dNBs retain the competency to undergo differentiation like wt NBs during metamorphosis . In contrast , a subset of differentiation-resisting dNBs persists in adult to propagate malignant tumors . Unlimited tumor growth could be propagated by a resetting of the temporal series in newly-born dNBs . However , of the four known temporal factors in VNC NBs ( Hb -> Kr -> Pdm -> Cas ) and Svp , only Cas was occasionally detected in tumors observed in late larvae and in adults ( Figure 2—figure supplement 1 ) . Nevertheless , removing ectopic Cas from prosRNAi tumors did not affect the ability to generate large tumors in adults ( Figure 2—figure supplement 2 ) . Therefore , the persistence of proliferating tumors in adults is neither caused by a reset nor by a stalling at an early stage of the temporal series in dNBs . We then concentrated on Chinmo , a transcription factor known to label a sub-population of early-born neurons in type-I lineages in larvae ( Maurange et al . , 2008; Zhu et al . , 2006 ) , ( Figure 2A ) . We find that Chinmo is not only expressed in early-born neurons of type-I and type-II lineages ( Figure 2—figure supplement 3 ) but also in early NBs from which they are generated . In the VNC , Chinmo is highly expressed in early larval NBs ( L1 and L2 ) and their progeny , but its expression in NBs progressively decreases from early L3 and is switched off in most NBs and their subsequent progeny by midL3 ( Figure 2A and Figure 2—figure supplement 4 ) . Moreover , when the late L2 pulse of Svp in NBs was abrogated , by inducing svp-/- MARCM clones in L1 , in order to block temporal patterning progression , we observed that NBs failed to silence Chinmo in L3 and adults ( Figure 2B ) . Therefore , Chinmo is a marker of early temporal identity and its window of expression in NBs is terminated in early L3 by progression of the temporal transcription factor series ( Figure 2H ) . 10 . 7554/eLife . 13463 . 007Figure 2 . Chinmo is ectopically expressed in tumors induced by dedifferentiation . All clones are induced in L1 ( 24 hr after larval hatching ) using MARCM and labeled with GFP . ( A ) NBs in wt clones ( arrows ) have silenced Chinmo at late L3 stages . Note that at this stage , Chinmo remains strongly expressed in early born neurons ( empty arrowheads ) . In adults , NBs are absents in wt clones . Chinmo is not expressed anymore in early-born neurons in adults . ( B ) NBs ( arrows ) in svp-/- clones maintain Chinmo in late L3 . Chinmo is also maintained in svp-/- NBs persisting in adults and their newly generated neurons ( arrow ) . ( C ) poxn NBs in late L3 have silenced Chinmo ( arrows ) while Chinmo expression is observed in early-born neurons ( empty arrowheads ) . ( D ) A subset of poxn>prosRNAi dNBs maintains Chinmo in late L3 and adults . ( E ) A subset of dNBs in VNC pros-/- MARCM clones maintains Chinmo at late L3 stages . All surrounding wt NBs have silenced Chinmo ( asterisks ) . Aberrant Chinmo expression is maintained in a subset of dNBs in adult pros-/- clones . ( F ) A subset of dNBs in nerfin-/- clones maintains Chinmo in late L3 and adult VNCs . ( G ) A subset of dNBs induced in brat-/- clones maintains Chinmo in late L3 and adult brains . ( H ) During early development ( from L1 to mid-L3 ) , Chinmo ( purple ) is expressed in NBs and early-born neurons . It is silenced in NBs during mid-larval stages by the progression of the temporal series . In svp-/- mutant NBs , Chinmo is maintained in NBs and their progeny up to adulthood . In pros-/- , nerfin-/-or brat-/- tumors , Chinmo escapes temporal regulation in a subpopulation of dNBs and remains expressed in tumors as development progresses . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 00710 . 7554/eLife . 13463 . 008Figure 2—figure supplement 1 . The temporal factors Hb , Kr and Pdm and the Svp nuclear receptor are not expressed in larval prosRNAi tumors . The embryonic temporal factors Hb , Kr and Pdm and the Svp receptor are not expressed in poxn>GFP wt NBs neither in poxn>prosRNAi tumors during late larval stages . wt NBs never express Cas during late larval stages , whereas dNBs occasionally express Cas . poxn-Gal4 is marked with GFP , NBs and dNBs are stained with Mira and delimited with dashed lines , and neurons are stained with Elav . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 00810 . 7554/eLife . 13463 . 009Figure 2—figure supplement 2 . Ectopic expression of the temporal factor Cas in early-induced pros-/- tumors does not contribute to persistence in adult . pros-/- , cas-/- MARCM clones were induced in late L2 . At this stage , the endogenous pulse of Cas occurring in early L2 NBs has passed . Therefore , loss of Cas does not block temporal progression in the NB from which the tumor originate , and only Cas expression in dNBs is eliminated . pros-/- , cas-/- MARCM clones are still able to generate large adult tumors , like pros-/- NB clones . Therefore , ectopic Castor in dNBs is not responsible for pros-/- tumor persistence and growth after metamorphosis . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 00910 . 7554/eLife . 13463 . 010Figure 2—figure supplement 3 . Chinmo is a marker of early-born neurons in type-II lineages . wt MARCM clone ( GFP+ ) labeling a type-II lineage . Ase is exclusively expressed in the INP arising from Type-II NB . Images represent three sections through a type-II lineage . Deep early-born neurons express Chinmo . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 01010 . 7554/eLife . 13463 . 011Figure 2—figure supplement 4 . Chinmo is expressed in young but not old NBs . Chinmo is expressed in young post-embryonic NBs and their progeny from L1 to early L3 ( arrows ) . Passed 60 hr after larval hatching ( ALH ) Chinmo is silenced in NBs and subsequently generated neurons ( arrows ) . Note that Chinmo is kept expressed in early-born neurons ( empty arrowheads ) until early pupal stages . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 01110 . 7554/eLife . 13463 . 012Figure 2—figure supplement 5 . Percentage of Chinmo-expressing dNBs in tumors . Percentage of Chinmo+ dNBs in poxn>prosRNAi larval and 6-day old adult tumors . Late L3 ( n = 6 VNCs , 2888 dNBs , m = 15 . 46 , SEM = 2 . 13 ) , 6 –day old adults ( n=8 VNCs , 5308 dNBs , m = 18 . 47 , SEM = 1 . 28 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 012 Interestingly , while Chinmo is silenced in poxn NBs in late L3 ( Figure 2C ) , we observed that a minor subset of dNBs in poxn>prosRNAi tumors retained Chinmo expression ( Figure 2D ) . Ectopic expression of Chinmo in about 20% of dNBs was also observed in tumors that persist in adults ( Figure 2D and Figure 2—figure supplement 5 ) . Moreover , aberrant expression of Chinmo in dNBs is not specific to the poxn lineage as it can be observed in L3 and adult pros-/- MARCM clones induced in L1 from random NBs ( Figure 2E ) . Thus , in pros-/- tumors induced during early larval stages , chinmo in a subset of dNBs appears to escape regulation by the temporal series and fails to be silenced at the appropriate time ( Figure 2H ) . Similar aberrant temporal expression of Chinmo was also observed in tumors originating from the dedifferentiation of immature neurons in nerfin1-/- MARCM clones ( Figure 2F ) and in tumors originating from the dedifferentiation of INPs in brat-/- MARCM clones from type-II NB lineages ( Figure 2G ) . Thus , ectopic Chinmo expression appears to be a common feature of neural tumors induced during early larval stages by dedifferentiation ( Figure 2H ) . To test whether Chinmo’s aberrant expression in dNBs contributes to tumorigenesis , we co-expressed chinmoRNAi and prosRNAi transgenes in random NB clones induced during early larval stages , or from different GAL4 drivers active in specific NB subsets . In all cases , despite efficient chinmo silencing ( Figure 3—figure supplement 1A ) , the presence of supernumerary Mira+ cells found dividing in late L3 larvae demonstrated that Chinmo was dispensable for dedifferentiation and initial NB amplification upon loss of Pros ( Figure 3A , B and Figure 3—figure supplement 1B , C ) . However , we find that prosRNAi , chinmoRNAi tumors are smaller than prosRNAi tumors in late L3 larvae ( Figure 3A , B , Figure 3—figure supplement 1C , D ) , due to smaller and fewer dNBs that exhibited a decreased mitotic index ( Figure 3D , Figure 3—figure supplement 1E–G ) . In contrast , forced expression of Chinmo ( using UAS-chinmo ) in a prosRNAi context enhanced the size of tumors in late L3 larvae , which contained larger dNBs ( Figure 3—figure supplement 1D , E ) . These results show that Chinmo promotes cell growth and proliferation within tumors . 10 . 7554/eLife . 13463 . 013Figure 3 . Chinmo sustains tumor growth beyond developmental stages . All clones were induced in L1/L2 . ( A ) Expression of prosRNAi in Flp-out clones induces malignant tumors , covering the VNC in adults . ( B ) Expression of both prosRNAi and chinmoRNAi in Flp-out clones induces tumors that fail to grow further in the adult VNC . ( C ) Mean tumor volume in Flp-out prosRNAi and prosRNAi;chinmoRNAi clones induced during early larval stages quantified in the VNC of wandering L3 ( wL3 ) , 1 day-old and 6 day-old adults . wL3: prosRNAi ( n = 6 VNCs , m = 7 . 4x104 , SEM=2 . 1x104 ) , prosRNAi;chinmoRNAi ( n = 7 VNCs , m = 2 . 3x104 , SEM = 5 . 7x103 ) ; 1 day-old adult: prosRNAi ( n = 3 VNCs , m = 4 . 7x105 , SEM = 2 . 5x104 ) , prosRNAi;chinmoRNAi ( n = 5 VNCs , m = 9 . 5x104 , SEM = 2 . 5x104 ) ; 6 day-old adult: prosRNAi ( n = 5 VNCs , m = 1 . 5x106 , SEM = 2 . 1x105 ) , prosRNAi;chinmoRNAi ( n = 6 VNCs , m = 2 . 4x105 , SEM = 1 . 9x105 ) . p-values are respectively 2 . 2x10-2 , 3 . 6x10-2 and 8 . 7x10-3 . ( D ) Mean percentage of PH3+ dNBs in late L3 and 1 day-old adult Flp-out prosRNAi and prosRNAi;chinmoRNAi induced during early larval stages . Late L3: prosRNAi ( n = 7 VNCs , m = 11 . 64 , SEM=0 . 99 ) , prosRNAi;chinmoRNAi ( n = 6 VNCs , m = 8 . 06 , SEM = 0 . 92 ) ; 1 day-old adult: prosRNAi ( n = 4 VNCs , m = 10 . 92 , SEM = 0 . 79 ) , prosRNAi;chinmoRNAi ( n = 4 VNCs , m = 1 . 97 , SEM = 0 . 50 ) . p-values are respectively 1 . 4x10-2 and 2 . 9x10-2; p-value between prosRNAi;chinmoRNAi wL3 and 1-day old adults is 9 . 5x10-3 . ( E ) Tumorigenic growth after transplantation of VNCs is assessed by the presence of GFP in the abdomen of transplanted flies after 7 days ( p-value is 6 . 0x10-6 ) . ( F ) MARCM brat-/- , chinmo-/- clones induced during early larval stages generate tumors ( Mira ) in late L3 . However , most clones undergo complete neuronal differentiation during metamorphosis , as shown with an absence of dNBs and large ectopic axonal bundles in adult clones ( inset ) . Occasional remaining dNBs are not able to reconstitute large tumors ( inset ) . Below , brat-/- , chinmo-/- clones are represented schematically during development . ( G ) MARCM brat-/- clones induced during early larval stages rapidly leads to large malignant tumors in the adult brain . ( H ) Mean tumor volumes in MARCM brat-/- and brat-/- , chinmo-/- clones induced during early larval stages quantified in 6 day-old adult central brains . MARCM brat-/- clones ( n = 4 brains , m = 3 . 0x106 , SEM = 1 . 0x106 ) and MARCM brat-/- , chinmo-/- clones ( n = 13 brains , m = 7 . 1x104 , SEM = 3 . 3x104 ) . p-value is 8 . 4x10-4 . ( I ) Overexpression of chinmo in Flp-out clones induces NB amplification in larvae ( yellow dotted line ) , giving rise to tumors composed of proliferating dNBs in adults . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 01310 . 7554/eLife . 13463 . 014Figure 3—figure supplement 1 . Chinmo knock-down leads to reduced tumor growth and increased differentiation . ( A ) chinmoRNAi expression in prosRNAi tumors using nab-GAL4 efficiently knocks down Chinmo expression ( remaining Chinmo+ cells are neurons in which chinmoRNAi is not expressed ) . ( B ) When prosRNAi is expressed in all larval NBs using nab-GAL4 , VNCs becomes very large in late L3 and are mainly composed of dNBs at the expense of neurons . In pharate adults , some dNBs have differentiated in neurons , but many dNBs continue proliferating giving rise to large and deformed VNCs . nab>prosRNAi , chinmoRNAi also induces dNBs but VNCs are smaller . In pharate adults , many dNBs have differentiated into neurons , leading to tumor regression in VNCs . Pharate adult pictures are projections of several confocal sections . ( C ) prosRNAi and prosRNAi , chinmoRNAi respectively expressed in a restricted subset of NBs using eg-GAL4 . eg>prosRNAi induces larger NB amplification than eg>prosRNAi , chinmoRNAi ( dashed line ) in late L3 . ( D ) Mean larval tumor volume in nab>prosRNAi ( n = 16 brains , m = 42753198 , SEM = 3323616 ) , nab>prosRNAi;chinmoRNAi ( n = 12 brains , m = 17419207 , SEM = 1127497 ) and nab>prosRNAi;chinmo ( n = 21 brains , m = 60341998 , SEM = 2099849 ) , p-values are respectively 2 . 6x10-7 and 7 . 9x10-5 . ( E ) dNB cell diameter is decreased when chinmo expression is knocked down ( prosRNAi; chinmoRNAi ) , and is increased when Chinmo is overexpressed ( prosRNAi; chinmo ) . prosRNAi ( n = 4 VNCs , 424 dNBs , m = 5 . 00 , SEM = 0 . 06 ) , prosRNAi; chinmoRNAi ( n = 4 VNCs , 550 dNBs , m = 4 . 08 , SEM = 0 . 05 ) , prosRNAi; chinmo ( n = 4 VNCs , 319 dNBs , m = 7 . 22 , SEM = 0 . 11 ) . p-values are respectively 1 . 8x10-29 and 3 . 3x10-54 . ( F ) The number of dNBs per larval tumor is reduced by Chinmo knockdown in eg>prosRNAi . prosRNAi ( n = 14 , m = 298 . 5 , SEM = 31 . 3 ) , prosRNAi; chinmoRNAi1 ( n = 12 , m = 80 . 7 , SEM = 6 . 5 ) , prosRNAi; chinmoRNAi2 ( n = 26 , m = 61 . 9 , SEM = 3 . 3 ) , p-values are respectively 1 . 7x10-5 and 2 . 6x10-7 . ( G ) Mean percentage of PH3+ dNBs in late L3 nab>prosRNAi ( n = 6 VNCs , m = 7 . 75 , SEM = 0 . 53 ) and nab>prosRNAi;chinmoRNAi ( n = 6 VNCs , m = 4 . 77 , SEM = 0 . 52 ) , p-value is 2 . 2x10-3 . ( H ) In nab>prosRNAi , chinmoRNAi pharates the proportion of dNBs ( Volume of Mira labeling/ Total volume of the VNC ) is decreased compared to nab>prosRNAi pharates , while the proportion of neurons ( Volume of Elav labeling/ Total volume of the VNC ) is increased . Vmira/Vtotal for prosRNAi ( n = 5 VNCs , m = 0 . 764 , SEM = 0 . 030 ) , Vmira/Vtotal for prosRNAi; chinmoRNAi ( n = 4 VNCs , m = 0 . 310 , SEM = 0 . 055 ) , p-value=1 . 6x10-2 . Velav/Vtotal for prosRNAi ( n = 5 VNCs , m = 0 . 236 , SEM = 0 . 030 ) , Velav/Vtotal for prosRNAi; chinmoRNAi ( n = 4 VNCs , m = 0 . 690 , SEM = 0 . 055 ) , p-value=1 . 6x10-2 . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 01410 . 7554/eLife . 13463 . 015Figure 3—figure supplement 2 . prosRNAi chinmoRNAi tumors fail to become malignant in adults . ( A ) Adults with prosRNAi Flp-out ( FO ) clones induced in L1/L2 more frequently contain tumors than adults with prosRNAi; chinmoRNAi clones . In the brain of 1 day-old adults , 79% of animals with prosRNAi clones ( n = 24 VNCs and brains ) and 53% of animals with prosRNAi; chinmoRNAi clones ( n = 17 VNCs and brains ) contain tumors , p-value is 0 . 048 . In the brain of 6 day-old adults , 78% of animals with prosRNAi clones and 33% of animals with prosRNAi; chinmoRNAi clones contain tumors ( n = 18 ) . In the VNC of 6 day-old adults , 80% of animals with prosRNAi clones ( n=18 VNCs and brains ) and 25% of animals with prosRNAi; chinmoRNAi clones ( n = 18 VNCs and brains ) contain tumors . p-value is 0 . 018 . Results are provided by at least 2 independent experiments . ( B ) Mean tumor volume in Flp-out prosRNAi and prosRNAi;chinmoRNAi induced in L1/L2 and quantified in the brain of late larvae , 1 day-old and 6 day-old adults . Late larval brain prosRNAi ( n = 7 brains , m = 3 . 3x104 , SEM = 4 . 2x103 ) , prosRNAi;chinmoRNAi ( n = 5 brains , m = 1 . 0x104 , SEM = 1 . 2x103 ) ; 1 day-old adult brain prosRNAi ( n = 8 brains , m = 8 . 5x105 , SEM = 3 . 0x105 ) , prosRNAi;chinmoRNAi ( n = 5 brains , m = 6 . 5x104 , SEM = 2 . 4x104 ) ; 6 day-old adult brain prosRNAi ( n = 8 brains , m = 1 . 7x106 , SEM = 3 . 0x105 ) , prosRNAi;chinmoRNAi ( n = 8 brains , m = 6 . 3x104 , SEM = 1 . 4x104 ) , p-values are respectively 2 . 5x10-3 , 1 . 6x10-3 and 1 . 6x10-4 . ( C ) In 6 day-old adult Flp-out prosRNAi;chinmoRNAi clones , remaining dNBs sometimes appear quiescent ( characterized by cytoplasmic extensions ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 01510 . 7554/eLife . 13463 . 016Figure 3—figure supplement 3 . Chinmo+ dNBs exhibit a higher mitotic index than Chinmo- dNBs within prosRNAi tumors . ( A ) Dividing Chinmo+ and Chinmo- ( red ) dNBs marked with anti-PH3 ( blue ) in late L3 and adults poxn>prosRNAi tumors . ( B ) Quantification of the mitotic index of wt NBs and poxn>prosRNAi dNBs in late L3 and adults . Chinmo+ dNBs possess a higher mitotic index than Chinmo- dNBs both in larval and adult tumors . Late L3 wt ( n = 3 VNCs , 464 NBs , m = 21 . 11 , SEM = 0 . 15 ) , late L3 poxn>prosRNAi Chinmo+ cells ( n = 11 VNCs , 812 dNBs , m = 11 . 47 , SEM = 0 . 09 ) , late L3 poxn>prosRNAi Chinmo- cells ( n = 11 VNCs , 4997 dNBs , m = 7 . 63 , SEM = 0 . 01 ) , adult poxn>prosRNAi Chinmo+ cells ( n = 12 VNCs , 303 dNBs , m = 12 . 48 , SEM = 0 . 14 ) , adult poxn>prosRNAi Chinmo- cells ( n = 12 VNCs , 1673 dNBs , m = 6 . 42 , SEM = 0 . 05 ) , p-values are respectively 1 . 1x10-5 and 8 . 0x10-5 . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 016 Importantly , a higher proportion of dNBs is converted to neurons in prosRNAi , chinmoRNAi tumors compared to prosRNAi tumors during metamorphosis ( Figure 3—figure supplement 1B , H ) . Consistently , adult persistence of tumors originating from prosRNAi , chinmoRNAi clones induced in L1/L2 is reduced by about 30% compared to prosRNAi clones ( Figure 3—figure supplement 2A ) . This indicates that Chinmo confers resistance to the differentiation cues operating during metamorphosis . Additionally , persisting prosRNAi; chinmoRNAi tumors are smaller and fail to grow further , or completely differentiate , in 6 day-old adults , while prosRNAi control clones rapidly invade the CNS ( Figure 3A–C and Figure 3—figure supplement 2B ) . Failure to grow is , at least , partly due to a low mitotic activity within prosRNAi; chinmoRNAi adult tumors , as already observed in larval tumors . Moreover , the mitotic rate in prosRNAi; chinmoRNAi tumors decreases from larval to adult stages , while remaining stable in prosRNAi tumors ( Figure 3D ) , in 6 day-old adults , prosRNAi; chinmoRNAi dNBs sometimes show cytoplasmic extensions characteristic of quiescence ( Chell and Brand , 2010; Truman and Bate , 1988 ) ( Figure 3—figure supplement 2C ) . Together , these experiments suggest that dNBs in prosRNAi; chinmoRNAi tumors progressively exhaust their proliferation potential . The decreased growth potential of prosRNAi; chinmoRNAi tumors was confirmed with the conventional transplantation assay ( Rossi and Gonzalez , 2015 ) with only 16% ( n=20 ) of VNCs developing large tumors in the abdomen of transplanted animals after 7 days , compared to 84% ( n=32 ) with prosRNAi VNCs ( Figure 3E ) . We also assessed the mitotic index of the Chinmo+ and Chinmo- dNBs present within the same prosRNAi tumor . We found that it was significantly higher for the subpopulation of Chinmo+ dNBs than for Chinmo- dNBs suggesting that two types of progenitors with different mitotic potential co-exist in tumors ( Figure 3—figure supplement 3 ) . We then tested whether Chinmo was also essential for the growth of brat-/- tumors . Like brat-/- MARCM clones ( Figure 2G ) , brat-/- chinmo-/- MARCM clones induced in L1 led to large clones with amplified NBs in late L3 ( Figure 3F ) . However , while brat-/- MARCM clones led to large tumors partly covering the adult brain in 6 day-old adults ( Figure 3G ) , brat-/- chinmo-/- MARCM clones differentiated during metamorphosis and occasional remaining dNBs were unable to reconstitute large tumors in adults ( Figure 3F , H ) . Thus , these results collectively indicate that ectopic Chinmo in a subset of dNBs boosts cell growth , counteracts neuronal differentiation during pupal stages , and is required for sustained proliferation once development is terminated . To further investigate Chinmo’s ability to promote cell growth and proliferation on its own , we over-expressed it in wt NBs clones from early larval stages . Although Chinmo does not induce proliferation when expressed in post-mitotic neurons ( Zhu et al . , 2006 ) , we found that clonal overexpression of UAS-chinmo in wt NBs and GMCs from early L1 is sufficient to induce NB amplification in 60% of NB clones ( 6 VNCs , 37 clones ) and the formation of tumors that resist differentiation during metamorphosis and continue proliferating in adults ( Figure 3I ) . Because adult flies containing UAS-chinmo clones rapidly die , we could not test the long-term growth potential of UAS-chinmo NB tumors . To circumvent this problem , we transplanted larval VNCs over-expressing Chinmo in all NBs ( nab>chinmo ) . We found that 26% of transplanted flies grow large tumors in the abdomen after 7 days ( Figure 3E ) showing that Chinmo over-expression in NBs and their GMCs is sufficient to induce sustained tumorigenic growth . Together , these data demonstrate that aberrant Chinmo expression in a subset of dNBs is oncogenic and drives sustained tumor growth beyond developmental stages . To explore Chinmo’s mode of action , we compared , by RNA-seq , the transcription profiles of late larval VNCs in which all NBs express the prosRNAi construct ( nab>prosRNAi ) with prosRNAi VNCs in which Chinmo was either knocked down or over-expressed ( respectively nab>prosRNAi , chinmoRNAi and nab>prosRNAi , chinmo ) ( Figure 4A ) . Two hundred and fourteen genes were both up-regulated in the nab>prosRNAi , chinmo condition and down-regulated in the nab>prosRNAi , chinmoRNAi condition when compared to the nab>prosRNAi control ( p-value < 0 . 05 ) . They were considered as putative targets positively regulated by Chinmo . On the other hand , 388 genes were both downregulated in the nab>prosRNAi , chinmo condition and up-regulated in the nab>prosRNAi , chinmoRNAi condition compared to the nab>prosRNAi control ( p-value<0 , 05 ) ( Figure 4—source data 1 ) . They were considered as putative targets negatively regulated by the presence of Chinmo . Gene ontology ( GO ) and KEGG pathway analyses indicate that negative targets are highly enriched in genes involved in neuronal differentiation consistent with the ability of Chinmo to prevent dNB differentiation during metamorphosis ( Figure 4B and Figure 4—figure supplement 1 ) . In contrast , positive targets are enriched in genes involved in ribosome biogenesis , RNA and DNA metabolism , DNA replication and cell-cycle progression ( Figure 4B and Figure 4—figure supplement 1 ) . These data support a role for Chinmo in promoting dNB growth and mitotic activity that is consistent with our genetic and proliferation assays ( Figure 3 ) . 10 . 7554/eLife . 13463 . 017Figure 4 . Chinmo promotes Imp and Lin-28 expression . ( A ) prosRNAi is expressed in all larval NBs using nab-GAL4 . RNA-seq indicated 214 genes to be commonly up-regulated , and 388 genes were found to be commonly downregulated , when comparing nab>prosRNAi vs . nab>prosRNAi , chinmoRNAi and nab>prosRNAi , chinmo vs . nab>prosRNAi ( adj p-value < 0 . 05 ) . ( B ) Graphical representation of the log2 fold change as a function of the base mean expression of Chinmo targets , comparing nab>prosRNAi , chinmo to nab>prosRNAi , chinmoRNAi . lin-28 , Imp and chinmo are highlighted in red . ( C ) Lin-28 , Imp ( cytoplasmic ) and Chinmo ( nuclear ) are co-expressed in the same subset of dNBs in poxn>prosRNAi larval and adult tumors ( delineated with the yellow dashed lines ) . ( D ) Clonal mis-expression of Chinmo in GFP+ Flp-out clones induced in L1 , delimited by yellow dashed lines , induces Imp and Lin-28 co-expression in dNBs . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 01710 . 7554/eLife . 13463 . 018Figure 4—source data 1 . Differentially expressed genes and enriched GO terms and KEGG pathways between prosRNAi tumors expressing various levels of Chinmo . ( A ) Differentially expressed genes between nab>prosRNAi and nab>prosRNAi , chinmoRNAi late larval VNCs ( p-value > 0 , 05 ) . ( B ) Differentially expressed genes between nab>prosRNAi , chinmo and nab>prosRNAi late larval VNCs ( p-value > 0 , 05 ) . ( C ) List of genes that are commonly up- and down-regulated when comparing nab>prosRNAi vs . nab>prosRNAi , chinmoRNAi VNCs and nab>prosRNAi , chinmo and nab>prosRNAi VNCs . ( D ) Differentially expressed genes between nab>prosRNAi , chinmo and nab>prosRNAi , chinmoRNAi late larval VNCs ( p-value > 0 , 05 ) . ( E ) Enriched KEGG pathways for commonly up-regulated genes ( see C ) . ( F ) Enriched GO for up-regaled and down-regulated genes . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 01810 . 7554/eLife . 13463 . 019Figure 4—figure supplement 1 . Transcriptional analysis summary . ( A ) GO enrichment for down-regulated genes ( negative Chinmo targets ) . ( B ) GO pathway enrichment for up-regulated genes ( positive Chinmo targets ) . ( C ) KEGG pathway enrichment for up-regulated genes ( positive Chinmo targets ) . ( D ) Top 12 up-regulated Chinmo targets when comparing poxn>prosRNAi , chinmo with poxn>prosRNAi , chinmoRNAi . See Figure 4—source data 1 for lists of differentially expressed genes . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 01910 . 7554/eLife . 13463 . 020Figure 4—figure supplement 2 . Chinmo is necessary for Imp expression in tumors . ( A , B ) Imp is expressed in prosRNAi and brat-/-tumors induced in L1/L2 . Imp is absent from prosRNAi chinmoRNAi and brat-/-chinmo-/- tumors . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 020 Among the 214 positive targets of Chinmo uncovered by RNA-seq , several genes are important regulators of malignancy in mammals such as lin-28 ( Molenaar et al . , 2012 ) , IGF-II mRNA-binding protein ( Imp ) ( Lederer et al . , 2014 ) , musashi ( msi ) ( Wang et al . , 2010 ) , Aldehyde dehydrogenase ( Aldh ) ( Ginestier et al . , 2007 ) , and snail ( sna ) ( Barrallo-Gimeno and Nieto , 2005 ) ( Figure 4—figure supplement 1 , Figure 4—source data 1 ) . Among them , the most highly up-regulated genes , when comparing the nab>prosRNAi , chinmo vs . nab>prosRNAi , chinmoRNAi conditions , are two mRNA-binding proteins Lin-28 ( 64 fold ) and Imp ( 10 fold ) ( Figure 4B , Figure 4—figure supplement 1 , Figure 4—source data 1 ) . Lin-28 and Imp are highly conserved oncofoetal genes in humans ( Bell et al . , 2013; Lederer et al . , 2014 ) and have recently been shown to be co-expressed in embryonic NSCs in mice ( Yang et al . , 2015 ) . We validated their expression in tumors by immunostaining . Interestingly , in larval and adult prosRNAi tumors , both Imp and Lin-28 are present in the cytoplasm of the subset of dNBs that co-express Chinmo ( Figure 4C ) . Furthermore , clonal overexpression of UAS-chinmo induces supernumerary NBs that express both Imp and Lin-28 , while small prosRNAi , chinmoRNAi or brat-/- chinmo-/- tumors in adults lack Imp ( not tested for Lin-28 ) ( Figure 4D and Figure 4—figure supplement 2 ) . Together , these experiments demonstrate that Imp and Lin-28 are , direct or indirect , positive targets of Chinmo in tumors . In mammals , the three orthologs of Imp ( IMP1-3 , also called IGF2BP1-3 ) are believed to be important regulators of tumorigenesis , but their function and targets are unclear ( Bell et al . , 2013 ) . We find that overexpression of Imp in NBs is not sufficient to induce their amplification and tumorigenesis ( Figure 5—figure supplement 1 ) . Moreover , efficient Imp knock-down using one or two different ImpRNAi transgenes ( ImpRNAi1 or ImpRNAi1;2 ) in poxn>prosRNAi larvae did not prevent initial NB amplification , although tumor growth appeared slowed ( Figure 5A , B and Figure 5—figure supplement 2A , B ) . Strikingly , most prosRNAi , ImpRNAi tumors failed to maintain continuous growth and remained small in 6 day-old adults ( Figure 5B–C ) . Thus Imp is required for sustained tumor growth . 10 . 7554/eLife . 13463 . 021Figure 5 . Imp sustains Chinmo expression in tumors . ( A ) Chinmo is expressed in a subset of dNBs in both larval ( dashed yellow lines ) and adult poxn>prosRNAi tumors ( see enlargement ) . ( B ) Chinmo is still expressed in a subset of dNBs in larval poxn>prosRNAi , ImpRNAi tumors but is progressively lost in adult tumors that remain small ( see enlargement ) . poxn>prosRNAi , Imp tumors tend to have an increased number of Chinmo+ cells ( not quantified ) . ( C ) Mean tumor volume of 6 day-old adults in poxn>prosRNAi , poxn>prosRNAi; ImpRNAi1;2 and poxn>prosRNAi; Imp . poxn>prosRNAi ( n = 6 VNCs , m = 1 . 1x106 , SEM = 1 . 3x105 ) , poxn>prosRNAi; ImpRNAi1;2 ( n = 8 VNCs , m = 9 . 5x104 , SEM = 2 . 4x104 ) , poxn>prosRNAi; Imp ( n = 6 VNCs , m = 1 . 0x106 , SEM = 1 . 3x105 ) . p-values are respectively 6 . 7x10-4 and 0 . 82 . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 02110 . 7554/eLife . 13463 . 022Figure 5—figure supplement 1 . Imp mis-expression is not sufficient to initiate tumors . Clonal mis-expression of Imp in larval GFP+ Flip-out clones induced in L1/L2 does not trigger supernumerary NB , neither ectopic expression of Chinmo in NBs . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 02210 . 7554/eLife . 13463 . 023Figure 5—figure supplement 2 . Imp knock-down decreases tumor growth . ( A ) ImpRNAiefficiently knocks down Imp in larval dNBs induced by poxn>prosRNAi , ImpRNAi1;2 . Note that a subset of dNBs continues expressing Chinmo . ( B ) Mean tumor volume per VNC in poxn>prosRNAi and poxn>prosRNAi; ImpRNAi1;2 in 5 day-old L3 reared on normal food . prosRNAi ( n = 6 VNCs , m = 482720 , SEM = 56988 ) , prosRNAi; ImpRNAi1;2 ( n = 5 VNCs , m = 217124 , SEM = 33046 ) , p-value=4 . 3x10-3 . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 02310 . 7554/eLife . 13463 . 024Figure 5—figure supplement 3 . Imp is necessary to sustain tumor growth and Chinmo expression . After 10 days in the sterol-free food , poxn>prosRNAi larval tumors become very large , invade the central brain and optic lobes ( yellow arrows ) and maintain Chinmo expression ( see enlargement ) . In contrast , tumor growth is strongly reduced and Chinmo is mostly absent ( see enlargement ) in poxn>prosRNAi , ImpRNAitumors . ( B ) Mean tumor volume per VNC in poxn>prosRNAi and poxn>prosRNAi; ImpRNAi1;2 in 10 day-old larvae reared on sterol-free food . prosRNAi ( n = 5 VNCs , m = 40859191 , SEM = 15192219 ) , prosRNAi; ImpRNAi 1;2 ( n = 6 VNCs , m = 6012717 , SEM = 822265 ) , p-value=4 . 3x10-3 . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 024 A recent study has shown that Imp post-transcriptionally promotes Chinmo expression in mushroom body neurons ( Liu et al . , 2015 ) . We thus sought to assess Chinmo expression in prosRNAi tumors lacking Imp . We find that Chinmo+ dNBs are still present in larvae . However , they are absent from many tumors in 6 day-old adults ( 12 out of 22 with ImpRNAi1 , and 11 out of 14 with ImpRNAi1;2 ) ( Figure 5A , B ) . Similar results are obtained with tumors grown for 10 days in larvae fed with a sterol-free diet that has been shown to prevent pupariation ( Katsuyama and Paro , 2013; Parkin and Burnet , 1986 ) . In such conditions , poxn>prosRNAi tumors grow extensively and invade the central brain and optic lobes , while poxn>prosRNAi , ImpRNAi1;2 tumors remain much smaller , stay localized to the VNC and show an almost complete loss of Chinmo+ dNBs ( Figure 5—figure supplement 3A , B ) . We also noted a tendency for an increased number of Chinmo+ dNBs in 4 day-old adults upon Imp over-expression ( Figure 5B , not quantified ) . These experiments indicate that Imp is not required to establish the initial population of Chinmo+ dNBs but is necessary for their long-term maintenance in tumors . Together with the transcriptional activation of Imp by Chinmo , these experiments reveal a positive feedback loop between Chinmo and Imp necessary for sustained tumor growth beyond developmental stages . In contrast to Imp , loss of Lin-28 does not appear to impair the self-renewal of Chinmo+ dNBs and the growth of pros-/- or brat-/- tumors induced in L1 ( tumor size quantified in adults ) ( Figure 6—figure supplement 1 ) . This suggests that Lin-28 is dispensable for Chinmo expression and for sustained tumor growth . However , while mis-expression of Drosophila lin-28 in wt larval NB clones is not sufficient to induce NB amplification and ectopic expression of Chinmo and Imp ( Figure 6—figure supplement 2 ) , we find that overexpression of Drosophila lin-28 in prosRNAi tumors from their initiation leads to a significant increase in the proportion of Chinmo+ dNBs and an overall increase in the intensity of Chinmo expression when observed in 6 day-old adults ( Figure 6A–C; Figure 6—figure supplement 3 ) . Moreover , Imp is co-expressed in all Chinmo+ dNBs . This suggests that overexpression of Lin-28 in the tumorigenic context is able to favor the self-renewing capacity of Chinmo+/Imp+ dNBs at the expense of Chinmo-/Imp- dNBs . Interestingly , overexpression of mammalian LIN28A and LIN28B is often associated with malignant tumors in human ( Carmel-Gross et al . , 2016; Molenaar et al . , 2012; Viswanathan et al . , 2009 ) . Remarkably , human LIN28A and LIN28B , mis-expressed in Drosophila poxn>prosRNAi tumors , have retained the ability to strongly enhance Chinmo expression within tumors , demonstrating evolutionary-conserved interactions between Lin-28 and Chinmo or its regulators ( Figure 6A–C ) . All together , these data identifies Chinmo and Imp as a core oncogenic module that sustains dNB proliferation and tumor growth beyond developmental stages . In addition , expression of this module is boosted by high levels of Lin-28 in tumors ( Figure 6D , Figure 6—figure supplement 4 ) . 10 . 7554/eLife . 13463 . 025Figure 6 . Chinmo , Imp and Lin-28 form an oncogenic loop . ( A ) Chinmo expression in 5 day-old adult VNCs color-coded relative to staining intensity . All transgenes are expressed with the poxn-GAL4 driver . Humanized tumors mis-express human LIN28A or LIN28B . ( B ) Ratio representing the volume of Chinmo+ cells over the total tumor volume in 5 day-old adult VNCs . poxn>prosRNAi ( n = 4 VNCs , m = 0 . 188 , SEM = 0 . 017 ) , poxn>prosRNAi , dlin-28 ( n = 6 VNCs , m = 0 . 583 , SEM = 0 . 072 ) and poxn>prosRNAi , LIN28A ( n = 8 VNCs , m = 0 . 474 , SEM = 0 . 060 ) . p-values are respectively 9 . 5x10-3 and 8 . 0x10-3 . ( C ) Mean Chinmo intensity in Chinmo+ cells . poxn>prosRNAi ( n = 7 tumors , m = 962 , SEM = 37 ) , poxn>prosRNAi , dlin-28 ( n = 8 tumors , m = 1413 , SEM = 125 ) and poxn>prosRNAi , UAS-LIN28A ( n = 9 tumors , m = 1236 , SEM = 110 ) . Each sample is the mean of 3 different focal sections of the same tumor . p-values are respectively 5 . 9x10-3 and 7 . 1x10-2 . ( D ) Representation of the observed cross-regulatory interactions composing the oncogenic module . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 02510 . 7554/eLife . 13463 . 026Figure 6—figure supplement 1 . Lin-28 is dispensable for tumor growth . ( A ) prosRNAi , Δlin28 MARCM clones induced in L1/L2 generate tumors that continue growing in adults and keep expressing Chinmo and Imp . ( B ) Mean tumor volumes in the brain of 6d adult Δlin28 homozygous flies containing brat-/- MARCM clones compared to MARCM brat-/-clones in otherwise wt background . MARCM brat-/- clones ( n = 5 brains , m = 6 . 2x106 , SEM = 9 . 4x105 ) and MARCM brat-/-clones inΔlin28 flies ( n = 5 brains , m = 6 . 1x106 , SEM = 8 . 5x105 ) . p-value is 0 , 82 . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 02610 . 7554/eLife . 13463 . 027Figure 6—figure supplement 2 . Lin-28 mis-expression is not sufficient to initiate tumors . Efficient clonal mis-expression of Lin-28 in larval GFP+ Flip-out clones does not induce NB amplification , neither ectopic expression of Chinmo and Imp in NBs . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 02710 . 7554/eLife . 13463 . 028Figure 6—figure supplement 3 . Lin-28 positively regulates chinmo and Imp in tumors . Overexpression of lin-28 in poxn>prosRNAi tumors leads in adults to an increase in the number of Chinmo+ cells , all of which also express Imp . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 02810 . 7554/eLife . 13463 . 029Figure 6—figure supplement 4 . Schematic conclusions . Scheme depicting the conclusions from the experiments in Figure 6 and Figure 6-figure supplements . Chinmo , Imp and Lin-28 are coexpressed in a subset of dNBs in both larval and adult tumors . Imp is necessary for long-term maintenance of Chinmo and persistent tumor growth . Lin-28 over-expression can increase the proportion of Chinmo+/Imp+ dNBs in the tumor . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 029 We then wondered whether Chinmo , Imp and Lin-28 ( CIL ) co-expression is specific to dNBs or can be found during development . Interestingly , in the VNC and central brain , both Imp and Lin-28 are co-expressed with Chinmo in normal NBs and their surrounding neuronal progeny from L1 to early L3 . From early L3 , they are then progressively downregulated together with Chinmo in NBs and subsequent progeny ( Figure 7A ) . Lin-28 and Chinmo remain transiently expressed in early-born neurons up to midL3 and early pupal stages respectively , while Imp expression in early-born neurons perdures in adult ( Figure 7—figure supplement 1 ) . However , Imp and lin-28 , like chinmo , fail to be silenced in late larval and adult NBs with a stalled temporal series ( svp-/- ) , as well as in their progeny ( Figure 7B ) . Thus , the temporal system silences chinmo , Imp and lin-28 in L3 NBs . We then tested the function of Chinmo in NBs during development . chinmo-/- MARCM clones were induced during embryogenesis in order to ensure the removal of Chinmo during the entire post-embryonic period ( Figure 7C ) . In chinmo-/- late L3 clones , we did not observe a difference in clone size in the VNC compared to the control wt clones ( Figure 7D ) and chinmo-/- NBs exhibit the same mitotic rate and possess the same size as wt NBs in late L3 ( Figure 7E , F ) . Thus Chinmo is dispensable for NB growth and proliferation during larval development . We then tested whether temporally blocked NBs persist in adults due to the maintenance of Chinmo , Imp and Lin-28 . Knocking-down Chinmo in temporally stalled svp-/- NBs is not sufficient to prevent adult persistence , although NB growth is impaired ( Figure 7G , H and Figure 7—figure supplement 2A ) . However , knocking down both Chinmo and Imp in svp-/- NBs restores their elimination before adulthood ( Figure 7G , H ) . This suggests that the Chinmo/Imp module is necessary to maintain the unlimited mitotic potential of NBs stalled in an early temporal identity ( Figure 7H ) . Together with our results showing that over-expression of Chinmo is sufficient to induce persistence of NBs in adults ( Figure 3H ) , these experiments also indicate that silencing of Imp and chinmo by the temporal patterning system may be necessary to limit NB mitotic activity and schedule the competency for NB terminal differentiation during metamorphosis ( Figure 7H ) . 10 . 7554/eLife . 13463 . 030Figure 7 . The temporal series silences chinmo , Imp and lin-28 in NBs for their timely termination during development . ( A ) Chinmo , Lin-28 and Imp are coexpressed in wt VNC NBs in L2 , and are silenced in NBs in late L3 . Note that in late L3 , surrounding early-born neurons keep expressing Imp and Chinmo whereas Lin-28 is down-regulated in all cells . ( B ) Imp is maintained in L1-induced MARCM svp-/- NBs ( GFP+ ) in late L3 . Lin-28 and Imp are maintained in MARCM svp-/- NBs ( GFP+ ) that persist in adult . ( C ) wt and chinmo-/- MARCM clones induced during embryogenesis . ( D ) Number of cells per clone in late L3 in VNC wt and chinmo-/- MARCM clones induced during embryogenesis . wt MARCM 40A ( n = 16 clones , m = 82 , SEM = 4 . 7 ) ; chinmo-/- MARCM clones ( n=17 clones , m = 83 , SEM = 4 . 8 ) . p-values is 0 . 88 . ( E ) Mean NB area in late L3 VNC wt and chinmo-/- MARCM clones induced during embryogenesis . wt MARCM 40A ( n = 45 NBs , m = 76 . 7 , SEM = 3 . 2 ) , chinmo-/- MARCM clones ( n = 46 NBs , m = 76 . 3 , SEM = 2 . 9 ) . p-value is 0 . 88 . ( F ) Mean percentage of PH3+ NBs in late L3 VNC MARCM wt and chinmo-/- clones induced during embryogenesis . wt clones ( n = 4 VNCs , m = 20 . 6 , SEM = 1 . 43 ) , chinmo-/- clones ( n = 4 VNCs , m = 20 . 8 , SEM = 2 . 23 ) , p-value is 0 , 88 . ( G ) NBs persist in adult MARCM svp-/- , chinmoRNAi clones induced in L1 . NBs are smaller ( Figure 7—figure supplement 1A ) and maintain Imp expression . Removing both Chinmo and Imp ( chinmoRNAi , ImpRNAi ) in a svp-/- MARCM clone is sufficient to restore NB elimination before the end of development . ( H ) Schematic recapitulation of the above experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 03010 . 7554/eLife . 13463 . 031Figure 7—figure supplement 1 . Lin-28 and Chinmo are respectively silenced in the CNS prior to and during metamorphosis , whereas Imp remains expressed in a subset of adult neurons . ( A ) Imp is expressed in a subset of wt late L3 Elav+ neurons , but is absent from wt L3 NBs . ( B ) Imp and Chinmo are expressed in early-born neurons in late L3 , but not in NBs . In contrast , Lin-28 is silenced in all neurons from late L3 ( note that the Lin-28::Venus construction is expressed at very low levels in a subset of late L3 neurons , see Figure 4C ) . ( C ) Chinmo and Lin-28 are silenced in the the adult CNS while Imp remains expressed in a subset of neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 03110 . 7554/eLife . 13463 . 032Figure 7—figure supplement 2 . Chinmo promotes the long-term growth of NBs stalled in an early temporal identity but is not required for Imp expression . ( A ) Mean NB area in MARCM svp-/- and svp-/- , chinmoRNAi adults . svp-/- ( n = 15 NBs , m = 58 . 8 , SEM = 3 . 18 ) , svp-/- , chinmoRNAi ( n = 11 NBs , m = 29 . 9 , SEM = 2 . 51 ) . p-value is 5 . 2x10-7 . ( B ) chinmoRNAi efficiently knocks down Chinmo in temporally blocked svp-/- , chinmoRNAi larval NBs . Note that Chinmo is silenced in surrounding wt late L3 NBs ( asterisk ) . Larval MARCM svp-/- , chinmoRNAi clones maintain Imp expression . Note that Imp is silenced in surrounding wt late L3 NBs ( asterisks ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 032 Surprisingly , in contrast to tumors , Chinmo knock-down in svp-/- MARCM clones does not lead to the down-regulation of Imp ( Figure 7G and Figure 7—figure supplement 2B ) . This shows that Imp expression in early-identity NBs does not require Chinmo . Thus , expression of chinmo and Imp in NBs during development and tumorigenesis are not subordinated to the same cross-regulatory interactions . We then tested whether ectopic expression of CIL proteins in pros-/- tumors was due to aberrant maintenance from an early CIL+ cell of origin . We induced pros-/- tumors by MARCM in the VNC either early ( during L1/L2 ) when Chinmo is still expressed in NBs and their progeny , or at midL3 after the CIL module has been switched off by temporal factor progression . One day after induction , ectopic Chinmo is observed in 91% of L1/L2-induced tumors ( n = 65 ) . In contrast , Chinmo is not detected 1 day after induction in midL3-induced tumors ( n = 54 ) ( Figure 8A ) , or 3 days after induction when larvae are grown in the sterol-free medium that prevent pupariation ( 26 out of 27 tumors still lack Chinmo ) ( Figure 8B , Figure 8—figure supplement 1 ) . Thus , Chinmo+ dNBs are only present in pros-/- tumors initiated during an early window of development ( before midL3 ) . 10 . 7554/eLife . 13463 . 033Figure 8 . pros-/- tumors induced in the VNC of midL3 larvae do not express Chinmo and do not grow in adults . ( A ) One day after early ( L1/L2 ) , induction most dNBs from pros-/- clones retain Chinmo expression . In contrast , Chinmo is absent from dNBs 1 day after late ( midL3 ) clonal induction . ( B ) Three days after L1/L2-induction in larvae raised on the sterol-free diet , pros-/- clones contain Chinmo+ dNBs ( 54 out of 59 ) . In contrast , 3 days after mid-L3 induction on the sterol-free diet , pros-/- clones do not contain Chinmo+ dNBs ( 26 out of 27 ) . ( C ) Adult VNC containing L1/L2-induced pros-/- MARCM clones are covered by tumors . In contrast , midL3-induced pros-/- MARCM clones are rare and remain small in adult VNCs . ( D ) On the sterol-free diet , 7 days after L1/L2-induction , pros-/- clones keep proliferating and generate large tumors of GFP+ dNBs ( Mira shown in red ) that cover the whole CNS . In contrast , pros-/- clones , 7 days after midL3-induction , rapidly stop growing and dNBs exhibit quiescence markers such as the loss of GFP and cytoplasmic extensions ( arrow emphasizes cytoplasmic extensions from dNBs , arrowhead emphasizes a cytoplasmic extension from a wt NB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 03310 . 7554/eLife . 13463 . 034Figure 8—figure supplement 1 . Schematic conclusion . Schematic representation recapitulating the conclusions from experiments in Figure 8 . When metamorphosis is prevented with a sterol-free diet , NBs and Chinmo- dNBs fail to terminally differentiate but exhibit a limited proliferation potential and enter quiescence ( marked by cytoplasmic extensions ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 03410 . 7554/eLife . 13463 . 035Figure 8—figure supplement 2 . The proliferation potential of normal NBs decreases over time while dNBs continue proliferating . ( A ) Mean % of PH3+ dNBs in L1/L2- and midL3-induced tumors after 10 days on the sterol-free diet . Early induction ( n = 5 VNCs , 5 clones , m = 5 . 7 , SEM = 1 . 30 ) , late induction ( n = 6 VNCs , 96 clones , m = 1 . 9 , SEM = 0 . 05 ) , p-value is 3 . 1x10-3 . ( B ) Larvae reared on the sterol-free diet do not pupariate . In 4 day-old larvae , wt NBs are large , highly active and Mira is cytoplasmic . In 10 day-old larvae , most wt NBs are small with nuclear Mira , possess cytoplasmic extensions ( inset , yellow arrowhead ) . Pictures are projections of several confocal sections . ( C ) Mean percentage of PH3+ NBs . 4 day-old larvae ( n = 4 VNCs , 319 NBs , m = 26 . 59 , SEM = 0 . 41 ) , 10 day-old larvae ( n = 7 VNCs , 701 NBs , m = 5 . 19 , SEM = 0 . 05 ) , p-value is 6 . 1x10-3 . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 035 We then investigated the malignant potential of L1/L2- and midL3-induced tumors . L1/L2-induced pros-/- clones persist in 100% of adults and rapidly progress to cover the totality of the VNCs ( n>100 ) . In contrast , midL3-induced tumors persist in only 36 . 4% of adult VNCs ( n = 19 ) ( Figure 8C ) , and make an average of 0 . 63 persisting tumor per adult VNC ( SEM = 0 . 23 ) , deriving from an average of 13 . 5 tumors per larval VNCs ( n= 16 , SEM = 1 ) . These rare persisting tumors ( 4 , 7% ) in adult VNCs remain small , containing no more than 100 dNBs . Thus , most midL3-induced tumors differentiated during metamorphosis , and the few tumors that persist in adults possess a limited growth potential ( Figure 8C ) . This limited growth potential of midL3-induced tumors was also observed when larvae were maintained in the sterol-free diet for 7 days after induction . In such conditions , midL3-induced tumors rapidly stop growing , with many dNBs progressively losing GFP expression , suggesting reduced transcriptional activity , and exhibiting reduced mitotic index and cytoplasmic extensions typical of quiescent NBs indicating a progressive exhaustion of their mitotic potential ( Chell and Brand , 2010; Sousa-Nunes et al . , 2011; Truman and Bate , 1988 ) ( Figure 8D , Figure 8—figure supplement 1 , 2A , B ) . Similar features are observed in surrounding wt NBs that do not undergo terminal differentiation in the absence of pupal signals ( Figure 8—figure supplement 1 , and Figure 8—figure supplement 2B , C ) . In contrast , L1/L2-induced tumors remain highly proliferative leading to massive dNB tumors covering the whole CNS ( Figure 8D ) . Together these results indicate that malignant pros-/- tumors can only be induced if they are initiated during an early window of development ( before midL3 ) while NBs and their progeny express the CIL module . In contrast , dNBs induced by the dedifferentiation of late-born CIL-negative GMCs cannot reactivate Chinmo expression , possess a limited proliferation potential and undergo terminal differentiation during metamorphosis like wt NBs , or persist as benign , slow-growing , tumors in adults . These results suggest that the silencing of the CIL genes in late-born neural cells by temporal patterning progression may abolish their malignant potential upon dedifferentiation . Alternatively , absence of CIL reactivation in late-induced tumors could be due to a non-permissive microenvironment subsequent to developmental progression . To distinguish these possibilities , we combined the GAL80ts system ( McGuire et al . , 2004 ) and MARCM to temporally control the induction of prosRNAi tumors in svp-/- NB clones blocked in an early temporal identity . Loss of Svp was induced in L1 by a 45 min heat-shock . Then larvae were kept at 18°C in the control experiment , or switched to 29°C from late L3 or pupal stages for late induction of prosRNAi . While adult flies raised at 18°C possess a single NB per clone due to the early loss of Svp , flies raised at 29°C from L3 or pupal stages contain large tumors of CIL+ dNBs in their VNCs ( Figure 9A , B ) . Thus , the GMCs of NBs blocked in an early temporal identity remain predisposed to generate CIL+ malignant tumors independently of the developmental stage . These results demonstrate that the malignant potential of a neural cell undergoing dedifferentiation is dictated by its temporal identity , and blocking temporal factor progression in NBs extends the window of malignant susceptibility . 10 . 7554/eLife . 13463 . 036Figure 9 . The temporal series regulates the malignant susceptibility of neural cells born during development . ( A ) Larvae were raised at 18°C and heat-shocked in L1 to induce svp-/- MARCM clones . Controls were kept at 18°C , to prevent prosRNAi expression , leading to the persistence of a single NB per clone in adults . In contrast , if larvae are switched to 29°C from late L3 , prosRNAi is expressed leading to large tumors in adults , exclusively composed of dNBs expressing Chinmo , Imp and Lin-28 . ( B ) Schematic recapitulation of the above experiments . Blocking temporal progression in NBs extends the window of malignant susceptibility . DOI: http://dx . doi . org/10 . 7554/eLife . 13463 . 036 We have demonstrated that the unlimited growth potential of pros-/- and brat-/- tumors induced during early larval development relies on the aberrant maintenance of an oncogenic module respectively co-opted from dedifferentiating early-born GMCs and INPs ( and possibly neurons in nerfin1-/- tumors ) . The core components of the module involve the BTB transcription factor Chinmo and the mRNA-binding protein Imp . Chinmo and Imp positively cross-regulate at the transcriptional and translational level respectively . We find that Chinmo is a proto-oncogene that is able to promote the transcription of a large set of genes that boost protein synthesis and cell-cycle progression , and is also a strong repressor of neural differentiation . Consequently , over-expression of Chinmo in NBs and GMCs , but not in neurons ( Zhu et al . , 2006 ) , is sufficient to cause NB amplification and tumor growth . This is consistent with a previously identified oncogenic activity of Chinmo when expressed in the hematopoietic and imaginal disc precursors ( Doggett et al . , 2015; Flaherty et al . , 2010 ) . In contrast , Imp mis-expression is not sufficient to trigger NB amplification but Imp is necessary to maintain pros-/- tumor growth beyond developmental stages , at least partly by allowing/sustaining Chinmo expression . Another mRNA-binding protein , Lin-28 , is also positively regulated by Chinmo in tumors . We found that Lin-28 is however dispensable for tumor growth although Lin-28 over-expression enhances the proportion of Chinmo+ dNBs in tumors . Thus Lin-28 overexpression may promote the self-renewal of Chinmo+ , Imp+ dNBs and therefore increase the growth potential of the tumor . This role is consistent with the observation that high expression of LIN28 isoforms in human tumors promotes tumor growth ( Viswanathan et al . , 2009 ) . Interestingly , a recent RIP-seq analysis in Drosophila embryos has uncovered both Imp and chinmo mRNAs among the most highly enriched targets of Lin-28 ( Chen et al . , 2015a ) , and mammalian Imp and Lin-28 orthologs share many common RNA targets ( Yang et al . , 2015 ) . Thus , Imp and Lin-28 may also directly co-regulate Chinmo translation in tumors . Further investigation is required to identify all components and the regulatory principles of this oncogenic network . Collectively , our results imply that , although representing a minor sub-population of dNBs , the subset of dNBs co-expressing Chinmo , Imp , and Lin-28 ( CIL+ dNBs ) is sustaining the unlimited growth of the tumor . It remains to be demonstrated whether these CIL+ dNBs act as cancer stem cells ( CSCs ) ( Beck and Blanpain , 2013 ) , able to self-renew and generate the bulk of the tumor , or represents a transient amplifying population of progenitors born from another population of rare and slow-proliferating CSCs . We have shown that Chinmo , Imp and Lin-28 are co-expressed in early NBs and their progeny and are silenced shortly after the L2/L3 transition in ageing NBs by progression of the temporal transcription factor series . Surprisingly , although Chinmo promotes tumor growth during both larval and adult stages , loss of Chinmo in normal NBs did not significantly affect the total number of progeny generated at the end larval stages . This indicates that the growth and proliferation of normal NBs and dNBs does not exhibit the same dependency on Chinmo . Although Chinmo is not required for proper NB mitotic activity in larvae , its over-expression in NBs is sufficient to promote tumorigenic growth in adults . Moreover , NBs blocked in an early temporal identity fail to persist in adults if lacking Imp and Chinmo . Thus , our results suggests that the silencing of Chinmo and Imp in early L3 by the temporal series is a way to limit the mitotic potential of NBs and ensure their timely termination during pupal stages . The role of Lin-28 in regulating NB mitotic activity during development is still unknown , but in mice , Imp1/Igf2bp1 and Lin28a/b are co-expressed in fetal cortical progenitors where they post-transcriptionally regulate growth genes involved in the PI3K/TOR pathway ( Yang et al . , 2015 ) . This suggests an evolutionary conserved growth-promoting role of these factors during neural development . Moreover , Lin28a and Imp1 are downregulated in mammals before birth ( Yang et al . , 2015 ) . This temporal pattern of expression is reminiscent to what is observed in NBs during Drosophila larval stages and suggests that the growth dynamic of fetal NSCs in mammals and larval NBs are controlled by evolutionary conserved gene networks . However , whether an analogous temporal transcription factor series in mammals similarly regulates Imp1 and Lin28a/b remains to be demonstrated . We noticed that loss of Chinmo in temporally blocked NBs does not induce the silencing of Imp . In contrast , Chinmo is necessary for Imp maintenance in tumors . Thus , chinmo and Imp are not subordinated to the same regulatory mechanisms whether in a developmental or in a tumorigenic context . This indicates that regulatory mechanisms upstream of chinmo and Imp may be different in normal NBs ( the temporal series ) and in dNBs after co-option . We have shown that progression throughout the NB-encoded series of temporal transcription factors terminates an early window of malignant susceptibility in the developing CNS around the beginning of the L3 stage . GMCs born during this early developmental window can generate malignant tumors upon loss of Pros . In contrast , GMCs born after the early window generate benign tumors that either completely differentiate during metamorphosis , demonstrating sensitivity to the growth-terminating cues operating in pupae , or persist in adults but fail to consistently grow , demonstrating a limited growth potential . Blocking progression of the temporal patterning system from early larval stages extends the window of susceptibility and late-born GMCs remain prone to malignant transformation , up to ( and possibly beyond ) metamorphosis . Our clonal analysis indicates that this window of malignant susceptibility correspond to the temporal window of expression of Chinmo and Imp . All together , our data strongly suggest that it is the presence of an activated Chinmo/Imp module in the cell ( GMC ) of origin that confers malignant potential . Thus , the temporal patterning regulates malignant susceptibility , at least partly , through the temporal control of Chinmo and Imp in newly-born GMCs during neurogenesis . Whether embryonic GMCs are prone to malignant transformation , like early larval GMCs , remains unclear . We could not in our various assays limit the loss of Pros to embryonically-born GMCs . This question therefore requires further investigation . Our work reveals a model for the rapid malignant progression of tumors with an early developmental origin . First , inactivation of genes that govern the terminal differentiation of intermediate progenitors or immature neurons triggers exponential NSC amplification . Second , if occurring at an early developmental stage , this dedifferentiation process interferes with the NSC-encoded temporal program that normally limits mitotic potential by silencing genes like Chinmo , Imp and Lin-28 in Drosophila . As a result , Chinmo , Imp and Lin-28 ( or their analogs in humans ) are co-opted in dedifferentiated cells to unleash an early mode of growth that is resistant to differentiation cues during late development and sufficient to sustain unlimited NSC amplification and malignant progression . In contrast to tumors initiated during adulthood , the cells of origin ( CIL+ GMCs in the case of pros-/- tumors ) already express the early oncogenic modules that are sufficient to sustain rapid growth . Consequently , early-induced tumors do not need to accumulate novel mutations to progress to malignancy . In contrast , later-born neural cells that have already silenced proto-oncogenic , modules may require the accumulation of more genetic and epigenetic alterations for the reestablishment of oncogenic combinations causing malignancy . All together , our data uncovers the developmental program regulating the malignant susceptibility of neural cells in Drosophila , and provides a model that may help to unveil the molecular basis underlying the rapid malignant growth of neural tumors with early developmental origins . A recent study in mice has shown that inactivation of Smarcb1 can most efficiently cause tumors resembling atypical teratoid/rhabdoid tumours ( AT/RTs ) , an agressive pediatric CNS cancer , when induced during an early window of pre-natal development ( E6 to E10 ) ( Han et al . , 2016 ) . In human , the embryonic/fetal origin of tumor cells is also suspected for a number of other pediatric neural cancers such as medulloblastoma in the cerebellum , retinoblastoma in the eye , and neuroblastoma in the sympathetic nervous system ( Marshall et al . , 2014 ) . Such tumors are typically composed of cells with immature characteristics , referred to as embryonal ( Marshall et al . , 2014 ) , and they usually carry few genetic alterations ( Vogelstein et al . , 2013 ) . However , the mechanisms underlying their rapid malignant transformation remains largely unsolved . We have demonstrated that the ability of human LIN28A/B to promote Chinmo expression in tumors is conserved in Drosophila , indicating that some regulatory aspects of the Chinmo/Imp/Lin-28 oncogenic module are likely to be evolutionary conserved . Interestingly , a recent survey of the scientific literature has revealed that LIN28A/B genes are much more frequently expressed in pediatric cancers than adult ones ( Carmel-Gross et al . , 2016 ) , a feature often associated with poor prognosis ( Molenaar et al . , 2012; Zhou et al . , 2013 ) . Human IMPs/IGF2BPs proteins have been less extensively investigated but are expressed in neuroblastoma and other cancers and are also associated with a poor outcome ( Bell et al . , 2013; 2015 ) . MYCN is a transcription factor of the MYC family , that like Chinmo in Drosophila , is expressed during early neurogenesis and promotes ribosome biogenesis and protein synthesis ( Boon et al . , 2001; Knoepfler et al . , 2002; Wiese et al . , 2013 ) . Interestingly , it is known to be up-regulated in many pediatric cancers of neural origin ( Huang and Weiss , 2013; Swartling et al . , 2010; Theriault et al . , 2014 ) , and is positively regulated by LIN28 and IMPs/IGF2BPs in neuroblastoma ( Bell et al . , 2015; Cotterman and Knoepfler , 2009; Molenaar et al . , 2012 ) . Thus , Chinmo/Imp/Lin-28 in insect may compose an ancestral oncogenic module with similar functions and regulatory interactions as mammalian MYCN/IMP/LIN28 during early neural development and tumorigenesis . Whether the temporal expression pattern of such modules , or oncofetal genes in general , delineates windows of malignant susceptibility in mammals is not clear . Interestingly , retinoblastoma is caused by the loss of the Retinoblastoma ( Rb ) protein that triggers dedifferentiation of photoreceptor cone cells ( Xu et al . , 2014 ) . Cone cells are among the earliest progeny to be generated by retinal progenitors ( from E12 to E16 in mice ) and their birth-order is dictated by a series of sequentially expressed temporal transcription factors ( Mattar et al . , 2015; Young , 1985 ) . In human , maturing cone cells in the retina , but not later born photoreceptors , express high levels of MYCN . Moreover , cooption of MYCN appears instrumental for retinoblastoma tumor growth ( Xu et al . , 2009 ) . Therefore , temporal patterning in retinal progenitors , as in Drosophila NBs , may dictate the malignant susceptibility of their progeny according to their birth order through the regulation of an early growth module involving MYCN . Our work suggests that deciphering temporal specification mechanisms in the different regions of the nervous system will help identify the cell types and gene networks at the origin of pediatric neural cancers . Drosophila stocks were maintained at 18°C on standard medium ( 8% cornmeal/8% yeast/1% agar ) . Sterol-free fly food was obtained by replacing classical yeast strain by a sterol-mutant erg2 knock-out strain in the medium ( Katsuyama and Paro , 2013; Parkin and Burnet , 1986 ) . Confocal images were acquired on Zeiss LSM510 and Zeiss LSM780 . ImageJ , FIJI and Photoshop were used to process confocal data . The area of individual tumors was measured from a z-projection of the entire tissue . For each experiment , at least 3 biological replicates were peformed . Biological replicates are defined as replicates of the same experiment with flies being generated by different parent flies . For all experiments , we performed a Mann-Whitney test for statistical analysis , except for Figure 7D and Figure 3—figure supplement 2 where a Fisher’s exact test was used . No data were excluded . Statistical test were performed with the online BiostaTGV ( http://marne . u707 . jussieu . fr/biostatgv/ ) . Results are presented as dot plots , also depicting the median in red and a boxplot in the background ( Whisker mode : 1 . 5IQR ) . The sample size ( n ) , the mean ( m ) , the standard error of the mean ( SEM ) , and the p-value are reported in the Figure legends . **: p-value ≤ 0 . 0001 , : p-value ≤ 0 . 001 , : p-value ≤ 0 . 01 and : p-value ≤ 0 . 05 . Experiments were performed at various temperatures as stated below . For generating MARCM clones ( Lee and Luo , 1999 ) , the following driver stocks were used: They were crossed with the following stocks: The progeny of the above crosses were heat-shocked 1 hr at 37°C just after larval hatching and raised at 25°C . The progeny of the above crosses were heat-shocked 1 hr at 37°C just after larval hatching and raised at 29°C . The progeny of this cross were raised at 18°C , heat-shocked 1 hr at 37°C just after larval hatching , and either maintained at kept at 18°C ( restrictive temperature ) for the rest of development ( controls ) or switched at 29°C ( permissive temperature ) in late L3 stage or early pupae ( Figure 9A ) . Flip-out clones were generated using hs-FLP; Act5c>CD2>GAL4 , UAS-GFP ( from N . Tapon ) with: The progeny of this cross were heat-shocked 1 hr at 37°C just after larval hatching and raised at 25°C . The progeny of this cross were heat-shocked 1 hr at 37°C during L2 stages and raised at 29°C . The GAL4 lines used were the following: The UAS lines used were: The lin28Δ1 , {lin-28::Venus} ( Lin-28-V ) stock contains a genomic rescue transgene encoding a fluorescently tagged Lin-28 ( Chen et al . , 2015a ) . GFP-Imp is the protein-trap line #G0080 ( Morin et al . , 2001 ) . The progeny of the above crosses were raised at 29°C . poxn-GAL4 is already active in embryonic NBs ( Boll and Noll , 2002 ) . However , we could show that inhibition of embryonic GAL4 , using a tub-Gal80ts transgene , does not significantly alter tumor formation demonstrating that prosRNAi expression during early larval stages is sufficient to cause malignant tumors in adults . Dissected tissues were fixed 5 min or more in 4% formaldehyde/PBS depending on the primary antibody . Stainings were performed in 0 . 5% triton/PBS with antibody incubations separated by several washes . Tissues were then transferred in Vectashield ( Clinisciences , France ) with or without DAPI for image acquisition . Primary antibodies were: chicken anti-GFP ( 1:1000 , Tebubio , France ) , mouse anti-Mira ( 1:50 , A . Gould ) , guinea-pig anti-Mira ( 1:1000 , A . Wodarz ) , guinea-pig anti-Asense ( 1:1000 , J . Knoblich ) , rabbit anti-PH3 ( 1:500 , Millipore , Billerica , MA ) , rat anti-PH3 ( 1:500 , Abcam , UK ) , rat anti-Elav ( 1:50 , DSHB , Iowa City , IA ) , rat anti-Chinmo ( 1:500 , N . Sokol ) , rabbit anti-Castor ( 1:500 , W . Odenwald ) , guinea-pig anti-Hunchback ( 1:500 , J . Reinitz ) , guinea-pig anti-Kruppel ( 1:500 , J . Reinitz ) , rabbit anti-Pdm ( 1:500 , X . Yang ) , mouse anti-Svp ( 1:50 , DSHB ) , rabbit anti-Imp ( 1:500 , P . Macdonald ) , rat anti-Lin-28 ( 1:500 , N . Sokol ) . Adequate combinations of secondary antibodies ( Jackson ImmunoResearch , West Grove , PA ) were used to reveal expression patterns . Females from the driver line nab-Gal4 , UAS-dicer2 were crossed to males carrying UAS-prosRNAiVDRC 2 ) UAS-prosRNAiVDRC; UAS-chinmoRNAiTRiPand UAS-prosRNAiVDRC; UAS-chinmo . Crosses were grown 7 days at 18°C and then switched at 29°C . Late L3 VNCs were dissected in cold PBS 3 days after the 29°C switch , during 30 min dissection rounds . Dissected VNCs were put in 500μL cold Lysis Buffer ( RA1 from the Total RNA Isolation kit , Macherey Nagel , Germany ) supplemented with 50 μL glass beads ( diameter 0 . 75-1mm , Roth , A554 . 1 ) and frozen in liquid nitrogen at the end of the dissection round . Sample tubes were then stored at -80°C up to RNA extraction . Biological triplicates were made for each condition with brain numbers as follows ( 1 ) n=79 , n=57 , n=65; ( 2 ) n= 70 , n=58 , n=78; ( 3 ) n=51 , n=60 , n=82 . For the RNA extraction , dissected brains stored in liquid nitrogen were thawed on ice . 10 μL TCEP were added to each tube following by 40s vortex . RNA extraction was then performed following the Total RNA Isolation NucleoSpin RNA XS protocol ( Macherey Nagel ) . RNA quality and quantity were checked by running samples on an Experion RNA HighSens Chip ( Biorad , 700–7105 , Hercules , CA ) and send to the Montpellier Genomix platform for RNA sequencing . Libraries were constructed using the Truseq stranded mRNA sample prep kit ( Illumina , ref . RS-122-2101 , San Diego , CA ) according to the manufacturer instructions . Briefly , poly-A RNAs were purified using oligo-d ( T ) magnetic beads from 300ng of total RNA . The poly-A+ RNAs were fragmented and reverse transcribed using random hexamers , Super Script II ( Life Technologies , ref . 18064–014 , Waltham , MA ) and Actinomycin D . During the second strand generation step , dUTP substitued dTTP . This prevents the second strand to be used as a matrix during the final PCR amplification . Double stranded cDNAs were adenylated at their 3' ends before ligation was performed using Illumia's indexed adapters . Ligated cDNAs were amplified following 15 cycles PCR and PCR products were purified using AMPure XP Beads ( Beckman Coulter Genomics , ref . A63881 , Brea , CA ) . Libraries were validated using a DNA1000 chip ( Agilent , ref . 5067–1504 , Santa Clara , CA ) on a Agilent Bioanalyzer and quantified using the KAPA Library quantification kit ( Clinisciences , ref . KK4824 ) . Four libraries were pooled in equimolar amounts per lane and sequencing was performed on an HiSeq2000 using the single read protocol ( 50nt ) . Image analysis and base calling were performed using the HiSeq Control Software and Real-Time Analysis component provided by Illumina . The quality of the data was assessed using FastQC from the Babraham Institute and the Illumina software SAV ( Sequence Analysis Viewer ) . Demultiplexing was performed using Illumina's sequencing analysis software ( CASAVA 1 . 8 . 2 ) . A splice junction mapper , TopHat 2 . 0 . 9 ( Kim et al . , 2013 ) ( using Bowtie 2 . 1 . 0 [Langmead and Salzberg , 2012] ) , was used to align RNA-Seq reads to the Drosophila melanogaster genome ( UCSC dm3 ) with a set of gene model annotations ( genes . gtf downloaded from UCSC on March 6 2013 ) . Final read alignments having more than 3 mismatches were discarded . Then , the gene counting was performed using HTSeq count 0 . 5 . 3p9 ( union mode ) ( Anders et al . , 2014 ) . The data is from a strand-specific assay , the read has to be mapped to the opposite strand of the gene . Before statistical analysis , genes with less than 15 reads ( cumulating all the analyzed samples ) were filtered and thus removed . Differentially expressed genes were identified using the Bioconductor ( Gentleman et al . , 2004 ) package DESeq ( Anders and Huber , 2010 ) 1 . 14 . 0 . Data were normalized using the DESeq normalization method . Genes with adjusted p-value to less than 5% ( according to the FDR method from Benjamini-Hochberg ) were declared differentially expressed . To perform the functional analysis of the resulting list of genes with the Gene Ontology ( GO ) annotations , the topGO ( Alexa et al . , 2006 ) package from Bioconductor was used . Overrepresented GO terms were identified using Fisher's exact test with the weight method . As confidence threshold we used a p-value of 1% . To realize this analysis the differentially expressed genes were compared with those of all known genes present in the annotation . The GO categories were found in the Org . Dm . eg . db package ( Carlson ) based on the gene reporter EntrezGeneID . Gene Ontology and Kegg pathway analysis were performed using Flymine ( Lyne et al . , 2007 ) . Raw RNA-seq data are available in the Gene Expression omnibus database ( accession number: GSE64405 ) . Transplantations of nab>GFP ( UAS-GFP/+; nab-GAL4 , UAS-dcr2/+ ) , nab>prosRNAi ( UAS-GFP/UAS-prosRNAi; nab-GAL4 , UAS-dcr2/+ ) , nab>prosRNAi; chinmoRNAi ( UAS-GFP/UAS-prosRNAi; nab-GAL4 , UAS-dcr2/UAS-chinmoRNAi ) , and nab>chinmo ( UAS-GFP; nab-GAL4 , UAS-dcr2/UAS-chinmo ) VNCs and brains have been performed in yw females according to ( Rossi and Gonzalez , 2015 ) .	Some aggressive brain tumors that affect children start to form before the child is even born . These tumors often develop much more rapidly than tumors found in adults , and require fewer genetic mutations to become dangerous and invasive . However , it is not known why this happens . Fruit flies are often used as animal models for cancer studies . As the fly brain develops , cells called neural stem cells divide several times , each time producing one stem cell and another cell known as the intermediate progenitor . The intermediate progenitor can itself divide one more time before maturing to become a neuron . Different types of neurons form in different stages of brain development . This is due to the sequential production of proteins called transcription factors in neural stem cells . Each transcription factor is inherited by a different set of intermediate progenitors and alters the activity of certain genes to determine the type of neuron the cells become . Some genetic mutations can prevent intermediate progenitors from maturing and cause them to revert to a stem-cell-like state , which allows them to rapidly divide and form tumors . Here , Narbonne-Reveau , Lanet , Dillard et al . use fruit flies to investigate why tumors that form early on in development progress so rapidly . The experiments uncover a ‘molecular clock’ in the neural stem cells that marks out a window of time in which they generate intermediate progenitors that are prone to becoming cancerous . This clock is represented by the sequential production of transcription factors that , in addition to determining neuronal identity , also turn off various growth-promoting genes in cells as brain development proceeds . These genes sustain normal cell division , but are silenced later on to prevent cells from dividing too many times . If the maturation of intermediate progenitors is disrupted early on in brain development while the growth-promoting genes are still active , the molecular clock fails to switch off the growth-promoting genes . As a result , these cells acquire an unlimited ability to divide , which drives tumor growth . However , later in development when the growth-promoting genes have already been switched off , disrupting the maturation of intermediate progenitors does not lead to these cells becoming cancerous . Therefore , Narbonne-Reveau , Lanet , Dillard et al . ’s findings explain why intermediate progenitors that mature early on in brain development are more prone to becoming cancerous than those that mature later , and why they need fewer mutations to become invasive . Most of the genes involved in this process are also found in humans . Therefore , the same mechanism might govern how aggressive childhood brain tumors are , which is a question for future studies to address .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cancer", "biology" ]	2016	Neural stem cell-encoded temporal patterning delineates an early window of malignant susceptibility in Drosophila
Murine studies have linked TGF-β signaling to emphysema , and human genome-wide association studies ( GWAS ) studies of lung function and COPD have identified associated regions near genes in the TGF-β superfamily . However , the functional regulatory mechanisms at these loci have not been identified . We performed the largest GWAS of emphysema patterns to date , identifying 10 GWAS loci including an association peak spanning a 200 kb region downstream from TGFB2 . Integrative analysis of publicly available eQTL , DNaseI , and chromatin conformation data identified a putative functional variant , rs1690789 , that may regulate TGFB2 expression in human fibroblasts . Using chromatin conformation capture , we confirmed that the region containing rs1690789 contacts the TGFB2 promoter in fibroblasts , and CRISPR/Cas-9 targeted deletion of a ~ 100 bp region containing rs1690789 resulted in decreased TGFB2 expression in primary human lung fibroblasts . These data provide novel mechanistic evidence linking genetic variation affecting the TGF-β pathway to emphysema in humans . Emphysema , that is pathologic destruction of lung parenchyma resulting in airspace enlargement , is one of the major manifestations of chronic obstructive pulmonary disease ( COPD ) . Emphysema occurs in distinct pathologic patterns , but these patterns are not captured by traditional quantitative measures of emphysema from lung computed tomography ( CT ) . In order to have more detailed radiographic measures of emphysema , we developed novel image extraction techniques to quantify the distinct patterns of emphysema based on the analysis of local lung density histograms ( Mendoza , 2012 ) . These local histogram emphysema ( LHE ) measures are more predictive of clinical outcomes than standard CT emphysema quantifications ( Castaldi et al . , 2013 ) , and in a previous genome-wide association study ( GWAS ) we identified genome-wide significant associations with these distinct LHE patterns ( Castaldi et al . , 2014 ) . However , the mechanisms by which these GWAS loci affect emphysema patterns are unknown . The majority of GWAS-identified loci for genetically complex diseases are located in non-coding DNA and influence gene regulatory elements ( Maurano et al . , 2012; Nicolae et al . , 2010 ) . Thus , for the functional characterization of emphysema GWAS loci , it is necessary to localize causal variants in regulatory elements and identify the gene ( s ) regulated by that element . Since multiple cell types contribute to emphysema , large-scale functional annotation projects such as the Genotype-Tissue Expression Project ( GTEx ) ( GTEx Consortium , 2015 ) and the Encyclopedia of Regulatory Elements ( ENCODE ) ( ENCODE Project Consortium , 2012 ) can be integrated with GWAS signals to identify candidate regulatory regions , tissues , and cell types of interest for more detailed functional characterization . In this study , we hypothesized that human emphysema is influenced by functional genetic variants that disrupt gene regulatory elements . As a screening approach , we cross-referenced GWAS results against large compendia of gene regulatory data from tissues and cell types to prioritize emphysema-associated loci for further functional study . This analysis identified rs1690789 as a high-probability functional variant in the GWAS-identified region near TGFB2 . Using chromatin conformation capture , we confirmed that the region spanning this SNP interacts with the TGFB2 promoter region . Via CRISPR/Cas-9 targeted deletion , we then demonstrated that a ~ 100 bp segment containing rs1690789 increases TGFB2 expression in primary human lung fibroblasts , providing novel evidence that genetic variation affecting TGF-β signaling contributes to the genetic predisposition to emphysema . In subjects from the COPDGene Study , we have previously demonstrated that LHE measures are associated with COPD-related phenotypes ( Castaldi et al . , 2013 ) and with common genetic variants at genome-wide significance ( Castaldi et al . , 2014 ) . To confirm these associations in an independent cohort and discover new genetic associations , we generated new LHE measures in 1519 subjects from the ECLIPSE Study , and we replicated the previously observed relationships between LHE pattern and GOLD ( Global Initiative for Obstructive Lung Disease ) spirometric grade ( Figure 1 ) . In the combined GWAS meta-analysis , we identified 10 independent regions with genome-wide significant associations to at least one LHE phenotype , six of which had been previously described ( Table 1 ) . One of the four novel associations is rs28929474 , the pathogenic Glu→Lys substitution in SERPINA1 which is known to be associated with COPD . There was no evidence of systematic inflation in the QQ-plots of these GWAS ( Figure 2 ) . Subject characteristics are shown in Supplementary file 1 Table 1 , and complete results by cohort are shown in Supplementary file 1 Table 2 . Since the more severe emphysema patterns ( severe centrilobular and panlobular emphysema ) are non-normally distributed , we performed a sensitivity analysis for these top results after performing inverse normal transformation of the LHE pattern phenotypes ( Supplementary file 1 Table 3 ) . In this analysis , four loci remained genome-wide significant ( loci on chromosome 15 , 14 , 11 , and 1 ) , two loci had p-values<5×10−7 , and four associations to the panlobular and severe centrilobular patterns had notably lower p-values suggesting that these specific associations are driven by extreme phenotype values and should be interpreted with caution . With regard to the association with the SERPINA1 Z-allele ( rs28929474 ) , subjects with known alpha-1 antitrypsin deficiency had been excluded from our primary analysis . However , when we examined the imputed genotypes of rs28929474 , we identified six individuals in ECLIPSE with imputed PiZZ genotypes . When we repeated the genetic analysis without these subjects , there was an increase in association p-value in ECLIPSE ( 0 . 003 versus 0 . 0004 , consistent direction of effect ) , and the meta-analysis association p-value was 1 . 6 × 10−7 . To determine whether these variants were associated with other COPD-related phenotypes , we queried the LHE GWAS significant associations against the results from two recent large GWAS studies for FEV1 , FEV1/FVC , and COPD status ( Shrine et al . , 2019; Sakornsakolpat et al . , 2019 ) . Five of the 10 LHE loci ( lead variants rs56113850 , rs796395 , rs17368659 , rs145770770 , and the 15q25 locus ) were associated to at least one of these outcomes at p<0 . 05 with a consistent direction of effect ( Supplementary file 1 Table 4 ) . Some loci associated with COPD and related phenotypes have also been associated with smoking behavior , raising the question of whether the COPD associations at these loci are mediated through smoking . To determine how many of our associations were also associated to smoking behavior , we queried our results against the UK Biobank Pheweb server GWAS for prior history of smoking , and we observed that the only associations that were nominally associated to smoking were the previously known smoking associations in the 15q25 and 19q13 loci ( Supplementary file 1 Table 5 ) . To generate functional hypotheses for emphysema-associated loci and prioritize regions for further functional study , we integrated our GWAS results with large-scale genome-wide eQTL and cell type epigenomic data , as shown in Figure 3 . To identify emphysema-associated loci that overlap with eQTL signals from multiple tissues , we cross-referenced our LHE GWAS results against eQTL results from 44 GTEx tissues and blood eQTLs from COPDGene . Since overlap between GWAS and eQTL signals can be due to chance , we used a Bayesian colocalization method ( Giambartolomei et al . , 2014 ) to quantify the probability that the local GWAS and eQTL signals were attributable to a shared causal variant . Four genome-wide significant LHE regions overlapped with eQTL regions with an estimated >80% probability of a shared causal variant responsible for the GWAS and eQTL associations ( Table 2 ) . To identify additional candidate colocalization loci that may be present below the stringent genome-wide significance threshold , we studied SNPs with a GWAS p<5×10−5 . At this threshold , the number of GWAS-eQTL overlap loci ranged from 78 ( panlobular pattern ) to 159 ( moderate centrilobular pattern ) , representing between 15% to 33% of the total number of loci with a GWAS p<5×10−5 . Of these loci , 32 had a > 80% estimated probability of having a shared causal GWAS-eQTL variant , and we identified the genes whose expression levels are altered by these loci ( Supplementary file 1 Table 6 ) . Full results of this analysis are available at https://cdnm . shinyapps . io/lhemphysema_eqtlcolocalization/ . To test for tissue-specific enrichment of LHE GWAS signals , we quantified the enrichment of LHE GWAS regions associated at p<5×10−5 in DNaseI peak regions from ENCODE and Roadmap cell types using the Garfield method ( Iotchkova et al . , 2019 ) . The most commonly enriched cell types were fibroblasts and fetal lung tissue , as can be seen in the enrichment results for moderate centrilobular emphysema ( Figure 4 ) . Out of 424 tested cell type annotations , there were 15 , 25 , and 1 cell type that exceeded the significance threshold for the moderate centrilobular , nonemphysematous , and severe centrilobular LHE phenotypes , respectively ( Supplementary file 1 Table 7 ) . One of the top GWAS-eQTL colocalization signals associated with the moderate centrilobular emphysema pattern spans a 200 kb region that includes the 3’ UTR of TGFB2 and extends 100 kb downstream . Multiple SNPs in this region were significantly associated with TGFB2 expression in human tissues from the GTEx project ( Figure 5 ) with the highest colocalization present with the eQTL signal in cultured fibroblasts . Given the essential roles of TGF-β signaling and fibroblasts in lung repair pathways , we selected this locus for further investigation . To confirm the colocalization results for TGFB2 , we performed a separate colocalization analysis using the same eQTL data but a separate colocalization methodology ( He et al . , 2013 ) . Sherlock analysis for the moderate centrilobular GWAS results and GTEx eQTL data from fibroblasts , lung tissue , and whole blood confirmed TGFB2 as a colocalization target for moderate centrilobular emphysema in fibroblasts , and a total of nine colocalizing genes or transcripts were identified at a p-value<1×10−4 ( Supplementary file 1 Table 8 ) . The GWAS signal in this region appears to demonstrate two independent peaks of association spanning a recombination hotspot , with the fibroblast eQTL signal appearing to colocalize with only one of these signals . We performed conditional genetic association analysis of this region , confirming the presence of two independent signals ( secondary association lead SNP rs3009942 p=4 . 4×10−7 , Figure 6 ) . To confirm that these are independent signals , we also performed conditional association adjusting for rs3009942 , which minimally attenuated the primary association ( rs796395 conditional p-value=3 . 3×10−7 ) . Focusing on the primary association peak which colocalized with the fibroblast eQTL signal , we estimated the causal probability ( i . e . the likelihood that each individual SNP is the causal variant ) of each SNP in this region using the PICS method ( Farh et al . , 2015 ) , identifying seven variants each with a > 5% estimated likelihood to be causal ( Supplementary file 1 Table 9 ) . We then queried whether any of these seven SNPs were predicted to alter transcription factor occupancy using the results of a previously published model developed from ENCODE data ( Maurano et al . , 2015 ) , identifying rs1690789 ( minor allele frequency of 0 . 48 in 1000 Genomes EUR population ) as the only variant in this set predicted to have allele-specific effects on transcription factor occupancy . Using the ENCODE uniformly processed DNaseI hypersensitivity dataset of 125 cell types , we observed that rs1690789 lies within a DNaseI hypersensitivity peak identified in 13 cell types ( Figure 5 , Panel C ) . Eight of these 13 cell types were fibroblasts , although this peak was not universally detected in all fibroblast DNaseI experiments , suggesting that this may be a context-specific regulatory element or that DNaseI accessibility may be influenced by genetic variation in these cell types . Since rs1690789 is located ~200 kb from the transcription start site of TGFB2 , we hypothesized that this region may regulate TGFB2 expression via a long-range chromatin interaction . Using publicly available 4C-Seq chromatin conformation data from IMR90 human lung fibroblasts ( Rao et al . , 2014 ) , we observed that the 10 kb region containing rs1690789 contacts multiple upstream and downstream regions around TGFB2 ( Figure 7A ) , suggesting that this region is a hotspot of chromosomal interaction . To confirm whether rs1690789 region indeed interacts with the promoter of TGFB2 in lung fibroblasts , we performed chromatin conformation capture ( 3C ) experiments in human lung fibroblasts ( IMR90 ) . Using the TGFB2 promoter region as the anchor region , we detected interaction between the rs1690789-containing region and the TGFB2 promoter in lung fibroblasts ( Figure 7B ) , suggesting long range regulation of TGFB2 by the region containing rs1690789 . To determine whether the DNA region near rs1690789 has regulatory effects on the expression of TGFB2 in human lung fibroblasts in the endogenous genomic context , we generated CRISPR/Cas-9 constructs containing gRNA pairs targeting the ~100 bp region spanning rs1690789 ( Figure 8A ) to generate genomic deletions in normal primary human lung fibroblasts . With sufficient deletion efficiency of the region spanning rs1690789 , we detected reduced expression of TGFB2 ( Figure 8B and C , Supplementary file 1 Table 11 ) , indicating that this distal genomic region has regulatory effects on the expression of TGFB2 in normal primary human lung fibroblasts . Previous GWAS studies have demonstrated that common genetic variation contributes to emphysema ( Cho et al . , 2015 ) , likely through the perturbation of gene regulatory mechanisms ( Castaldi et al . , 2014 ) . In order to identify putative causal variants and regulatory mechanisms for these loci , we used a screening approach that leverages large compendia of gene regulatory information in the GTEx and ENCODE projects . Using Bayesian colocalization , we identified 32 emphysema-associated loci at p<5×10−5 where it is likely that colocalized GWAS and eQTL signals arise from the same causal variant . It should be noted that these are putative and not confirmed disease variants due to our use of a relaxed GWAS significance threshold and the inherent complexities of colocalization , which continues to be an area of active methodological development . For the genome-wide significant locus near TGFB2 , multiple sources of publicly available and newly generated experimental data link a functional variant , rs1690789 , to TGFB2 expression in fibroblasts . These data suggest that naturally occurring genetic variability in TGF-β signaling plays a causal role in the development of emphysema . The TGF-β family of proteins constitutes a set of highly conserved signaling pathways that play a key role in human development and many other cellular functions ( Huminiecki et al . , 2009; Massagué , 2012 ) . With respect to the lung , TGF-β family proteins participate in normal lung development and are dysregulated in COPD , emphysema , asthma , and pulmonary fibrosis ( Verhamme et al . , 2015; Morris et al . , 2003; Thomas et al . , 2016 ) . Genetic variants near TGF-β superfamily members TGFB2 ( Castaldi et al . , 2014; Cho et al . , 2014 ) , ACVR1B ( Boueiz et al . , 2019 ) , LTBP4 ( Wain et al . , 2017 ) , and BMP6 ( Loth et al . , 2014 ) have been identified in GWAS for lung function and COPD , but prior to this study the region near ACVR1B was the only one linked to a gene in the TGF-β pathway through functional studies ( Boueiz et al . , 2019 ) . Our findings demonstrate that the emphysema-associated variant rs1690789 is located in an active gene regulatory region in human lung fibroblasts that interacts with the promoter region of TGFB2 and regulates TGFB2 expression . These analyses highlight the genetic and gene regulatory complexity of this region . Conditional association analyses identified two independent associations with moderate centrilobular emphysema near TGFB2 , and both associations are in linkage equilibrium ( i . e . low linkage disequilibrium ) with the lead variant identified in a previous GWAS of severe COPD ( Cho et al . , 2014 ) . In addition , the region containing rs1690789 has multiple interactions with other DNA regions , including the TGFB2 promoter and other downstream regions , indicating that this is a region of active chromatin interaction in human lung fibroblasts . While our analyses provide evidence that the emphysema-associated GWAS region downstream from TGFB2 interacts with the promoter of TGFB2 and regulates the expression of TGFB2 in human primary lung fibroblasts , many important questions remain about the function of the emphysema-associated locus near TGFB2 . First , the rs1690789 variant appears to be an eQTL for expression of TGFB2 in fibroblasts , but it is also strongly associated with TGFB2 expression in thyroid tissue in GTEx with an opposite direction of effect , suggesting complex and possibly context-dependent activity of this region . This is further supported by the observation that rs1690789 lies within a DNaseI peak in some but not all fibroblasts in the ENCODE and Roadmap projects , suggesting that the regulatory element in this region may be active only in certain fibroblast subsets , under certain conditions , or that the regulatory activity of this region is influenced by common ( but unmeasured ) genetic variation in these cells . Additional investigations are warranted to examine the context-specific function of this region . Second , our studies do not explain the function of the secondary association signal in this region , and it is also possible that both association regions may have functional effects in other cell types that contribute to COPD susceptibility . Third , it is possible that even within a single , statistically independent association peak , there may be multiple functional variants in tight linkage disequilibrium that contribute to the emphysema-related effects of this region . Future functional screening studies of this region can address this question . Finally , gene-level functional studies will be required to characterize the functional consequences of increased and decreased TGFB2 expression on lung fibroblast function . In summary , integrative GWAS-eQTL analysis of emphysema patterns identified 32 candidate loci with strong evidence of harboring gene regulatory variants responsible for the GWAS signal , including a locus near TGFB2 . Functional investigation of the associated region near TGFB2 confirmed the presence of a functional variant , rs1690789 , that likely contributes to the genetic predisposition to emphysema by regulating TGFB2 expression in fibroblasts . This region has multiple independent association signals and an extensive pattern of chromosomal interaction , indicating that additional investigations are required to fully characterize the gene regulatory activity at this locus . In addition to the association near TGFB2 , we identified dozens of other high confidence regions in our colocalization analysis , indicating additional functional variants that could be identified by high-throughput functional characterization approaches such as massively parallel reporter assays or CRISPR-mediated mutagenesis . The Evaluation of COPD Longitudinally to Identify Predictive Surrogate Endpoints Study ( ECLIPSE; SCO104960 , NCT00292552 , www . eclipse-copd . com ) is a longitudinal study with three-year follow-up data available for 2501 smoking subjects ( 2164 subjects with COPD and 337 smoking controls ) . The detailed study protocol and inclusion criteria have been previously published ( Vestbo et al . , 2008 ) . For this analysis , 1519 subjects with COPD ( defined as GOLD spirometric stages 2–4 ) and available CT scans were analyzed . COPD was defined by FEV1 <80% of predicted and FEV1/FVC < 0 . 7 . Genotyping was performed using the Illumina HumanHap 550 V3 ( Illumina , San Diego , CA ) . Subjects and markers with a call rate of <95% were excluded . Subjects with alpha-1 antitrypsin deficiency based on serum protein levels were excluded from this analysis . Population stratification and genotype imputation was performed using the same procedures and software as described above for COPDGene . GWAS models were adjusted for age , gender , pack-years of smoking history , and genetic ancestry via principal components . We performed GWAS analyses of the 5 LHE measurements separately in the three cohorts ( COPDGene NHW , COPDGene AAs , and ECLIPSE , total N = 11 , 282 subjects , 18 , 383 , 174 SNPs imputed to the 1000 Genomes reference panel , version 3 , hg19 ) . Analysis was limited to imputed SNPs with an imputation r2 >0 . 3 . Imputed genotypes were analyzed using the --dosage command in PLINK v1 . 9 ( Chang et al . , 2015 ) , though for SNPs with genotyped data the observed genotypes were used . GWAS models were adjusted for age , gender , pack-years of smoking history , and genetic ancestry via principal components ( Price et al . , 2006 ) . Results were meta-analyzed using the METAL ( Willer et al . , 2010 ) program using fixed effects meta-analysis with inverse variance weighting using SNP effect sizes and standard errors . We analyzed SNPs with a MAF >1% , and we meta-analyzed SNPs with results in at least two of the three cohorts . The generation of LHE measures in COPDGene has been previously described ( Castaldi et al . , 2013 ) . For the current studies , additional LHE measures were generated in ECLIPSE CT scans using the same method . LHE measurements have been previously associated with key COPD-related measures ( e . g . spirometry , MMRC ) ( Castaldi et al . , 2013 ) . To test if this relationship was consistent in the measurements generated in ECLIPSE , we visualized the median percentage of each emphysema pattern by GOLD stage . For the 14 lead SNPs associated with one or more of the LHE phenotypes at genome-wide significance , we queried other COPD-related GWAS for these variants or variants in linkage disequilibrium with these variants ( r2 >0 . 8 in the 1000 Genomes EUR reference panel ) . The queried GWAS studies were published studies of FEV1 and FEV1/FVC ( Shrine et al . , 2019 ) , COPD status ( Sakornsakolpat et al . , 2019 ) , or history of smoking . The smoking GWAS results were obtained from the UK Biobank Pheweb server ( http://pheweb . sph . umich . edu:5000/ ) on July 7 , 2019 for the phenotype ‘20116_1: Smoking status: Previous . ’ For colocalization and cell type enrichment analyses , GWAS SNPs significant at p<5×10−5 were considered . GTEx version six full results for 44 tissues were downloaded from the GTEx portal ( https://www . gtexportal . org/home/datasets ) , and eQTLs were calculated from blood RNAseq data in 385 NHW subjects from the COPDGene study using the same methods used in the GTEx Study v6 analysis . Details on the generation of COPDGene RNAseq data have been previously described ( Parker et al . , 2017 ) . GWAS-eQTL integrative analysis was performed according to the approach previously described in Castaldi et al . ( 2015 ) . Briefly , for each set of eQTL results , SNPs with a significant cis eQTL association at a 10% FDR threshold were extracted from each of the five sets of LHE GWAS results . Q-values were calculated for each subset of GWAS SNPs separately using the q-value package ( Storey et al . , 2019 ) , and SNPs demonstrating both significant eQTL and GWAS associations were retained for subsequent analysis ( i . e . eQTL-GWAS SNPs ) . Within each set of eQTL-GWAS SNPs , association regions for colocalization were defined by selecting all SNPs within 250 kilobases ( kb ) of each independent GWAS association . Colocalization of the GWAS and eQTL signals in these regions was calculated using the Bayesian colocalization method implemented in the R package coloc ( Giambartolomei et al . , 2014 ) using the default settings for the prior probability of a SNP being associated to target gene expression , the GWAS phenotype , and both measures ( prior probability 1 × 10−4 , 1 × 10−4 , and 1 × 10−5 , respectively ) . To confirm the colocalization results for TGFB2 , colocalization was also performed for the GWAS results for moderate centrilobular emphysema using the Sherlock method ( He et al . , 2013 ) . This analysis was performed using all the moderate centrilobular GWAS results referenced against three GTEx v6 eQTL datasets ( transformed fibroblasts , lung , and whole blood ) . The following parameter settings were used: cis eQTL significance threshold p<0 . 001 , trans eQTL significance threshold p<1×10−5 . To narrow the list of putative causal variants for the primary association near TGFB2 , we used the probabilistic inference of causal SNPs algorithm ( PICS ) ( Farh et al . , 2015 ) which infers per SNP causal probabilities from the strength of association of the lead SNP and linkage disequilibrium information from 1000 Genomes reference populations . The EUR reference population was used for this analysis , which was conducted via the PICS web interface ( https://pubs . broadinstitute . org/pubs/finemapping/pics . php ) . For SNPs with a PICS causal probability of 5% or greater , we queried these SNPs against their Contextual Analysis of Transcription Factor Occupancy ( CATO ) model predictions ( Maurano et al . , 2015 ) , which was trained on deep DNaseI sequencing data from the Roadmap project to predict per-SNP effects on transcription factor occupancy based on the predicted effects of each SNP on the binding energy of overlapping TF motifs and a number of factors related to local genomic sequence content . SNPs exceeding a CATO score of 0 . 1 were considered likely to alter TF occupancy . To determine whether LHE GWAS association were enriched in gene regulatory annotations from ENCODE and Roadmap Epigenomics data , we performed enrichment analysis for the LHE phenotypes with genome-wide significant results using the Garfield program and its pre-processed epigenomic annotations ( Iotchkova et al . , 2019 ) . The GWAS significance threshold was set at p<5×10−5 , and the default parameters were used for LD pruning ( r2 >0 . 1 ) , LD proxy threshold ( r2 >0 . 8 ) , minor allele frequency binning ( five bins ) , LD tag binning ( five bins ) , and TSS distance binning ( five bins ) . The significance threshold was set at p<0 . 0001 corresponding to Bonferroni adjustment for the effective number of independent annotations . Imputed DNaseI hypersensitivity peaks from Roadmap Epigenomics cell types or cell lines ( Ernst and Kellis , 2015 ) were downloaded from http://egg2 . wustl . edu/roadmap/data/byFileType/peaks/consolidatedImputed/narrowPeak/ . The overlap of rs1690789 with DNaseI peaks and enhancer marks was identified using the GoShifter program ( Trynka et al . , 2015 ) , and the raw DNaseI data for these cell types was visualized using the UCSC Genome browser . IMR-90 fibroblasts were purchased from ATCC and cultured in Eagle's Minimal Essential Medium ( EMEM , #12–611F , Lonza ) supplemented with 10% fetal bovine serum , penicillin and streptomycin . The cells tested negative for mycoplasma by MycoAlert Detection Kit ( #LT07-418 , Lonza ) . Primary human lung fibroblast cells were isolated from the lung tissue of healthy individuals ( Marsico Lung Institute , University of North Carolina at Chapel Hill , North Carolina ) as previously described ( Fulcher et al . , 2005 ) . Briefly , lung tissue samples were cut into small pieces and seeded onto culture dishes supplemented with DMEM/F12 medium , 10% fetal bovine serum , penicillin , streptomycin , amphotericin B and gentamicin . Amphotericin B and gentamicin were removed from the medium after the cells were passaged . The primary human lung fibroblasts were passaged twice and grown to 90% confluence prior to subsequent experiments . Human lung tissue was obtained under protocol #03–1396 approved by the University of North Carolina at Chapel Hill Biomedical Institutional Review Board . 4C chromosome conformation interaction results from the paper by Rao et al . ( 2014 ) were queried from the Yue Lab public website ( http://promoter . bx . psu . edu/ ) using the following search parameters: Species = human , Assembly = hg19 , Tissue = IMR90 , Type = Lieberman VC-norm , Resolution = 10 kb , SNP = rs1690789 , Extended Region = 500 kb . Human lung fibroblasts IMR90 cells were cultured to 80% confluency then cross-linked and lysed followed by digestion with BglII overnight . DNA fragments were then ligated with T4 ligase ( New England Biolabs , #M0202L ) for 6 hr at 16°C . After purification , 3C templates were used in PCR detection with unidirectional primers to indicate specific chromatin interaction by comparing relative band intensity from targeted regions against negative and positive control regions with three technical replicates ( i . e . same 3C templates , multiple PCR repeats ) . Primer sequences used for 3C-PCR are listed in Supplementary file 1 Table 10 . Detailed description of our methods has been published previously ( Zhou et al . , 2012 ) . To generate the rs1690789 CRISPR/Cas9 regional knockout primary human lung fibroblast cells , two guide RNAs ( u1 forward: 5’- GATACTCCAGTACATTGAGAAGG-3’; u2 forward: 5’-TGGAGTATCATTTCAGTGTTAGG-3’ ) located upstream from the SNP and two guide RNAs ( d1 forward: 5’-CAGCAGCGAGTTTGGCACTCAGG-3’; d2 forward: 5’-TGTCTCATTGCACACTCATGGGG-3’ ) located downstream from the SNP were cloned into pSpCas9 ( BB ) −2A-Puro ( PX459 ) V2 . 0 vectors ( Addgene plasmids #62988 ) , individually . Plasmids were verified by DNA sequencing . FuGENEHD was applied to transfect three pairs of gRNA plasmids ( u1 and d1 , u1 and d2 , u2 and d2 ) into primary normal human lung fibroblast ( NHLF ) cells according to the manufacturer’s instructions . PX459 empty vectors were transfected as control . Forty-eight hours after transfection , cells were selected with 1 . 2 µg/mL puromycin . After 2–3 weeks of recovery and expansion , cells were collected for DNA , RNA extraction and qPCR . Four biological replicates were performed ( i . e . same donor , four different transfections ) . DNA samples from human lung fibroblast cells were extracted using QuickExtract DNA Extraction solution ( #QE0905T , Lucigen , WI ) following manufacturer’s instructions . SYBRGreen dye-based quantitative RT-PCR was performed using the same equipment system and analysis method mentioned above , with the following primers to assess editing efficiency ( forward: 5’- GTTACCGATGCTTAAATGCCAC-3’; reverse: 5’- AGAATATCCCCATGAGTGTGC-3’ ) . The control was cells transfected with PX459 empty vector . Human lung fibroblast cell RNA was extracted using RNeasy Mini Kit ( #74106 , Qiagen , MD ) , and reverse transcription was performed by using High-Capacity cDNA Reverse Transcription Kit ( #4374966 , Applied Biosystems , MA ) . Quantitative RT-PCR was performed on QuantStudio 12K Flex Real-Time PCR System ( Applied Biosystems ) with gene-specific TaqMan probes ( Hs . PT . 58 . 24824921 ) from IDT ( Integrated DNA technologies , IA ) for detecting TGFB2 expression . Relative gene expression level was calculated based on the standard 2−ΔΔCT method , using GAPDH as a reference gene . For both the TGFB2 expression and editing efficiency tests , qPCR values were normalized against the mean qPCR value for the control cells for each experiment . Comparisons were performed using unpaired t-tests . Written , informed consent was obtained for all participants , and all study and consent forms were approved by the institutional review boards of the participating institutions .	It is well known that smoking is bad for the lungs . Not only can smoking cause lung cancer , it can also lead to conditions such as emphysema . This is the gradual damage to lung tissue that occurs when the walls of the tiny air-sacs in the lungs where the blood takes up oxygen , called the alveoli , weaken and break . Emphysema causes shortness of breath and difficulty pushing air out of the lungs , and it is part of chronic obstructive pulmonary disease ( also known as COPD ) . Genetic differences mean that certain people are more likely to develop emphysema than others . As an example , if someone has genetic mutations that alter the activity of a gene called TGFB2 , their risk of developing emphysema increases . However , the specific genetic mutations that modify the activity of TGFB2 were previously unknown . Parker et al . analyzed the genetic sequences of TGFB2 from patients with emphysema and compared them to those from healthy individuals . This revealed that certain mutations near the TGFB2 gene were more common in patients with emphysema . Next , Parker et al . showed that , in healthy lung cells called fibroblasts , the stretch of DNA that was mutated in patients with emphysema touched the part of TGFB2 that controls when the gene is activated . Deleting that same stretch of DNA in the fibroblasts meant the cells could no longer activate the TGFB2 gene as efficiently . Together , these results reveal a genetic difference that increases the risk for emphysema . COPD affects approximately 175 million people worldwide , causing over three million deaths each year . The findings of Parker et al . suggest that developing drugs that safely and efficiently target TGFB2 may be a way to help patients with early signs of emphysema .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "genetics", "and", "genomics" ]	2019	Identification of an emphysema-associated genetic variant near TGFB2 with regulatory effects in lung fibroblasts
Generating mammalian cells with specific mitochondrial DNA ( mtDNA ) –nuclear DNA ( nDNA ) combinations is desirable but difficult to achieve and would be enabling for studies of mitochondrial-nuclear communication and coordination in controlling cell fates and functions . We developed ‘MitoPunch’ , a pressure-driven mitochondrial transfer device , to deliver isolated mitochondria into numerous target mammalian cells simultaneously . MitoPunch and MitoCeption , a previously described force-based mitochondrial transfer approach , both yield stable isolated mitochondrial recipient ( SIMR ) cells that permanently retain exogenous mtDNA , whereas coincubation of mitochondria with cells does not yield SIMR cells . Although a typical MitoPunch or MitoCeption delivery results in dozens of immortalized SIMR clones with restored oxidative phosphorylation , only MitoPunch can produce replication-limited , non-immortal human SIMR clones . The MitoPunch device is versatile , inexpensive to assemble , and easy to use for engineering mtDNA–nDNA combinations to enable fundamental studies and potential translational applications . Mitochondrial DNA ( mtDNA ) and nuclear DNA ( nDNA ) genome coordination regulates metabolism , epigenome modifications , and other processes vital for mammalian cell survival and activity ( Patananan et al . , 2018; Ryan and Hoogenraad , 2007; Singh et al . , 2017 ) . Together , these genomes encode >1100 mitochondrial proteins , with only 13 essential electron transport chain ( ETC ) proteins encoded within the mtDNA ( Calvo and Mootha , 2010 ) . The mitochondrial proteins encoded in the mtDNA and the nDNA must be compatible to support mitochondrial ETC activity . Mutations in mtDNA can impair the ETC by altering nDNA co-evolved ETC complex protein interactions , causing defective cellular respiration and debilitating diseases ( Greaves et al . , 2012 ) . Furthermore , the coordination of these two genomes to transcribe , translate , and potentially modify appropriate levels of their respective gene products to maintain energetic and metabolic homeostasis is essential to the proper functioning of the ETC ( Wolff et al . , 2014 ) . As a result , methods that enable pairing of specific mtDNA and nDNA genotypes in tractable systems are key to understanding the basic biology of mitonuclear interactions and their implications for health and disease . Our current inability to edit mtDNA sequences is a roadblock for many studies and potential applications . For example , endonucleases targeted to the mitochondrion inefficiently eliminate and cannot alter mtDNA sequences ( Bacman et al . , 2018 ) . An exciting new bacterial cytidine deaminase toxin generates a limited repertoire of point mutations in the mtDNA; however , its efficiency remains low and it is unable to knock-in new gene sequences ( Mok et al . , 2020 ) . Mitochondrial transfer between cells in vitro and in vivo provides a potential path forward for transplanting existing mtDNA sequences; however , the mechanisms controlling such transfers remain unknown ( Dong et al . , 2017; Torralba et al . , 2016 ) . Isolated mitochondrial transfer has been used to deliver mitochondria to a range of recipient cell types in vitro and even in vivo ( Caicedo et al . , 2015; Emani et al . , 2017; Kitani et al . , 2014 ) ; however , most studies using these methods observe only short-term changes to cell or organ performance and function . A small number of these studies have coincubated mitochondria with recipient cells and observed permanent retention of the exogenous mtDNA in mtDNA-deficient ( so-called ‘ρ0’ ) cells using large doses of mitochondria or antibiotic selection schemes ( Clark and Shay , 1982; Patel et al . , 2017 ) , although these approaches may not be possible when mitochondrial donor material is limited or does not possess a suitable selection marker . Methods that deliver mitochondria directly into ρ0 cells can increase stable mitochondrial transfer efficiency and employ a wider range of mitochondrial donor sources . Such methods include membrane disruption ( King and Attardi , 1988; Wu et al . , 2016 ) or fusion with enucleated cytoplasts ( Wilkins et al . , 2014 ) . However , these methods are typically laborious , low-throughput , or depend on cancerous , immortal recipient cells lacking physiologic mitochondrial activity . An interesting recent study did report one desired mtDNA–nDNA clone and 11 false-positive clones using cybrid fusion with replication-limited cells , an achievement hampered by a low generation rate with unknown reproducibility or generalizability ( Wong et al . , 2017 ) . There exist clinically relevant methods to replace the mtDNA of human cells , such as somatic cell nuclear transfer and pronuclear transfer that involve delivering nuclear genetic material from patients with mtDNA diseases into enucleated oocytes with non-mutant mtDNA genotypes ( Hyslop et al . , 2016; Tachibana et al . , 2013 ) . These methods hold potential for replacing deleterious mtDNA for the unborn , but they are technically challenging , low-throughput , dependent on high-quality patient samples , and prone to contamination by mutant mtDNA from the affected nuclear source material ( Kang et al . , 2016 ) . Higher-throughput techniques that exchange non-native for resident mtDNAs in non-immortal somatic cells in tissue culture could enable studies of mtDNA–nDNA interactions and replace deleterious mtDNAs within cells with therapeutic potential ( Patananan et al . , 2016 ) . Thus , a higher throughput , reproducible , and versatile mtDNA transfer approach to generate multiple desired ‘stable isolated mitochondrial recipient’ ( SIMR ) clones in replication-limited cells remains essential for statistically valid studies and potential translation of mitochondrial transplantation . We developed ‘MitoPunch’ as a simple , high-throughput mitochondrial transfer device consisting of a lower polydimethylsiloxane ( PDMS ) reservoir loaded with a suspension of isolated mitochondria , covered by a polyethylene terephthalate ( PET ) filter seeded with ~1 × 105 adherent cells ( Figure 1A , Figure 1—figure supplement 1 ) . MitoPunch uses a solenoid-activated plunger to transfer isolated mitochondria in a holding chamber by force into the cytosol of mammalian cells . Upon actuation , a mechanical plunger deforms the PDMS from below , which , as calculated by numerical simulation , generates pressure up to 28 kPa inside the PDMS chamber , propelling the suspension through numerous 3 µm pores in the PET filter . This pressure cuts the plasma membrane of recipient cells sitting atop the pores and delivers mitochondria into the cytoplasm of the cut cells ( Figure 1B ) . To assess performance , we compared MitoPunch to mitochondrial coincubation ( Kitani et al . , 2014 ) and to MitoCeption ( Caicedo et al . , 2015 ) , a method that uses centripetal force generated in a centrifuge to localize mitochondria to recipient mammalian cells ( Figure 1C ) . In MitoCeption , a 1500 × g centripetal force draws isolated mitochondria to a recipient cell monolayer . We calculate that the suspended mitochondria exert a pressure of ~1 . 6 Pa on recipient cell membranes ( Figure 1D ) ( see Materials and methods ) . We isolated and delivered dsRed-labeled mitochondria from ~1 . 5 × 107 HEK293T cells ( Miyata et al . , 2014 ) into ~1 × 105 143BTK– ρ0 osteosarcoma cells and replication-limited BJ ρ0 foreskin fibroblasts in technical triplicate and measured the fraction of recipient cells positive for dsRed fluorescence by ImageStreamx MarkII imaging flow cytometry ( Figure 2A ) . We define technical replication as independently performed mitochondrial deliveries using the same isolated mitochondrial preparation into recipient cells of the same passage . For 143BTK– ρ0 cells at ~2 hr post-delivery , imaging flow cytometry showed that MitoPunch yielded the lowest fraction of dsRed-positive cells compared to coincubation or MitoCeption . Similarly , for BJ ρ0 recipient cells , MitoPunch yielded the lowest fraction of dsRed-positive cells compared to coincubation or MitoCeption , although at lower levels relative to 143BTK– ρ0 recipients . This measurement assesses colocalization of mitochondria with recipient cells , and not necessarily the occurrence or mechanism of internalization of delivered mitochondria . These data suggest that the method of delivery and target cell type affect the efficiency of initiating mitochondria–recipient cell interactions . We quantified the number of discreet dsRed-spots in each cell ~2 hr following delivery from this data ( George et al . , 2004; Figure 2B and Figure 2—figure supplement 1 ) . ImageStream spot count analysis of 143BTK– ρ0 recipient cells showed MitoPunch delivered a lower mean and median number of dsRed spots per cell than coincubation or MitoCeption . MitoPunch transfers into BJ ρ0 recipient cells yielded fewer mean spots/cell compared to coincubation and MitoCeption with an equivalent median number of spots/cell for MitoPunch and MitoCeption and slightly more for coincubation . Next , we used confocal microscopy to observe dsRed mitochondrial fluorescence in 143BTK– ρ0 recipients fixed 15 min post-transfer , which we chose for its robust mitochondrial acquisition ( Figure 2C ) . We visualized mitochondrial localization with confocal microscopy by detecting dsRed protein from the donor mitochondria , shown in red , and labeling the recipient cell plasma membranes with either CellMask Green ( coincubation and MitoCeption ) or wheat germ agglutinin ( MitoPunch ) , shown in green . Following MitoPunch , mitochondrial dsRed appeared to localize to pores in the filter insert and within the cytoplasm of cells , whereas coincubation and MitoCeption uniformly coated recipient cells with mitochondria , with greater mitochondrial association with recipient cells following MitoCeption . While all three methods initiate physical interactions between mitochondria and recipient cells , MitoPunch delivers mitochondria to the basal membranes of recipient cells at regions associated with the PET membrane pores , compared to a diffuse membrane association pattern seen with coincubation and MitoCeption . To investigate the capacity of these methods to disrupt recipient cell plasma membranes , we delivered the membrane impermeant dye propidium iodide ( PI ) by coincubation , MitoPunch , and MitoCeption to measure membrane disruption from delivery and quantified uptake by flow cytometry ( Figure 2D; Novickij et al . , 2017 ) . Delivery into 143BTK– ρ0 cells by MitoPunch and MitoCeption resulted in similar percentages of PI-positive recipient cells , and both were greater than coincubation . Interestingly , BJ ρ0 cells showed comparable fractions of PI-positive cells to the 143BTK– ρ0 after coincubation and MitoCeption . However , MitoPunch yielded an approximately fivefold increase in the PI-positive fraction compared to all other conditions . These data show that MitoPunch and MitoCeption disrupt the plasma membranes of recipient cells for potential mitochondrial transfer , and the degree of disruption is cell type and delivery method dependent . After verifying mitochondrial interaction with recipient cells by coincubation , MitoPunch , and MitoCeption , we next determined whether these methods result in permanent retention of exogenous mtDNA to generate SIMR cells . ρ0 cells cannot synthesize pyrimidines and therefore cannot proliferate or survive without supplemented uridine because of ETC impairment , so we used nucleotide-free medium prepared with dialyzed fetal bovine serum ( SIMR selection medium ) to select for SIMR cells with transplanted mtDNA and restored ETC activity ( Grégoire et al . , 1984; Figure 3A and B ) . BJ ρ0 cells survive longer under this selection scheme compared to the 143BTK– ρ0 ( data not shown ) , so we included an additional selection phase by culturing these cells in nucleotide-free , glucose-free , galactose supplemented medium ( galactose selection medium ) ( Robinson et al . , 1992 ) . We isolated and transferred HEK293T dsRed mitochondria into 143BTK– ρ0 and BJ ρ0 cells by coincubation , MitoPunch , and MitoCeption , performed SIMR selection in cell-type appropriate medium for 7 days , and quantified the number of viable clones by crystal violet staining ( Figure 3C ) . Coincubation did not generate SIMR clones in 143BTK– ρ0 cells , in contrast to MitoPunch and MitoCeption , which each generated dozens of clones . BJ ρ0 cells with delivered HEK293T mitochondria by coincubation or MitoCeption did not form SIMR clones . MitoPunch generated numerous SIMR clones in both cell types , although fewer BJ ρ0 SIMR clones than in 143BTK– ρ0 cells , whereas MitoCeption only generated clones in 143BTK– ρ0 cells and was unable to form stable clones in replication-limited BJ cells . To assess the risk of mitochondrial donor cells surviving disruption during mitochondrial isolation and generating false positive SIMR clones , we performed three independent mitochondrial isolations , plated an aliquot from each isolation representing mitochondria isolated from ~1 . 5 × 107 HEK293T dsRed cells on 10 cm dishes , and carried these plates through the 10-day selection with SIMR selection medium before crystal violet staining for visual assessment ( Figure 3—figure supplement 1 ) . We observed no cell growth on any of the three plates , indicating a minimal incidence of donor cell survival through the mitochondrial isolation protocol . We next investigated whether differences in SIMR clone generation between 143BTK– ρ0 and BJ ρ0 cells were driven by sensitivity to differences in delivery pressure . We developed a MitoPunch device with adjustable plunger acceleration modulated by changing the circuit voltage ( ImmunityBio ) . We generated independent voltage titration curves in 1 V increments ( 0 V – 5 V ) for each cell type in technical triplicate at each voltage and used the same mitochondrial preparation for all samples for each cell type . All prior experiments in this study are controlled by delivering DPBS with calcium and magnesium to recipient cells by MitoPunch , but here we included a 0 V condition in which the seeded filter insert was positioned atop the PDMS reservoir and pressed against an aliquot of isolated mitochondrial suspension similar to deliveries with force , but without actuating the piston . We achieved maximum 143BTK– ρ0 SIMR clone generation with this tunable MitoPunch at 1 V , with a sharp reduction to background with increasing voltage ( Figure 3D ) . The BJ ρ0 recipient also showed maximal SIMR generation at 1 V , with a shallow decline in SIMR generation efficiency to 5 V . Surprisingly , the 0 V condition consistently yielded a few SIMR clones in the 143 BTK– ρ0 recipients and inconsistently in the BJ ρ0 recipients . This result suggests that the pressure generated by sealing the filter insert against the PDMS reservoir is sufficient to generate SIMR clones at a low frequency . For all forthcoming MitoPunch trials we use the variable voltage MitoPunch device set to 1 V . We performed a similar force titration with MitoCeption by varying the maximum centripetal force , using a common mitochondrial preparation for all samples of both cell types . In 143BTK– ρ0 cells , we observed maximum clone generation at 1000 × g and 1500 × g , and we did not generate BJ ρ0 SIMR clones greater than the 0 × g background at any acceleration tested ( Figure 3E ) . This background , present in both 143BTK– ρ0 and BJ ρ0 conditions at 0 × g , is likely from rare un-lysed donor cells from mitochondrial preparations directly pipetted into the culture medium of recipient cells during MitoCeption . We have infrequently observed imperfect donor cell lysis , usually in larger mitochondrial preparations , that results in rare , persistant dsRed fluorescent colonies as observed by fluorescence microscopy . True SIMR clones cannot produce the dsRed protein from donor mitochondria and lose fluorescence with time over selection , while these persistent dsRed colonies maintain their fluorescence over the same period ( data not shown ) . Despite this occasional low-level contaminating donor cell background , MitoCeption yielded a strong dose-dependent response in SIMR clone generation from 143BTK– ρ0 recipients above the background . Additionally , MitoPunch deliveries into B16 ρ0 mouse melanoma cells ( Tan et al . , 2015 ) yielded maximal SIMR generation at a different voltage than in the human cell lines tested , showing that optimal mitochondrial delivery pressure may be cell type dependent ( Figure 3—figure supplement 2 ) . These data suggest that MitoPunch is uniquely able to generate SIMR clones in replication-limited fibroblasts and SIMR generation efficiency depends on delivery pressure . We next quantified the reproducibility of our mitochondrial preparation technique and the MitoPunch procedure by performing triplicate MitoPunch transfers using three independent mitochondrial preparations from equal numbers of HEK293T dsRed biological replicate populations ( Figure 3—figure supplements 3 and 4 ) . We define biological replication here as mitochondrial preparations derived from independently cultured populations of mitochondrial donor cells . Mitochondrial preparations 1 , 2 , and 3 ( same as those pictured in Figure 3—figure supplement 1 ) generated consistent protein concentrations , and each preparation yielded dozens of SIMR clones in all three technical replicate MitoPunch deliveries with the exception of Prep 3 , which resulted in two lower efficiency replicates . We quantified the number of SIMR clones generated per microgram of mitochondrial mass loaded into the MitoPunch apparatus and observed a similar trend . These results showed that our mitochondrial isolation technique produced consistent levels of isolated mitochondrial mass and that the MitoPunch technique yielded high numbers of SIMR clones . To enable desirable mtDNA–nDNA clone generation using limited starting material , such as mitochondria from rare cell subpopulations , we determined the minimal mass of mitochondrial isolate required to generate SIMR clones . We performed coincubation , MitoPunch , and MitoCeption transfers into ~1 × 105 143BTK– ρ0 recipient cells using decreasing concentrations of dsRed mitochondria isolated from HEK293T cells and plated half of the recipient cell population on 10 cm plates . We observed a similar dose-dependent relationship between mitochondrial mass delivered and SIMR clones observed for MitoPunch and MitoCeption across 0 . 16 µg , 1 . 6 µg , and 16 µg total mitochondrial protein suspended in 120 µL of 1× DPBS , pH 7 . 4 transfer buffer ( Figure 3—figure supplement 5 ) . These results showed that although MitoPunch and MitoCeption generate SIMR clones from transformed recipient cells with similar efficiency per microgram of mitochondrial isolate delivered , the differences inherent to the two protocols rendered direct comparisons of their relative efficiencies less meaningful . Moving the seeded PET filter from a 12-well dish to the MitoPunch apparatus often resulted in excess medium being carried to the PDMS reservoir . Combined with the small volume of mitochondrial preparation delivered to the recipient cells , we observed that MitoPunch resulted in diluted residual mitochondrial isolate left in the reservoir post-transfer . In the interest of conserving mitochondrial material , we tested whether a used 120 µL aliquot of isolated mitochondria can be applied to repeated MitoPunch transfers to generate SIMR clones ( Figure 3—figure supplement 6 ) . We performed 11 sequential deliveries into 143BTK– ρ0 cells using one aliquot of mitochondrial isolate and found maximal SIMR clone generation from the first and second deliveries , after which we observe a sharp reduction in SIMR cell formation and inconsistent SIMR generation rate up to the 11th transfer . These data showed that multiple MitoPunch transfers can be performed using a single aliquot of mitochondrial suspension when material is limited . Finally , we measured mitochondrial function in SIMR cells by quantifying the rate of oxygen consumption and assessing mitochondrial morphology . We isolated three independent 143BTK– ρ0 SIMR clones generated by MitoPunch or MitoCeption transfer of isolated HEK293T mitochondria and measured each clone’s oxygen consumption rate ( OCR ) using a Seahorse Extracellular Flux Analyzer mitochondrial stress test ( Figure 4A , Figure 4—figure supplement 1 ) . To determine whether SIMR clone respiration remained stable through time , we grew the clones through two freeze/thaw cycles in uridine supplemented medium and measured cellular respiration . We found that one MitoCeption clone lost its respiratory capacity and one MitoPunch clone was not viable after freezing and thawing ( data not shown ) . In the remaining clones , basal and maximal respiration , spare respiratory capacity , and ATP generation remained stable throughout both freeze-thaw cycles . We have performed numerous similar experiments using a range of recipients and mitochondrial donors and observed successful clone viability after freeze-thaw ( data not shown ) . We then immunostained the freeze-thawed SIMR clones with anti-TOM20 and anti-double-stranded DNA ( dsDNA ) antibodies to detect mitochondria and mtDNA content , respectively , by confocal microscopy ( Figure 4B and Figure 4—figure supplement 2 ) . The MitoCeption clone that lost respiratory capacity showed a fragmented mitochondrial network with no detectable mtDNA ( Figure 4—figure supplement 2 ) , whereas the other SIMR clones generated by MitoPunch and MitoCeption contained mtDNA with filamentous mitochondrial network morphologies . These data show that the majority of 143BTK– ρ0 SIMR clones generated by either MitoPunch or MitoCeption have retained mtDNA , restored respiratory profiles , and filamentous mitochondrial network morphologies . Stability of the mitochondrial genome is essential for studying the long-term effects of mtDNA–nDNA interactions and for potential therapeutic applications of mitochondrial transfer . MitoPunch generates up to hundreds of SIMR clones in both transformed and Hayflick-limited recipient cells by exerting a pressure sufficient to perforate mammalian cell membranes in regions small enough to be repaired within minutes , which sustains cell viability and resumed cell growth and proliferation ( Boye et al . , 2017 ) . We have generated SIMR clones by MitoPunch with mitochondria isolated by a commercially available kit , as performed here , as well as by using standard mitochondrial isolation buffers . Additionally , we achieved similar results by disrupting mitochondrial donor cells using Dounce homogenization ( data not shown ) but found the commercially available kit with syringe disruption is advantageous due to its ease of use , reproducibility , and a reduced number of steps to isolate mitochondria . We generated dozens of SIMR clones by MitoPunch and MitoCeption using these mitochondrial isolation methods and anticipate that other mitochondrial preparation techniques will also yield SIMR clones . Interestingly , we do not observe SIMR clone generation by coincubation in our study . Few reports show limited stable clone formation by coincubation techniques , but these studies used up to 100-fold higher levels of exogenous mitochondria in coincubation experiments than required for MitoPunch or MitoCeption in our hands , or antibiotic selection schemes to achieve stable mitochondrial transfer ( Clark and Shay , 1982; Patel et al . , 2017 ) . High levels of mitochondrial protein are easily isolated from fast-growing immortalized cell lines but may not be available when using human donor-derived or other limiting starting material . Additionally , mitochondrial donor cells of interest nearly exclusively lack antibiotic selection markers , making such selection schemes unfeasible . Particularly in those cases , the greatly enhanced SIMR generation capacity of MitoPunch and MitoCeption is strongly enabling for generating desired mtDNA–nDNA combinations . The distinct mechanisms and procedures of MitoPunch and MitoCeption make direct comparisons of their relative efficiencies challenging . Despite this , our results demonstrate that both techniques generate SIMR clones from ρ0 transformed cells in a mitochondrial dose-dependent fashion and can be readily adopted by laboratories studying mtDNA–nDNA interactions . Strikingly , in the cell types we have tested , we find that only MitoPunch generates SIMR clones from ρ0 primary , non-immortal cells . Studies in our laboratory suggest that the transcriptome and metabolome of replication-limited SIMR clones differ significantly from un-manipulated control clones but can be recovered and reset to un-manipulated control levels by cellular reprogramming to induced pluripotent stem cells and subsequent differentiation ( Patananan et al . , 2020 ) . These results indicate that SIMR clone generation in replication-limited , reprogrammable cells is crucial for studies of mtDNA–nDNA interactions involving mitochondrial transplantation into ρ0 cells , and that MitoPunch is uniquely capable of efficiently generating enough clones for statistically valid studies in such work . We have circumvented the need for ρ0 recipient cells by using the MitoPunch technology to completely replace mutant mtDNA in mouse cells without mtDNA depletion . This was done by delivering mitochondria containing mtDNA with a chloramphenicol resistant point mutation and selecting for SIMR clones containing only rescue mtDNA using antibiotic supplemented nucleotide-free medium ( Dawson et al . , 2020 ) . However , this workflow is dependent upon using antibiotic resistant mitochondrial donor cells and is not applicable to investigating the full spectrum of mtDNA sequences required for robust studies of mtDNA–nDNA interactions . Future work with MitoPunch and other isolated mitochondrial transfer modalities will be improved by developing techniques to avoid fully depleting the mtDNA of recipient cells of interest before generating SIMR clones for downstream analysis and applications . Human ρ0 cells were grown in DMEM ( Fisher Scientific , Waltham , MA , Cat . # MT10013CM ) supplemented with 10% FBS , non-essential amino acids ( Gibco , Waltham , MA , Cat . #11140–050 ) , GlutaMax ( Thermo Fisher Scientific , Waltham , MA , Cat . # 35050–061 ) , penicillin and streptomycin ( VWR , Radnor , PA , Cat . # 45000–652 ) , and 50 mg/L uridine ( Thermo Fisher Scientific , Cat . # AC140770250 ) . All other human cell lines were grown in DMEM ( Fisher Scientific , Cat . # MT10013CM ) supplemented with 10% FBS , non-essential amino acids , GlutaMax , and penicillin and streptomycin . B16 ρ0 cells were grown in RPMI ( Thermo Fisher Scientific , Cat . # MT-10–040 CM ) supplemented with 10% FBS , non-essential amino acids , GlutaMax , penicillin and streptomycin , pyruvate ( Corning , Corning , NY , Cat . # 25000 CI ) , and 50 mg/L uridine . L929 cells were grown in RPMI supplemented with 10% FBS , non-essential amino acids , GlutaMax , penicillin and streptomycin , and pyruvate . All mammalian cells were cultured in a humidified incubator maintained at 37°C and 5% CO2 . The following cells were used in this study: HEK293T dsRed ( female ) , 143BTK– ( female ) , 143BTK– ρ0 ( female ) , BJ ρ0 ( male ) , B16 ( male ) , and L929 ( male ) . We have not formally identified these cell lines; however , we have sequenced their mitochondrial and nuclear DNA for polymorphisms and find unique sequences which we use for genotyping our cultures ( unpublished data ) . BJ ρ0 cells were used as mitochondrial recipients within three passages of thaw for all mitochondrial transfer experiments in this work to avoid the onset of senescence . All lines were routinely tested for mycoplasma with negative results . Mitochondria were isolated from ~1 . 5 × 107 mitochondrial donor cells per mitochondrial transfer using the Qproteome Mitochondrial Isolation Kit ( Qiagen , Hilden , Germany , Cat . #37612 ) with slight alterations to the manufacturers protocol . Mitochondrial donor cells were harvested using a cell scraper ( Fisher Scientific , Cat . # 08-100-241 ) and collected in 50 mL conical tubes at approximately 6 × 107 cells per tube ( Thermo Scientific , Cat . #12-565-271 ) . Cells were pelleted by centrifugation at 500 × g for 10 min at 4°C and washed with DPBS before pelleting again by centrifugation at 500 × g for 10 min at 4°C . Cells were resuspended at 1 × 107 cells/mL in ice-cold Lysis Buffer with Protease Inhibitor Solution and incubated for 10 min at 4°C in 2 mL tubes on an end-over-end shaker . Lysates were centrifuged at 1000 × g for 10 min at 4°C and supernatant was aspirated . Pellets were resuspended in 1 . 5 mL ice-cold Disruption Buffer with Protease Inhibitor Solution and mechanical disruption was accomplished by 10 passes through a 26 G blunt ended needle ( VWR , Radnor , PA , Cat . # 89134–164 ) attached to a 3 mL syringe ( VWR , Cat . # BD309657 ) . The subsequent lysates were centrifuged at 1000 × g for 10 min at 4°C and the supernatants were transferred to new 2 mL tubes . The resultant supernatants were centrifuged again at 1000 × g for 10 min at 4°C to remove any remaining intact cells , and the supernatants were transferred to clean 1 . 5 mL tubes . These supernatants were centrifuged at 6000 × g for 10 min at 4°C and the supernatants were aspirated . The resulting mitochondrial pellets were resuspended in mitochondrial storage buffer and pelleted by centrifugation at 6000 × g for 20 min at 4°C . The isolated mitochondrial pellets were resuspended in 120 µL per transfer replicate 1× DPBS with calcium and magnesium ( Thermo Fisher Scientific , Cat . # 14040133 ) immediately prior to mitochondrial transfer and kept on ice . ~1 × 105 143BTK– ρ0 or BJ ρ0 cells were seeded into wells of 6-well dishes ~ 24 hr prior to delivery . Mitochondria isolated from ~1 . 5 × 107 HEK293T dsRed cells resuspended in 120 µL 1× DPBS with calcium and magnesium were pipetted into the culture medium of each well containing recipient cells and incubated at 37°C and 5% CO2 for 2 hr . Cells were then released from the dish using Accutase ( Thermo Fisher Scientific , Cat . # A1110501 ) and seeded into 10 cm plates for SIMR cell selection or harvested for additional analyses . A 5 V solenoid ( Sparkfun , Boulder , CO , Cat . # ROB-11015 ) is screwed into a threaded plug ( Thor Labs , Newton , NJ , Cat . # SM1PL ) and inserted into a bottom plate ( Thor Labs , Cat . # CP02T ) ( Figure 1—figure supplement 1 ) . The solenoid is regulated by a Futurlec mini board ( Futurlec , New York , NY , Cat . # MINIPOWER ) and powered by a MEAN WELL power supply ( MEAN WELL , New Taipei City , Taiwan , Cat . # RS-35–12 ) . Optomechanical assembly rods ( Thor Labs , Cat . # ER3 ) are inserted into the bottom plate . The middle and top plates ( Thor Labs , Cat . # CP02 ) are threaded through the assembly rods . The middle plate is fitted with a retaining ring , which supports an aluminum washer ( outer diameter , 25 mm; inner diameter , 10 mm ) . The middle plate is secured along the assembly rods using the included screws . The retaining ring is adjusted such that the top surface of the washer is at the same height as the piston surface in its retracted state . A flexible PDMS ( 10:1 ratio of Part A base: Part B curing agent ) ( Fisher Scientific , Cat . #NC9644388 ) reservoir consisting of a bottom layer ( 25 mm diameter , 0 . 67 mm height ) bonded to an upper ring ( outer diameter , 25 mm; inner diameter , 10 mm; height , 1 . 30 mm ) is placed on top of the washer . This reservoir can contain up to ~120 µL of liquid . To perform MitoPunch delivery , a 3 µm membrane transwell insert ( Corning , Cat . # 353181 ) seeded with 1 × 105 adherent cells is lowered through the top plate and rested atop one retaining ring . The insert is secured to the top plate by an additional retaining ring . This assembly is lowered until the base of the insert contacts the top surface of the PDMS reservoir and is secured in place with screws to form a tight seal . In addition , a variable voltage version of this device based on the same principles with identical delivery procedures as MitoPunch , but with tunable plunger acceleration achieved by varying actuator voltage , was engineered by ImmunityBio and is available upon request to the corresponding author . Optimal MitoPunch delivery voltage for individual cell lines is determined empirically by performing a voltage-response curve in technical triplicate across a range of voltages from 1 V to 5 V using the piston acceleration control software . Filter inserts with 3 µm pores ( Corning , Cat . # 353181 ) are placed in wells of a 12-well dish . 1 . 5 mL warm uridine supplemented medium is dispensed in the wells outside of the filter insert , and 1 × 105 adherent cells suspended in 0 . 5 mL warm uridine supplemented medium are seeded within the filter inserts and placed in a humidified incubator maintained at 37°C and 5% CO2 1 day prior to mitochondrial delivery . Following mitochondrial isolation , the MitoPunch apparatus is sterilized with 70% ethanol and entered into the biological safety cabinet and an autoclaved PDMS reservoir is placed in the device as indicated in Figure 1—figure supplement 1 . The PDMS reservoir is washed 3× with 120 µL sterile DPBS with calcium and magnesium after being set in the MitoPunch apparatus . 120 µL mitochondrial suspension from ~1 × 107 donor cells in DPBS with calcium and magnesium is loaded into the PDMS reservoir . Mitochondrial transfer is performed by securing the seeded membrane to the PDMS reservoir and actuating the solenoid for 3 s . The mechanical plunger strikes the middle of the PDMS chamber , displacing the base layer by ~1 . 3 mm . This displacement pressurizes the mitochondrial suspension and propels it through the membrane and into the cells ( Figure 1B ) . Once the solenoid has returned to its starting position , the insert is removed from the apparatus , placed back in the 12-well dish in its original medium , and incubated at 37°C and 5% CO2 for 2 hr . Cells were then released from the dish using Accutase and seeded into 10 cm plates for SIMR cell selection or harvested for additional analyses . Following MitoPunch mitochondrial transfer and 2 hr incubation , medium is aspirated from within the transwell filter with care taken not to disrupt the cells on the membrane , and then from outside and underneath the filter insert . The well and insert are washed 1× with DPBS ( 0 . 5 mL inside the insert and 1 mL outside the insert ) with DPBS aspirated as before . Cells are released from the membrane by 5 min incubation at 37°C and 5% CO2 with Accutase ( 0 . 5 mL inside the insert and 1 mL outside the insert ) . Following incubation , the cells are suspended in the Accutase within the filter insert using a P1000 pipette , being careful not to puncture the PET membrane , and directly pipetted into 10 cm plates with 10 mL warm uridine supplemented medium . As described previously ( Caicedo et al . , 2015 ) , 1 × 105 recipient cells were seeded in each well of a 6-well dish and incubated at 37°C and 5% CO2 overnight . Mitochondrial isolate from ~1 × 107 donor cells suspended in 1× DPBS with calcium and magnesium was pipetted into the well and the plate was centrifuged at 1500 × g for 15 min at 4°C . Cells were removed from the centrifuge and incubated for 2 hr at 37°C and 5% CO2 before being centrifuged a second time at 1500 × g for 15 min at 4°C . Cells were then released from the dish using Accutase and seeded into 10 cm plates for SIMR cell selection or harvested for additional analyses . The pressure generated by the MitoCeption method was estimated by calculating the force exerted per unit area of the cell membrane during centrifugation . The force induced by the centrifugation of a single mitochondrion on the cell membrane was equal to the centripetal force of the mitochondria under the acceleration of 1500 × g minus the buoyancy force , Fcentrifugation=mmito- mwaterawhere mmito and mwater are the mass of mitochondria and water , and a is the acceleration rate of centrifugation . The equivalent pressure induced by mitochondria during centrifugation was approximated byp=FcentrifugationS=mmito-mwateraS= ρmito-ρwaterVaS≈ρmito-ρwateradwhere ρmito ( 1 . 1 g/cm3 ) and ρwater ( 1 . 0 g/cm3 ) are the density of mitochondria and water , V and S are the volume and cross-sectional area of mitochondria , and d is the thickness of a mitochondrion ( ~1 µm ) . Using values for the geometry and properties of a mitochondrion , the pressure induced by MitoCeption centrifugation was ~1 . 6 Pa . The finite element method ( COMSOL Inc , Burlington , MA , Multiphysics 5 . 3 ) was used to simulate the pressure inside the MitoPunch PDMS chamber . We constructed the simulation geometry according to real device dimensions . Piston movement was applied as initial displacement in the y direction . Considering the incompressibility of the aqueous medium inside the PDMS chamber , the volume of the chamber was maintained constant while solving for the stress distribution of all the materials . Mitochondrial recipient and vehicle delivery control 143BTK– ρ0 cells were grown in complete medium supplemented with 50 mg/L uridine for 3 days following mitochondria or vehicle transfer . After 3 days , the medium was changed to SIMR selection medium ( complete medium with 10% dialyzed FBS ( Life Technologies , Carlsbad , CA , Cat . # 26400–044 ) ) and medium was exchanged daily . After the vehicle delivery control sample died and clones emerged on mitochondrial transfer plates ( ~7 days SIMR selection medium ) , clones were isolated using cloning rings or plates were fixed and stained with crystal violet for counting . Mitochondrial recipient and vehicle delivery control BJ ρ0 and B16 ρ0 cells were grown in complete media supplemented with 50 mg/L uridine for 3 days following mitochondria transfer . After 3 days , the medium was changed to SIMR selection medium and exchanged daily . On day 5 post-delivery , cells were shifted to galactose selection medium ( glucose-free , galactose-containing medium [DMEM without glucose , Gibco , Cat . # 11966025] supplemented with 10% dialyzed FBS and 4 . 5 g/L galactose [Sigma-Aldrich , Cat . #G5388-100G] ) . After the vehicle delivery control sample died and clones emerged on mitochondrial transfer plates ( ~36 hr in galactose selection medium ) , clones were isolated using cloning rings or plates were fixed and stained with crystal violet for counting . Media was aspirated from 10 cm plates before fixation with 1 mL freshly diluted 4% paraformaldehyde in 1× DPBS for 15 min at RT . Fixative was removed and 1 mL 0 . 5% w/v crystal violet solution ( Thermo Fisher Scientific , Cat . # C581-25 ) dissolved in 20% methanol in water was applied to each plate and incubated for 30 min at RT . Crystal violet was removed and plates were washed 2× with deionized water before drying overnight at RT . Dried plates were photographed and crystal violet stained clones were counted manually using FIJI ( Schindelin et al . , 2012 ) . Mitochondria were transferred to recipient cells , which were harvested and collected in 1 . 5 mL tubes . Samples were centrifuged 5 min at 1000 × g , supernatant was aspirated , and cells were washed 3× with 0 . 5 mL 1× DPBS , pH 7 . 4 . The DPBS was aspirated and cells were fixed in 100 µL freshly diluted 4% paraformaldehyde ( Thermo Fisher Scientific , Cat . # 28906 ) for 15 min on ice . Fixative was diluted with 1 mL of 1× DPBS , pH 7 . 4 , and 5% FBS and centrifuged for 10 min at 500 × g . Supernatant was removed and cells resuspended in 1× DPBS , pH 7 . 4 , with 5% FBS . Imaging flow cytometry was performed using an ImageStream MarkII platform and analyzed using the IDEAS 6 . 2 software package ( Luminex , Austin , TX ) . 1 × 105 cells were plated in 6-well dishes with 2 mL of media on glass coverslips ( Zeiss , Oberkochen , Germany , Cat . # 474030–9000 ) ~24 hr prior to sample preparation . Medium was aspirated and samples were fixed with 0 . 5 mL freshly diluted 4% paraformaldehyde in 1× DPBS , pH 7 . 4 pipetted onto samples and incubated for 15 min at RT . Paraformaldehyde was removed and samples were washed 3× with 5 min 1× DPBS incubations . Samples were then permeabilized by 10 min RT incubation in 0 . 1% Triton-X 100 ( Sigma , St . Louis , MO , Cat . # X100 ) . Permeabilized samples were washed 3× with 1× DPBS and then incubated for 1 hr at RT with 2% bovine serum albumin ( BSA ) dissolved in 1× DPBS blocking buffer . Blocking buffer was aspirated and cells incubated for 1 hr at RT with a 1:1000 dilution of primary antibodies in 2% BSA blocking buffer against dsDNA ( Abcam , Cambridge , United Kingdom , Cat . # ab27156 ) and TOM20 protein ( Abcam , Cat . # ab78547 ) , and then washed 3× with 5 min 1× DPBS incubations . Cells were then incubated with secondary antibodies ( Invitrogen , Cat . # A31573 and A21202 ) diluted 1:100 in 2% BSA blocking buffer protected from light for 1 hr at RT . After incubation with secondary antibodies , samples were washed 3× with 5 min 1× DPBS incubations and mounted on microscope slides . To mount , samples were removed from the 6-well dish and rinsed by dipping in deionized water , dried with a Kimwipe , and mounted using ProLong Gold Antifade Mountant with DAPI ( Invitrogen , Carlsbad , CA , Cat . # P3691 ) or ProLong Glass Antifade Mountant with NucBlue Stain ( Thermo Fisher Scientific , Cat # P36985 ) on microscope slides ( VWR , Cat . # 48311–601 ) . Samples were dried at RT protected from light for 48 hr prior to confocal imaging with a Leica SP8 confocal microscope ( Leica , Wetzlar , Germany ) and later analyzed with either LAS X Lite 3 . 7 . 1 . 21655 ( Leica ) for two-dimensional image preparation or Imaris File Converter 9 . 5 . 1 ( Oxford Instruments , Abingdon , United Kingdom ) and Imaris Viewer 9 . 5 . 1 ( Oxford Instruments ) for Z-stack analysis . To perform confocal imaging on cells immediately following mitochondrial transfer , 1 × 105 cells were plated in 6-well dishes with 2 mL of media on glass coverslips for coincubation and MitoCeption or seeded onto 12-well filter inserts as described above for MitoPunch ~24 hr prior to delivery . Immediately prior to mitochondrial transfer , coincubation and MitoCeption samples were stained with 1× CellMask Green PM ( Molecular Probes , Eugene , OR , Cat . # C37608 ) diluted in warm medium for 10 min and washed twice in DPBS , and MitoPunch samples were stained with 5 µg/mL Alexa Fluor 488 conjugated Wheat Germ Agglutinin ( Invitrogen , Cat . # W11261 ) diluted in warm media for 10 min and washed twice in DPBS . Following delivery , culture medium was removed and 1 mL freshly diluted 4% paraformaldehyde in 1× DPBS , pH 7 . 4 , was pipetted onto samples and incubated for 15 min at RT . Paraformaldehyde was aspirated and samples were washed 3× with 1× DPBS , pH 7 . 4 . Samples were further washed with DPBS 3× with 5 min RT incubation per wash . MitoPunch filters were removed from the plastic insert using an inverted P1000 pipette tip . Samples were mounted and imaged as described above . OCR measurements were performed using a Seahorse XFe96 Extracellular Flux Analyzer ( Agilent , Santa Clara , CA ) . 2 × 104 cells were seeded into each well of a V3 96-well plate ( Agilent , Cat . # 101085–004 ) and cultured 24 hr before measuring OCR . The Agilent Seahorse mitochondrial stress test was used to quantify OCR for basal respiration and respiration following the sequential addition of the mitochondrial inhibitors oligomycin , carbonyl cyanide-p-trifluoromethoxyphenylhydrazone ( FCCP ) , and antimycin A . Data were analyzed using the Wave 2 . 6 . 2 software package ( Agilent ) . Cells ( 1 × 105 ) were plated for delivery and incubated overnight . Media was changed to FluorBrite DMEM media ( ThermoFisher Scientific , Cat . # A1896701 ) with 3 µM PI ( Thermo Fisher Scientific , Cat . # P1304MP ) immediately before transfer . MitoCeption and coincubation were carried out as described above , and MitoPunch was performed with PI FluorBrite medium loaded into the PDMS reservoir and incubated for 15 min at 37°C and 5% CO2 . All samples were washed with 1× DPBS and collected using Accutase . Samples were collected in flow cytometry tubes and centrifuged 5 min at 500 × g . Samples were washed with 1× DPBS with 5% FBS three times and analyzed on a BD Fortessa flow cytometer ( BD Biosciences , San Jose , CA ) and data were processed using FlowJo 10 . 6 . 2 ( BD Biosciences ) . All information pertaining to experimental replication are found in the figure legends . Mitochondrial transfer experiments were performed in technical triplicate to enable calculation of standard deviation unless otherwise indicated , and oxygen consumption measurements were collected in technical quadruplicate or quintuplicate indicated in the legend of Figure 4 . Investigators were blinded for SIMR colony counting analysis . All column heights represent the mean of technical triplicate results unless noted otherwise . All error bars in this manuscript represent standard deviation of three technical replicates unless otherwise specified in the figure legend .	Mitochondria are specialized structures within cells that generate vital energy and biological building blocks . Mitochondria have a double membrane and contain many copies of their own circular DNA ( mitochondrial DNA ) , which include the blueprints to create just thirteen essential mitochondrial proteins . Like all genetic material , mitochondrial DNA can become damaged or mutated , and these changes can be passed on to offspring . Some of these alterations are linked to severe and debilitating diseases . Both the double membrane of the mitochondria and their high number of DNA copies make treating such diseases difficult . A successful therapy must be capable of correcting almost every copy of mitochondrial DNA . However , the multiple copies of mitochondrial DNA create a problem for genetic research as current techniques are unable to reliably introduce particular mitochondrial mutations to all types of human cells to investigate how they may alter cell function . Sercel , Patananan et al . have developed a method to deliver new mitochondria into thousands of cells at the same time . This technique , called MitoPunch , uses a pressure-driven device to propel mitochondria taken from donor cells into recipient cells without mitochondrial DNA to reestablish their function . Using human cancer cells and healthy skin cells that lack mitochondrial DNA , Sercel , Patananan et al . showed that cells that received mitochondria retained the new mitochondrial DNA . The technique uses readily accessible parts , meaning it can be performed quickly and inexpensively in any laboratory . It further only requires a small amount of donor starting material , meaning that even precious samples with limited material could be used as mitochondrial donors . This new technique has several important potential applications for mitochondrial DNA research . It could be used in the lab to create large numbers of cell lineswith known mutations in the mitochondrial DNA to establish new systems that test drugs or probe the interaction between mitochondrial and nuclear DNA . It could be used to study a broad spectrum of biological questions since mitochondrial function is essential for several processes required for life . Critically , it could also be used as a starting point to develop next-generation therapies capable of treating inherited mitochondrial genetic diseases in severely affected patients .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "tools", "and", "resources" ]	2021	Stable transplantation of human mitochondrial DNA by high-throughput, pressurized isolated mitochondrial delivery
Although individuals of many species inexorably age , a number of observations established that the rate of aging is modulated in response to a variety of mild stresses . Here , we investigated how heat stress promotes longevity in yeast . We show that upon growth at higher temperature , yeast cells relax the retention of DNA circles , which act as aging factors in the mother cell . The enhanced frequency at which circles redistribute to daughter cells was not due to changes of anaphase duration or nuclear shape but solely to the downregulation of the diffusion barrier in the nuclear envelope . This effect depended on the PKA and Tor1 pathways , downstream of stress-response kinase Pkc1 . Inhibition of these responses restored barrier function and circle retention and abrogated the effect of heat stress on longevity . Our data indicate that redistribution of aging factors from aged cells to their progeny can be a mechanism for modulating longevity . Many cell types divide asymmetrically to generate a naive daughter cell that renews the division potential of the lineage , and a committed daughter cell that progresses toward differentiation and generally shows a limited division potential ( Knoblich , 2010; Chen et al . , 2016; Ouellet and Barral , 2012 ) . This is the case for many stem cells , which have the dual function of maintaining an eternal division potential and of generating differentiating daughters that eventually integrate themselves both structurally and functionally into organs ( Chen et al . , 2016; Fisher and Sozzani , 2016 ) . Accordingly , while the stem cell remains young during most of the lifespan of the individual , differentiating daughters age and need to be replaced over time . Current knowledge indicates that aging in metazoans follows a progressive loss of stem cells proliferation , and hence a loss in the regeneration potential of the organs ( Van Zant and Liang , 2003; Beerman and Rossi , 2015; Ahmed et al . , 2017 ) . However , how aging takes place at the cellular level is not well understood . Particularly , we do not yet understand how asymmetric division generates both one naive but rejuvenated and one committed but aging daughter cell . The unicellular fungus Saccharomyces cerevisiae is an excellent model for studying this process ( Higuchi-Sanabria et al . , 2014; Denoth Lippuner et al . , 2014 ) . Indeed , these cells proliferate through budding small , rejuvenated daughter cells from the surface of the larger , mother cell ( Mortimer and Johnston , 1959; Hartwell and Unger , 1977; Kennedy et al . , 1994; Henderson and Gottschling , 2008 ) . Strikingly , with each daughter produced , the mother cell ages and progressively loses its division potential until it eventually stops proliferating and dies . This process is called replicative aging and the replicative lifespan , that is , the number of daughters a mother cell generates before dying , is limited , reaching about 25 generations in average for haploid wild-type cells ( Henderson and Gottschling , 2008; Denoth Lippuner et al . , 2014 ) . Beyond limiting the lifespan , yeast aging also manifests itself through a number of additional traits , such as the formation of protein aggregates ( Aguilaniu et al . , 2003; Erjavec et al . , 2007; Hill et al . , 2014; Saarikangas and Barral , 2015 ) , the neutralization of the vacuolar pH ( Hughes and Gottschling , 2012; Henderson et al . , 2014 ) , the fractionation of mitochondrial organization ( Hughes and Gottschling , 2012 ) and the decreased sensitivity of the cell to signaling pheromone ( Smeal et al . , 1996; Caudron and Barral , 2013; Schlissel et al . , 2017 ) reviewed in Denoth Lippuner et al . ( 2014 ) . In contrast , the daughter cells reset their vacuolar pH , mitochondrial organization , pheromone response and division potential . They then become mother cells themselves; they start budding-off daughters and aging . The progressive decline of cellular fitness with age is thought to be driven by the retention and accumulation of so-called aging factors in the mother cell . Three types of aging factors have been described . First , plasma-membrane proteins such as the proton-exporter Pma1 and several multi-drug transporters remain in the mother cell as it divides and contribute to its fitness decay ( Eldakak et al . , 2010; Henderson et al . , 2014; Thayer et al . , 2014 ) . Second , aging yeast mother cells also form a deposit that accumulates protein aggregates ( Aguilaniu et al . , 2003; Erjavec et al . , 2007; Hill et al . , 2014; Saarikangas and Barral , 2015 ) . Cells that fail to form this aggregate are long-lived ( Hill et al . , 2014; Saarikangas and Barral , 2015 ) . Third , intra-chromosomal recombination between repeated rDNA units excise extrachromosomal rDNA circles ( ERCs ) that segregate to and accumulate in the mother cell nucleus ( Szostak and Wu , 1980; Sinclair and Guarente , 1997; Shcheprova et al . , 2008 ) . Except for the endogenous two micron plasmid , ERCs and actually all DNA circles tested so far accumulate in the mother cell with age and accelerate aging ( Murray and Szostak , 1983; Falcón and Aris , 2003 ) . Old mother cells contain up to thousand ERCs and this load , which increases exponentially with successive divisions , might be what ultimately kills the cell ( Sinclair and Guarente , 1997 ) . High-fidelity retention in the mother cell of the DNA circles and of the precursors of protein aggregation is facilitated by the formation of lateral diffusion barriers in the ER membrane and the outer nuclear membrane at the bud neck ( Luedeke et al . , 2005; Shcheprova et al . , 2008; Clay et al . , 2014; Saarikangas et al . , 2017 ) . These barriers limit exchange of membrane-proteins between mother and bud . Therefore , retention of aging factors in the mother cell relies on their anchorage into the ER-membrane . Retention of the aggregation precursors relies on their membrane attachment through the farnesylated chaperone Ydj1 ( Saarikangas et al . , 2017 ) . DNA circles attach to the nuclear envelope through the SAGA complex and nuclear pore complexes ( NPCs ) ( Shcheprova et al . , 2008; Denoth-Lippuner et al . , 2014 ) . Remarkably , yeast cells show an extended life span when subjected to mild stresses such as calorie restriction and growth at 37°C ( Shama et al . , 1998a; Shama et al . , 1998b; Swieciło et al . , 2000; Kapahi et al . , 2017 ) . Similar effects take place in organisms as distinct as nematodes , flies and mice , indicating that the regulation of longevity involves similar regulatory pathways in all these organisms , at least upon calorie restriction , namely the TOR and PKA pathways ( Steinkraus et al . , 2008; Kapahi et al . , 2010; Wasko and Kaeberlein , 2014 ) . How these regulatory pathways actually modulate ageing progression itself is largely unknown . The fact that yeast cells are able to modulate their longevity in response to environmental signals suggests that they have some control on the generation and accumulation of aging factors , or on the impact that these have on the physiology of the cell . We reasoned that one potential mechanism for increasing longevity could be the down-regulation of the retention of aging factors in the mother cell . Indeed , mutants affecting the retention of DNA circles in the mother cell are long-lived ( Shcheprova et al . , 2008; Denoth-Lippuner et al . , 2014 ) . Thus , we set here out to test whether physiological stresses affect the retention of aging factors in the mother cell , and how . In order to investigate whether the confinement of aging factors in the mother cell is affected upon conditions that promote longevity , we asked whether a model DNA circle ( Shcheprova et al . , 2008; Denoth-Lippuner et al . , 2014 ) was more likely to propagate to the bud upon heat stress or calorie restriction than under optimal growth conditions . This model DNA circle carries a centromere flanked with LoxP sites and hence , is turned into a non-centromeric circle upon expression of the Cre-recombinase and excision of the centromeric sequence . It also carries an array of repeated TetO sequences . Expression of the protein TetR , which binds to the TetO sequence , fused to GFP ( TetR-GFP ) allow the visualization of the circle as a fluorescent dot in vivo . An autonomously replicating sequence ( ARS ) promotes the replication of the DNA circle during S-phase . We transfected this model circle into cells co-expressing Cre fused to an estradiol-binding domain ( Cre-EBD ) and TetR-GFP . EBD mediates the retention of Cre in the cytoplasm until β-Estradiol is added to the growth medium ( Lindstrom and Gottschling , 2009 ) . Upon β-Estradiol treatment and consequent centromere excision ( see Materials and methods for details ) the non-centromeric DNA circles detaches from the Spindle ( Spc42-CFP marks the spindle pole bodies - SPB , Figure 1A , B ) , and no-longer segregate symmetrically , unlike their centromeric counterparts . Using this system , we asked whether cells grown in conditions of calorie restriction ( 0 . 1% glucose , 30°C ) or heat stress ( 2% glucose , 37°C ) affected the retention of the circles in the mother cell compared to cells maintained in optimal growth conditions ( 2% glucose , 30°C ) . In cells grown under optimal conditions , the circles passed very infrequently to the bud ( frequency of propagation to the bud: 0 . 04 ± 0 . 01 ) . Although calorie restriction had no effect ( propagation frequency: 0 . 04 ± 0 . 01 ) , the frequency at which individual plasmids passed to the bud was increased four folds in cells grown at 37°C ( propagation frequency: 0 . 17 ± 0 . 01 ) . This value is very similar to what we observed in cells lacking the diffusion barrier in the outer nuclear membrane ( bud6Δ mutant cells , propagation frequency: 0 . 13 ± 0 . 01; Shcheprova et al . ( 2008 ) ; Denoth-Lippuner et al . , 2014; Figure 1B , C ) . Thus , these data suggested that , at least under heat stress , cells might relax their ability to confine DNA circles in the mother cell . Three parameters affect the retention of DNA circles in the yeast mother nucleus . First , increased anaphase duration leaves more time to circles to diffuse from the mother into the bud and therefore promotes their propagation ( Gehlen et al . , 2011 ) . Second , a failure to efficiently narrow down the median constriction of the dividing nucleus leaves opportunity to nucleoplasmic , but not membrane attached , material to exchange between mother and bud parts of the nucleus ( Gehlen et al . , 2011; Boettcher et al . , 2012 ) . Third , the presence of a diffusion barrier in the outer nuclear membrane restricts the movement of circles through the bud neck , provided that they are attached to nuclear pore complexes ( NPCs , Shcheprova et al . , 2008; Denoth-Lippuner et al . , 2014 ) . Thus , we wondered which of these parameters is affected in cells grown under heat stress . First , we characterized the effect of heat on both anaphase duration and changes in nuclear morphology . In a strain carrying our model DNA circle , we tagged the outer nuclear membrane protein Nsg1 with GFP . Using this marker , we quantified the duration of nuclear division ( Figure 2A , Figure 2—figure supplement 1A ) and the morphology of the nuclei ( length of longitudinal axis and diameter of their scission constriction ) throughout anaphase ( Figure 2—figure supplement 2A ) . We defined anaphase as the time window starting with the entry of a nuclear lobe into the bud and finishing with the completion of karyokinesis ( the two separate nuclei move slightly toward each other , Figure 2A , Figure 2—figure supplement 1A ) . Compared to optimal growth conditions , exposure to heat shortened anaphase duration by about 20% ( from 19 . 7 ± 0 . 3 to 15 . 7 ± 0 . 5 min , Figure 2B ) . The length of the nucleus was increased in the first 100 seconds of anaphase but the scission constriction at the bud neck was unaffected ( Figure 2—figure supplement 2B–D ) . If anything , an increased length of the nucleus would reduce , and not increase , the DNA circle propagation to the bud . Thus , at first sight the increased propagation of circles in populations of yeast cells grown under mild heat stress was not due to an overall prolongation of anaphase duration or change in nuclear morphology . However , the morphology of the nucleus strongly changes between early and late anaphase . To address whether the duration of one of the stages of anaphases was particularly affected and how this could influence DNA circle exchange between mother and daughter cell , we studied when exactly during anaphase DNA circles are exchanged between the mother and daughter part of the nucleus . Based on this , we then examined whether heat stress prolonged that particular stage . We took advantage of the strain described above and followed now simultaneously nuclear division and the segregation of the DNA circles ( Figure 2—figure supplement 1B ) . In movies of wild type cells grown at 30°C in medium containing 2% glucose , we determined the propagation flux of DNA circles through the bud neck ( number of passages to the bud per minute ) and compared the values obtained during early versus late anaphase . This frequency was six to eight folds higher in early anaphase than in late anaphase ( 0 . 03 ± 0 . 01 and 0 . 004 ± 0 . 003 passage per minute , respectively , Figure 2C ) . Thus , DNA circles exchange between mother and bud essentially during early anaphase . An extension of early anaphase duration at the cost of late anaphase could increase the frequency at which circles pass to the bud , while maintaining a constant duration of the total anaphase . Therefore , using the same movies , we determined the time cells spent in early anaphase . Remarkably , early anaphase was shorter , certainly not longer , in cells grown at 37°C , compared to 30°C ( 3 . 4 ± 0 . 1 versus 3 . 9 ± 0 . 14 min , Figure 2D ) . The duration of late anaphase was reduced as well ( 12 . 9 ± 0 . 46 versus 14 . 5 ± 0 . 46 min , Figure 2—figure supplement 1C ) . Furthermore , early anaphase lasted the longest upon calorie restriction ( 5 . 2 ± 0 . 36 min , Figure 2D ) , although this condition did not affect the retention of DNA circles ( Figure 1C ) . Thus , the increased propagation of the DNA circle to the bud in cells grown at 37°C was not due to an extension of the duration of early anaphase . The high-fidelity retention of DNA circles into the mother cell requires their anchorage to NPCs , in order to subject them to confinement by the diffusion barrier in the nuclear envelope ( Shcheprova et al . , 2008; Denoth-Lippuner et al . , 2014 ) . Thus , we tested whether DNA circle-NPC interaction was affected in cells grown at 37°C . The co-localization of DNA circles with NPCs was measured as previously described ( Denoth-Lippuner et al . , 2014 ) . Intensity profiles of Nup82 labeled with 3x super folder GFP ( Nup82-3x sfGFP ) were obtained along the nuclear envelope in equatorial focal sections of the nuclei containing a single mCherry-labeled DNA circle at the rim ( Figure 3A ) . Nup82-3x sfGFP intensity profiles from at least 40 cells were aligned relative to the maximum intensity of the DNA circle and averaged . When the circle anchors to the NPC , a local Nup82-3x sfGFP intensity peak correlates with the intensity peak of the DNA circle ( Figure 3B , C ) . If the DNA circle-NPC interaction is compromised , for example by the knock-out of the acetyltransferase Gcn5 in the SAGA complex ( Denoth-Lippuner et al . , 2014 ) , then the Nup84 fluorescence is not in phase with the DNA circle fluorescence and the signal correlation is lost ( Figure 3—figure supplement 1 ) . Interestingly , the correlation between DNA circle and NPC remained intact upon heat shock . Thus , the increased frequency of circle propagation into the daughter upon growth at higher temperature is not due to circles detachment from NPCs . Together , these data indicated that the increased frequency at which circles propagate to daughter cells in cells grown at 37°C was not due to changes in nuclear morphology , an increase of anaphase duration or circle detachment from NPCs . Therefore , we envisioned the possibility that it might be due to the diffusion barrier , normally present in the outer nuclear membrane , being impaired in cells grown at 37°C . Thus , we probed the diffusion barrier in the outer nuclear membrane using Fluorescence Loss In Photobleaching ( FLIP [Bolognesi et al . , 2016] ) in cells grown at 30°C or 37°C in medium containing 2% or 0 . 1% glucose . In cells expressing the nucleoporin Nup49 tagged with GFP ( Nup49-GFP , nuclear membrane reporter ) , a small region of the mother part of anaphase nuclei was constantly photo-bleached over time and the fluorescence decay in both mother and daughter nuclear compartments was measured ( Figure 4A , B ) . The ratio between the time it took to lose 25% of the signal in the non-bleached compartment ( bud ) to the time it took to lose 25% of the signal in the bleached compartment ( mother ) was computed and is defined as the Barrier index ( BI , Shcheprova et al . , 2008 , Figure 4C ) . A weaker membrane compartmentalization between mother and bud results in a faster signal decay in the bud , thus a weaker barrier ( low BI ) . These experiments established that barrier strength in the nuclear membrane was reduced by roughly half in cells grown at 37°C compared to those grown at 30°C ( BI = 24 . 6 ± 2 . 5 vs 41 . 9 ± 4 . 1; Figure 4D ) . As a positive control , the barrier index in the bud6Δ mutant cells grown at 30°C , which bear strong barrier defects ( Shcheprova et al . , 2008 ) , was similarly decreased ( BI = 25 . 3 ± 3 . 2 ) . In contrast , and to our surprise , the barrier was significantly strengthened in calorie-restricted cells grown in 0 . 1% glucose ( BI = 68 . 6 ± 15 , Figure 4D ) . The diffusion barrier in the cortical ER was not affected by heat stress and calorie restriction ( Figure 4—figure supplement 1A–C ) . Thus , the effect of temperature and calorie restriction targeted specifically the diffusion barrier in the outer nuclear membrane . We conclude that both heat stress and calorie restriction affect the nuclear diffusion barrier , but in opposite manners . These results suggest that the increased propensity with which daughters inherit circles upon growth at 37°C might be due to a reduction in barrier strength . Furthermore , the measured increase of barrier strength upon calorie restriction ( Figure 4D ) might explain why these cells , while spending more time in early anaphase ( Figure 2D ) , do not segregate circles more frequently to their daughters ( Figure 1C ) . We expect that during early anaphase , the propagation flux of individual DNA circles through the bud neck ( number of passages to the bud per unit of time , independent of anaphase duration ) is increased at 37°C and decreased in calorie restricted cells , assuming that the DNA circles retention depends on the diffusion barrier . Indeed , the propagation flux in early anaphase was increased two folds in cells grown at 37°C compared to 30°C ( 0 . 06 ± 0 . 004 versus 0 . 03 ± 0 . 01 passage per minute; Figure 4E ) and decreased three folds in calorie restricted cells ( 0 . 01 ± 0 . 01 passage per minute ) . The decreased propagation flux in combination with a longer anaphase duration ( Figure 2D ) yielded an unaltered DNA circle propagation frequency in calorie restricted cells ( Figure 1C ) . The increased propagation flux in heat-stressed cells is comparable to that in the bud6Δ mutant cells ( 0 . 10 ± 0 . 02 passage per minute , Figure 4E ) . Thus , heat stress affected the permeability of the bud neck for DNA circles . We conclude that the permeability of the diffusion barrier emerged as the tightest and most direct determinant of circle retention in the mother cell . The observation that heat stress had a specific effect on the nuclear barrier and not the cortical ER barrier , hinted toward a regulated process instead of a general effect of temperature for example on membrane fluidity . To address this possibility , we investigated whether stress response pathways regulate barrier strength . We particularly focused on the possible role of the cell wall integrity pathway ( CWI ) , which is activated upon heat stress . At the top of this pathway , the Pkc1 kinase responds to plasma-membrane and cell wall stress and activates a MAP-kinase cascade to promote cell wall remodeling and repair ( Levin , 2005 , Figure 5A ) . Thus , we asked whether constitutively activating this pathway affected the strength of the barrier in the nuclear membrane . To this end , we introduced the constitutively active allele of the PKC1 gene , PKC1-R398P ( Nonaka et al . , 1995 ) in our wild-type strain expressing our reporter nucleoporin ( Nup49-GFP ) and applied FLIP to determine its effect on barrier strength in cells grown at 30°C . Supporting the idea that Pkc1 regulates the nuclear barrier , cells expressing this constitutively activated form of Pkc1 showed a much weaker barrier compared to wild type cells ( BI 19 . 1 ± 2 versus 41 . 9 ± 4 . 1 , Figure 5B , Figure 5—figure supplement 1 ) and a significant reduction in DNA circle retention ( Figure 5C ) . Strikingly , this effect did not require the MAP-kinase cascade downstream of Pkc1 . Constitutive activation of the MAP kinase kinase kinase Bck1 , using the BCK1-20 allele , did not change barrier strength ( BI = 38 . 4 ± 7 . 1 , Figure 5B ) , compared to wild-type cells . Furthermore , inactivating the MAP kinase Slt2 , which acts most downstream in the CWI pathway ( Figure 5A ) , did not revert the effect of the PKC1-R398P mutation ( BI = 17 . 9 ± 2 . 1 in the PKC1-R398P slt2∆ double mutant cells , Figure 5B , Figure 5—figure supplement 1 ) . Finally , instead of promoting barrier strength , the slt2Δ mutation by itself tends to slightly weakening the barrier strength ( BI = 28 . 5 ± 4 . 2 ) , fitting with the cell wall defects observed in these cells . Thus , these data indicate that activation of the Pkc1 kinase inhibits the diffusion barriers in the nuclear envelope and that this effect depends on a distinct signaling branch than the MAP-kinase cascade . These data support the conclusion that the weakening of the diffusion barrier in heat-treated cells corresponds to a regulatory response of the cells and not a direct effect of temperature on barrier structure or function . Among others , the two kinases PKA and Tor1 also contribute to the cellular responses to stress ( Causton et al . , 2001; Castells-Roca et al . , 2011; Loewith and Hall , 2011; Pautasso and Rossi , 2014 ) . They also promote cell growth in response to nutrients availability and are down-regulated in response to calorie restriction ( Thevelein and de Winde , 1999; Loewith and Hall , 2011 ) , Figure 5A ) . Based on this and the fact that the cells increase barrier strength upon calorie restriction ( Figure 4D ) , we tested whether PKA and Tor1 inhibited the nuclear diffusion barrier . Constitutive activation of PKA through deleting the BCY1 gene that encodes its inhibitory subunit ( Toda et al . , 1987 ) significantly reduced barrier strength compared to wild type cells grown in the same conditions ( BIn19 . 9 ± 1 versus 41 . 9 ± 4 . 1 , Figure 5D , Figure 5—figure supplements 1–2 ) . Likewise , expression of the constitutive active allele of Tor1 , TOR1-A1957V ( Reinke et al . , 2006 ) had the same effect ( BIn = 20 . 75 ± 1 . 9 , Figure 5D , Figure 5—figure supplements 1–2 ) . In reverse , partial inhibition of Tor1 by addition of rapamycin ( 200 ng/ml for 16–18 h ) , a TORC1-specific inhibitor , to the growth medium ( Brunn et al . , 1996; Loewith and Hall , 2011 ) increased barrier strenght , even more than lowering glucose concentration ( BIn up to 107 ± 24 and 68 . 6 ± 15 , respectively Figure 5D , Figure 5—figure supplement 1 ) . These data indicate that the Tor1 and PKA kinases act in pathways inhibiting the nuclear diffusion barrier in rich medium and perhaps in response to stress , similarly to Pkc1 . To investigate whether Tor1 and PKA activity contribute to barrier weakening in response to heat stress , we took advantage of calorie restriction inhibiting both PKA and Tor1 ( Steinkraus et al . , 2008; Fontana et al . , 2010 ) . Thus , we tested whether cells grown in low glucose , that is , with low PKA and Tor1 activity , were still able to repress barrier function in response to heat stress . Whereas cells grown at 37°C in 2% glucose decreased the barrier strength compared to wild-type cells grown in the same medium at 30°C ( BI = 24 . 6 ± 2 . 5 , versus 41 . 9 ± 4 . 1 , Figure 5E ) , calorie restriction abrogated this effect . In fact , cells grown at 37°C in the calorie restricting medium ( 0 . 1% glucose ) formed a stronger barrier ( Figure 5E , Figure 5—figure supplements 1–2 ) . We could not measure the BI in these cells , because in average the fluorescence failed to decay significantly in the bud . Calorie restriction had a similar effect on the PKC1-R398P mutant cells grown at 30°C . Here again , calorie restriction did not simply restore the barrier of the mutant cells , but enhanced it compared to growth under optimal conditions ( BI = 100 ± 40 in 0 . 1% glucose versus 17 . 9 ± 2 . 1 in 2% glucose , Figure 5E , Figure 5—figure supplements 1–2 ) . Interestingly , PKC1-R398P mutant cells grown at 30°C ( in 2% glucose ) and treated with rapamycin ( 200 ng/ml ) , restored the barrier strength to levels similar to those observed in wild-type cells ( BI = 42 . 2 ± 7 . 7 , Figure 5E , Figure 5—figure supplements 1–2 ) . These data suggest that 37°C and Pkc1 regulate barrier strength upstream of Tor1 and PKA , and that Tor1 and/or PKA activity is required in order to repress barrier function in response to heat stress . Strikingly , calorie restriction also restored the barrier in the bud6Δ mutant cells compared to growth in 2% glucose ( BI = 57 ± 9 . 9 versus 25 . 3 ± 3 . 2 , respectively; Figure 5E , Figure 5—figure supplements 1–2 ) . Thus , the barrier defect of bud6Δ mutant cells is at least in part suppressed by inhibiting the Tor1 and PKA pathways . Collectively , our data indicate that the strength of the nuclear diffusion barrier is a regulated trait under the control of Tor1 and PKA . We next sought to directly test whether the increased DNA circle propagation frequency observed upon heat stress is indeed due to their weaker diffusion barrier in the nuclear envelope . We reasoned that if it were the case , strengthening the barrier by calorie restriction ( as in Figure 4D ) in cells grown at 37°C should reduce the propagation of DNA circles to the bud , normally observed upon heat stress . Thus , we examined the propagation frequency of the DNA circle in cells grown at 37°C in medium containing 0 . 1% glucose . At 30°C ( 0 . 1% glucose ) the propagation frequency was 0 . 04 ± 0 . 005 , similar to the 2% glucose condition ( 0 . 04 ± 0 . 008 , Figure 6A , B ) . Interestingly , in cells grown at 37°C calorie restriction restored the propagation frequency to the levels normally observed at 30°C ( 0 . 06 ± 0 . 01 , compared to 0 . 16 ± 0 . 02 in 2% glucose , Figure 6A , B ) . In line with the reinforcing effect of calorie restriction on the barrier , bud6Δ mutant cells grown at 30°C in 0 . 1% glucose containing medium also retrieved their ability to confine the DNA circle into the mother cell ( propagation frequency = 0 . 04 ± 0 . 007 , compared to 0 . 12 ± 0 . 01 in 2% , Figure 6A , B ) . We conclude that restoring barrier strength under heat stress conditions rescues the retention of DNA circles in the mother cell , indicating that the increased propagation of DNA circles at elevated temperatures ( Figure 1C ) is indeed caused by a down regulation of the diffusion barrier . Collectively , our data indicate that heat stress , which causes lifespan extension ( Shama et al . , 1998a; Swieciło et al . , 2000 ) , relaxes the confinement of DNA circles in the yeast mother cell through weakening of the diffusion barrier . Thus , our data predicts a reduction in DNA circle content when the aged mother cells were grown at 37°C instead of 30°C . Therefore , we monitored by Southern blotting the levels of rDNA circles in aged cells , incubated at both 30°C and 37°C , as described ( Denoth-Lippuner et al . , 2014 ) . We enriched for aged yeast mother cells ( 15% of aged cells , 10 000 fold enrichment; average age of the aged cells: 16 generations old ) , using the mother enrichment program ( Lindstrom and Gottschling , 2009 ) . We extracted total DNA from young and aged cell populations . Similar amounts of DNA ( 300 ng ) were used , from young cell populations and aged cell populations with same fraction of aged cells . A 32P-labeled probe specific for the rDNA locus was used to detect the ERCs . While no DNA circles were observed in young cells , a substantial amount could be seen in the old ones . And indeed , cells cultured at 37°C showed 3 . 3-fold less ERCs than cells cultured at 30°C ( Figure 7A ) . We concluded that cells grown at 37°C indeed decrease the amount of ERCs that they accumulate , consistent with DNA circle retention being relaxed . Thus , our data opened the possibility that barrier weakening is a mechanism through which cells regulate the accumulation of DNA circle with age , and hence , their longevity in response to heat stress . We reasoned that if it where the case then restoring the diffusion barrier in heat stressed cells , using calorie restriction ( see Figure 4D ) , should reduce their longevity to the level of unstressed cells . Thus , we compared the longevity of heat stressed and calorie restricted cells , and cells subjected to both conditions at the same time , using standard micro-dissection technics and pedigree analysis . Whereas wild-type cells grown under optimal conditions showed a median lifespan of 23 generations , calorie restriction increased their median lifespan to 30 generations at 30°C ( Figure 7B ) , as reported ( Fontana et al . , 2010 ) . As published ( Shama et al . , 1998a; Shama et al . , 1998b; Swieciło et al . , 2000 ) , growth at 37°C increased the longevity to a median lifespan of 32 . 5 generations ( Figure 7B ) . Calorie restriction of cells grown at 37°C shortened their lifespan back to 27 generations ( Figure 7B ) . While the bud6Δ mutant cells grown at 30°C in rich medium showed an increased lifespan ( median lifespan = 39 ) , as reported ( Shcheprova et al . , 2008 , Figure 7C ) , calorie restriction shortened their longevity back to a value close to that of wild type cells under the same conditions ( median lifespan = 28 . 5; Figure 7C ) . Thus , aging of the bud6∆ mutant and heat stressed cells was restored upon strengthening the diffusion barrier back to wild type or higher levels . This is consistent with the idea that heat stress increases the longevity of the mother cell at least in part through the inhibition of the diffusion barrier in the nuclear envelope and consequently through releasing aging factors , such as DNA circles , to their progeny . Collectively , our data suggest that the redistribution of age load , such as DNA circles , between mother and daughter is a mechanism for modulating the longevity of the cell in response to stresses such as heat ( Figure 7D ) . The efficient retention of DNA circles in the mother cell during mitosis is a key determinant of the replicative lifespan of S . cerevisiae cells . How cells achieve this retention , has been a topic of debate ( Gehlen et al . , 2011; Khmelinskii et al . , 2011; Ouellet and Barral , 2012; Denoth-Lippuner et al . , 2014 ) . Our data provide evidence confirming the nuclear diffusion barrier as being a key parameter contributing to DNA circle retention and reveals that the strength of the barrier is regulated . First , we observed that the barrier loses permeability in cells grown at 37°C , increasing propagation of DNA circles to the bud . Second , the barrier permeability is restored in these cells upon calorie restriction which leads to full restoration of circle retention in the mother cell . These results are consistent with the observation that the attachment of DNA circles to NPCs is required for their efficient retention in the mother cell , in a barrier-dependent manner ( Shcheprova et al . , 2008; Denoth-Lippuner et al . , 2014 ) . Importantly , heat stress did not impair NPC-DNA circle association , indicating that it does not affect DNA circle propagation via their detachment from the nuclear envelope . Furthermore , our data indicate that regulation of barrier permeability under heat stress results in the redistribution of the age load during mitosis , resulting in a longer life span for the mother cell . In other words , the yeast cell is able to modulate longevity by regulating the diffusion barrier . We observed a two-fold reduction of barrier strength during heat stress ( Figure 4D ) and a roughly four-fold increase of DNA circle propagation to the daughter cells at 37°C ( Figure 1C ) . Effectively , this represents a decrease of circle retention from 96% to 83% for what concerns the model DNA circle . Is such an apparently mild decrease sufficient to explain the lifespan increase that we observe ( Figure 7B ) ? Mathematical modeling studies predicted that a retention probability above 99% is required to simulate the aging curves experimentally established for wild type cells ( Gillespie et al . , 2004 ) . This is in part due to the fact that the replication origin on the rDNA circles is activated only in about 60% of the cell cycles ( Gillespie et al . , 2004 ) , such that the loss of ERCs to the daughter cells is inefficiently compensated by replication of the DNA circles . Accordingly , previous studies confirmed that a slight reduction of this retention efficiency has a strong effect on the longevity of the cell ( Shcheprova et al . , 2008 ) reviewed in Denoth Lippuner et al . ( 2014 ) . This fits with the observation here that ERC levels decrease with roughly 3 . 5-fold in 16 generations old mother cells grown at 37°C compared to 30°C , despite the apparently high retention efficiency of 83% ( Figure 7A ) . Our study establishes that the modulation of barrier strength is not merely a direct consequence of nutrients limitation or temperature increase on barrier formation or stability , but a process elicited by specific regulatory pathways in response to environmental disturbances . The effect of heat stress on barrier permeability could be mimicked by activation of Pkc1 and prevented by calorie restriction . Remarkably , inhibition of both PKA and Tor1 ( via calorie restriction ) in cells either grown at 37°C or expressing a constitutively active form of Pkc1 prevented the weakening of nuclear barrier that is normally observed under these conditions . These data suggest that Pkc1 might modulate barrier strength through Tor1 and PKA . To our knowledge , this is first evidence that Pkc1 might regulate TORC1 and PKA , revealing an intriguing link between these pathways that now needs to be substantiated . Besides S . cerevisiae ( Shama et al . , 1998a; Shama et al . , 1998b; Swieciło et al . , 2000 ) , heat stress promotes longevity in several other species , ranging from flies ( Smith , 1958; Lithgow et al . , 1995; Khazaeli et al . , 1997; Le Bourg et al . , 2001 ) to worms ( Butov et al . , 2001; Cypser and Johnson , 2002; Gems and Partridge , 2008; Rodriguez et al . , 2012 ) and rats ( Holloszy and Smith , 1986 ) . Heat stress is also a protein denaturant agent . Prokaryotic and eukaryotic species exposed to elevated temperatures induce the expression of heat shock proteins ( HSPs ) that are well conserved among species and help the cell maintain proteostasis ( Lindquist and Craig , 1988; Morano et al . , 2012 ) . In budding yeast , several HSPs are induced upon heat stress ( Boy-Marcotte et al . , 1999 ) and are implicated in refolding heat-denatured proteins , preventing their aggregation or , if severely damaged , targeting them for degradation ( Hilt and Wolf , 2006; Craig et al . , 1994; Parsell et al . , 1994; Glover and Lindquist , 1998 ) . Accumulation of damaged yeast ( Aguilaniu et al . , 2003; Erjavec et al . , 2007; Hill et al . , 2014; Saarikangas and Barral , 2015 ) . Additionally , the lifespan extension observed in S . cerevisiae upon heat stress , requires the heat shock protein Hsp104 ( Shama et al . , 1998a; Shama et al . , 1998b ) . HSPs expression decreases with ageing in mammals , flies and worms ( Kurapati et al . , 2000; Wyttenbach et al . , 2002; Hsu et al . , 2003 ) and positively correlates with maximum longevity in mammals and birds ( Salway et al . , 2011 ) . Heat stress was also proposed to induce the oxidative stress response , thus protecting cells against Reactive Oxygen Species ( ROS ) , also associated with ageing in S . cerevisiae ( Morano et al . , 2012 ) . These observations have suggested that heat stress promotes longevity via induction of “repair” and/or a “clearance” mechanisms that refold denatured molecules , prevent their aggregation and target damaged/toxic species for degradation ( Estruch , 2000; Verbeke et al . , 2001; Calderwood et al . , 2009 ) . In striking contrast , we show here that a potent mechanism to promote longevity is the opening of the diffusion barrier in the nuclear envelope and the ‘dumping’ of at least some of the mother’s burden onto their daughters . Barrier weakening is clearly not the mechanism promoting longevity in response to calorie restriction , establishing that it is not the only mechanism through which cells can modulate the effects of age . Most likely , the ‘repair’ , ‘clearance’ and ‘sharing/dumping’ mechanisms co-exist and their relative impact might vary between conditions . In any case , the fact that rescuing DNA circle confinement largely abolishes the effect of heat stress on lifespan extension indicates that at least under that condition ‘dumping’ age load onto the population is a prominent mechanism of longevity regulation . Here , we have focused on the characterization of a single type of aging factor . However , the fact that other factors , such as deposit precursors ( Saarikangas et al . , 2017 ) also depend on the diffusion barrier for their confinement in the mother cell suggest that the regulation we have uncovered might affect more than only DNA circles . Furthermore , other mechanisms might help redistribute aging factors . Our study is probably only scratching the top of a larger panel of possibilities . Indeed , the nature of the aging factors being redistributed to the progeny could depend on the nature of the physiological signals that the cell faces . In addition , the fact that calorie restriction extends the replicative lifespan of yeast cells without affecting circle confinement indicates that barrier opening is not the only mechanism of longevity modulation . The idea that stress does not lead to higher fidelity but rather less is actually quite satisfying . Indeed , the widespread idea that yeast cells generate less damage or eliminate them better when they are under stress is somehow paradoxical . It seems somewhat counterintuitive that cells would do better when they are stressed than when they are not , although we agree that there might be selection advantages to such a situation . In any case , our data open the question of why do cells activate a pathway for living longer and aging slower in response to stress , and this potentially at the cost of their progeny . Several lines of thoughts merit attention here . One possibility is that dispersing the pre-existing damage through a large population of cells allows the individual cells to better survive . Here , the gain in longevity would merely be a contingent effect . However , another line of thoughts could be that older yeast mother cells have acquired with age something beneficial , helping them to better cope with stress . Sharing this with their progeny might be important for the survival of the population . Remarkably , middle-age mother cells do cope better with heat than young ones ( reviewed in Denoth Lippuner et al . , 2014 ) . If the yeast mother cells carry some components that provide them with a selective advantage under stress , then opening the barrier might have two advantages at once: it might not only allow them to share this advantage with their daughters , but by allowing them to dump damage , it could also help keeping these experienced mothers alive longer . Both effects would be to the benefit of the survival of the genotype . In favor of such a model , it has been recently shown that around 30% of the genome can recombine out of the chromosomes and form DNA circles ( Møller et al . , 2015 ) . In some cases , these circles carry genes likely to promote survival in response to specific stresses , such as cupper intoxication and salt stress ( Møller et al . , 2015 ) . Relaxing barrier permeability would still allow circle accumulation in the mother , but at the same time increase the number of daughters that inherit at least few circles and can accumulate them as well . Another supportive scenario relates to the observation that protein aggregates that are retained in the mother cell can provide adaptive advantages . This is for example the case when the Whi3 mnemon aggregates in response to futile pheromone exposure and mediates adaptation of the cell to the lack of a partner ( Caudron and Barral , 2013 ) . Provided that mnemon asymmetry also depends on ER compartmentalization , opening the barrier would allow the mother to share adaptive circles and mnemons with her progeny and at the same time to soften of the pro-aging effect that these factors have on her . Whether the ‘dumping/sharing’ mechanism is conserved beyond yeast is unknown . However , it has recently become clear that diffusion barriers are conserved in metazoans ( Moore et al . , 2015; Lee et al . , 2016 ) . Furthermore , at least the diffusion barrier of neural stem cells changes strength during development and with age ( Moore et al . , 2015 ) . Thus , barrier regulation might be a recurring way to modulate the distribution of age and fate determinants between sister cells in many organisms , and hence a widely used mechanism for modulating the longevity of diverse cell types . All used strains were constructed according to standard genetic techniques ( Janke et al . , 2004 ) and are isogenic to S288c ( Winzeler et al . , 1999 , Table 1 ) . The strains carrying Nsg1-GFP , Nup49-GFP , and Nup82-3GFP are from the genome-wide GFP collection ( Huh et al . , 2003 ) . The plasmids expressing the PKC1-R398P , Bck1-20 and TOR1-A1957V alleles were already described ( Nonaka et al . , 1995; Helliwell et al . , 1998; Reinke et al . , 2006 ) and the first two were a kind gift of Michael Hall . Concerning the experiments testing the effect of Pkc1 , Bck1 and Tor1 on the diffusion barrier ( Figure 5 ) : the plasmids expressing PKC1-R398P , Bck1-20 , and the empty vector ( pRS315 ) were introduced as centromeric plasmids in the strain carrying Nup49-GFP ( plasmids list in Table 2 ) . The allele expressing constitutively active Tor1 ( TOR1-A1957V ) was introduced in the TOR1 genomic locus , replacing the endogenous TOR1 copy . The strain used for the DNA circles propagation assay was described ( Shcheprova et al . , 2008 ) , but we introduced in that background SPC42-CFP:kanMX4 and the Estradiol Binding Domain ( EBD ) fused to the GAL4 activator ( EBD-GAL4:TRP1 ) ( Takahashi and Pryciak , 2008 ) . We backcrossed this strain five times into the S288C background . Medium was supplemented as indicated with Rapamycin at 200 ng/ml for 18–20 hr or with β-Estradiol at 200 ng/ml for 3 hr and glucose at 2% . All reagents were bought from Sigma-Aldrich ( St . Louis , MO ) . The amino acid mixes , yeast nitrogen base and ammonium sulfate used to prepare Synthetic Dropout ( SD ) medium were purchased from FORMEDIUM ( United Kingdom ) . The Agar was purchased from SERVA ( Germany ) and the yeast extract and the bactopeptone from BD ( USA ) . Unless otherwise stated , all strains were grown on plates , at 30°C in YPD ( Yeast extract , Peptone , Dextrose ) medium . They were kept on the condition of interest ( e . g . 30°C , 2% glucose ) from the culture prepared the day before the experiment and then during the experiment itself . Cells were grown overnight to low density on plates lacking uracil ( -URA ) . The next morning cells still exponentially growing were streaked on YPD plates containing 1 µM β-Estradiol . β-Estradiol binds to the Estradiol Binding Domain ( EBD ) in EBD-Gal4 , which consequently enters the nucleus and triggers the expression of a recombinase . This is responsible for recombination between the recombination sites ( Figure 1A ) , thus excising the centromere from the DNA circle . After 4 hr in β-Estradiol , cells were suspended in medium lacking tryptophan ( -TRP ) and imaged with an Olympus BX50 microscope , equipped with a piezo motor , a monochromatic light source and a CCD camera ( Andor885 ) . The microscope was controlled with the TillVision software ( Till Photonics/FEI Munich GmbH , Germany ) . Images were acquired with a 100x/1 . 4 NA oil immersion objective , 2 × 2 binning and 20 focal slices ( 0 . 3 µM spacing ) . Maximum intensity projection was used to localize the DNA circles and the SPBs . Cells in anaphase containing 1 , 2 and 4 circles not co-localizing with the SPBs were analyzed for the presence of the circle in the mother and/or the bud . The DNA circle propagation frequency was calculated as:1− ( n1×p1+2n2×p2+4n4×p44n1+2n2+4n4 ) where n1 , n2 and n4 are the number of cells respectively containing 1 , 2 and 4 circles , and p1 , p2 and p3 the percentage of cells retaining respectively 1 , 2 or 4 circles in the mother . For the experiments in Figures 2C and 4E and Figure 2—figure supplement 1A after 4 hr in β-Estradiol , cells were suspended in SD -TRP medium , immobilized on a cover slip with a SD -TRP pad ( 2% agar ) and imaged using a Deltavision microscope ( Applied Precision , Slovakia ) . The microscope was equipped with a CCD HQ2 camera ( Photometrics , Arizona ) , 250W Xenon lamps , Softworx software ( Applied Precision , Slovakia ) and a temperature chamber , set to the desired temperature . A 100x/1 . 40 NA U plan S Apochromat oil immersion objective ( Olympus , Japan ) was used . 30-min time lapse movies ( 1 frame per minute ) with 20 stacks ( 0 . 25 µM spacing ) were acquired . Maximum intensity projection was performed . The propagation flux of individual DNA circles through the bud neck in early and late anaphase was measured as a DNA circle passage frequency per minute , considering the moment when a circle is observed for the first time in the daughter compartment as the passage event . For the data in Figure 2 and Figure 2—figure supplement 1 , cells were grown and samples prepared and imaged as explained in the previous paragraph . Early anaphase was defined as the time window starting with the entry of the nucleus into the bud and finishing with the formation of a dumbbell-shaped nucleus . Late anaphase corresponds to the further elongation of the nucleus , and particularly of the bridge between the two future daughter nuclei , and finishes with the resolution of that bridge at karyokinesis ( Figure 2—figure supplement 2A , B ) . For the experiments shown in Figure 2—figure supplement 2 , cells were grown overnight to low density on YPD plates . The next morning cells still exponentially growing , were suspended in -TRP medium and immobilized on a -TRP pad ( 2% agar ) . Cells were imaged with a Deltavision microscope as described in the previous paragraph . A 2 × 2 binning and an auxiliary 1 . 6x magnification were used . Time lapse movies of 25 s interval for 25 min and 7 stacks ( 0 . 3 µM spacing ) were acquired . After 3D iterative deconvolution , neck width and M-D axis length were measured considering only cells where we could follow the entire nuclear division process . These measurements were performed in an equatorial focal section of each nucleus and plotted as mean distance over time . Cells were grown overnight as in ‘DNA circles propagation assay’ . The cells contained the model circle and expressed TetR fused to mCherry ( TetR-mCherry ) , a Nup82 tagged with 3 copies of super folder GFP ( Nup82-3x sfGFP ) to label the NPCs and Spc42p with GFP ( Spc42-GFP ) to label the SPB . The cells were grown at either 30°C or 37°C . After 4 hr of β-Estradiol treatment , the cells were suspended in low fluorescent SD -TRP medium and immobilized on a cover slip with a SD -TRP pad ( 2% agar ) . For rapid imaging , a Nikon Eclipse T1 microscope was used , with acquisition times of 25 and 50 ms for GFP and mCherry respectively . The microscope was equipped with a LUDL BioPrecision2 stage with Piezo Focus , two 200 mW laser lines ( DPSS 488 nm and DIode561 nm ) , a sCMOS camera ( Orca Flash 4 . 0 V2 ) and a temperature chamber ( Oko-lab ) , set to the desired temperature . A 100x/1 . 49 CFI Apochromat TIRF oil immersion objective was used . The microscope was controlled with the VisiVIEW software ( Metamorph/Molecular Devices , California ) . For each channel ( mCherry and GFP ) , 9 stacks ( 0 . 3 µM spacing ) were acquired , with one bright field image in the center of the stack . Cells in anaphase containing one DNA circle dot distinct from the SPB were analyzed using the software Fiji ( imagej . net/Fiji , Schindelin et al . , 2012 ) . 2 pixel wide GFP and mCherry fluorescence intensity profiles were measured for each cell along the nuclear rim ( excluding the SPB ) . These measurements were performed in the focal plane for both Nup82-3x sfGFP and the DNA circle ( tetR-mCherry ) . Cells where the DNA circle localized at , or close to the SPB were disregarded . All the single-cell Nup82-3x sfGFP traces were aligned relative to the corresponding brightest tetR-mCherry pixel ( DNA circle ) and averaged to obtain a mean profile , subsequently plotted for both channels . The mean Nup82-3x sfGFP intensity per position was only calculated for positions where the number of cells > 10 . After background subtraction , the mean GFP intensity at the rim was set to 1 . The GFP intensity at the tetR-mCherry peak was measured as fold induction compared to the normalized value of the rim . Cells were grown as explained in 'Anaphase duration and nuclear morphology analysis' . Time lapse movies of 3–5 s interval for 4–5 min were acquired . For the experiments shown in Figures 4 and 5 and Figure 5—figure supplements 1 and 2 ( except bcyΔ mutant cells ) , cells were imaged with a confocal LSM 510 microscope , controlled by ZEN 2010 ( Carl Zeiss Microimaging Inc , Germany ) . We used a Plan-Apochromat 63x/1 . 4 NA oil immersion objective and 3% of laser intensity with 25% laser output ( Argon laser , 488 nm ) . Bleaching pulses were iterated ( as indicated in the figures ) for 80 times with 60% laser intensity . For bcyΔ mutant cells ( Figure 5 and Figure 5—figure supplement 2 ) all was as for Figure 4 but the following: a confocal LSM 780 microscope ( controlled by ZEN 2011 , Carl Zeiss Microimaging Inc , Germany ) , 3 . 5% of laser intensity ( laser output of 40% , Argon laser , 488 nm ) and a multi-array 32PMT GaAsP detector were used . Bleaching pulses were iterated for 50 times with 100% laser intensity . For the experiments in Figure 4—figure supplement 1 , everything was as for bcyΔ mutant cells but the following: 20% of laser intensity; bleaching pulses were iterated for 100 times with 100% laser intensity . For all FLIP experiments , bleaching pulses were iterated at every frame and quantification was performed using Fiji as follows: the total integrated fluorescence intensity was measured in the mother and bud compartments and in 3–5 neighboring control cells . After background subtraction , the fluorescence intensity of mother and bud compartments was normalized to the mean intensity of the control cells ( to correct for fluorescent decay due to exposure ) and set to 100% . The resulting single-cell fluorescence profiles were pooled to obtain a single profile . This was fit to a one phase decay function , using the software Prism 6 ( GraphPad software , GraphPad Software , Inc . , California ) . The initial Y0 value was constrained to 100% . The resulting best fit values for plateau and K and their relative errors were used to measure the Barrier Index ( BI , see main text for BI definition ) . The standard error of the BIs was calculated by error propagation on the errors obtained from the fit . bcy1Δ mutant cells and a wild type strain were analyzed in parallel , using the LSM 780 microscope . This showed a significantly lower BI in bcy1Δ mutant cells . For comparison purposes , this BI was normalized relative to the wild type analyzed with the LSM 510 microscope . This allowed us to compare bcy1Δ with the experiments performed with the LSM 510 microscope and shown in Figure 5 . Cells were streaked from −80°C and grown for 2 days on YPD plates . After 2 days they were streaked again on YPD plates and grown at either 30°C or 37°C overnight . The next morning they were streaked on fresh and pre-warmed YPD plates and grown for 2 hr at 30°C or 37°C . After 2 hr , virgin daughters were separated from mother-virgin daughter pairs and placed on defined spots on the plate , using a Zeiss Axioscope 40 microdissection microscope . The microscope was equipped with a 10X objective . Every 1 . 5 hr , the isolated daughters were visited and newly born daughters removed and counted . Dissection was performed for 10–14 hr per day and at room temperature for all experiments . Between dissection rounds , cells were kept in a wet box at either 30°C or 37°C . Overnight , the wet boxes was stored at +4°C , to slow down the cell cycle . Two independent experiments per condition with 20–40 virgin daughters per experiment were analyzed . A Southern blot was performed , as before ( Denoth-Lippuner et al . , 2014 ) but with DNA from young and aged cells both cultured at 30°C or 37°C . Young cells were harvested from an exponentially growing culture . Aged cells were purified according to the protocol of the mother enrichment program ( Lindstrom and Gottschling , 2009 ) , with some adaptations . Briefly , 5 × 107 cells were washed with PBS and labeled with Sulfo-NHS-LC-Biotin ( Pierce/Thermo Fisher Scientific , Massachusetts ) and recovered for 2 hr at 30°C or 37°C prior to the addition of β-Estradiol ( 1 µM final concentration ) . The cells were harvested after 26 hr of incubation at 30°C or 37°C . After a wash with PBS , batches of 2 × 109 cells were resuspended in 1 mL PBS supplemented with 50 µl streptavidin-coated magnetic beads ( MicroMACS , Miltenyi Biotec , Germany ) , incubated for 30 min at 4°C and loaded onto LS MACS columns ( Miltenyi Biotec ) for purification . Cells were eluted using 1 × PBS containing 2 mM EDTA and split into two fractions: ( 1 ) 10% of the cells were fixed with paraformaldehyde and the bud scars labeled with 5 µg/ml calcofluor white and visualized by microscopy . The fraction of aged cells in the population and the bud scar count per aged cells were quantified . ( 2 ) Cells were lysed and DNA was purified using standard methods . DNA content was quantified performing qPCR amplifying ACT1 from aged and young cells in triplicate . The fraction of DNA from aged cells per sample was equilibrated with DNA from young cells , to have in both samples same amount of DNA originating from aged cells . The amount of DNA loaded into the gel was equilibrated based on qPCR reads . A 0 . 6% agarose gel contained ethidium bromide and was run in TBE with ethidium bromide for 25 hr at 50 V at 4°C . The gel was blotted to a cationized nylon transfer membrane ( Zeta-Probe GT , Bio-RAD , California ) using standard protocols . Membranes were hybridized with a probe generated by conventional PCR , amplifying the rDNA locus of genomic DNA extracted from wild type yeast ( Lindstrom et al . , 2011 ) and 5’ end-labeled with 32P . The blot was visualized using a Typhoon phosphoimager ( GE healthcare , UK ) . The relative band intensity of the different ERC bands per sample were measured in Fiji ( imagej . net/Fiji , Schindelin et al . , 2012 ) and summed for plotting . A two-tailed unpaired student’s t-test was used to test for significance , for the nuclear morphology a two-way ANOVA followed by Tukey’s multiple comparison test was used , for the replicative lifespan experiments a log-Rank ( Mantel-Cox ) test was used . For the anaphase durations , the non-parametric Mann-Whitney U test was used . Unless otherwise indicated , all the analyzed mutants/conditions were always compared to wild type cells ( 30°C , 2% glucose ) .	Aging is often an inevitable part of life . Yet , it seems like the aging process can be sped up or slowed down in organisms as distinct as fruit flies , mice and budding yeast in response to changes in the environment . Yeast , for example , lives 40% longer when grown at the elevated – and mildly stressful – temperature of 37°C instead of 30°C . But how can yeast , or any other organism , change its lifespan in response to stressful conditions ? Budding yeast ( Saccharomyces cerevisiae ) , as its name suggests , divides by budding small daughter cells from its surface . The mother cell gets older with each division , whereas the age of each daughter cell is reset to zero . The mother cell protects the daughters by keeping some harmful aging factors for itself . Many aging factors – like toxic DNA circles – are anchored to the membranes of the endoplasmic reticulum ( the compartment in the cell where many proteins are made ) and the nucleus ( the compartment where the cell's genetic information is stored ) . Before the cells divide , a diffusion barrier keeps molecules in the membranes of the mother cell , preventing them from entering the membranes of the daughter cell . The aging factors only leak into the daughter cells if they detach from the membrane in the mother or if the diffusion barrier becomes permeable . The loss of aging factors causes the mother cell to live much longer than normal . Baldi , Bolognesi , Meinema et al . asked if something similar occurred when cells experience stress , which could explain why stressed yeast cells live longer . Indeed , mother cells did redistribute toxic DNA circles to their daughters when grown at a higher temperature . This did not happen because the DNA circles detached from the membrane . Instead the diffusion barrier became more permeable . Baldi et al . then went on to show that it was not just the heat that weakened the barrier . Rather the diffusion barrier was specifically down-regulated by one of the yeast’s normal responses to stress . Lastly , when Baldi et al . took steps to make the diffusion barrier in mildly stressed cells less permeable again , the cells largely resumed aging like unstressed cells . Together these data suggest that yeast cells undergoing mild stress might not repair damage or clear out the aging factors but rather dispose of the factors by passing them on to their offspring . It is possible that this helps the population to cope with the stress , by sharing the burden of age – or , as Baldi et al . also discuss , the wisdom of age – with the other individuals . Stress-response pathways are conserved among many other organisms , and similar diffusion barriers occur in worms and mammals too . Thus , this newly discovered process might also happen in other cells that divide asymmetrically , including human stem cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2017	Heat stress promotes longevity in budding yeast by relaxing the confinement of age-promoting factors in the mother cell
Sleep is necessary for the optimal consolidation of newly acquired procedural memories . However , the mechanisms by which motor memory traces develop during sleep remain controversial in humans , as this process has been mainly investigated indirectly by comparing pre- and post-sleep conditions . Here , we used functional magnetic resonance imaging and electroencephalography during sleep following motor sequence learning to investigate how newly-formed memory traces evolve dynamically over time . We provide direct evidence for transient reactivation followed by downscaling of functional connectivity in a cortically-dominant pattern formed during learning , as well as gradual reorganization of this representation toward a subcortically-dominant consolidated trace during non-rapid eye movement ( NREM ) sleep . Importantly , the putamen functional connectivity within the consolidated network during NREM sleep was related to overnight behavioral gains . Our results demonstrate that NREM sleep is necessary for two complementary processes: the restoration and reorganization of newly-learned information during sleep , which underlie human motor memory consolidation . There is now ample evidence that sleep plays a crucial role in the consolidation of newly-acquired procedural memory , particularly for explicitly instructed sequential motor skills ( Walker et al . , 2002; Korman et al . , 2003; Doyon and Benali , 2005; Korman et al . , 2007; Debas et al . , 2010 ) . Several mechanistic hypotheses have also been proposed regarding the contribution of sleep in this memory process ( see [Frankland and Bontempi , 2005; Rasch and Born , 2007; Tononi and Cirelli , 2014] for comprehensive reviews ) . Yet , the dynamic neural changes that drive motor memory consolidation during sleep still remain controversial ( Frankland and Bontempi , 2005; Rasch and Born , 2013; Tononi and Cirelli , 2014 ) . One pioneering sleep-dependent consolidation model , the trace reactivation hypothesis assumes that the repeated reactivation of a recently formed memory representation during sleep leads to a gradual strengthening of the learning-related connections , and thus to long-term storage of the memory trace ( Rasch and Born , 2007 , 2013 ) . There is now mounting evidence in support of this hypothesis including the replay of hippocampal place cell firing ( Skaggs and McNaughton , 1996; Lee and Wilson , 2002 ) in rodents , as well as human studies employing targeted memory reactivation paradigms using auditory or olfactory cues ( Rasch et al . , 2007; Cousins et al . , 2014; Laventure et al . , 2016 ) , and neuroimaging studies showing the reactivation of learning-related brain regions during sleep or awake rest ( Maquet et al . , 2000; Rasch et al . , 2007; Deuker et al . , 2013; Staresina et al . , 2013; Tambini and Davachi , 2013 ) . Another model , built in part upon the trace reactivation , the systems consolidation hypothesis ( Frankland and Bontempi , 2005; et al . , 2005; Rasch and Born , 2013 ) proposes that sleep engages an active reorganization process that stabilizes the labile neural representation of a novel skill into a consolidated memory trace . For instance , a systematic transfer in memory representations from hippocampal to neocortical areas has been reported for non-procedural forms of memories ( Frankland et al . , 2004; Maviel et al . , 2004; Frankland and Bontempi , 2005 ) . On the other hand , a systemic shift from cortical ( e . g . , motor , parietal cortex ) to subcortical regions ( e . g . , striatum ) has been proposed to underlie the consolidation of procedural memory , and motor sequence learning in particular ( Doyon and Benali , 2005; Yin et al . , 2009; Debas et al . , 2010; Kawai et al . , 2015 ) . Yet in humans , the systems consolidation model has only been inferred indirectly by comparing the effect of motor practice on offline gains in behavioral performance and changes in neural activity between the initial learning and retention sessions separated by either diurnal or nocturnal sleep ( Walker et al . , 2002; Fischer et al . , 2005; Gais et al . , 2007; Takashima et al . , 2009; Debas et al . , 2010 ) . Thus , direct evidence in support of this hypothesis from human neuroimaging studies is lacking . Finally , an alternative and potentially complementary model , the synaptic homeostasis hypothesis ( Tononi and Cirelli , 2003 , 2006 , 2014 ) proposes that local neuronal networks are potentiated and eventually become saturated during learning . In order for new information to be encoded the following day , sleep would be involved in the restoration of these local networks by downscaling the strength of synaptic connections ( Tononi and Cirelli , 2003; Huber et al . , 2004; Tononi and Cirelli , 2006 ) . However , direct experimental evidence to support the synaptic homeostasis hypothesis in humans remains limited and controversial ( Frank , 2012 ) . It is thus unclear whether and how these different sleep-dependent mechanisms of memory consolidation may be reconciled and contribute to motor skill learning in humans . Here , for the first time , we used simultaneous EEG and fMRI in order to identify the relative contributions of the trace reactivation , systems consolidation , and synaptic homeostasis hypotheses to the consolidation of procedural memory in humans . Specifically , we tested the hypothesis that the memory trace of motor sequence learning involves network-wide reactivation and further reorganization into a more stable representation during non-rapid eye movement ( NREM ) sleep periods . In order to directly examine the off-line periods during which the motor memory trace is being consolidated , we acquired blood-oxygen-level dependent ( BOLD ) fMRI data during motor sequence task practice , wake resting-state and post-training sleep conditions . Brain functional images were recorded while thirteen participants performed two different finger movement tasks using a response pad one week apart . In the motor sequence learning ( MSL ) task , subjects practiced a self-paced , explicitly known 5-item finger sequence task , which was compared with performance on a motor control task ( CTL ) in which participants were asked to produce simultaneous movements of all four fingers at the same average frequency , and for the same number of times as in the MSL task . These two conditions were administered in a counterbalanced order ( Figure 1a ) . For both MSL and CTL tasks , participants underwent an initial training session at 10:30 PM ( i . e . , learning session; S1 ) , followed by a retest session at 9:00 AM the next morning ( i . e . , retest session; S2 ) ( Figure 1a ) . Resting-state conditions , during which subjects stayed awake with eyes opened , were also acquired before and after each practice session in the evening ( RS1 and RS2 ) and the following morning ( RS3 and RS4; Figure 1a ) . Immediately following the training session ( i . e . , around 11:00 PM ) , a simultaneous EEG-fMRI recording scan lasting a maximum of 2 . 5 hr took place while subjects slept in the scanner . This design allowed us to investigate MSL memory trace reactivation and further transformation during off-line periods , including both resting-state and sleep conditions , from its initial state to a consolidated trace that was later recruited during performance at retest . 10 . 7554/eLife . 24987 . 003Figure 1 . Experimental design , behavioral performance , and task activation maps . ( A ) Experimental procedure . On Day 1 , subjects first experienced a screening and adaptation night in the mock scanner , which mimicked conditions experienced in both experimental and control nights . Subjects returned and underwent fMRI scans ( Day seven and Day 14 ) while training ( S1 ) on either the motor skill learning ( MSL ) or motor control ( CTL ) task in a counterbalanced order , interleaved by resting-state conditions ( RS1 and RS2 ) . This was followed by simultaneous EEG-fMRI sleep recording for up to ~2 . 5 hr . Subjects were then allowed to sleep for the remainder of the night in the sleep lab . Finally , on the following morning subjects underwent retest fMRI sessions ( S2 ) on the same take as the previous training session ( Day eight and Day 15 ) , interleaved by resting-state conditions ( RS3 and RS4 ) . Arrows shows the experiment’s timeline . ( B ) Performance speeds ( i . e . , inter-key interval ) averaged across all subjects show that the learning curves differed between the MSL ( red ) and CTL ( blue ) conditions during the learning session ( S1 ) . ( C ) Only the MSL task was consolidated overnight , as indicated by performance gain averaged across subjects ( asymptotic performance at the end of S1 compared to the beginning of S2 ) . ( D ) Color-coded activation maps representing motor sequence-related areas during the learning ( S1 ) and retest ( S2 ) sessions ( corrected for multiple comparisons using Gaussian random field theory , cluster level threshold p<0 . 05 ) . Bar plots illustrate the volume of cortical and subcortical activation in each map . As expected , the connectivity index ( CI ) within the learning ( E ) and the consolidated ( F ) patterns was significantly higher in the MSL compared to the CTL condition . Error bars represent s . e . m . ; and * indicate p<0 . 01 and p<0 . 001 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 24987 . 00310 . 7554/eLife . 24987 . 004Figure 1—source data 1 . Summary of activation peaks related to the learning pattern . The learning pattern represents all brain areas with greater activation in the MSL compared to the CTL condition during the learning practice session ( S1 ) . For each peak of activity , the anatomical label , MNI coordinates , the corrected cluster-level p value using GRF , and the associated Z-score are reported . DOI: http://dx . doi . org/10 . 7554/eLife . 24987 . 00410 . 7554/eLife . 24987 . 005Figure 1—source data 2 . Summary of activation peaks related to the consolidated pattern . The consolidated pattern represents all brain areas with greater activation in the MSL compared to the CTL condition during the retest practice session ( S2 ) . Reporting conventions are as in Figure 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 24987 . 00510 . 7554/eLife . 24987 . 006Figure 1—figure supplement 1 . Task-related activation maps during the learning session ( S1 ) . Figure shows the results of block-design analysis related to the MSL ( color-coded in red ) and CTL tasks ( color-coded in blue ) . Color-coded activation maps indicate Z-score values and are corrected for multiple comparisons using GRF , p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 24987 . 00610 . 7554/eLife . 24987 . 007Figure 1—figure supplement 2 . Differences in the activation of motor sequence-related areas between the learning ( S1 ) and retest ( S2 ) sessions . Only subcortical areas including bilateral putamen and cerebellar cortex ( lobules V-VI ) revealed significantly greater activation during the retest compared to the learning session ( contrast: retest [MSL-CTL] – learning [MSL-CTL]; top row ) , while two cortical clusters , including the superior parietal lobule and anterior intraparietal sulcus bilaterally , showed significantly greater activation during the learning compared to the retest session ( contrast: learning [MSL-CTL] – retest [MSL-CTL];bottom row ) . Color-coded activation maps indicate Z-score values and are corrected for multiple comparisons using GRF , p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 24987 . 007 Motor performance was measured using the speed at which key presses for correct responses were executed in both MSL and CTL tasks ( Albouy et al . , 2012; Fogel et al . , 2014; Lungu et al . , 2014 ) . As expected , a two-factor repeated measures ANOVA across practice blocks ( 14 blocks ) and task conditions ( MSL , CTL ) during the training session ( S1 ) revealed that performance speed evolved differently over the course of learning between the two tasks ( Figure 1b; significant condition × practice block interaction; F13 , 156=4 . 94 , p<0 . 0001 ) . Yet given that we intentionally matched the speed at which both MSL and CTL tasks had to be performed , as expected , average speed did not differ between tasks ( no main effect of task; F1 , 12=3 . 14 , p>0 . 1 ) . Also , consistent with previous studies ( Walker et al . , 2002; Albouy et al . , 2015 ) , only performance on the motor learning task revealed evidence of consolidation overnight , as indicated by off-line improvements in the MSL , but not the CTL task , in the absence of additional practice ( Figure 1b ) . Specifically , the improvement in task performance between the end of the training ( mean of last three blocks in S1 ) and beginning of the retest session ( mean of first three blocks in S2 ) significantly differed across tasks , as revealed by a two-factors [session ( end S1 , beginning S2 ) × task ( MSL , CTL ) ] repeated measures ANOVA ( significant interaction; F1 , 12=16 . 77 , p=0 . 001; follow-up paired t-tests: MSL[beginning S2 – end S1] , t12=2 . 43 , p<0 . 05; and CTL[beginning S2 – end S1] , t12=−1 . 65 , p>0 . 1; Figure 1c ) . We also examined changes in performance accuracy by measuring the percentage of incorrect key presses in each block of the MSL task . Given that the simple 5-item sequence was explicitly known to the participants , as expected ( Walker et al . , 2002; Debas et al . , 2010; Albouy et al . , 2012; Fogel et al . , 2014 ) , performance error was very low overall , and did not show any significant change overnight ( average error in the last three blocks of S1 compared to the first three blocks of S2 , t12=0 . 95 , p>0 . 35 ) . This further confirms that the performance speed was a suitable measure to quantify off-line improvements in motor performance . Furthermore , we investigated performance variability by calculating the standard deviation of inter-key-press intervals in each block of MSL and CTL tasks . In line with the performance speed results , performance variability significantly decreased overnight only in the MSL task , as revealed by a two-factors [session ( end S1 , beginning S2 ) × task ( MSL , CTL ) ] repeated measures ANOVA ( significant interaction; F1 , 12=15 . 9 , p=0 . 0018; follow-up paired t-tests: MSL [beginning S2 – end S1] , t12=3 . 98 , p<0 . 002; and CTL [beginning S2 – end S1] , t12=0 . 36 , p>0 . 7 ) . We identified distinct brain activation patterns recruited during the learning and retest sessions following a night of sleep . For comparison purpose , the task-related activation maps during MSL and CTL sessions are presented in Figure 1—figure supplement 1 . Using a two-factor [practice session ( S1 , S2 ) × task ( MSL , CTL ) ] repeated-measures ANOVA at the group level , we identified motor sequence-related brain areas that were either activated during the learning ( the ‘learning pattern’; S1[MSL – CTL] , Figure 1d top ) or retest session ( the ‘consolidated pattern’; S2[MSL - CTL] , Figure 1d bottom ) . Although , both the learning and consolidated patterns comprised similar sensorimotor core regions ( see Figure 1—source data 1 and 2 ) , the relative activation levels of different cortical and subcortical areas were mostly altered across the two maps; that is , the consolidated pattern revealed increased activity in sub-cortical structures and decreased activity in cortical regions ( Figure 1d ) . Specifically , despite the fact that the total volume of motor sequence-related activity was preserved across sessions ( 37 . 48 cm3 in learning versus 36 . 10 cm3 in consolidated pattern ) , the volume of all cortically activated voxels ( including mostly the fronto-parietal sensorimotor regions ) in the consolidated pattern was almost reduced by half ( from 29 . 34 cm3 to 15 . 58 cm3 ) , while the volume of sub-cortical activations ( including mostly the basal ganglia and cerebellar regions ) was nearly quadrupled ( from 4 . 01 cm3 to 15 . 32 cm3; Figure 1d middle bar plots ) . Similarly , a contrast between the learning and retest sessions strongly confirmed our volume-based analysis results; that is , two cortical clusters ( including the superior parietal lobule and anterior intraparietal sulcus bilaterally ) showed significantly greater activation during the learning compared to the retest session , while only subcortical regions ( including the putamen and cerebellar cortex ) revealed significantly greater activation during the retest compared to the learning session ( Figure 1—figure supplement 2 ) . Consistent with recent work in both human and animal models ( Debas et al . , 2010; Kawai et al . , 2015 ) , these results suggest a topological shift of activity from cortical to subcortical regions that might underlie sequence memory consolidation . To investigate whether regions within these distinct patterns became more highly interconnected ( reflecting a strengthening of the memory trace ) , we examined the changes in functional connectivity within the learning and the consolidated patterns in both MSL and CTL conditions . The strength of functional connectivity within a given brain network ( i . e . , the ‘connectivity index’ , ( CI ) ) was estimated using a straightforward approach that measured the overall co-activation level of brain areas within that network during different fMRI runs ( Vahdat et al . , 2014 ) . Specifically , the CI was defined as the power of a time series of normalized coefficients in a spatial regression model , which estimated the co-activation level of areas within a given network over different scanning times ( see Materials and methods for the formulation ) . This connectivity measure was selected as it provides a hypothesis-driven multivariate approach specifically suited to study dynamics of changes in connectivity within a network of areas across the whole brain ( see Materials and methods for more details ) . As a validation check , we first evaluated CI during task performance in the learning ( S1 ) and retest ( S2 ) sessions . It was expected that since the learning and consolidated patterns are extracted from the [MSL – CTL] contrast , we would observe greater levels of CI during the MSL as compared to the CTL task periods . Consistently , a two-factors ( session × task ) repeated measures ANOVA reported a significant main effect of task for both the learning ( F1 , 12=31 . 2 , p<0 . 0005; Figure 1e ) and consolidated patterns ( F1 , 12=63 . 4 , p<0 . 000005; Figure 1f ) . Notably , there was also a significant effect of session for the consolidated pattern , showing greater CI during the retest ( S2 ) compared to the learning ( S1 ) session ( significant interaction; F1 , 12=9 . 33 , p=0 . 01; also significant main effect of session in MSL , t12=3 . 31 , p=0 . 006 ) . This analysis confirms that CI is a sensitive measure to detect changes in within-network functional connectivity across experimental conditions . In order to investigate whether memory reorganization from the learning to the consolidated trace occurred during the off-line periods ( dependent upon either simple passage of time or sleep ) , or whether the consolidated trace merely manifested itself during retest , we calculated CI within the learning and consolidated patterns during different resting-state periods ( RS1 , RS2 , RS3 ) , as well as during NREM sleep . A two-factor ( resting-state condition ( RS1 , RS2 , RS3 ) × task ( MSL , CTL ) ) repeated-measures ANOVA revealed that the CI changed as a function of motor task condition ( MSL vs . CTL ) across resting-state periods for the consolidated pattern ( significant interaction; F2 , 24=6 . 06 , p=0 . 007 , Figure 2b ) . Interestingly , CI within the consolidated pattern was significantly enhanced for the MSL task only during RS3 ( significant MSL [RS3-RS1] , t12=3 . 14 , p<0 . 01 , significant MSL [RS3-RS2] , t12=2 . 26 , p<0 . 05 , and significant main effect of task during RS3 , t12=3 . 78 , p<0 . 005 , Figure 2b ) . Thus , despite the fact that the consolidated pattern’s CI was not yet increased immediately after training ( MSL [RS2-RS1] , t12=0 . 94 , p>0 . 35 ) , it was already significantly elevated before the retest session ( i . e . RS3 ) , hence suggesting that the consolidation process took place during the preceding interval filled with sleep , and did not manifest itself as a result of practice during retest . 10 . 7554/eLife . 24987 . 008Figure 2 . Connectivity index ( CI ) during resting-state and NREM sleep . ( A , B ) show the CI within the learning and the consolidated patterns , respectively , averaged across subjects during the resting-state conditions before ( RS1 ) and after ( RS2 ) the training session ( S1 ) , as well as on the following morning before retest ( RS3 ) . The learning pattern’s CI was significantly increased in the MSL immediately following training ( RS2 ) , while the consolidated pattern’s CI increased significantly only post-sleep ( RS3 ) in the MSL as compared to the CTL condition . ( C , D ) show , respectively , CI within the learning and the consolidated patterns averaged across subjects during NREM sleep . Only the consolidated pattern’s CI differed significantly between the MSL and CTL nights . Error bars represent s . e . m . ; * and ** indicate p<0 . 05 and p<0 . 01 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 24987 . 00810 . 7554/eLife . 24987 . 009Figure 2—figure supplement 1 . The group-level spatial maps of four highly-reproducible brain networks extracted during the MSL task . Each row shows a spatial map extracted from the application of independent component analysis ( ICA ) on the time-concatenated data of the learning ( S1 ) and retest ( S2 ) sessions during the MSL task across all subjects . The label for each component is provided on the left . Color-coded activation maps indicate Z-score values . DOI: http://dx . doi . org/10 . 7554/eLife . 24987 . 00910 . 7554/eLife . 24987 . 010Figure 2—figure supplement 2 . Connectivity index ( CI ) during resting-state periods within the four control networks reported in Figure 2—figure supplement 1 . ( a–d ) show the CI within the default mode , visual , left and right fronto-parietal networks , respectively , averaged across subjects during the resting-state conditions before ( RS1 ) and after ( RS2 ) the training session ( S1 ) , as well as on the following morning before retest ( RS3 ) . No significant difference in CI between MSL and CTL tasks was observed in any of the four control networks ( p>0 . 2 , in all cases ) . Error bars represent s . e . m . ; n . s . indicates not significant interaction . DOI: http://dx . doi . org/10 . 7554/eLife . 24987 . 01010 . 7554/eLife . 24987 . 011Figure 2—figure supplement 3 . Connectivity index ( CI ) during non-REM ( NREM ) sleep within the four control networks reported in Figure 2—figure supplement 1 . ( a–d ) show the CI within the default mode , visual , left and right fronto-parietal networks , respectively , averaged across subjects during NREM sleep . No significant difference in CI between MSL and CTL tasks was observed in any of the four control networks ( p>0 . 3 , in all cases ) . Error bars represent s . e . m . ; n . s . indicates not significant paired t-statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 24987 . 011 By contrast , the CI analysis within the learning pattern yielded an opposite pattern of findings . Although the effect of motor task condition ( MSL vs . CTL ) across all resting-state conditions was only marginally significant ( interaction; F2 , 24=3 . 14 , p=0 . 06 , Figure 2a ) , there was a significant effect of task on the learning pattern’s CI across the resting-state conditions before and after learning ( repeated measured ANOVA ( resting-state condition ( RS1 , RS2 ) × task ( MSL , CTL ) ; F1 , 12=6 . 99 , p=0 . 02 , Figure 2a ) . A follow-up paired t-test revealed that CI increased immediately following learning in the MSL task only ( significant MSL [RS2-RS1] , t12=3 . 05 , p=0 . 01 , and significant RS2 [MSL - CTL] , t12=2 . 37 , p=0 . 035 ) . However , the learning pattern’s CI dropped after sleep , so that it was no longer significantly different from baseline in the following morning ( MSL [RS3-RS1] , t12=1 . 6 , p>0 . 1 ) , nor was it different across MSL and CTL tasks in the post-sleep resting-state condition ( RS3 [MSL - CTL] , t12=1 . 1 , p>0 . 25 ) . The resting-state analyses suggest that the sequence-related memory trace was likely reorganized from the learning to the consolidated pattern between RS2 and RS3 runs , that is , during sleep . To directly test this hypothesis , we calculated CI during stage 2 and slow wave sleep ( SWS ) of NREM sleep , as classified by simultaneous EEG recordings during the intervening sleep session . Importantly , this analysis confirmed that only the consolidated pattern’s CI was significantly elevated during NREM sleep in the MSL compared to the CTL night ( t12=3 . 47 , p<0 . 005; Figure 2d ) . By contrast , the learning pattern’s CI during NREM sleep did not differ significantly between the two tasks ( t12=1 . 26 , p>0 . 2; Figure 2c ) . In order to test the specificity of our findings with respect to other activation patterns that would be presumably unrelated to the experimental conditions , we extracted four highly-reproduced canonical brain networks ( Damoiseaux et al . , 2006 ) using the application of independent component analysis ( ICA ) on the task fMRI data ( see Materials and methods ) . These networks included the default mode , visual , and the left and right fronto-parietal networks ( Figure 2—figure supplement 1 ) . We then calculated CI within each of these networks during different resting-state periods ( RS1 , RS2 , RS3 ) , as well as during NREM sleep . Two-factor ( resting-state condition ( RS1 , RS2 , RS3 ) × task ( MSL , CTL ) ) repeated-measures ANOVAs revealed no significant change in CI in any of these networks as a function of motor task across resting-state periods ( no interaction; p>0 . 25 , and no main effect of task or resting-state condition; p>0 . 2 , in all four networks; Figure 2—figure supplement 2 ) . Likewise , no significant change in CI between the two tasks was found during NREM sleep in any of the four networks ( paired t-statistics , p>0 . 3; Figure 2—figure supplement 3 ) . Overall , these analyses suggest that the observed changes in the learning and consolidated patterns were not due to some global epiphenomena of time , learning or sleep on resting state connectivity . The CI analyses allowed us to specify whether a pattern of brain areas as a whole showed changes in functional connectivity across different conditions . However , in order to specify the brain areas within the consolidated pattern that are primarily responsible for modulating the strength of connectivity in the MSL night , we performed a dual regression analysis ( Filippini et al . , 2009 ) . For each subject , this approach projects a group-level activation map ( i . e . , as a spatial regressor ) onto a selected fMRI condition ( e . g . , RS1 or RS3 ) , in order to identify brain areas within that spatial map or pattern that are specifically recruited during the given fMRI run ( see Materials and methods for more details ) . Separate dual regression analyses were carried out using the learning and the consolidated patterns as spatial regressors . For each pattern , a group-level repeated-measures GLM was performed to evaluate the contribution of different areas within each pattern during NREM sleep periods across tasks . We found that the ventrolateral putamen was the primary brain region within the consolidated pattern whose functional connectivity was significantly elevated during NREM sleep in the MSL compared to the CTL night ( corrected for multiple comparisons using Gaussian random field theory ( GRF ) , p<0 . 05 , Figure 3 ) . Importantly , changes in functional connectivity between the putamen and the rest of the structures in the consolidated pattern were significantly related to the amount of sleep-dependent behavioral gains in motor performance ( r=0 . 72 , p=0 . 005 ( N = 13 ) , Figure 3 scatter plot ) . This indicates that individuals with greater increases in functional connectivity with the putamen had greater overnight improvements in performance . By contrast , none of the brain areas within the learning pattern showed a significant change in connectivity during NREM periods between the two tasks . 10 . 7554/eLife . 24987 . 012Figure 3 . Neural correlates of motor sequence memory consolidation during NREM sleep . The ventrolateral putamen functional connectivity within the consolidated pattern differed significantly between the MSL and CTL nights during NREM sleep . Bar plot illustrates the functional connectivity of the putamen with the rest of structures in the consolidated pattern averaged across subjects during NREM sleep . Importantly , the putamen connectivity within the consolidated pattern during NREM sleep was significantly related to the extent of sleep-dependent performance speed gains on a per subject basis , as depicted in the scatter plot . The color-coded activation map indicates Z-score values and is corrected for multiple comparisons using GRF , p<0 . 05 . Error bars represent s . e . m . ; * indicates p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 24987 . 01210 . 7554/eLife . 24987 . 013Figure 3—source data 1 . Regions of interest ( ROI ) used in the seed-based functional connectivity analysis . Each ROI is selected based on the location of activation peaks in the learning pattern ( Le ) , the consolidated pattern ( Co ) , or both ( Le/Co ) . The table shows seeds’ coordinates in MNI space ( given in mm ) and their anatomical labels . DOI: http://dx . doi . org/10 . 7554/eLife . 24987 . 01310 . 7554/eLife . 24987 . 014Figure 3—figure supplement 1 . Changes in brain functional connectivity related to motor sequence learning during non-REM sleep . Two ROIs within the consolidated pattern , including the cerebellar cortex ( top row ) and the putamen ( bottom row ) , showed significant changes in functional connectivity between tasks during non-REM sleep ( paired t test ( MSL−CTL ) , corrected for multiple comparisons using GRF , p<0 . 05 ) . In each row , the left and middle brain slices show the locations of each ROI and the clusters showing significant change in connectivity with that ROI , respectively . The color-coded activation map indicates Z-score values . Each bar plot displays the average values of functional connectivity across subjects during the MSL and CTL nights between the corresponding ROI and the specified clusters . The scatter plots illustrate the linear trend between individual changes in functional connectivity ( MSL−CTL ) and the associated offline gains in performance speed on a per subject basis . The value of r represents the Pearson correlation coefficient . DOI: http://dx . doi . org/10 . 7554/eLife . 24987 . 014 Likewise , we used the dual regression method to identify the brain areas within each of the learning and consolidated patterns whose connectivity changed during either of the post-learning resting-state conditions ( i . e . , RS2 or RS3 ) as compared to the baseline ( i . e . , RS1 ) across the two tasks ( resting-state conditions × task , repeated-measure ANOVA ) . Consistent with the active role of the putamen found during NREM sleep , this analysis revealed that the ventrolateral putamen ( Figure 4a ) and lobules V-VI of the cerebellar cortex ( Figure 4b ) were the principal brain regions responsible for the elevated connectivity within the consolidated pattern during the post-sleep resting-state condition ( significant ( RS3−RS1 ) × ( MSL−CTL ) interaction , corrected for multiple comparisons using GRF , p<0 . 05 ) . Figure 4—figure supplement 1 shows the average amounts of connectivity in each cluster for each session and task separately . As shown in the figure , the amplitude of connectivity was significantly elevated in RS3 compared to RS1 only in the MSL task ( paired t-statistics , p<0 . 01 in MSL for both clusters , p>0 . 2 in CTL for both clusters ) . Yet among these two structures , only the putamen connectivity was significantly correlated with the amount of overnight gains in performance speed ( r=0 . 69 , p<0 . 01 ( N = 13 ) , Figure 4a , right scatter plot ) . Again , no area within the learning pattern showed a significant change in connectivity during the post-sleep resting-state condition across tasks . However , when the learning pattern was examined during the pre-sleep resting-state condition immediately following training ( i . e . , RS2 ) , connectivity was elevated in the posterior parietal lobule with respect to the baseline condition in the MSL compared to the CTL task ( significant ( RS2−RS1 ) × ( MSL−CTL ) interaction , p<0 . 05; Figure 4—figure supplement 2 ) . Consistent with the CI analyses , none of the areas within the consolidated pattern showed changes in connectivity in the pre-sleep resting-state condition compared to the baseline , hence suggesting again that changes within the consolidated pattern were initiated during sleep . 10 . 7554/eLife . 24987 . 015Figure 4 . Neural correlates of motor sequence memory consolidation during post-sleep resting-state periods . The ventrolateral putamen ( A ) and the cerebellar cortex ( lobules V-VI ) ( B ) functional connectivity within the consolidated pattern differed significantly between the MSL and CTL conditions during post-sleep resting-state periods ( RS3 ) as compared to baseline ( RS1 ) . Bar plot illustrates the change in functional connectivity of putamen within the consolidated pattern between RS3 and RS1 scans averaged across subjects in each task condition . The scatter plot in ( A ) shows that only the putamen functional connectivity was significantly related to the extent of overnight behavioral gains in performance speed . The color-coded activations maps indicate Z-score values and are corrected for multiple comparisons using GRF , p<0 . 05 . Error bars represent s . e . m . ; * and ** indicate p<0 . 05 and p<0 . 01 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 24987 . 01510 . 7554/eLife . 24987 . 016Figure 4—figure supplement 1 . Functional connectivity within the consolidated pattern during post-sleep resting-state periods ( RS3 ) and baseline ( RS1 ) in each task . ( A and B ) illustrate , respectively , the average functional connectivity values of the putamen and cerebellum clusters ( see Figure 4 ) within the consolidated pattern during RS3 and RS1 scans in each task condition . Error bars represent s . e . m . ; * indicates significant ( p<0 . 05 ) interaction between task and resting-state condition . DOI: http://dx . doi . org/10 . 7554/eLife . 24987 . 01610 . 7554/eLife . 24987 . 017Figure 4—figure supplement 2 . Neural correlates of motor sequence learning during resting-state periods immediately following training . The posterior parietal lobule functional connectivity within the learning pattern significantly differed between the MSL and CTL tasks during the resting-state condition immediately following training ( RS2 ) as compared to baseline ( RS1 ) . Bar plot illustrates the change in functional connectivity of this area within the learning pattern between RS2 and RS1 scans averaged across all subjects in each task condition . The color-coded activation map indicates Z-score values and is corrected for multiple comparisons using GRF , p<0 . 05 . Error bars represent s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 24987 . 01710 . 7554/eLife . 24987 . 018Figure 4—figure supplement 3 . Changes in brain functional connectivity related to motor sequence learning during the post-sleep ( top row ) and the pre-sleep ( bottom row ) resting-state conditions . An ROI located in the let putamen showed a significant change in functional connectivity between tasks during the post-sleep resting state condition ( RS3 ) as compared to the baseline ( RS1 ) ; significant interaction ( RS3−RS1 ) × ( MSL−CTL ) , top row . In contrast , an ROI located in the right posterior parietal cortex ( brodmann area , BA7 ) showed a significant change in functional connectivity between tasks during the pre-sleep resting state condition immediately following learning ( RS2 ) as compared to the baseline ( RS1 ) ; significant interaction ( RS2−RS1 ) × ( MSL−CTL ) , bottom row . As shown in the scatter plot , only changes in functional connectivity in the post-sleep condition were correlated with the amount of offline gains in performance speed . All maps are corrected for multiple comparisons using GRF , p<0 . 05 . Display conventions are as in Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 24987 . 018 Given that we first observed an elevation of connectivity within the consolidated pattern during NREM sleep following MSL , we further examined the pattern of dynamic changes in CI over the course of the post-training night as compared to the intermittent periods of wakefulness . To do so , we employed a sliding-window approach over the temporally-concatenated fMRI data of each state and condition . Specifically , we separated NREM stage two and SWS periods , and calculated CI values in NREM stage 2 , as well as , the intermittent periods ( epochs ) of wake , for which we had sufficient data in our group of participants . On average 22 min of data ( 600 volumes ) for each subject and condition were selected ( see Materials and methods for more details , and Table 1 for sleep architecture information in the scanner ) . The mean epoch duration and the number of epochs selected for data analysis are reported in Figure 5—source data 1 . Note that there was no significant difference in the characteristics of the selected epochs between MSL and CTL conditions ( Figure 5—source data 1 ) . Additionally , the mean epoch duration and the number of concatenated epochs were not significantly different between the wakefulness and NREM stage two sleep periods ( p=0 . 12 , paired t-test for the mean duration; and p=0 . 40 , Wilcoxon signed rank test for the number of epochs ) . These analyses revealed a gradual increase in the strength of connectivity within the consolidated pattern during stage 2 NREM sleep following MSL as compared to the CTL task , as supported by a repeated measure ANOVA ( significant time×task interaction , F10 , 120=1 . 98 , p<0 . 05; and paired-samples t test comparing mean of the first three and the last three time points [corresponding to the first and the last 7 min and 12 s periods of recorded NREM stage-2 fMRI data , respectively] , t12=2 . 23 , p<0 . 05; Figure 5b ) . Two follow-up repeated measures ANOVAs during stage two sleep also revealed a significant main effect of time on the consolidated pattern’s CI in the MSL ( F10 , 120=2 . 35 , p<0 . 05 ) , but not in the CTL task ( F10 , 120=0 . 98 , p>0 . 4 ) . Importantly , however , the results showed that the consolidated pattern’s CI did not change during the intermittent periods of wake distributed throughout the sleep session ( no time×task interaction , F10 , 120=0 . 93 , p>0 . 5; no main effect of task or time , p>0 . 8 for both; Figure 5e ) . 10 . 7554/eLife . 24987 . 019Figure 5 . Temporal dynamics of memory trace during NREM stage two sleep . ( A , B ) illustrate the time course of CI change within the learning and the consolidated patterns during NREM stage two sleep , respectively . The consolidated pattern’s CI gradually increased during NREM sleep only in the MSL condition ( red curve ) , while the learning pattern’s CI decreased . CI did not differ significantly over the course of NREM stage two sleep in the CTL night ( blue curves ) . Furthermore , CI within the learning ( D ) and consolidated ( E ) patterns did not change significantly over the course of intermittent awakenings distributed throughout the sleep session in the CTL ( blue curves ) or MSL ( red curves ) night . ( C ) , ( F ) illustrate the time course of ventrolateral putamen functional connectivity within the consolidated pattern during NREM stage two sleep and intermittent bouts of awakenings , respectively . Likewise , the putamen functional connectivity with the rest of structures in the consolidated pattern gradually increased only during NREM sleep in the MSL condition . Each data point is calculated using 100 fMRI volumes . Shaded area represents s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 24987 . 01910 . 7554/eLife . 24987 . 020Figure 5—source data 1 . Average duration and number of epochs used in the temporal dynamics analysis ( Figure 5 ) . Table reports the average epoch duration and the number of epochs which were used in data analysis during the sleep scanning session in the MSL or CTL condition nights . Mean and SEM duration values are reported in minutes . p values are calculated based on paired t-statistics , and Wilcoxon signed rank test for the average duration and the number of epochs , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 24987 . 02010 . 7554/eLife . 24987 . 021Figure 5—figure supplement 1 . Robustness of temporal dynamics analysis with respect to the window size . ( A , B ) illustrate the CI values , averaged over subjects , within the learning and the consolidated patterns respectively during non-REM stage two sleep ( including on average 600 volumes ) using different window sizes ranging from n = 50 to 300 volumes . Each curve is calculated using different time windows comprising either the first n fMRI volumes ( purple , green ) or the last n fMRI volumes ( red , blue ) in the MSL ( purple , red ) or the CTL ( green , blue ) condition . Shaded area represents s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 24987 . 02110 . 7554/eLife . 24987 . 022Table 1 . Sleep architecture during post-training EEG-fMRI recording session on CTL and MSL condition nights . Mean and SEM values are reported in minutes . Sleep onset is calculated relative to the start of simultaneous EEG-fMRI recording . See also Materials and methods and Results for additional details . Slow wave sleep ( SWS ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24987 . 022CTLMSLMSL vs . CTLMeanSEMMeanSEMTPWake53 . 28 . 9866 . 97 . 021 . 680 . 12Stage 110 . 82 . 168 . 61 . 580 . 760 . 46Stage 242 . 75 . 8534 . 15 . 131 . 750 . 11SWS17 . 04 . 8810 . 23 . 770 . 950 . 36Sleep onset16 . 966 . 1215 . 803 . 010 . 170 . 87 Likewise , CI within the learning pattern changed as a function of motor task ( MSL vs . CTL ) over the period of NREM sleep ( significant time×task interaction , F10 , 120=2 . 00 , p<0 . 05; Figure 5a ) , but not during wakefulness ( F10 , 120=0 . 73 , p>0 . 6; Figure 5d ) . In contrast to the consolidated pattern , however , the CI within the learning pattern showed a significant difference between tasks only at the beginning of the night . In fact , when early in the NREM stage two period was analyzed separately , we found a significant increase in connectivity within the learning pattern in the MSL compared to CTL ( paired t test , mean of the first three time points ( corresponding to the first 7 min and 12 s of NREM stage-2 fMRI data ) in MSL compared to CTL , t12=2 . 32 , p<0 . 05 ) . However , there was no significant difference across tasks during later stage two sleep periods ( mean of the last three time points in MSL compared to CTL; paired t test , t12=0 . 47 , p>0 . 6 ) . In order to test the robustness of our findings with respect to the window size , we repeated our sliding window analyses iteratively by varying the size of the window . Consistently , a significant difference in the consolidated pattern connectivity was found between early and late periods in NREM sleep only following MSL using different window sizes ranging from 50 to 300 volumes ( significant window size×time period ( early vs . late ) interaction , F5 , 60=5 . 67 , p<0 . 0005; and significant effect of time period ( early vs . late ) in the MSL , F1 , 12=5 . 75 , p<0 . 05; Figure 5—figure supplement 1 ) ; thus suggesting that the pattern of results was robustly detectable over a range of time-windows . In order to further examine the role of the ventrolateral putamen in shaping the functional connectivity of the consolidated pattern as supported by the dual regression analysis , we employed a similar sliding window approach to investigate the dynamic changes in functional connectivity of the putamen ( same region as reported in Figure 3 ) within the consolidated pattern during NREM stage-2 sleep ( see Materials and methods ) . We found that functional connectivity of the putamen was gradually enhanced during NREM two sleep following MSL as compared to the CTL task ( significant time×task interaction , F10 , 120=3 . 21 , p=0 . 001; follow-up paired t tests comparing mean of the first three time points in MSL and CTL , t12=0 . 89 , p>0 . 3; and mean of the last three time points in MSL compared to CTL , t12=2 . 69 , p<0 . 05;Figure 5c ) , in line with CI analyses within the consolidated pattern . Again , there was no effect of task on the connectivity of putamen during the intermittent periods of wakefulness ( no time×task interaction , F10 , 120=0 . 59 , p>0 . 8; Figure 5f ) . Finally , a series of conventional seed-based functional connectivity analyses ( see Materials and methods ) on the resting-state and NREM sleep data largely confirmed our findings using the dual regression approach ( Figure 3—figure supplement 1 , and Figure 4—figure supplement 3 ) . The results revealed that increased connectivity of areas mainly within the consolidated pattern ( including the putamen and cerebellar cortex; Figure 1—figure supplement 2 , top row ) during both NREM sleep ( Figure 3—figure supplement 1 ) and the post-sleep resting-state condition ( [RS3 – RS1]; Figure 4—figure supplement 3 , top row ) were significantly associated with the amount of overnight gains in performance . Also , the functional connectivity within the posterior parietal cortex ( an area more activated during the learning compared to the retest; Figure 1—figure supplement 2 , bottom row ) was increased in the resting-state condition immediately following the MSL as compared to the baseline ( [RS2 – RS1]; Figure 4—figure supplement 3 , bottom row ) . Our findings demonstrate a gradual shift in motor memory representations following motor sequence learning; a transiently activated cortical trace is downscaled back to baseline levels and a subcortically-dominant and more interconnected trace , emerges during NREM sleep . These findings suggest that sleep supports both a homeostatic restoration of the memory trace potentiated during learning , and also actively reorganizes the memory trace at a systems-level . Specifically , our findings reveal that the ventrolateral putamen plays a central role in the emergence of the consolidated pattern during NREM sleep . Thirteen healthy right-handed adults ( seven female , age 27 . 4 ± 3 . 6; mean ± std ) passed the inclusion/exclusion criteria ( see below ) and completed the full experimental protocol . Ethical and scientific approval was obtained from the Research Ethics Board at the Institut Universitaire de Gériatrie de Montréal ( IUGM ) , Montreal , Quebec , Canada and informed written consent was obtained prior to entering the study . Subjects were included in the study based on the following inclusion/exclusion criteria . They had to be a non-smoker , medication free and to have normal body weight ( BMI ≤25 ) . They also had to present no history of psychiatric or neurologic disorders and to score ≤8 on the Beck Depression ( Beck et al . , 1974 ) and Anxiety ( Beck et al . , 1988 ) Inventories . Participants who had previous formal training as a typist or musician and who were categorized as extreme morning or evening types ( Horne Ostberg Morningness-Eveningness Scale [Horne and Ostberg , 1976] ) , worked at night , or had taken a trans-meridian trip ≤3 months prior to the experiment were also excluded from this study . Finally , subjects were included if they did not exhibit signs of excessive daytime sleepiness ( ≤9 on the Epworth Sleepiness Scale [Johns , 1991] ) and if the quality of their sleep was normal as assessed by the Pittsburgh Sleep Quality Index questionnaire ( Buysse et al . , 1989 ) . Participants were required to keep a regular sleep-wake cycle ( bed-time between 10:00 PM – 1:00 AM , wake-time between 07:00 AM – 10:00 AM ) and to abstain from consuming alcohol , caffeine or nicotine and from taking daytime naps throughout their participation in the study . Compliance to the schedule was assessed using both sleep diaries and wrist actigraphy ( Actiwatch 2 , Philips Respironics , Andover , MA , USA ) worn on the non-dominant wrist . Moreover , one week prior to the first experimental session ( Figure 1a ) , each participant experienced a screening night beginning at 11:00 PM in the mock scanner located at the Functional Neuroimaging Unit , Montreal , Quebec , Canada . Participants were given a two-hour opportunity to sleep . The mock scanner noise ( recorded from the scanner and presented at same approximate sound level ) and lighting conditions ( i . e . , lights off ) were similar to those of the experimental nights in the actual MRI scanner . EEG signal was recorded using the same MR-compatible electrode cap as that during the experimental nights . Following this two-hour sleep opportunity in the mock scanner , EEG electrodes were removed and subjects were permitted to sleep in the nearby sleep laboratory until 7:30AM . In order to be included in the study , a minimum of five minutes of consolidated NREM sleep ( i . e . , the minimum amount of data necessary for data analysis purposes ) during the two-hour screening period was required . Finally , 13 subjects met these inclusion criteria and completed the study , and were thus included in the data analyses ( see power analysis at the end of Materials and methods ) . Subjects were tested using a version of the motor sequence learning task ( Karni et al . , 1995 ) , in which they were required to perform self-generated finger movements with their non-dominant ( left ) hand as quickly and accurately as possible . A custom MR-compatible ergonomic response pad comprising four push buttons located in a row was used . Each participant was scanned under two different conditions including motor sequence learning ( MSL ) and control ( CTL ) , which were performed on two separate nights and the following mornings one week apart ( Figure 1A ) . The order of the MSL and the CTL conditions was counterbalanced across participants . In each condition , subjects first practiced the corresponding task in the evening ( learning session , S1 ) , and were tested again later on the same task in the following morning ( retest session , S2 ) . On the MSL night , subjects first explicitly memorized and slowly demonstrated to the experimenter the 5-item sequence of finger movements ( 4-1-3-2-4 , where one stands for the index finger and four for the little finger ) , until they could produce 3 consecutive correct 5-item sequences using an MR-compatible response pad . During the experiment , subjects lay supine in the scanner and executed the task following color-coded cues , which appeared on a screen visible via a mirror attached to the head coil . A green cross displayed in the center of the screen indicated the start of the task block , which terminated after 60 key presses . Each practice block was separated by a 15 s rest period ( indicated by a red cross ) during which subjects were instructed to keep their fingers immobile . Subjects were administered 14 blocks of practice during each of the learning and retest sessions . All subjects performed the sequence with an average accuracy of more than 83% , corresponding to more than 10/12 correct sequences per block . The timing of all key presses was recorded and speed was measured by the inter-key-press interval for correct responses only . On the CTL night , subjects were required to press all four fingers of the left hand simultaneously at the same average rhythm as the MSL task . This task was designed to have the same motor performance characteristics of the MSL condition ( e . g . , same number of finger flexion movements , same average inter-key press interval , all in the same amount of time ) , but importantly , without any sequence to learn . Similar to a previous study ( Orban et al . , 2010 ) , random individual key presses were not used as we intended to employ a control task that was uncued and thus internally generated and explicitly known , similar to the MSL task . The use of self-generated random sequences was not possible either , as it has been shown that people are not able to reliably produce random sequences of movements ( Figurska et al . , 2008 ) . Subjects were first instructed to press all four keys simultaneously following the rhythm of an auditory tone ( presented monotonically at 3 Hz ) as long as a green cross was displayed on the screen . This first pre-training step was intended to entrain subjects to the average speed of performance ( ~3 Hz ) normally observed during practice of the MSL task . After three blocks of practice ( 60 movements each ) in this pre-training step , subjects performed the task in the absence of the audio tones . Here , participants were instructed to maintain the same rhythm as practiced in the first pre-training step . Once performance was maintained at 3 Hz ( ±0 . 25 Hz ) for three consecutive blocks , this step of the pre-training phase was terminated . This pre-training phase ensured that subjects could reliably press all four keys simultaneously at the target rhythm . Similar to the MSL task , the pre-training was not included in subsequent analyses . For the practice sessions of the CTL task , participants were instructed to follow the same rhythm as practiced during the pre-training phase , and to rest during the presentation of the red cross . Similar to the MSL task , subjects were administered 14 blocks of practice , where each practice block terminated after 60 simultaneous 4-key presses , and each intervening rest period lasted 15 s . Again to be consistent with the MSL task , performance in the CTL task was measured as the inter-response interval between consecutive key presses ( i . e . , simultaneous flexion of all four fingers ) . The onset of the first of four finger presses was used in the subsequent analyses if the four fingers did not precisely touch their respective keys instantaneously . Images were collected using a 3T TIM TRIO Siemens scanner with a 12-channel head coil . A structural volume was acquired in the sagittal plane using a magnetization prepared rapid gradient echo ( MPRAGE ) sequence ( TR = 2300 ms , TE = 2 . 98 ms , FA = 9° , 176 slices , FoV = 256 × 256 mm² , voxel size = 1 × 1 × 1 mm³ ) . For functional acquisitions , an echo-planar imaging ( EPI ) gradient echo sequence was used with the following parameters: TR = 2160 ms; TE = 30 ms; FA = 90°; FoV = 220 × 220 mm²; matrix size = 64 × 64; 40 transverse slices , slice thickness = 3 mm; 10% inter-slice gap; inplane resolution = 3 . 44 × 3 . 44 mm² . In order to minimize the effects of gradient artifact on electroencephalography recordings , the sequence parameters were chosen so that the MR scan repetition time ( 2160 ms ) matched a common multiple of the EEG sample time ( 0 . 2 ms ) , the product of the scanner clock precision ( 0 . 1 μs ) and the number of slices ( 40 slices ) . Imaging parameters were the same during the resting-state scanning periods ( RS1 to RS4; Figure 1A ) , the practice sessions of the MSL and CTL tasks ( S1 and S2 ) , as well as post-training sleep where EEG measurements were simultaneously recorded with fMRI acquisitions . The number of acquired functional volumes during practice sessions was variable depending on the participant’s speed during the task . Each resting-state scan , however , lasted for 150 volumes or 6 min and 24 s . The sleep session was terminated when the maximum possible number of volumes for a single fMRI session ( 4000 volumes , lasting a maximum of 2 . 5 hr ) in the Siemens 3 . 0T TIM TRIO MRI system was reached , or if subjects voluntarily terminated the session . Similar to the acclimatization night , EEG electrodes were removed after this sleep opportunity and subjects were then allowed to sleep in the nearby sleep laboratory . The retest sessions were administrated at least 1 . 5 hr after awakening at 7:30 AM to ensure the dissipation of sleep inertia . Preprocessing of the imaging data was carried out using the FSL software package ( Beckmann et al . , 2003 ) and in-house programs developed in MATLAB . This included ( 1 ) removal of the first two volumes in each scan series , ( 2 ) slice time correction , ( 3 ) non-brain tissue removal ( 4 ) motion correction , ( 5 ) global intensity normalization , ( 6 ) spatial smoothing ( Gaussian kernel of FWHM 6 mm ) and ( 7 ) temporal high-pass filtering ( σ = 100 s ) . To achieve the transformation between the low-resolution functional data and the standard stereotaxic space ( MNI152: average T1 brain image constructed from 152 normal subjects ) , we performed two transformations . The first was from the functional image to the T1-weighted structural image ( using a 6 degree of freedom ( DOF ) transformation ) , and the second was from T1-weighted structural image to the average standard space ( using a 12 DOF linear affine transformation , voxel size = 2 × 2 × 2 mm ) . Also to minimize the effect of head motion during the sleep session data acquisition , those fMRI volumes that were scored as movement artefact in the EEG scoring ( less than 1% of total volumes in all subjects , see Polysomnographic Recording and Analysis ) were not included in the fMRI analyses . In our functional connectivity analysis on sleep data during NREM periods as classified by the EEG scoring ( see the EEG analysis for more details ) and intermittent wakefulness periods in between sleep stages , we selected the longest continuous segment of data to avoid discontinuities/sudden jumps in our data analysis . Also , because we performed a repeated measures analysis , we selected for each subject the minimum number of sleep or wakefulness volumes available in each task night ( MSL and CTL ) , so that the differences in the number of volumes across tasks did not affect our between-task contrast . These criteria resulted in the selection of 316 ± 32 ( mean ± s . e . m . ) fMRI volumes during NREM stage two and SWS , and 358 ± 17 ( mean ± s . e . m . ) volumes during wakefulness periods . For each subject , each condition ( MSL or CTL ) , and each practice session ( S1 or S2 ) , changes in brain regional responses were estimated using a model including responses to the task practice blocks weighted by each block’s performance speed ( inversely related to block duration ) . This regressor was then convolved with a double-gamma hemodynamic response function ( HRF ) . Six rotation and translation motion parameters were also included in the model as confounds . Because we were interested to extract a single map for each practice session , we did not separate areas related to the main effect of practice and those related to the modulation by performance speed . Hence , the use of a weighted regressor made this analysis more sensitive to identify all brain regions that are related to both task execution and performance improvements over the course of training . The subject-level regression coefficients ( COPEs in FSL ) and their variance maps ( VARCOPEs in FSL ) were then input to a group-level analysis , which used a mixed-effects ( FLAME1 , FSL ) general linear model ( Z > 3 . 5 , corrected family-wise error using Gaussian random field theory , cluster significance threshold of p<0 . 05 ) . In order to extract group-level maps of learning ( S1 ) and retest ( S2 ) practice sessions related to motor sequence learning as compared to simple motor performance , linear contrasts were performed to compare the difference between tasks during learning S1[MSL - CTL] and retest S2[MSL - CTL] . In this repeated-measures analysis , several regressors modeled the mean of each subject across different practice sessions . We name these thresholded Z-statistics maps ( Z > 3 . 5 ) corresponding to S1 and S2 sessions the ‘learning’ and the ‘consolidated’ patterns , respectively . In each of the learning and consolidated patterns , we calculated the total volume of activated voxels , the volume of cortical activation using a mask extracted from the Harvard-Oxford cortical structural atlas ( Desikan et al . , 2006 ) ( more than 25% tissue probability ) , and the volume of subcortical activation using a mask extracted from the Harvard-Oxford subcortical structural atlas and the probabilistic cerebellar atlas ( Diedrichsen et al . , 2009 ) ( more than 25% tissue probability ) . To assess the strength of functional connectivity within a network of brain regions , we used a ‘connectivity index’ ( or CI ) as proposed in Vahdat et al . ( Vahdat et al . , 2014 ) . Assume that X ( v ) specifies a vector of voxel intensities for a given brain network including n voxels , vi , i=1 , . . , n , and Y ( v , t ) represents a matrix of preprocessed BOLD data for a given subject and run ( e . g . during resting-state or sleep condition ) including m volumes , t=1 , . . , m . Data preprocessing included the regular steps explained above , as well as normalization of each voxel time series by the standard deviation of all the voxels’ time series inside the brain mask for each fMRI run and each subject . In this way , we made sure that between-run comparisons were not confounded by differences in total variation of BOLD signal across all brain voxels and time . We then first employed a spatial general linear model ( GLM ) to estimate the time course of activation of X ( v ) in the BOLD data Y ( v , t ) over time ( Equation 1 a ) , and then normalized the resulting regression coefficients β ( t ) using the standard deviation of the residuals ε ( v , t ) ( Equation 1 b , c ) . The normalized factor η ( t ) follows a student-t distribution , so that we could compare it over time and across different runs for a given network ( Kruggel et al . , 2002 ) . ( 1 ) Y ( v , t ) =X ( v ) β ( t ) +ϵ ( v , t ) ( a ) σ ( t ) =std ( ε ( v , t ) ) ( b ) η ( t ) =β ( t ) σ ( t ) ( c ) Finally , the CI was calculated as the power ( variance ) of the resulting normalized coefficients η ( t ) for each subject and each run . This index simply represents the strength of functional connectivity ( or functional integration ) of spatial pattern X ( v ) in a given fMRI data Y ( v , t ) . The normalization by the regression residuals makes this index specific to the co-activation of the desired pattern ( X ( v ) ) , and not to any general activation in all or parts of brain areas . This connectivity measure was selected as opposed to other methods such as ICA and seed-based analysis due to the following reasons . First , unlike the application of data driven approaches such as ICA , we sought to examine specific hypotheses regarding the reactivation and reorganization of pre-defined activation patterns recruited during the learning and retest task performances . Second , we sought to investigate the reactivation of the entire learning/consolidated pattern using a multivariate approach , as opposed to univariate methods such as seed-based analysis , which are suited to examine distinct connections at each time . Third , although the representational similarity analysis ( Kriegeskorte et al . , 2008 ) offers a unique approach to study reactivation ( Staresina et al . , 2013; Tambini and Davachi , 2013 ) , this approach is well-suited in designs aiming at studying activity patterns within a specific region of interest in the brain . As in this study we sought to investigate the reactivation patterns across all brain areas , a more general connectivity measure was desirable . Lastly , we sought to examine the dynamics of change in each specified network over time during sleep; hence estimation of a connectivity measure time-series was applicable . Nevertheless , we also report the results of seed-based functional connectivity analysis for comparison purposes . We performed two separate ANOVA to compare changes in CI within each of the learning and the consolidated patterns . One analysis compared CI during NREM sleep between tasks ( MSL – CTL ) , and the other one compared it across different resting-state conditions ( RS1 , RS2 , and RS3 ) within each pattern . As the aim of the current study was to capture off-line changes in connectivity ( whether related to simple passage of time or to sleep ) that happens following the acquisition of a new motor sequence , thus , we did not include the final resting-state condition following retest session ( RS4 ) in our analyses because it involved changes related to the retest practice session , which was beyond the scope and hypotheses of this study . Furthermore , in a control analysis , we investigated changes in CI with respect to several highly-reproducible brain networks ( Damoiseaux et al . , 2006 ) during resting-state periods and NREM sleep . In this analysis , the preprocessed fMRI data registered to the MNI standard space during the MSL condition in both S1 and S2 sessions were temporally-concatenated across all subjects . This time-concatenated matrix was then fed to the fast-ICA algorithm ( Hyvärinen , 1999 ) to extract 30 group-level spatial components . These Z-score spatial maps were correlated with the templates of four highly-reported resting-state networks including the default mode , visual , and left and right fronto-parietal networks ( Damoiseaux et al . , 2006 ) , and the corresponding group-level spatial maps with the highest correlation with templates were selected ( Figure 2—figure supplement 1 ) . Similar to the learning and consolidated patterns , these group-level spatial maps were then used as the pattern of interest within which CI was calculated during each of the resting-state and NREM sleep conditions in both MSL and CTL nights . Similar ANOVA models as explained above were conducted to investigate changes in CI in each spatial map during resting-state periods and NREM sleep . In order to identify the brain areas within a given network whose functional connectivity is significantly changed across different experimental conditions , we used dual regression analysis . The dual regression method is based on two levels of regression , the first level is similar to the spatial regression carried above ( Equation 1 a ) , where for each subject and run , a time series of regression coefficients is estimated from a group-level spatial map ( the given brain pattern ) . This time series is then normalized by its standard deviation , and is entered in a second level temporal regression as a predictor , where a spatial map is estimated for the same subject and run . These subject-level regression coefficients ( COPEs in FSL ) and their variance maps ( VARCOPEs in FSL ) were then input to a mixed-effects group-level analysis ( FLAME1 , FSL ) , where linear contrasts assessed the difference between task conditions ( corrected for family-wise error using Gaussian random field theory , cluster significance threshold of p<0 . 05 ) . Again , we performed two separate group-level GLMs for each of the learning and the consolidated patterns . One analysis compared functional connectivity within each pattern during NREM sleep between tasks ( MSL – CTL ) , while the other one used a repeated-measures ANOVA to assess changes in functional connectivity across different resting-state conditions ( RS1 , RS2 , and RS3 ) . To evaluate changes in the strength of functional connectivity in a given brain pattern over time , we calculated CI in a sliding window over the sleep fMRI session . In this analysis , for each subject , we selected NREM stage two sleep segments ( epochs ) of more than 50 volumes and concatenated all epochs in each run ( note that we did not have enough data during SWS to perform this analysis ) . Table 1 reports sleep architecture information inside the MRI scanner , including the average time spent awake and asleep , and in specific sleep stages , and sleep onset time relative to the start of simultaneous EEG-fMRI recording . We also applied the same concatenation procedure to the intermittent wakefulness periods ( epochs ) during the sleep run . This resulted in 954 ± 156 ( mean ± s . e . m . ) fMRI volumes during stage 2 of NREM sleep , and 1517 ± 235 ( mean ± s . e . m . ) volumes during wakefulness periods for each subject and each condition . We selected the first 600 volumes in each condition , so that we had enough and comparable number of fMRI volumes across subjects during stage two sleep or wakeful periods for group-level averaging ( the mean duration and the number of selected epochs averaged across subjects for each of wakefulness and NREM stage two sleep in MSL and CTL nights are reported in Figure 5—source data 1 ) . Then , we selected a window size of 100 volumes , which corresponded to 3 min and 36 s of data , and slid the window by 50-volumes steps ( half the window size overlap ) , which resulted in 11 data points in each of the NREM stage two and wakefulness periods . We then calculated CI in each time window for the learning and the consolidated patterns . Finally , we performed two-factor repeated measures ANOVA ( time by task ) to assess changes in CI over time and experimental tasks ( MSL and CTL ) . Similarly , in a region of interest ( ROI ) based analysis , we applied a similar sliding window approach to measure changes in the functional connectivity of a given ROI and the rest of consolidated pattern over time during NREM stage two sleep and wakefulness . Hence , in each time window , we calculated the Pearson’s correlation between the mean time series of the ROI and each voxel inside the mask of the consolidated pattern ( excluding the ROI voxels ) , and averaged the correlation values across all voxels . The average correlation values were then tested in a time ( 11 points ) by task ( two levels ) repeated measures ANOVA . To ensure that the method employed here was reproducible using more conventional seed-based approaches ( Vahdat et al . , 2011 , 2014 ) , we defined 12 ROIs based on peaks of activity during either the learning ( 6 ROIs ) or retest ( 6 ROIs ) practice sessions ( Figure 3—source data 1 ) . We defined a spherical mask ( radius = 6 mm ) around each seed in standard space . We re-sampled this mask first to the T1 weighted structural image of each subject and from there to the low-resolution functional space of that subject . For each subject , the average time course of the BOLD signal within the transformed mask during the resting-state ( RS1 , RS2 , and RS3 ) and NREM sleep ( stage two and SWS ) was calculated . The mean BOLD time-course of each ROI was used as a predictor in a subject-level GLM to assess the functional connectivity of that ROI with every other voxel in the brain . Physiological noise was removed from the fMRI data based on a procedure described in Vahdat et al . ( Vahdat et al . , 2011 ) . We calculated the following regressors: the average white-matter BOLD signal ( WM ) , cerebro-spinal fluid ( CSF ) , and the global signal . In total , nine nuisance regressors were used: WM , CSF , global signal and six motion parameters ( x , y , and z translations and rotations obtained from the motion correction step in preprocessing ) . Hence , for each subject and each run a separate multiple regression analysis was carried out using the time series of nuisance signals as confound regressors and the time series of the ROI as the regressor of interest . We included the time derivative of each ROI’s signal as a regressor in the GLM to account for possible time differences in the haemodynamic response function ( HRF ) of different cortical areas , as well as the latency for signal propagation from one cortical area to another ( Vahdat et al . , 2011 ) . This analysis produced maps of all voxels that were positively or negatively correlated with an ROI’s mean time-course . This was followed by between-subjects analyses that were carried out using a mixed-effects model ( FLAME1 ) implemented in FSL ( Beckmann et al . , 2003 ) . As in Vahdat et al . ( Vahdat et al . , 2011 , 2014 ) , we used each subject’s behavioral outcome ( overnight improvements in performance ) as a regressor to obtain a weighted average of the difference between conditions ( MSL compared to CTL ) . Corrections for multiple comparisons at the cluster level were carried out using Gaussian random field theory ( min Z > 2 . 7; cluster significance: p<0 . 05 , corrected ) . To correct for multiple ROIs we identified as statistically significant those clusters that had a probability level of better than p=0 . 05/12 ( 12 = number of ROIs ) . We then examined the correspondence between the behavioral regressor and the changes in functional connectivity ( for the resting-state data the change between RS3 or RS2 and RS1 , and for the sleep data the difference between the MSL and CTL nights ) . We constructed a vector for each connection between an ROI and target cluster whose elements were each subjects’ change in functional connectivity ( ΔFC ) . This vector was correlated with a vector of associated overnight improvements in motor performance . EEG was recorded by using an MR-compatible EEG cap ( Braincap MR , Easycap , Herrsching , Germany ) with 64 ring-type electrodes and two MRcompatible 32-channel amplifiers ( Brainamp MR plus , Brain Products GmbH , Gilching , Germany ) . EEG caps included 62 scalp electrodes referenced to FCz . Two bipolar ECG recordings were taken from V2-V5 and V3-V6 using an MR-compatible 16-channel bipolar amplifier ( Brainamp ExG MR , Brain Products GmbH , Gilching , Germany ) . Electrode-skin impedance was reduced to <5 KOhm using high-chloride abrasive electrode paste ( Abralyt 2000 HiCL; Easycap , Herrsching , Germany ) . In order to reduce movement-related EEG artifacts , subjects' heads were immobilized in the head-coil by surrounding the subject’s head with foam cushions . EEG was digitized at 5000 samples per second with a 500-nV resolution . Data were analog filtered by a low pass filter at 250 Hz and a high pass filter at 0 . 0159 Hz . Data were transferred via fiber optic cables to a personal computer where Vision Recorder Software , Version 1 . x ( Brain Products , Gilching , Germany ) was synchronized to the scanner clock . Sleep EEG was monitored online with Brain Products RecView software using online artifact correction . EEG data were preprocessed by a low-pass filter ( 60 Hz ) , down-sampled to 250 samples/sec and re-referenced to averaged mastoids . Scanner artifacts were removed using the ‘fMRI Artifact rejection and Sleep Scoring Toolbox ( FASST ) ’ for MATLAB ( Mathworks , Natick , Massachusetts , USA [Leclercq et al . , 2011] ) , using an adaptive average subtraction method . Ballistocardiographic artifacts were then removed using an algorithm based on a combination of artifact template subtraction and event-related independent component analysis ( Leclercq et al . , 2009 ) for artifacts time-locked to the R-peak of the QRS complex of the cardiac rhythm . Following gradient artifact and ballistocardiographic artifact correction , EEG recordings were sleep stage scored according to standard criteria ( Berry et al . , 2012 ) to identify periods of NREM sleep , free of any movement artifact , during which the EEG and fMRI data were analyzed . Results are shown as mean ± s . e . m . Based on our previous resting-state analysis results on neurologically healthy subjects ( Debas et al . , 2010; Vahdat et al . , 2011 ) , on average , a difference of 0 . 73 ± 0 . 56 ( mean ± std ) of baseline functional connectivity ( Z score units ) within the sensorimotor network have been detected following motor learning . Thirteen subjects would thus provide 90% power to detect changes across experimental conditions at a significance level of α = 0 . 05 ( Rigby and Vail , 1998 ) . Data were checked for normality and equality of variance across conditions . Unless otherwise indicated , statistical significance was determined using repeated measures two-tailed t tests ( when comparing two conditions ) or repeated measures ANOVAs ( when comparing more than two conditions ) . Results were considered to be significant at p<0 . 05 .	The idea that , while you sleep , you could be honing skills such as the ability to play a musical instrument may sound like science fiction . But studies have shown that sleep , in addition to being beneficial for physical and mental health , also enhances memories laid down during the day . The process by which the brain strengthens these memories is called consolidation , but exactly how this process works is unclear . Memories are thought to persist as altered connections between neurons , often referred to as memory traces . When we practice a skill , we activate the neurons encoding that skill over and over again , strengthening the connections between them . However , if this process were to continue unchecked , eventually the connections would become saturated and no further increases in strength could occur . One possible solution to this problem is that sleep enhances skill learning by downscaling connections across the brain as a whole , thereby freeing up capacity for further learning . Alternatively , sleep may reorganize an initially unstable memory trace into a more robust form with the potential to last a lifetime . To test these possibilities , Vahdat et al . asked healthy volunteers to practice a finger-tapping task while lying inside a brain scanner , and then to sleep inside that scanner for 2–3 hours . When the volunteers returned to the scanner the next morning and attempted the task again , they performed better than they had the previous night . Their brains also showed a different pattern of activity when performing the task after a night’s sleep . So what had happened overnight ? As the volunteers lay awake inside the scanner , their brains reactivated the memory trace formed during learning . However , as they entered a stage of non-dreaming sleep called non-REM sleep , this activity became weaker . At the same time , a new pattern of activity – the one that would dominate the scan the next morning – began to emerge . Whereas the post-learning activity was mainly in the brain’s outer layer , the cortex , the new pattern included other areas that are deeper within the brain . The activity of one deeper region in particular , the putamen , predicted how well the volunteers would perform the task the next day . Non-REM sleep thus strengthens memories via two complementary processes . It suppresses the initial memory trace formed during learning , and reorganizes the newly-learned information into a more stable state . These results might explain why people who are sleep-deprived often have impaired motor skills and memories . The findings also open up the possibility of enhancing newly learned skills by manipulating brain circuits during non-REM sleep .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Network-wide reorganization of procedural memory during NREM sleep revealed by fMRI
The ultimate overexpression of a protein could cause growth defects , which are known as the protein burden . However , the expression limit at which the protein-burden effect is triggered is still unclear . To estimate this limit , we systematically measured the overexpression limits of glycolytic proteins in Saccharomyces cerevisiae . The limits of some glycolytic proteins were up to 15% of the total cellular protein . These limits were independent of the proteins’ catalytic activities , a finding that was supported by an in silico analysis . Some proteins had low expression limits that were explained by their localization and metabolic perturbations . The codon usage should be highly optimized to trigger the protein-burden effect , even under strong transcriptional induction . The S–S-bond-connected aggregation mediated by the cysteine residues of a protein might affect its expression limit . Theoretically , only non-harmful proteins could be expressed up to the protein-burden limit . Therefore , we established a framework to distinguish proteins that are harmful and non-harmful upon overexpression . Protein overexpression is sometimes harmful to cellular growth ( Makanae et al . , 2013; Sopko et al . , 2006 ) , and a few mechanisms that could result in overexpression-triggered growth defects have been proposed ( Moriya , 2015 ) . Resource overload , stoichiometric imbalance , promiscuous interaction , and pathway modulation are triggered upon overexpression of , respectively , ( i ) a protein that has a high demand of cellular resources ( Dong et al . , 1995; Kintaka et al . , 2016; Stoebel et al . , 2008 ) , ( ii ) a protein that is part of a protein complex ( Kaizu et al . , 2010; Makanae et al . , 2013; Papp et al . , 2003 ) , ( iii ) a protein with a nonspecific interaction domain ( Ma et al . , 2010; Vavouri et al . , 2009 ) , and ( iv ) a protein that catalyzes a pathway ( Prelich , 2012; Youn et al . , 2017 ) . The mechanism of protein overexpression-triggered growth defects depends on the protein’s structural and functional characteristics , which are not always fully understood yet . Therefore , it is still difficult to predict whether the overexpression of a particular protein will be harmful to cellular growth and which mechanisms cause the harmful effect . The ultimate overexpression of a protein could be harmful for cellular growth , because it monopolizes and depletes limited resources that are involved in protein production , such as ribosomes and aminoacyl-tRNAs ( Gong et al . , 2006; Shachrai et al . , 2010; Vind et al . , 1993 ) . This phenomenon is known as the protein burden/cost effect ( Kafri et al . , 2016; Snoep et al . , 1995 ) . Proteins that have no harmful effects on cellular functions can be overexpressed up to a level that causes protein-burden–triggered growth defects . Conversely , if a protein cannot be overexpressed up to that level because it adversely affects cellular functioning , then overexpression of that protein will cause growth defects at relatively low expression levels , and we should consider mechanisms causing the defects . We previously developed a genetic tug-of-war ( gTOW ) method that can be used to estimate the expression limit of a target protein that triggers growth defects in the yeast Saccharomyces cerevisiae ( Makanae et al . , 2013; Moriya et al . , 2006 , 2012 ) . We estimated that the expression limit of a green fluorescent protein ( GFP ) was about 15% of the total cellular protein in S . cerevisiae ( Kintaka et al . , 2016 ) . Because GFP is a highly structured cytoplasmic protein unrelated to the cellular functions of yeast and thus harmless , this level could be considered the expression limit for any protein that causes growth defects triggered by the protein-burden effect . We predicted that the expression limits of some native highly expressed glycolytic proteins would be similar ( >15% of the total cellular protein ) ( Moriya , 2015 ) , suggesting that overexpression of these proteins would be harmless even though they have metabolic functions in yeast . The prediction was performed by the calculation of the proteins’ native expression levels ( Kulak et al . , 2014 ) and their gene copy number limits as determined by gTOW analysis , and has not yet been experimentally validated ( Moriya , 2015 ) . In this study , therefore , we tried to measure the expression limits of 29 glycolytic proteins to assess whether they are expressed up to levels that cause growth defects triggered by the protein-burden effect . There are five reasons why we chose glycolytic proteins: ( 1 ) because they are generally highly expressed and thus considered non-harmful upon high-level expression , they are excellent targets for examining whether they are expressed up to the protein-burden limit; ( 2 ) because they have been intensely studied , we have information that can allow us to manipulate their catalytic activities; ( 3 ) because the glycolytic pathway is one of the best-known metabolic pathways , we can predict and measure metabolic changes upon overexpression of these proteins; ( 4 ) they include a heteromer ( Pfks ) , a mitochondrially localized protein ( Adh3 ) , and membrane proteins ( Hxts ) , so we can assess how these characteristics affect expression limits; and ( 5 ) they include paralogs whose expressions are differently regulated , so that we can test how their differences affect their expression limits . We found that the expression limits of most of the 29 proteins were comparable to that of GFP and were independently determined by their catalytic activities , as suggested by a kinetic model of yeast glycolysis , confirming that their overexpression was harmless . Also , some of the proteins had far lower expression limits than those that would create a protein burden ( the expression limit of GFP ) , and their harmful effects were derived from their localization and metabolic perturbations . Owing to their codon optimality , native poorly expressed isozymes were not produced at levels sufficient to cause growth defects , even when they were expressed from the strong TDH3 promoter on the multicopy plasmids . Some glycolytic proteins formed S–S-bond-mediated aggregates when overexpressed , and this aggregation also seemed to restrict their expression limits . Figure 1A shows the experimental system ( plasmid ) used to express glycolytic proteins to limits that cause growth defects . The target glycolytic proteins analyzed in this study and their characteristics are summarized in Supplementary file 1 . We cloned each target gene on the gTOW plasmid ( pTOW40836 ) ( Moriya et al . , 2012 ) , such that the target protein was expressed under the control of the strong TDH3 promoter . PFK1 and PFK2 were exceptionally expressed from the less-strong PYK1/CDC19 promoter because their expression from the TDH3 promoter was too strong and consequently the growth of the transformants was very poor ( data not shown ) . The plasmids were used to transform the S . cerevisiae strain BY4741 ( ura3Δ leu2Δ ) . Copy numbers of the plasmid within the cell were controlled by changing growth conditions: up to 35 copies per cell in +leucine ( –uracil ) conditions ( low-copy conditions ) and up to 150 copies per cell in –leucine conditions ( high-copy conditions ) due to the biases 2 µm ORI and leu2-89 ( Moriya et al . , 2012 ) . In this experimental system , the maximum growth rates of the cells with the vector in +leucine conditions are much greater than those in –leucine conditions ( see Figure 1B–C ) , probably because the copy number of leu2-89 is not sufficient tosupport fully the leucine requirement in –leucine conditions . We measured the expression limits of most of the 29 target proteins in low-copy conditions because the expression levels produced under these conditions were already sufficient to cause growth defects . We first measured the growth rates of cells harboring gTOW plasmids . As shown in Figure 1B , all cells expressing glycolytic proteins , with the exceptions of those expressing HXT1 , HXT3 , and HXT4 , showed significant growth retardation compared to the vector control cells in low-copy conditions ( p<0 . 01 , Welch's t-test , Figure 1—source data 1 ) , indicating that expression of most of the glycolytic proteins caused growth defects . This observation was confirmed by the growth measurement in the high-copy conditions shown in Figure 1C , as cells expressing most of the glycolytic proteins did not grow in these conditions . Cells expressing GLK1 , FBA1 , GPM1 , PYK2 , PDC6 , ADH5 , and ADH4 could grow in high-copy conditions , although their growth rate was significantly lower than that of the vector control ( p<0 . 01 , Welch's t-test , Figure 1—source data 1 ) . As previously reported , the copy number of the gTOW plasmid inside the cell inversely reflects the deleterious effect of protein expression from the plasmid due to the gTOW effect: the plasmid copy number is low if expression of the target protein is harmful to cellular growth , and high if expression of the target protein is less harmful ( Kintaka et al . , 2016; Makanae et al . , 2013; Moriya et al . , 2006 , 2012 ) . Figure 1D and E show the copy numbers of gTOW plasmids in low- and high-copy conditions . In high-copy conditions , the copy numbers of gTOW plasmids expressing only GLK1 , FBA1 , GPM1 , PYK2 , PDC6 , and ADH4 were determined because yeast containing plasmids expressing the other protein-coding genes failed to grow . The copy numbers of all gTOW plasmids containing target genes were significantly lower than those containing the empty vector ( p<0 . 05 , Welch's t-test , Figure 1—source data 1 ) , confirming that they were expressed up to levels that caused growth defects in this experimental system . Because the copy numbers of plasmids expressing most of the glycolytic proteins tested here , other than GLK1 , PYK2 , and ADH4 , were not greater than the copy number of GFP , the expression of most glycolytic proteins in this experimental system seemed no less defective than that of GFP . We concluded that most glycolytic proteins were expressed close to their upper limits , even in low-copy conditions , and that a copy number increase in high-copy conditions was required to express GLK1 , FBA1 , GPM1 , PYK2 , PDC6 , and ADH4 to their limits . Next , we measured the expression levels of proteins within the cells , overexpressing them from the gTOW plasmid . Figure 2A shows how protein abundance was estimated . As reported previously , when GFP is expressed up to its expression limit ( and probably to the level required to trigger the protein-burden effect ) , the protein is visible within whole cellular proteins separated by sodium dodecyl sulfate polyacrylamide gel electrophoresis ( SDS-PAGE ) ( Kintaka et al . , 2016 ) . Because most glycolytic proteins were also expressed to similar levels by our experimental system , we measured the expression levels in arbitrary units ( AU ) as the relative intensities of target protein bands within the total protein separated by SDS-PAGE . The AU is considered to reflect the total number of amino acids within the band , and the relative number of protein molecules can be estimated by dividing AU by the protein length . When two proteins of different sizes give bands of the same number of AUs , the molecule number of the larger protein in the band should be lower than that of the smaller protein . The relationship between the AU and the percentage of total protein that we previously reported ( Kintaka et al . , 2016 ) was estimated , as shown in Figure 2—figure supplement 1 , as % total protein = 5 . 5 × AU . Representative images of SDS-PAGE-separated total proteins from cells harboring gTOW plasmids containing the target glycolytic protein genes are shown in Figure 2—figure supplement 2 . As shown in Figure 2B , most proteins were expressed at levels high enough to make them visible within the SDS-PAGE–separated whole cellular proteins , and the expression levels of Pgk1 , Gmp1 , Eno2 , and Eno1 were higher than that of GFP . By contrast , the expression levels of Pfk1 , Adh3 , and Hxts were almost undetectable with this experimental system . The x-fold increase in the expression of each target protein over its native level is shown in Figure 2—figure supplement 3 . The expression of some proteins was increased more than 10 , 000-fold in this experimental system . The expression of Tdh3 and Gpm1 further increased in –leucine conditions ( Figure 2C ) , but the cells in this condition had stunted growth ( Figure 1C ) . There could be two reasons that explain why the expression level of a protein is low: ( i ) its strong overexpression is harmful to cellular growth and ( ii ) its expression is repressed . We can distinguish these two possibilities by comparing the copy numbers and protein abundance as shown in Figure 2D because the copy number of the plasmid inversely reflects the deflective effect of protein expression as described above . Overexpression of Pfk1 , Adh3 , and Hxts seemed harmful because their copy numbers were lower than the those of the other proteins ( red circles in Figure 2D ) . By contrast , the expression of Glk1 , Pyk2 , and Pdc6 seemed to be repressed because their copy numbers were higher than those of the other proteins ( blue circles in Figure 2D ) . The relationship between protein expression levels and copy numbers ( shown in Figure 2D ) also suggested that expression levels are not solely determined by the promoter , because there was no significant correlation between the expression levels and plasmid copy numbers ( Pearson r=0 . 28 , p=0 . 12 ) . Next , we tried to reveal the factors causing harmful effects that restrict expression limits , and the mechanisms that repress protein expression . Overproduction of glycolytic proteins might cause metabolic perturbations by accelerating the reactions that these proteins catalyze . To test whether growth inhibition caused by metabolic perturbations limits the expression levels of glycolytic proteins , we analyzed the expression limits of mutant proteins with reduced enzymatic activities by introducing mutations into the catalytic centers ( here we call the mutant a ‘CC mutant’ ) . The mutations introduced into the glycolytic proteins are summarized in Supplementary file 1 . Figure 3A shows the expression levels of wild-type and mutant proteins in low-copy conditions . The expression levels of all proteins except Pfk1 , Fba1 , Tdh3 , and Eno1 were not significantly changed by introducing mutations . The expression levels of mutant Pfk1 and Tdh3 were significantly higher than those of wild-type proteins , and the expression levels of mutant Fba1 and Pgk1were significantly lower than those of wild-type proteins ( p<0 . 05 , Welch's t-test , Figure 3—source data 1 ) . For Pfk1 , Tdh3 , and Pfk2 ( which catalyzes the same reaction with Pfk1 ) , we further analyzed the expression levels in high-copy conditions ( Figure 3B ) . The expression level of mutant Pfk2 significantly increased when compared with that of wild-type Pfk2 ( p=0 . 046 , Welch's t-test ) . The expression levels of both wild-type and mutant Pfk1 were almost undetectable in these conditions , probably because their high-level expression was too toxic to the yeast cells . Because the expression level of wild-type Tdh3 was greater than that of mutant Tdh3 , the enzymatic activity of Tdh3 probably did not restrict its protein expression limit . We concluded that the expression limits of most of the glycolytic proteins studied here are not restricted by metabolic perturbations triggered by their overproduction , whereas the expression limits of Pfk1 and Pfk2 are exceptionally restricted by metabolic perturbations . The expression levels of mutant Pfk1 and Pfk2 , however , remained markedly lower than those of other glycolytic proteins , suggesting that other factors also influence their expression limits . Next , we focused on Adh3 , whose expression level was lower than those of the other glycolytic proteins , probably because high-level expression of Adh3 is harmful ( Figure 2D ) . This harmful effect , however , is not triggered by metabolic perturbations , because the expression level of the mutant Adh3 with reduced enzymatic activity was almost the same as that of wild-type Adh3 ( Figure 3A ) . Among the glycolytic proteins tested in this study , Adh3 alone is a mitochondrial protein ( Young and Pilgrim , 1985 ) . To test whether the mitochondrial localization of Adh3 restricts its protein expression limit , we constructed a mutant without the mitochondrial targeting sequence ( ΔMTS-Adh3 , Figure 3—figure supplement 1 ) and compared its expression level to that of wild-type Adh3 . As shown in Figure 3C and D , the expression level of ΔMTS-Adh3 was about three times higher than that of wild-type Adh3 . We concluded that the mitochondrial localization of Adh3 restricts its expression limit , probably because the high-level expression of this mitochondrial protein causes growth defects due to overloading of mitochondrial transport resources ( Kintaka et al . , 2016 ) . The results suggested that the overexpression of most glycolytic proteins do not cause serious metabolic perturbations . To test whether this speculation is theoretically supported , we used a kinetic model of the yeast glycolytic pathway ( Smallbone et al . , 2013 ) ; a schematic diagram of which is shown in Figure 4—figure supplement 1 . Figure 4A–D shows the x-fold change of glycolytic metabolites in simulations in which each glycolytic protein is overexpressed up to 128-fold compared with the wild-type simulation . Overproduction of 14 of 20 glycolytic proteins did not cause more than a two-fold metabolic change ( gray lines in Figure 4A ) , indicating that overexpression of most glycolytic proteins does not cause serious metabolic perturbations . By contrast , the overproduction of Hxk1 and Hxk2 affected glycolytic metabolism throughout , and the overproduction of Pdc1 and Cdc19 affected metabolism locally ( Figure 4B–C ) . Because the experimental results using CC mutants suggested that their overexpression did not trigger metabolic perturbations leading to the growth defects , unknown mechanisms to explain the discrepancy might exist . Overproduction of Pfk1 or Pfk2 did not cause a metabolic change , because , in the model , these individual enzymes did not catalyze the Pfk reaction whereas the Pfk1–Pfk2 complex did . Simultaneous overproduction of both Pfk1 and Pfk2 caused severe metabolic changes ( Figure 4D ) , whose pattern was quite similar to the changes caused by Hxk1 and Hxk2 overexpression ( Figure 4—source data 1 ) ( except G6P and F6P levels did not change as these metabolites are upstream of the Pfk reaction ) . Although metabolic changes upon overexpression of Pfks and Hxks showed a similar pattern , overexpression of Pfks but not Hxks caused growth defects ( Figures 1 and 2 ) , and catalytic mutations of only Pfks increased the expression limit of this protein ( Figure 3 ) . Hence , the metabolic changes observed in the simulation do not by themselves explain the growth defects triggered by the overexpression of Pfks . To further characterize physiological conditions that are triggered by the overexpression of Pfks , we next analyzed metabolic changes in yeast cells overexpressing wild-type and CC mutant Pfk2 over the vector control by measuring 35 metabolites ( Figure 5—source data 1 ) , because the CC mutants showed increased expression limits ( Figure 3B ) . Figure 5A shows changes in the levels of nine glycolytic metabolites . Overexpression of both wild-type and CC mutant Pfk2 triggered significant reductions in some metabolites ( p<0 . 05 , Welch’s t-test , Figure 5—source data 1 ) . Moreover , the patterns of metabolic changes were inconsistent with those predicted by the model ( Figure 5—figure supplement 1 ) . These metabolic reductions were thus not triggered by the catalytic activity of Pfk2 . We noticed , however , that the level of F16bP in the cells overexpressing wild-type Pfk2 was >3-fold higher than that in the CC mutant Pfk2 ( Figure 5A , p<0 . 05 , Welch’s t-test , Figure 5—source data 1 ) . F16bP is the product of Pfk catalysis and the simulation predicted an increase in the F16bP level upon overexpression of Pfks ( Figure 4D ) , suggesting that the catalytic activity of Pfk2 triggers this metabolic difference . We next measured metabolic changes in 29 metabolites in cells overexpressing wild-type Pfk1 and Tdh3 and their CC mutants because these CC mutants also showed increased expression limits ( Figure 3A ) . As shown in Figure 5B and C , levels of glycolytic metabolites in the cells overexpressing wild-type Pfk1 and Tdh3 were not changed more than three-fold over the vector control . We did not observe any reproducible increase in F16bP level in the cells overexpressing wild-type Pfk1 over levels in its CC mutant . Moreover , overall metabolic changes were higher in the cells overexpressing CC mutant than in those expressing wild-type Pfk1 ( Figure 5B ) . We did not observe any reproducible difference in the metabolic changes between the cells overexpressing wild-type Tdh3 and its CC mutant ( Figure 5C ) . We thus concluded that overexpression of Pfk1 and Tdh3 did not trigger significant metabolic changes through their catalytic activities , at least in the detected glycolytic metabolites . We next focused on Glk1 , Pyk2 , and Pdc6 , as their expression levels were lower than those of other glycolytic proteins in low-copy conditions , while they did not seem to be harmful ( Figure 2D ) . Moreover , the expression levels of Glk1 and Pyk2 were significantly elevated in high-copy conditions ( Figure 6A ) . These results raised the possibility that expressed protein levels per single gene copy are lower than those for other genes either because protein synthesis rates are low or because protein degradation rates are high . Codon optimality strongly contributes totranslational elongation rate and mRNA stability ( Presnyak et al . , 2015 ) . Therefore , we analyzed the tRNA adaptation index of a gene ( tAIg ) ( Tuller et al . , 2010 ) for the the glycolytic genes studied here ( Figure 6B and Figure 6—figure supplement 1 ) and noticed that GLK1 , PYK2 , and PDC6 had a much lower tAIg than the other glycolytic genes . To test whether the codon optimality of GLK1 affects the protein expression level , we constructed codon-optimized GLK1 ( CoGLK1 ) and measured its protein expression level ( Figure 6A ) . Glk1 expressed from CoGLK1 was present at levels 3 . 6 and 4 . 7 times higher than that expressed from native GLK1 in low- and high-copy conditions , respectively . We concluded that Glk1 expression was low due to its low codon optimality . Glk1 expression increases after a diauxic shift—a growth-phase shift triggered by the carbon source alteration from glucose to ethanol ( Zampar et al . , 2013 ) . We speculated that GLK1 might have a codon usage that is optimized for the tRNA pool after a diauxic shift and its translational rate might be higher after the shift . To investigate this possibility , we monitored the expression levels of GFPs with different codon usages under different growth conditions . We constructed two GFP genes whose codons were differently optimized: ( i ) oG-GFP , whose codons were selected at random with probabilities obtained from the codon usage table of GLK1 , and ( ii ) oT-GFP , whose codons were substituted by the synonymous codon used most frequently in TDH3 . We added the ornithine decarboxylase degron ( Jungbluth et al . , 2010 ) to the C-terminus of these GFP genes to allow accurate monitoring of the timings of their syntheses . Figure 6C shows the GFP fluorescence and the growth of cells expressing the GFP genes . The GFP fluorescence of both genes peaked during their exponential growth phases . Next , we measured the lag time between the inflection points of the GFP fluorescence curve and the growth curve ( where the diauxic shift is supposed to happen ) , as shown in Figure 6D . Because the lag times were not significantly different ( p=0 . 44 ) , we concluded that the codon usage of GLK1 was not optimized to maximize their translation after the diauxic shift . When we measured the expression levels of Eno2 and Pgk1 proteins , we unexpectedly observed high-molecular-weight bands whose sizes ( ~125 and 100 kDa ) were different from the sizes of the monomers or dimers of Eno2 and Pgk1 ( 45 and 90 kDa , respectively ) ( Figure 7A ) . The band formation was independent of the catalytic activities of Eno2 because the bands were also observed in the experiment with Eno2 CC mutant ( Figure 7A ) . The band in the Eno2 experiment seemed to be S–S-bond-connected protein aggregates because it disappeared after treatment with the reducing agent dithiothreitol ( DTT ) ( Figure 7B ) . We confirmed that cysteines were responsible for creating these bands because they disappeared when cysteine residues were removed from Pgk1 and Eno2 ( Figure 7—figure supplement 1 ) . To identify the protein species in the bands , we analyzed them by liquid chromatography-tandem mass spectrometry ( LC-MS/MS ) . As shown in Figure 7C and D , we mainly detected glycolytic proteins , translational elongation factors , and translation initiation factors , in addition to each overexpressed protein . Most of the detected proteins were also detected in the CC mutant experiment ( Figure 7C ) . This aggregation did not seem to affect the expression limits of Eno3 and Pgk1 because the expression limits of wild-type proteins and cysteine-less mutants ( Eno2-C248S and Pgk1-C98S ) were indistinguishable ( Figure 7—figure supplement 2 ) . Next , we focused on Tpi1 because it was detected in both Pgk1 and Eno2 aggregates ( Figure 7C–D ) and because its expression limit ( 1 . 7 U ) was lower than that of the highest-limit proteins such as Pgk1 and Gpm1 ( >2 . 0 U ) ( Figure 2B ) . As shown in Figure 8A , Tpi1 constituted many aggregation bands upon its overexpression . The majority of these bands disappeared when cysteine residues were removed from Tpi1 ( C41S , C126S ) , or after DTT treatment . These results suggested that non-specific S-S-bond-connected aggregation occurred upon overexpression of Tpi1 . To test whether the aggregation restricts the Tpi1 expression limit , we measured the expression limits of cysteine-less Tpi1 . As shown in Figure 8B , the expression levels of cysteine-less Tpi1 significantly increased above those of wild-type Tpi1 . Because mutant Tpi1 levels were higher than wild-type Tpi1 levels , even in +DTT conditions , the removal of cysteine residues would not only prevent the formation of aggregates but would also increase the expression limit of Tpi1 . According to the protein-burden concept ( Dong et al . , 1995; Kafri et al . , 2016; Shah et al . , 2013; Snoep et al . , 1995; Stoebel et al . , 2008 ) , the ultimate overexpression of any protein could cause growth defects by overloading basic protein production resources . But only non-harmful proteins can be overexpressed up to the ultimate level , or the protein-burden limit , because the expression limit of harmful proteins should be restricted by their harmful effects . Knowing the protein-burden limit itself is thus essential when seeking to determine whether the overexpression of a protein is harmful to cellular functions . We previously estimated the protein-burden limit of S . cerevisiae cells by measuring the expression level of GFP that causes growth defects . This was 15% of the total cellular protein ( Kintaka et al . , 2016 ) . In this study , we first tried to measure the expression limits of yeast glycolytic proteins in order to confirm whether the protein-burden limit measured using GFP applies to endogenous proteins . Most of the glycolytic proteins studied here caused growth defects when they were expressed from a strong promoter on a multicopy plasmid ( Figure 1 ) . The expression levels of some glycolytic proteins in these conditions were , indeed , comparative or even higher than that of GFP ( Figure 2 ) . Also , their expression levels did not increase due to mutations in their catalytic centers ( Figure 3A ) . These results strongly suggest that the protein-burden effect largely determines the expression limit , and that the limit is around 15% of the total cellular protein . Among the glycolytic proteins studied here , Pgk1 , Gpm1 , and Eno2 had the highest expression limits . Although Pgk1 ( 44 . 7 kDa ) and Eno2 ( 46 . 9 kDa ) are 1 . 5-fold larger than Gpm1 ( 27 . 6 kDa ) , their expression limits were similar to those of Gpm1 ( Figure 2B ) . These results suggest that a protein's size does not affect its expression limit , at least for proteins in this molecular weight range . These data also suggest that the expression limits of proteins are not determined by the molar concentrations of those proteins but by the cost of the protein production . Some other glycolytic proteins , such as Pfk1 , Pfk2 , Adh3 , and Hxts , showed expression limits far below the protein burden limit of 15% ( Figure 2 ) , suggesting that overexpression of these proteins is harmful . Of the 18 glycolytic proteins studied , Pfk1 and Pfk2 were the only ones whose expression limits were significantly increased by mutations in their catalytic centers ( Figure 3A–B ) , suggesting that their metabolic functions restrict their expression limits . We think , however , that the metabolic perturbations that trigged the overexpression only partially affect the expression limits because the expression limits of the mutant proteins were still far below those of other glycolytic proteins ( Figure 3A–B ) . Pfk1 and Pfk2 form a hetero-octameric complex , and their stoichiometric imbalance leads to the formation of filamentous Pfk1 structures in the cytosol ( Schwock et al . , 2004 ) . This stoichiometry-imbalance-triggered protein aggregate might cause growth defects upon overexpression of Pfk1 ( and Pfk2 ) , although we could not confirm this hypothesis because simultaneous overexpression of Pfk1 and Pfk2 did not increase the expression limits of these proteins ( our unpublished observation ) . The CC mutants of Fba1 and Pgk1 showed lower expression limits than their wild-type proteins ( Figure 3A ) . We currently do not have any substantial and consistent explanation of why these CC mutants have lower expression limits . We can assume some general mechanisms: CC mutant proteins sequester the wild-type enzymes into inactive complexes; CC mutant proteins sequester the substrate molecules for the wild-type enzymes; or mutation in the catalytic center destabilizes the structure of the enzyme . For example , Fba1 is an essential homodimeric enzyme ( UniProtKB: P14540 ) . Overexpression of CC mutant Fba1 molecules might sequester active wild-type Fba1 molecules into inactive complexes . The limit of CC mutant Tdh3 was higher than that of the wild-type in low-copy conditions whereas it was lower in high-copy conditions ( Figure 3A–B ) . This strange behavior might be related to its moonlighting function . The catalytic activity of Tdh3 did not seem to explain the difference in the expression limits of wild-type and CC mutant Tdh3 ( Figure 5C ) . Beside its metabolic function , Tdh3 directly binds to Sir2 protein to promote transcriptional silencing , and a mutation in the catalytic center ( C150G ) reduces the silencing ( Ringel et al . , 2013 ) . It is thus possible that the CC mutant Tdh3 ( C150S ) causes silencing in a dose-dependent manner by competing with wild-type Tdh3 for binding with Sir2 . We speculated that the localization of Adh3 to the mitochondria and of Hxts to the plasma membrane restricted their expression limits because localized proteins overload more-limited localization resources ( Kintaka et al . , 2016 ) . This hypothesis was confirmed because the removal of the mitochondrial signal from Adh3 increased its expression limit ( Figure 3C , D ) . We also speculated that the expression limits of membrane proteins such as Hxts should be restricted by their localization , although there is no experimental evidence to support this hypothesis yet . The fact that the expression limits of most glycolytic proteins were not affected by mutations in their catalytic centers ( Figure 3A ) suggests that their overexpression does not cause metabolic perturbations . This finding was theoretically confirmed by simulations using a kinetic model of glycolytic metabolism ( Figure 4 ) . The reason why their overexpression does not cause metabolic perturbations is probably that they are bidirectional enzymes: the metabolic flux should be determined only by the availability of substrates when the concentrations of these enzymes are more than a certain level . To support this idea , the overexpression of 14 bidirectional enzymes showed minor metabolic changes , whereas the overexpression of 6 unidirectional enzymes ( including Hxks , Pfks , Cdc19 , and Pdc1 ) showed strong metabolic changes in the simulation ( Figure 4 ) . The expression limits of Hxks in the cells , however , were close to the protein burden limit ( Figure 2B ) and were not affected by mutations in the catalytic center ( Figure 3A ) . These results suggest an additional mechanism that is not implemented into the model that allows cells to avoid the effects of big metabolic changes upon overexpression of Hxks: a mechanism that prevents these metabolic perturbations from occurring , or a mechanism that prevents these metabolic perturbations from causing growth defects . Through the metabolic analysis , we realized that we currently do not have any systematic way to identify metabolic changes that are directly triggered by the overexpression of an enzyme , because metabolism is interconnected and the overexpression of a protein could cause non-specific perturbations that ultimately affect metabolism . Moreover , we know very little about how much change in which metabolite triggers a growth defect . Comparison of the metabolic changes in cells overexpressing wild-type and CC-mutant enzymes could be one solution for this . In fact , we observed a three-fold difference between cells expressing wild-type and CC mutant Pfk2 ( Figure 5A ) . Nevertheless , once again , we cannot conclude from our current knowledge that this difference causes the difference in the expression limits of these two forms of Pfk2 . By using a mathematical model , we tried to predict the potential metabolic changes that would be triggered by overexpression of an enzyme without considering unknown effects other than the enzyme's metabolic activity . In the simulations , overexpression of Pfks and Hxks triggered divergent and almost catastrophic metabolic changes ( ~1000-fold increase in some metabolites , Figure 4B , D ) , suggesting that their overexpression would cause growth defects due to these strong metabolic perturbations . We thus expected to obtain similar metabolic changes upon overexpression of Pfks , whose CC mutants had higher expression limits . We did not , however , observe such great changes ( Figure 5A–B and Figure 5—figure supplement 1 ) . To answer these issues precisely , we need a much deeper understanding of the connections between metabolite levels and cellular growth . The translational rate of some glycolytic proteins , including Glk1 , seemed low because of their lower codon optimality ( Figure 6 ) . Actually , the codon optimality of Glk1 ( tAIg = 0 . 38 ) is close to the average for all the yeast genes ( tAIg = 0 . 37 ) , and the codon optimality of other glycolytic proteins studied here is exceptionally high ( Figure 6—figure supplement 1 ) . These observations suggest that the codon optimality of most yeast genes is not high enough to allow expression of their proteins up to the protein-burden limit , even if they are expressed from a strong promoter on a multicopy plasmid . Overexpression of Eno2 , Pgk1 , and Tpi1 triggered S–S-bond-connected aggregation ( Figures 7 and 8 ) , and the aggregates that are formed contain other glycolytic proteins and translational factors ( Figure 7C–D ) . We think that this aggregation is triggered by spontaneous non-specific S–S bond formation among proteins existing in high concentrations . Interestingly , we also detected the same proteins within the gel of the corresponding molecular weight in the vector control , although the amounts estimated by LC-MS/MS were lower and cannot be identified as visible protein bands ( Figure 7—source data 1 ) . Therefore , we speculated that the S–S-bond-mediated protein aggregation occurs even in normal physiological conditions , but it is accelerated by an increase in the concentration of cytoplasmic proteins upon overexpression of glycolytic proteins . This aggregation might affect the expression limits of cysteine-containing glycolytic proteins , because changing the cysteine residues of Tpi1 into serine residues increases the protein's expression limit ( Figure 8B ) . As the amount of protein corresponding to the Tpi1 monomer was not changed by DTT treatment , the expression level of Tpi1 should not be reduced simply by aggregation but by the harmful effect of spontaneous S–S bond formation . This hypothesis is supported by the fact that the most highly expressed glycolytic protein Gpm1 , which has a molecular weight similar to that of Tpi1 , does not have a cysteine residue . The deleterious effect of this aggregation , however , seems protein-specific because the expression limits of Pgk1 and Eno1 were among highest measured ( Figure 2A ) , and removal of their cysteine did not increase their expression limits ( Figure 7—figure supplement 1 ) . As described above , we revealed mechanisms that restrict the expression limits of some glycolytic proteins . We do not think , however , that these mechanisms are the sole factors restricting the expression limits of these proteins . The expression limits of ΔMTS-Adh3 ( 0 . 45 AU , Figure 3D ) and CoGlkl ( 1 . 07 AU , Figure 6A ) are still lower than those of other high limit proteins such as Pgk1 and Gpm1 ( 2 . 26 AU and 2 . 63 AU , respectively , Figure 2B ) . It is thus likely that multiple mechanisms restrict the expression limits of these proteins . Protein misfolding or misinteraction is considered to cause toxicity upon high-level expression of a protein with low translational robustness , low folding stability , or a high propensity for misinteraction ( Drummond and Wilke , 2009; Zhang and Yang , 2015 ) . In general , highly expressed proteins such as glycolytic proteins are thus evolved to avoid these characteristics ( Zhang and Yang , 2015 ) , and that should be a requirement for a protein to be expressed up to the protein-burden limit . Cdc19 , one of the glycolytic proteins studied here , aggregates in a stress-induced and reversible manner through a region of low compositional complexity ( Saad et al . , 2017 ) . This aggregation capacity of Cdc19 might explain why its expression limit ( 0 . 42 AU ) is lower than the protein burden limit ( >2 . 0 AU ) ( Figure 2B ) . Our finding in Figure 8 suggested that the high-level expression of a cysteine-containing protein could also cause a misinteraction-triggered toxic effect; hence unimportant cysteines should be avoided in highly expressed proteins . Concentration-dependent liquid phase separation is also considered to cause toxicity upon overexpression of structurally disordered and nucleic-acid-binding proteins ( Bolognesi et al . , 2016 ) . We do not think that this mechanism caused growth defects upon overexpression of the glycolytic proteins studied here because they are less structurally disordered ( Moriya , 2015 ) and not nucleic-acid-binding proteins . We summarize our analysis in Supplementary file 1 . In conclusion , we established the ultimate expression level that causes cellular growth defects due to the protein-burden effect as around 15% of the total cellular protein . The next interesting theme is to identify characteristics of proteins that can be overexpressed up to the protein-burden limit because such proteins are considered non-harmful to cellular functions . Those characteristics should conversely imply the properties of proteins that are harmful when they are overexpressed . BY4741 ( MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0 ) ( Brachmann et al . , 1998 ) was used as the host strain for the experiments . Yeast culture and transformation were performed as previously described ( Amberg et al . , 2005 ) . A synthetic complete ( SC ) medium without uracil ( Ura ) or leucine ( Leu ) , as indicated , was used for yeast culture . The plasmids used in the study are listed in the Key Resources Table ( Supplementary file 2 ) . The plasmids were constructed by the homologous recombination activity of yeast cells ( Oldenburg et al . , 1997 ) , and their sequences were verified by DNA sequencing . The plasmid copy number was measured by real-time polymerase chain reaction , as previously described ( Moriya et al . , 2006 ) , using a LightCycler480 system ( Roche ) . The LEU2 ( LEU2-2F and LEU2-2R ) and LEU3 primer sets ( LEU3-3F and LEU3-3R ) were used to amplify DNA fragments of the pTOW40836 plasmid and genomic DNAs , respectively . Mean values , standard deviations ( SD ) , and p-values of Welch's t-test were calculated from biological triplicates . The total protein was extracted from log-phase cells with an NuPAGE LDS sample buffer ( ThermoFisher ) after 0 . 2N NaOH treatment ( Kushnirov , 2000 ) . For each analysis , the total protein extracted from two optical density ( OD ) units of cells with OD600 was used . For total protein visualization , the extracted total protein was labeled with Ezlabel FluoroNeo ( ATTO ) , as described in the manufacturer’s protocol , and separated by 4–12% SDS-PAGE . Proteins were detected and measured using the LAS-4000 image analyzer ( GE Healthcare ) in SYBR–green fluorescence detection mode and Image Quant TL software ( GE Healthcare ) . The expression of each target protein ( AU ) was calculated , as shown in Figure 2 . Average values , SD , and p-values of Welch's t-test were calculated from biological triplicates . For detection of Tpi1 , the SDS-PAGE-separated proteins were transferred to a PVDF membrane ( ThermoFisher ) . Tpi1 was detected using an anti-Tpi1 antibody ( RRID:AB_11130951 ) , a peroxidase-conjugated secondary antibody ( Nichirei Biosciences ) , and a chemiluminescent reagent ( ThermoFisher ) . The chemiluminescent image was acquired with an LAS-4000 image analyzer in chemiluminescence detection mode . Cellular growth and GFP fluorescence were measured by monitoring OD595 and Ex485 nm/Em 535 nm , respectively , every 30 min using an Infinite F200 microplate reader ( Tecan ) . The maximum growth rate ( MGR ) was calculated as described previously ( Moriya et al . , 2006 ) . Average values , SD , and p-values of Welch's t-test were calculated from biological triplicates . We define growth defect based on a significant reduction in the maximum growth rate of the cells overexpressing a target protein compared with that of cells overexpressing the control vector ( p<0 . 01 , Welch’s t-test ) . We used a kinetic model of the yeast glycolytic pathway developed previously ( Smallbone et al . , 2013 ) . To predict metabolic changes upon overexpression of glycolytic proteins , we changed the initial concentration of each target protein 128-fold over the original concentration , and calculated the concentration of each metabolite at the steady state . We did not analyze the metabolism for the overproduction of Pyk2 , Adh2 , Adh3 , Adh4 , and Adh5 , because they were not included or because their turnover ratios were set to 0 in the model . We also did not analyze Hxts overexpression , because its concentration was not changeable in the model . Yeast cells were aerobically cultivated at 30°C for 24–48 hr in an SC–Ura medium . The cells were inoculated into 200 mL of the medium at an OD600 of 0 . 5 and then aerobically cultured at 30°C for 3 hr . 1 . 0 mL of culture containing cells with an of OD600 of 50 was mixed with 1 . 4 mL of methanol solution pre-cooled at –80°C . The sample was centrifuged at 5 , 000 g at –20°C for 5 min . After the removal of the supernatant , 1 . 0 mL of 75% ethanol pre-heated at 95°C was added to the sample , which was then incubated for 3 min at 95°C . 10 µL of 17 µM D-camphor sulfonic acid was added to the sample as an internal standard for liquid chromatography triple–stage quadrupole-mass spectrometry ( LC-QqQ-MS ) analysis . After placing on ice for 5 min , the sample was centrifuged at 5 , 000 g at 4°C for 5 min to remove cell debris . 950 µL of the supernatant was transferred to a new tube and centrifuged at 15 , 000 rpm at 4°C for 5 min . 300 µL of the supernatant collected as cell extract was dried under vacuum , and then stored at –80°C until the mass spectrometry analysis . All metabolites were measured using LC-QqQ-MS . LC-QqQ-MS analysis was performed according to the method given by Kato et al . ( 2012 ) . We calculated the normalized internal standard peak areas for each metabolite . Samples from three independent cultures were analyzed for the cells overexpressing Pfk2 , Pfk2 CC mutant , and the vector control . Samples from two independent cultures were analyzed for the cells overexpressing Pfk1 , Pfk1 CC mutant , Tdh3 , Tdh3 CC mutant , and the vector control . The total protein extracts in the overexpression of Eno2 , Eno2 CC mutant , and Pgk1 were separated by SDS-PAGE and stained by Coomassie staining solution ( ThermoFisher ) . Proteins of interest were excised from the gels and digested using trypsin . The tryptic peptides were analyzed by LC-MS/MS consisting of an LTQ-Orbitrap mass spectrometer ( ThermoFisher ) and a DiNa nano LC ( KYA Technologies ) system according to the method described previously ( Kito et al . , 2016 ) . The peptide mixture was separated with reverse-phase chromatography . Mobile phase A contained 0 . 1% formic acid , and mobile phase B contained 0 . 1% formic acid/80% acetonitrile . Peptides were eluted at a flow rate of 200 nL/minute using a 55 min gradient as follows: from 0% to 32% solvent B over 45 min , from 32% to 40% solvent B over 5 min , and from 40% to 80% solvent B over 5 min . The acquired MS/MS spectra were subjected to a database search against the protein sequences of S . cerevisiae . The aggregating protein species in Figure 7 are those for which the number of peptide hits in the database search was five or more and was 1 . 5-fold more than that of the vector control .	If a cell makes too much of a given protein , it can sometimes cause problems and impair the cell’s growth . Overproducing some proteins may deplete the cell’s limited resources , meaning it does not have enough to make other more essential proteins . This phenomenon is known as the protein burden effect . Theoretically , only harmless proteins can be overproduced up to a level where growth would be impaired in this way . Conversely , if an overproduced protein causes harm before it becomes a burden on resources , scientists must consider other mechanisms to explain the cell’s problems , namely that the protein itself is harmful . Knowing the ultimate level of protein production that could cause the protein burden effect – the protein burden limit – would allow scientists to distinguish between harmful and non-harmful proteins . However , to date , this limit had not been defined for any cell . Eguchi et al . have now tried to estimate the protein burden limit for budding yeast – one of the best-studied experimental organisms . The experiments first focused on enzymes involved in alcoholic fermentation because they were expected to be non-harmful . Some of these enzymes were overproduced to the level were the made up 15% of all the cell’s proteins before they started to cause growth defects . The same results were seen with versions of the enzymes that had been mutated to be less active , leading Eguchi et al . to conclude that this level is the protein burden limit . In other experiments , harmful enzymes could only be overproduced to levels that were far less than this proposed protein burden limit . These enzymes caused problems for the yeast in several ways , including interfering with biochemical reactions and forming large aggregates in the cell . Lastly , Eguchi et al . looked at the yeast’s genetic code and saw that most of its genes seemed to have evolved to specifically limit the production of proteins to a level that would avoid the unwanted protein burden effect . Together these findings establish a framework to clearly distinguish between harmful and non-harmful proteins . This framework will be useful to understand the different reasons why the overproduction of certain proteins , which is seen in neurodegenerative diseases and cancer cells , can cause problems for cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology" ]	2018	Estimating the protein burden limit of yeast cells by measuring the expression limits of glycolytic proteins
The emergence of cooperation is a central question in evolutionary biology . Microorganisms often cooperate by producing a chemical resource ( a public good ) that benefits other cells . The sharing of public goods depends on their diffusion through space . Previous theory suggests that spatial structure can promote evolution of cooperation , but the diffusion of public goods introduces new phenomena that must be modeled explicitly . We develop an approach where colony geometry and public good diffusion are described by graphs . We find that the success of cooperation depends on a simple relation between the benefits and costs of the public good , the amount retained by a producer , and the average amount retained by each of the producer’s neighbors . These quantities are derived as analytic functions of the graph topology and diffusion rate . In general , cooperation is favored for small diffusion rates , low colony dimensionality , and small rates of decay of the public good . Public goods dilemmas are frequently observed in microbes . For example , the budding yeast Saccharomyces cerevisiae cooperates by producing the enzyme invertase , which hydrolyzes sucrose into monosaccharides , when yeast colonies are grown in glucose-limited media ( Greig and Travisano , 2004; Gore et al . , 2009 ) . Other examples include the production of chemical agents that scavenge iron ( Griffin et al . , 2004; Buckling et al . , 2007; Cordero et al . , 2012; Julou et al . , 2013 ) , enable biofilm formation ( Rainey and Rainey , 2003 ) , eliminate competition ( Le Gac and Doebeli , 2010 ) , induce antibiotic resistance ( Chuang et al . , 2009; Lee et al . , 2010 ) , or facilitate infection of a host ( Raymond et al . , 2012 ) . In many cases , the benefits of public goods go primarily to cells other than the producer . For example , in a S . cerevisiae population subject to continuous mixing , only ∼1% of monosaccharides are imported into the cell that hydrolyzes them , with the remainder diffusing away ( Gore et al . , 2009 ) . Furthermore , production of public goods typically involves a metabolic cost , which may exceed the direct benefit to the producer . In this case , absent some mechanism to support cooperation ( Nowak , 2006 ) , public goods production is expected to disappear under competition from cheaters , resulting in the tragedy of the commons ( Hardin , 1968 ) . There is growing evidence from experiments ( Griffin et al . , 2004; Kümmerli et al . , 2009; Julou et al . , 2013; Momeni et al . , 2013 ) and simulations ( Allison , 2005; Misevic et al . , 2012 ) that spatial or group clustering can support cooperation in microbial public goods dilemmas , although this effect depends on the nature of competition for space and resources ( Griffin et al . , 2004; Buckling et al . , 2007 ) . These findings agree with insights from mathematical models ( Nowak and May , 1992; Durrett and Levin , 1994; Santos and Pacheco , 2005; Ohtsuki et al . , 2006; Szabó and Fáth , 2007; Taylor et al . , 2007; Perc and Szolnoki , 2008; Fletcher and Doebeli , 2009; Korolev and Nelson , 2011 ) suggesting that spatial structure can promote cooperation by facilitating clustering and benefit-sharing among cooperators . However , these mathematical results focus largely on pairwise interactions rather than diffusible public goods . On the other hand , previous theoretical works that specifically explore microbial cooperation ( West and Buckling , 2003; Ross-Gillespie et al . , 2007; Driscoll and Pepper , 2010 ) use a relatedness parameter in place of an explicit spatial model , obscuring the important roles of colony geometry and spatial diffusion in determining the success of cooperation . Here we present a simple spatial model of a diffusible public goods dilemma . Our model is inspired by the quasi-regular arrangements of cells in many microbial colonies ( Figure 1A , B ) . The geometry of these arrangements depends on the shapes of cells and the dimensionality of the environment . For example , approximately spherical organisms such as S . cerevisiae arrange themselves in a hexagonal lattice-like structure when the colony is constrained to a two-dimensional plane ( Figure 1A ) . This differs from the arrangements of rod-shaped organisms such as the bacterium Escherichia coli ( Figure 1B ) . 10 . 7554/eLife . 01169 . 003Figure 1 . Colony geometry and public goods sharing in microbes of different shapes . ( A ) A two-dimensional colony of S . cerevisiae self-organizes into approximate hexagonal geometry due to the spherical shape of yeast cells . ( B ) A two-dimensional colony of E . coli , expressing green fluorescent protein , exhibits transient regular-graph-like structure . Panels C and D show idealized graph representations of colony spatial structure , and the consequent sharing of public goods , for sphere-shaped and rod-shaped organisms , respectively . Background colors show the stationary distributions ψi of public goods resulting from a single cooperator ( center ) . In each case , the diffusion parameter is set as λ = 3 . ( C ) Two-dimensional colonies of spherical organisms can be represented by triangular lattices with uniform edge weights . ( D ) Two-dimensional colonies of rod-shaped organisms can be represented using a triangular lattice with unequal weights . In this case , the weights are chosen as 0 . 1 , 0 . 15 and 0 . 25 , roughly proportional to the shared surface area between E . coli cells when arranged as shown . DOI: http://dx . doi . org/10 . 7554/eLife . 01169 . 003 To allow for a maximum variety of possible arrangements , we represent space as a weighted graph G ( Figure 1C , D; Lieberman et al . , 2005 ) . Edges join cells to their neighbors , with edge weights eij proportional to the frequency of diffusion between neighboring cells . The graph structure thereby captures all features of cell arrangement that are relevant to the diffusion of public goods . The edge weights are normalized to satisfy Σj eij = 1 , so that they represent relative frequencies of diffusion to each neighbor . Since we are modeling intercellular diffusion , we set eii = 0 for each i . We also suppose that G has bi-transitive symmetry ( Taylor et al . , 2007 ) , which implies that space is homogeneous ( i . e . , that the colony looks the same from each cell ) . Our model therefore applies primarily to the interiors of colonies rather than their boundaries . Bi-transitive symmetry also requires that pairwise relationships are symmetric—in particular eij = eji for every pair i and j . This captures the reasonable assumption that public goods diffuse as frequently from cell i to cell j as they do from j to i . To characterize local structure , we introduce the Simpson degree κ= ( ∑j∈Geij2 ) −1 . This quantity can be understood as the Simpson diversity ( Simpson , 1949 ) of neighbors per cell , and coincides with the usual notion of degree on regular unweighted graphs . By symmetry , κ does not depend on which vertex i is used in the above sum . We consider two cells types: cooperators , C , that produce the public good , and defectors , D , that do not . These traits are passed to offspring upon reproduction . Production of the public good inflicts a cost c on its producer , and generates a total benefit b that is distributed among cells according to a diffusion process described below . Because our model is inspired by public goods that directly increase cell growth rate ( such as hydrolyzed monosaccharides ) it is less applicable to public goods with indirect benefits , such as quorum-sensing molecules ( Waters and Bassler , 2005 ) . Cooperators produce one unit of public good per unit time . The public goods in the vicinity of a given cell either are utilized for the benefit of this cell or diffuse toward neighboring cells in proportion to edge weight . ( The possibility of public goods decay is discussed below . ) We quantify diffusion by the ratio λ of the diffusion rate to the utilization rate . The dynamics of the local public goods concentration ψi at each node i ∈ G are given by ( 1 ) ψ˙i=si−ψi−λψi+λ∑j∈Gejiψj . Above , si = 0 , 1 indicates the current type , D or C respectively , of cell i . The term si in Equation 1 represents public goods production , −ψi represents utilization , −λψi represents diffusion outward , and the remaining term represents diffusion inward . Equation 1 is equivalent to supposing that each particle of public good performs a random walk among cells ( with step probabilities equal to edge weights ) , and has probability 1/ ( 1+λ ) of being utilized at each cell it encounters , including its producer . In this interpretation , λ equals the expected number of steps a particle travels before being utilized . For most empirical systems , diffusion and utilization occur much faster than cell division . We therefore suppose that the local public goods concentrations ψi reach stationary equilibrium levels between reproductive events ( ‘Materials and methods’ ) . Two key quantities in our analysis are the fractions , ϕ0 and ϕ1 , of public goods that are retained by its producer and the producer’s immediate neighbors , respectively ( Figure 2 ) . For a state in which only a single cell , i , is a cooperator , we have ϕ0 = ψi and ϕ1 = Σj∈G eij ψj . 10 . 7554/eLife . 01169 . 004Figure 2 . The success of cooperation depends on the amounts of public good retained by a cooperator and its neighbors . Of the public good that a cooperator produces , a fraction ϕ0 is retained by the producer , a fraction ϕ1 is absorbed by each of the cooperator’s nearest neighbors , and the remainder diffuses to cells further away . ( For graphs with unequal edge weights , ϕ1 is the edge-weighted average fraction received by each neighbor . ) Cooperation is favored if b/c > 1/ ( ϕ0 + ϕ1 ) , that is , if the benefit bϕ0 received by producer , plus the average benefit bϕ1 received by each neighbor , exceeds the cost c of production . This figure shows a triangular lattice with equal edge weights and diffusion parameter λ = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 01169 . 004 Turning now to the dynamics of evolution , we suppose that the fecundity ( reproductive rate ) of cell i is given by Fi = 1 + bψi − csi . In words , each individual has baseline fitness 1 , plus the benefit , bψi , of public goods utilization , minus the cost , csi of public goods production . We suppose b > 0 and 0 < c < 1 , so that benefits , costs , and overall fecundity are always positive . Some of our results apply to all such b and c values , while others apply only in the weak selection regime , b , c ≪ 1/κ . Reproductions and deaths follow the Death–Birth update rule ( Ohtsuki et al . , 2006 ) . At each time step , a cell is selected randomly to die , with uniform probability . A neighbor of the now-vacant position is randomly selected to reproduce , with probability proportional to fecundity times edge weight . The new offspring fills the vacancy . For the moment , we suppose that reproduction follows the same edge weights as diffusion ( we will relax this assumption later ) . We also consider other update rules in Supplementary file 1 . We quantify the evolutionary success of cooperation in terms of the fixation probabilities ρC and ρD , defined as the probability that the cooperator or defector type , respectively , will fix , upon starting from a single mutant in a population initially of the opposite type . Cooperation is favored if ρC > ρD . This is equivalent to the condition that , for small mutation rates , cooperators have greater time-averaged frequency than would be expected from mutational equilibrium alone ( Allen and Tarnita , 2012 ) . The assortment of cell types due to local reproduction can be studied using coalescing random walks ( Wakeley , 2009; Allen et al . , 2012 ) , which represent the ancestral lineages of chosen individuals as the coalesce into the most recent common ancestor . By applying random walk theory to both diffusion and assortment , we are able to obtain exact conditions for the success of cooperation ( ‘Materials and methods’; Supplementary file 1 ) . We find that public goods cooperation is favored , for any graph and diffusion rate , if and only if ( 2 ) bc>1ϕ0+ϕ1 . In words , cooperation is favored if , of the public goods a cooperator produces , the benefits received by the producer , bϕ0 , plus the ( edge-weighted ) average benefits received by each neighbor , bϕ1 , outweigh the cost c of production ( Figure 2 ) . This result is strikingly simple , given the complex patterns of public goods sharing that result from diffusion ( Figure 1 ) . Condition ( 2 ) holds for arbitrary selection strength on complete graphs and one-dimensional lattices , and for weak selection on other graphs . This condition also holds for a variety of other diffusion processes ( Supplementary file 1 ) —including diffusion that follows a different graph structure from reproduction . ( In this case , the neighbor average ϕ1 is computed using the weights for the reproduction graph . ) Condition ( 2 ) can alternatively be expressed as b/c > λ/[ϕ0 ( 1 + 2λ ) − 1] ( ‘Materials and methods’ ) , showing how the success of cooperation depends on the relationship between the retention fraction ϕ0 and the diffusion parameter λ . We have derived this relationship exactly for simple graph structures ( Table 1 ) , and present a general method for obtaining this relationship in the ‘Materials and methods’ . Figure 3A , B illustrates how the critical b/c ratios vary with the diffusion parameter λ and the graph topology . 10 . 7554/eLife . 01169 . 005Table 1 . Fraction of public goods retained by producer for different graph structures and diffusion ratesDOI: http://dx . doi . org/10 . 7554/eLife . 01169 . 005Graph structureFraction ϕ0 of public goods retainedComplete ( well-mixed ) 11+λ1D lattice11+2λ2D square lattice†1agm ( 1 , 1+2λ ) n-dimensional lattice ( general ) ‡1 ( 2π ) n∫−ππ⋯∫−ππdny1+λ−λ χ ( y ) k-Bethe lattice§ ( k−2 ) 2 ( 1+λ ) 2+4 ( k−1 ) ( 1+2λ ) − ( k−2 ) ( 1+λ ) 2 ( 1+2λ ) These results are for large populations . Corrections for finite population size are given in Supplementary file 1 . †agm denotes the arithmetic-geometric mean . ‡This result applies to any mathematical lattice , including triangular and von Neumann lattices . χ ( y ) denotes the structure function of the lattice in question , defined in the ‘Materials and methods’ . §A Bethe lattice ( a . k . a . infinite Cayley tree ) , is an infinite regular graph with no cycles . In the formula , k denotes the graph degree . 10 . 7554/eLife . 01169 . 006Figure 3 . Cooperation becomes harder to achieve with increasing λ , graph degree and dimensionality , and public goods decay rate . ( A ) The critical b/c ratio for public goods production to be favored for various graph structures , plotted against the diffusion rate λ . These results are derived from Equation 2 and the expressions for ϕ0 in Table 1 . For a well-mixed population ( complete graph ) , cooperation is favored if and only if b/c > 1 + λ; for other graph structures , the critical b/c ratio is a increasing , convex function of λ . In general , the conditions for cooperation become increasingly stringent with both the degree and the dimensionality of the graph . ( B ) Our results are confirmed by simulations on a 15 × 15 periodic triangular lattice with uniform edge weights and cost c = 5% . The critical b/c threshold from Equation 2 is plotted in black . A plus ( + ) indicates that the frequency of cooperator fixation exceeded the frequency of defector fixation ( ρC > ρD ) , while a minus ( − ) indicates the opposite . In all cases the results were statistically significant ( two-proportion pooled z-test , p<0 . 05 ) . ( C ) Adding decay of rate d effectively reduces both λ and b by the factor 1/ ( 1 + d ) , reflecting greater locality in sharing but reduced overall concentration of public good . On a graph of b/c versus λ , this moves each point ( b/c , λ ) along a straight line toward the origin . Since the increase in the critical b/c ratio with λ is in all cases sublinear , this change always hinders cooperation . The critical b/c ratio for a planar triangular lattice is plotted in black . Adding a decay rate equal to the utilization rate ( d = 1 ) changes favorable ( b/c , λ ) combinations ( marked by circles ) to unfavorable ones ( arrowheads ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01169 . 006 Above , we have assumed that diffusion and replacement are both described by the same graph structure . However , this may not be the case for all microbes . In E . coli colonies , for example , it is reasonable to conjecture that diffusion occurs more frequently among cells that have a long side in common , whereas replacement may occur more frequently among end-to-end neighbors ( Figure 1A , C ) . Additionally , some systems may follow a public goods diffusion process other than that modeled by Equation 1 . To account for these variations , we consider a more general model in which diffusion is described by the fractions ϕij of public goods which , if produced by cell i , would be utilized by cell j . Probabilities of replacement are described by a graph with edge weights eij as before . The diffusion fractions ϕij are normalized so that ∑j ϕij = 1 for each i , and they have the same symmetries as the replacement graph; within these restrictions , they may be specified arbitrarily . Remarkably , our main result , Equation 1 , remains valid in this generalized setting , with the neighbor average ϕ1 defined as ϕ1 = ∑j eij ϕij . Our results suggest three qualitative regimes for diffusible public goods scenarios . For λ ≪ 1 , the benefits are almost all retained by producer , and production is favored whenever b/c > 1 . Conversely , for λ ≪ 1 , public goods are shared indiscriminately , and production is favored only if public goods are essential for survival , in which case b is effectively infinite . Between these extremes , public goods are shared locally , and the spatial arrangement of cells plays a critical role in the success of cooperation ( Figure 3A ) . At the smaller end of this critical regime , the expansion b/c>1+λ ( κ−1 ) /κ+O ( λ2 ) of condition ( 2 ) , derived in Supplementary file 1 , shows how the difficulty of cooperation increases with the diffusion parameter λ and the Simpson degree κ . For the hydrolysis of monosaccharides in S . cerevisiae , we estimate λ ∼ 3 ( ‘Materials and methods’ ) ; thus we expect the success of invertase production to be strongly affected by colony geometry . Interestingly , this diffusion length is of the same order of magnitude as those reported in other recent experiments with diffusible public goods ( Julou et al . , 2013; Momeni et al . , 2013 ) . Our model predicts that the advantage of cooperation decreases with colony dimensionality; for example , less cooperation would be expected in three-dimensional structures than in flat ( 2D ) colonies ( Figure 3A ) . It also predicts that cooperation becomes more successful with increased viscosity of the environment and/or rate of public goods utilization , both of which would decrease λ . A more subtle question is how cooperation is affected if the public good may decay ( or equivalently , escape the colony ) instead of being utilized . Decay reduces the absolute amount of public goods to be shared , but also restricts this sharing to a smaller circle of neighbors; thus the net effect on cooperation is at first glance ambiguous . We show in the ‘Materials and methods’ that incorporating decay effectively decreases λ by a factor 1/ ( 1 + d ) , reflecting the smaller neighborhood of sharing , and also effectively decreases b by the same factor , reflecting the diminished absolute amount of public goods . Here d represents the ratio of the decay rate to the utilization rate . Since the critical benefit-to-cost ratio always increases sublinearly with λ , the net effect is to make cooperation more difficult ( see Figure 3C ) . Thus decay of the public good has a purely negative effect on cooperation . Our results help elucidate recent emiprical results on microbial cooperation in viscous environments . For example , Kümmerli et al . ( 2009 ) found that increased viscosity promotes the evolution of siderophore production in Pseudomonas aeruginosa , while Le Gac and Doebeli ( 2010 ) found that viscosity had no effect on the evolution of colicin production in E . coli . In both cases , increased viscosity restricted cell movement , effectively leading to fewer neighbors per cell ( lower graph degree ) . The crucial difference lies in the effect on public goods diffusion . In the study of Kümmerli et al . ( 2009 ) , the diffusion rate decreased significantly as viscosity increased , while for Le Gac and Doebeli ( 2010 ) , the diffusion rate remained large even with high viscosity . Thus the divergent outcomes can be understood as a consequence of differences in the diffusion rate , captured in our model by λ . Here we have considered homotypic cooperation—cooperation within a single population . Momeni et al . ( 2013 ) , published previously in eLife , investigate heterotypic cooperation between distinct populations of S . cerevisiae , in the form of exchange of essential metabolites . Type R produces adenine and requires lysine , type G produces lysine and requires adenine , and type C ( a cheater ) requires adenine but does not produce adenine . While such heterotypic cooperation is not incorporated in our model , the results are qualitatively similar , in that spatial structure promoted the cooperative strategies G and R over the cheater C . This similarity can be understood by noting that heterotypic cooperation also entails a form of second-order homotypic cooperation . For example , G-cells aid nearby R-cells , which in turn aid nearby G-cells , so the benefit produced by a G-cell indirectly aids other G-cells nearby . Thus the conclusion that spatial structure aids cooperative strategies can apply to heterotypic cooperation as well . Finally , our model can also represent the spread of behaviors via imitation on social networks ( Bala and Goyal , 1998; Bramoullé and Kranton , 2007; Christakis and Fowler , 2007 ) . Suppose an action generates a benefit b0 for the actor , and additionally generates further benefits that radiate outward according to some multiplier m , so that first neighbors receive a combined benefit mb0 , second neighbors receive m2b0 , and so on . Education , for example , exhibits this kind of social multiplier in its effect on wages ( Glaeser et al . , 2003 ) . This effect can be captured using the parameter change b = b0/ ( 1 − m ) , λ = m/ ( 1 − m ) . For non-well-mixed social networks , the action becomes more likely to spread as the multiplier increases , and can spread even if there is a net cost to the actor ( Figure 4 ) . 10 . 7554/eLife . 01169 . 007Figure 4 . The spread of behaviors on social networks increases with their social multipliers . In an alternate interpretation of our model , an action has benefits that radiate outward from the actor according to some multiplier m . Individual receiving a large amount of benefit are more likely to be imitated by social contacts . The likelihood of the action to spread—and the benefits to the network as a whole—both increase with the multiplier m . DOI: http://dx . doi . org/10 . 7554/eLife . 01169 . 007 We obtain a recurrence relation for the stationary public goods distribution in a given state by setting ψ˙i=0 in Equation 1 . This yields ( 3 ) ( 1+λ ) ψi=si+λ∑j∈Gejiψj . In particular , for a state in which only cell i is a cooperator , we have ( 1 + λ ) ϕ0 = 1 + λϕ1 . Combining this identity with ( 2 ) yields the equivalent condition b/c > λ/[ϕ0 ( 1 + 2λ ) − 1] . We analyze the distribution of public goods and the assortment of cell types using the generating function for random walks ( Montroll and Weiss , 1965; Lawler and Limic , 2010 ) . For a given graph G , this generating function is given by the power seriesGij ( z ) =∑n=0∞pij ( n ) zn . Above , pij ( n ) denotes the probability that a random walk of n steps starting at i will terminate at j . We prove in Supplementary file 1 that the stationary concentration of public goods in a particular state are given byψi=11+λ∑j∈Gsj Gji ( λ1+λ ) . In particular , the fraction ϕ0 that a cooperator retains of its own public good can be written ( 4 ) ϕ0=11+λGii ( λ1+λ ) . Spatial assortment of types can be quantified using identity-by-descent IBD probabilities ( Rousset and Billiard , 2000; Taylor et al . , 2007 ) . For this , we introduce a small probability u that each new offspring is a mutant . Then , two given cells are IBD if no mutation separates them from their most recent common ancestor . Based on the theory of coalescing random walks ( Allen et al . , 2012 ) , the probability that i and j are IBD can be writtenqij=Gij ( 1−u ) Gjj ( 1−u ) . Considering the dynamics of Death–Birth updating , and applying established properties of generating functions , we derive ( Supplementary file 1 ) the success condition ( 2 ) . To obtain the expressions in Table 1 , we combine ( 4 ) with previously established expressions for Gij on the graphs in question . A general expression is available for a lattice of any dimension . Such a lattice is defined by a finite collection of vectors v1 , … , vk ∈ Rn with associated weights w1 , … , wk . The nodes of the lattice are all points of the form x=m1v1+…+mkvk∈Rn , where m1 , … , mk are integers . The edges from a node x consist of the vectors v1 , … , vk , positioned to start at the point x , with weights given by w1 , … , wk , respectively . The generating function of a random walk on such a lattice , starting from the lattice origin 0 , can be expressed as ( Montroll and Weiss , 1965 ) ( 5 ) G0x ( z ) =1 ( 2π ) n∫−ππ⋯∫−ππe−i x⋅y1−z χ ( y ) dny . Above , χ ( y ) is the ‘structure function’ of the lattice , defined as ( 6 ) χ ( y ) =∑j=1kwkei vj⋅y . The argument y = ( y1 , … , yn ) of χ ( y ) is a vector in Rn . For example , for an n-dimensional square lattice , we haveχ ( y ) =1n∑i=1ncos ( yi ) . For a two-dimensional triangular lattice , χ ( y ) =13[cos ( y1 ) +cos ( y2 ) +cos ( y1+y2 ) ] . Similar expressions for other lattices , including the square lattice with von Neumann neighbors and lattices with unequal edge weights ( e . g . , Figure 1B ) , can be readily obtained from ( 6 ) . We suppose that glucose uptake follows Michaelis–Menten kinetics , so that the uptake rate is given by Vmaxψ/ ( ψ+K ) , where ψ is the concentration of glucose , Vmax is the maximal uptake rate , and K is the concentration at which the uptake rate reaches half of its maximum . We treat fructose as equivalent to glucose . Since we are interested in the case that glucose is limited , we assume ψ≪K , and the uptake rate therefore simplifies to Vmaxψ/K . Gore et al . ( 2009 ) estimated the uptake kinetics to be Vmax ∼ 2 × 107 molecules per second and K ∼ 1mM . We calculate the lifetime L of a glucose molecule prior to absorption as the reciprocal of the fraction of glucose absorbed per unit time:L=# glucose molecules per unit excluded volume ( Uptake rate per cell ) × ( # cells per unit excluded volume ) , where ‘excluded volume’ refers to the volume of water excluded by the yeast cells . Supposing that each yeast cell has volume v∼4π ( 2μm ) 3/3 , and that yeast cells in a tightly-packed colony occupy approximately half of the available volume , we obtainL=ψ ( Vmax ψ/K ) × ( 1/v ) =KvVmax∼1 sec . The diffusion length before uptake is calculated as D/L , where D is the diffusion constant , which we estimate as 100 μm2/sec in the colony environment . Combining with the above calculation of L gives a diffusion length of ∼10 μm , which is ∼3 cell lengths . We therefore estimate λ = 3 for this system . Decay or escape of the public good can be incorporated into our model by adding a decay term to the right-hand side of Equation 1 . This yieldsψ˙i=si−ψi−dψi−λψi+λ∑j∈Gejiψj . Above , d represents the ratio of the decay rate to the utilization rate . Setting ψ˙i=0 and rearranging , we obtainψi ( 1+d ) ( 1+λ1+d ) =si+λ1+d∑j∈Gejiψj ( 1+d ) . Defining the effective quantities ψ˜i=ψi ( 1+d ) and λ˜=λ/ ( 1+d ) , we recover the recurrence relation ( 3 ) . All of our results then carry forward using these effective quantities , except that b must also be reduced by the factor 1 + d to compensate for the rescaling of ψi by this same factor .	The natural world is often thought of as a cruel place , with most living things ruthlessly competing for space or resources as they struggle to survive . However , from two chimps picking the fleas off each other to thousands of worker ants toiling for the good of the colony , cooperation is fairly widespread in nature . Surprisingly , even single-celled microbes cooperate . Individual bacterial and yeast cells often produce molecules that are used by others . Whilst many cells share the benefits of these ‘public goods’ , at least some cells have to endure the costs involved in producing them . As such , selfish individuals can benefit from molecules made by others , without making their own . However , if everyone cheated in this way , the public good would be lost completely: this is called the ‘public goods dilemma’ . Allen et al . have developed a mathematical model of a public goods dilemma within a microbial colony , in which the public good travels from its producers to other cells by diffusion . The fate of cooperation in this ‘diffusible public goods dilemma’ depends on the spatial arrangement of cells , which in turn depends on their shape and the spacing between them . Other important factors include rates of diffusion and decay of the public good—both of which affect how widely the public good is shared . The model predicts that cooperation is favored when the diffusion rate is small , when the colonies are flatter , and when the public goods decay slowly . These conditions maximize the benefit of the public goods enjoyed by the cell producing them and its close neighbors , which are also likely to be producers . Public goods dilemmas are common in nature and society , so there is much interest in identifying general principles that promote cooperation .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology", "physics", "of", "living", "systems" ]	2013	Spatial dilemmas of diffusible public goods
Presenilin 1 ( PS1 ) is an essential γ-secretase component , the enzyme responsible for amyloid precursor protein ( APP ) intramembraneous cleavage . Mutations in PS1 lead to dominant-inheritance of early-onset familial Alzheimer’s disease ( FAD ) . Although expression of FAD-linked PS1 mutations enhances toxic Aβ production , the importance of other APP metabolites and γ-secretase substrates in the etiology of the disease has not been confirmed . We report that neurons expressing FAD-linked PS1 variants or functionally deficient PS1 exhibit enhanced axodendritic outgrowth due to increased levels of APP intracellular C-terminal fragment ( APP-CTF ) . APP expression is required for exuberant neurite outgrowth and hippocampal axonal sprouting observed in knock-in mice expressing FAD-linked PS1 mutation . APP-CTF accumulation initiates CREB signaling cascade through an association of APP-CTF with Gαs protein . We demonstrate that pathological PS1 loss-of-function impinges on neurite formation through a selective APP gain-of-function that could impact on axodendritic connectivity and contribute to aberrant axonal sprouting observed in AD patients . Alzheimer’s disease ( AD ) is a progressive neurodegenerative disease pathologically characterized by a cerebral deposition of β-amyloid peptide ( Aβ ) in senile plaques and neuronal loss . Inheritance of dominant forms of mutations in genes encoding amyloid precursor protein ( APP ) and presenilins ( PSEN1 and PSEN2 genes ) cause aggressive forms of early onset familial AD ( FAD ) . A mutation in presenilin genes causes autosomal dominant early-onset familial Alzheimer’s disease ( FAD ) ( Tanzi and Bertram , 2001 ) . Gene knockout studies in mice reveal that PS1 acts as the catalytic core of the multisubunit γ-secretase complex that is responsible for regulated intramembranous proteolysis of several type-I transmembrane protein substrates . Over 90 substrates have been identified so far including amyloid precursor protein ( APP ) , Notch and Eph receptors and ligands , cadherins , and deleted in colorectal cancer ( DCC ) ( Haapasalo and Kovacs , 2011; Kopan and Ilagan , 2004; McCarthy et al . , 2009; Parks and Curtis , 2007 ) . Several of these substrates are known for their diverse functions during neuronal development including axon guidance , neurite outgrowth , and synaptogenesis . In the case of APP , the most studied γ-secretase substrate , sequential ectodomain shedding by α-secretase , which occurs mainly at the cell surface , or β-secretase is required before subsequent cleavage by γ-secretase ( Deyts et al . , 2016; Haass et al . , 2012; Thinakaran and Koo , 2008 ) . Therefore , cleavage of full-length APP ( APP-FL ) by α- or β-secretases releases the entire ectodomain ( soluble APPα or soluble APPβ , respectively ) , leaving behind membrane-bound C-terminal fragments ( CTF ) , comprising the transmembrane and cytoplasmic domain ( APP-C83 and APP-C99 , respectively ) . Subsequent cleavage of APP-CTF by γ-secretase releases the cytosolic domain from the membrane ( APP intracellular domain , AICD ) and either p3 fragment or the neurotoxic Aβ peptide . Consequently , inhibiting γ-secretase activity would prevent accumulation of Aβ and AICD in favor of accumulation of APP-C83 or APP-C99 depending on their prior non-amylogenic α-CTF or amylogenic β-CTF cleavage products , respectively . It has been proposed that γ-secretase complex activity may serve as the membrane proteasome that removes C-terminal stubs generated after ectodomain shedding and prevents further cell-surface signaling by a variety of substrates ( Kopan and Ilagan , 2004 ) . No evidence has been described that such a role may be an important contributor of disease states associated with γ-secretase components . The contribution of APP holoprotein and other PS1-dependent substrates in the AD etiology has not been fully examined . The present work focuses on determining the importance of PS1-dependent modulation of APP-CTF accumulation and the subsequent effects on associated signaling that promote neurite outgrowth . We report here that alteration of γ-secretase activity , through pharmacologic PS1 inhibition , genetic Psen1 ablation , or expression of FAD-linked PS1 variants , enhances neurite outgrowth . Our findings indicate that APP is required for PS1-dependent changes in neurite outgrowth . Ablation of APP expression prevented aberrant axonal sprouting observed in the hippocampal dentate gyrus of PSEN1 knock-in mouse model harboring FAD-linked PS1 variant . APP-CTF accumulation contributed to γ-secretase-dependent increases of CREB signaling cascade seen in neurons that exhibit PS1 loss-of-function , an effect that was prohibited through adenylate cyclase inhibition . Our results provide the first demonstration that a pathological loss of PS1 function leads to a selective gain of APP function that may impact axodendritic connectivity . We previously demonstrated that accumulation of APP-CTF in the raft produced a marked increase of neurite extension in a variety of neuronal cells including cortical neurons ( Deyts et al . , 2012 ) . We also observed that overexpression of APP and concurrent γ-secretase inhibition , that produce an accumulation of APP-CTF and other γ-secretase-dependent CTF substrates , leads to an increase in neurite outgrowth ( Deyts et al . , 2012 ) . Given the fact that APP plays important roles in neurite outgrowth through the function of its extra- and intracellular fragments ( review by ( Chasseigneaux and Allinquant , 2011; Muller and Zheng , 2012; Nicolas and Hassan , 2014 ) ) , we wanted to establish the physiological significance of endogenous APP and its metabolites that are generated through modulation of γ-secretase activity . First , we have examined neurite outgrowth in cortical neurons lacking APP using APP knockout ( APP KO ) mice ( Figure 1 ) . Lack of APP expression modestly ( but significantly ) and selectively altered axonal outgrowth ( Figure 1b1; WT , 1 . 00 ± 0 . 03; APP KO , 0 . 77 ± 0 . 05 , P<0 . 001 ) , whereas the dendritic outgrowth was not affected ( Figure 1b2 , see also Figure 1—figure supplement 1 ) . Second , we confirmed the physiological consequence of γ-secretase inhibition on neurite outgrowth , using γ-secretase inhibitor Compound E ( Seiffert et al . , 2000 ) . We observed that inhibition of γ-secretase in WT neurons enhances axonal and dendritic outgrowth ( Figure 1b1 and b2 ) , with a more predominant effect on the axon as reflected by an increase of the axon/dendrite area ratio ( see Figure 1b3; WT , 1 . 78 ± 0 . 07; WT+CompE , 2 . 52 ± 0 . 18 , P< 0 . 05 ) . Next , we examined whether γ-secretase activity influenced neurite outgrowth in neurons lacking APP . Interestingly , we observed that neurons treated with γ-secretase inhibitor did not exhibit axonal or dendritic outgrowth in the absence of APP expression ( Figure 1b1 and b2 ) , suggesting that APP is the main contributing factor in γ-secretase-mediated neurite outgrowth . 10 . 7554/eLife . 15645 . 003Figure 1 . Exuberant axodendritic outgrowth associated with γ-secretase inhibition requires APP expression . ( a ) WT or APP KO primary cortical neurons ( 8 DIV ) coexpressing YFP and EV or APP-FL were treated with Compound E ( CompE 10 nM , 24 hr ) and immunostained with MAP2 antibody . ( a ) Representative overlay images of YFP and MAP2 staining reveal axon ( green ) and dendrites ( yellow ) . ( b ) Quantitative analysis of neurite outgrowth is represented as relative changes in total axonal area ( b1 ) , dendritic area ( b2 ) , and axon/dendrite ratio ( b3 ) in treated or untreated neurons with Compound E , expressing either EV or APP-FL as compared to WT-EV control group . ( c1 ) Endogenous APP full-length ( APP-FL ) and accumulation of APP-CTF were detected by immunoblotting brain lysates from WT and APP KO mice with CTM1 antibody , which recognizes the C-terminus of APP ( von Koch et al . , 1997 ) . ( c2 ) Accumulation of APP-CTF in WT neurons overexpressing empty vector ( EV ) or APP-FL is detected by Western blot using a Tris-Tricine gel , before or after treatment with Compound E ( CompE: 10 nM , 24 hr ) . GAPDH was used as loading control . Statistical analysis was performed using ANOVA Kruskal-Wallis test followed by Dunn’s post hoc multiple comparison analysis . p<0 . 05 , p<0 . 001 compared to WT-EV; yellow p<0 . 05 , p<0 . 001 compared to APP KO , and ##p<0 . 001 compared to WT treated with CompE . The total number of quantified cells is shown in parentheses ( WT , n = 13 embryos; APP KO , n = 6 embryos; Compound E , n = 5 embryos ) . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 15645 . 00310 . 7554/eLife . 15645 . 004Figure 1—figure supplement 1 . Changes in axodendritic outgrowth associated with γ-secretase inhibition requires APP expression . Quantitative analysis of neurite outgrowth is represented as raw data ( pixel number ) in total axonal area ( a ) , and dendritic area ( b ) , in treated or untreated neurons with Compound E ( CompE: 10 nM , 24 hr ) , expressing either EV or APP-FL as compared to WT-EV control group . Statistical analysis was performed using ANOVA Kruskal-Wallis test followed by Dunn’s post hoc multiple comparison analysis . p<0 . 001 compared to WT-EV; yellow *p<0 . 001 compared to APP KO , and ##p<0 . 001 compared to WT treated with CompE . The total number of quantified cells is shown in parentheses ( WT , n = 13 embryos; APP KO , n = 6 embryos; Compound E , n = 5 embryos ) . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 15645 . 00410 . 7554/eLife . 15645 . 005Figure 1—figure supplement 2 . Identification of APP-CTF through differential processing of APP by secretases . ( a ) HEK293 cells and ( b ) COS-7 cells , coexpressing empty vector ( EV ) , APP-FL , APP-M596V ( APP β-site cleavage mutant ) , APP-F615P ( APP α-site cleavage mutant ) , or APPswe , were treated with or without γ-secretase inhibitor Compound E ( CompE , 24 hr , 10 nM ) . Lysates from primary cortical neurons ( 14 DIV ) generated from WT and PS1 KO mouse embryos were loaded adjacent to COS-7 transfected cells . Western blotting analysis from Tris-Tricine separation gel was performed to identify APP-CTF using rabbit polyclonal CTM1 antibody . Detection of APP-CTF α , +11 and β species is notified , as previously described ( Vetrivel et al . , 2011 ) . GAPDH was used as loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 15645 . 005 We subsequently decided to overexpress APP in an attempt to reinstate the effect seen in APP KO neurons to the WT level . We observed that concomitant expression of APP-FL and inhibition of γ-secretase are sufficient to reestablish the enhancement of axodendritic outgrowth at the level seen in WT neurons treated with a γ-secretase inhibitor ( Figure 1a and b ) . In our neuronal cultures , fractionation of lysates on Tris-Tricine gels showed that the inhibition of γ-secretase activity produced an overall increase of APP-CTF whereas APP-FL was not changed ( Figure 1c2 and Figure 1—figure supplement 2 ) . As expected , we confirmed that γ-secretase dependent processing of APP affects accumulation of both APP-CTFα and APP-CTFβ , at least in HEK293 and COS-7 cells ( Figure 1—figure supplement 2 ) . Overexpression of APP-FL was not sufficient to induce axodendritic outgrowth , although accumulation of APP-CTF through γ-secretase inhibition did promote axodendritic outgrowth . This set of data confirmed that inhibition of γ-secretase activity favors a preferential augmentation of axonal outgrowth in comparison to the dendritic compartment , as shown by the increase of the axon/dendrite ratio ( Figure 1b3 ) . This effect was abolished in the absence of endogenous APP expression and restored by APP overexpression . Our findings emphasize that both γ-secretase inhibition and APP expression are required to induce axodendritic outgrowth . It is well established that inhibition of γ-secretase activity , or lack of presenilin 1 ( PS1 ) expression , produces an endogenous accumulation of APP-CTF in the brain , neuronal cultures , or cell lines ( Bentahir et al . , 2006; De Strooper et al . , 1998; Parent et al . , 2005; Walker et al . , 2005; Woodruff et al . , 2013 ) . In earlier studies , it has been proposed that expression of FAD-linked PS1 mutant proteins may attenuate γ-secretase activity , consequently limiting proteolysis of full-length protein substrates and , therefore , accumulating their CTF ( Bentahir et al . , 2006; De Strooper , 2007; Shen and Kelleher , 2007; Walker et al . , 2005; Wolfe , 2007; Woodruff et al . , 2013 ) . Based on our results reported in Figure 1 , we wondered whether pathogenic PS1 mutations , that have been described to affect γ-secretase activity , might influence neurite outgrowth . We focused on three PS1 variants that carry a PS1-M146L substitution and PS1-ΔE9 deletion of exon 9 , which is known to cause earlier age-of-onset FAD , and a known dominant negative PS1-D385A substitution . We stably transfected HEK293 cells with wild-type human PS1 ( PS1-WT ) or with PS1 mutations . As shown in Figure 2a1 , pools of established stably transfected HEK293 cells expressed comparable levels of PS1 protein , as shown by cell lysates immunoblotted with a PS1 polyclonal antibody that detects PS1-FL and PS1-NTF . Next , we analyzed the ability of PS1 mutation to modulate the accumulation of APP-CTF in PS1 stable HEK293 cells that overexpressed APP-FL ( Figure 2a2 ) . We observed that APP-CTF accumulates in PS1 mutants , as compared to PS1-WT ( Figure 2a2 ) , confirming previous findings that pathogenic PS1 mutations might reduce γ-secretase activity ( Bentahir et al . , 2006; De Strooper , 2007; Heilig et al . , 2013; Shen and Kelleher , 2007; Walker et al . , 2005; Woodruff et al . , 2013; Xia et al . , 2015 ) . Interestingly , we observed that neurons expressing PS1-ΔE9 mutation exhibit a significant increase in axodendritic outgrowth ( Figure 2b and c ) , an effect that is exacerbated by the presence of APP-FL in neurons overexpressing PS1-M146L and PS1-D385A mutations . Taken together , our results demonstrate that pathogenic PS1 mutations favor axodendritic outgrowth as seen under a condition that reduces γ-secretase activity . 10 . 7554/eLife . 15645 . 006Figure 2 . Expression of FAD-linked PS1 variants promotes neurite outgrowth . FAD-linked PS1 mutations enhance axonal and dendritic arborization that correlates with APP C-terminal fragment ( APP-CTF ) accumulation . FAD-linked PS1 mutations affect APP processing in cells expressing APP full-length ( APP-FL ) . ( a ) Stable HEK293 cells overexpressing FAD-linked PS1 variants ( a1 ) were transiently transfected with APP-FL ( a2 ) . ( a1 ) The PS1NT polyclonal antibody was used to detect PS1 full-length ( PS1-FL ) and PS1 N-terminal fragment ( PS1-NTF ) . ( a2 ) The CTM1 polyclonal antibody was used to detect APP-FL and APP-CTF accumulation . ( b ) GAPDH was used as loading control . ( b ) WT primary cortical neurons ( 8 DIV ) coexpressing YFP , APP-FL and PS1 mutants were immunostained with MAP2 antibody . Representative overlay images of YFP and MAP2 staining reveal axonal ( green ) and dendritic ( yellow ) arbors . ( c ) Quantification of the total axonal ( b1 ) and dendritic ( b2 ) areas is shown in cortical neurons ( 7–8 DIV ) 24 hr following transfection of PS1 variants ( in gray ) with APP-FL ( in light green ) . The total number of quantified cells is shown in parentheses ( n = 5 embryos for each transfected condition ) . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 15645 . 006 To substantiate the impact of PS1 function on γ-secretase activity , we took advantage of knockout and knock-in mouse models with targeted deletion of Psen1 ( PS1 KO ) and replaced Psen1 gene with human FAD-linked PSEN1-M146V variant ( PS1 KI ) , respectively . In this later strain , PS1-M146V protein is expressed at a level equivalent to WT protein level , therefore mimicking human FAD patients carrying this mutation . These PS1 KI mice do not exhibit overt Aβ accumulation or deposition; thus , the main interest of using this mouse model is to understand the contribution of early development of AD process . As shown in Figure 3a , Western blot analysis of steady-state APP expression levels in brains of E15 mouse embryos reveals a significant increase of APP-CTFα in PS1 KI , and larger increase of both APP-CTFα and APP-CTFβ in PS1 KO as compared to WT . Analysis of brain lysates revealed similar increases of APP-CTF in total membrane fractions of heterozygous ( PS1 KI/+ ) and homozygous ( PS1 KI/KI ) brains harvested during synaptogenesis ( Figure 3b and Figure 3—figure supplement 1a1 , postnatal day 10 – P10 ) or in adulthood ( Figure 3—figure supplement 1; 6 months-old – 6M ) , as compared to their WT littermates . Levels of APP-FL were not significantly affected among these groups ( Figure 3b and Figure 3—figure supplement 1a ) . Consequently , ratios of APP-CTF/APP-FL were increased in PS1 KI brains ( Figure 3b2 and Figure 3—figure supplement 1b2 ) , suggesting that the FAD-linked PS1 mutation may result in loss of activity toward APP proteolysis , leading to accumulation of membrane-tethered APP-CTF . However , APP-CTF accumulated to a lesser extent in PS1 KI as compared to PS1 KO ( Figure 3a ) , therefore supporting the notion of partial loss-of-function mutant . 10 . 7554/eLife . 15645 . 007Figure 3 . Neurons expressing FAD-linked PS1 mutations exhibit partial loss of γ-secretase activity associated with APP-dependent increases of neurite outgrowth . ( a ) Levels of APP full-length ( APP-FL ) and proteolytic C-terminal fragments APP-CTF ( APP-CTFα and APP-CTFβ ) accumulation were detected by high resolution Tris-Tricine Western blot analysis of lysates generated from WT , PS1 KO , APP KO , PS1 KI/+ , and PS1 KI/KI mouse brains using CTM1 antibody . Lysate from HEK cells overexpressing APPswe and GAPDH antibody detection were used as control . Half amount of proteins was loaded for PS1 KO brain lysate . NS indicates a non-specific cross-reacting band . ( b1 ) Endogenous APP-FL and accumulation of APP-CTF were detected with CTM1 antibody by immunoblotting brain lysates from postnatal day 10 ( P10 ) transgenic PSEN1-M146V knock-in mice ( PS1 KI ) . ( b2 ) Quantitative analysis of APP-FL , APP-CTF accumulation and the ratio APP-CTF/APP-FL in PS1 KI is shown . Values are reported as a relative change in the intensity of the protein as compared with the WT littermates . ( c ) Primary cortical neurons ( 7–8 DIV ) from WT , PS1 KO , PS1 KI ( heterozygous PS1 KI/+ and homozygous PS1 KI/KI ) , APP KO , APP KO x PS1 KI , DCC KO and DCC KO x PS1 KI mice were transfected with YFP and immunostained with MAP2 antibody . Representative resulting overlay images reveal differences in axons ( green ) and dendrites ( yellow ) . ( d ) Quantitative analysis of total axonal area ( d1 ) and dendritic area ( d2 ) are shown . Results are reported as relative values as compared to WT . ( e ) Western blot analysis of P10 brain lysates was performed to detect endogenous DCC-CTF fragments from WT and PS1 KI . ( e1 ) Representative immunoblot lysate samples are shown . ( e2 ) Quantitative analysis is shown as a relative change in the intensity of DCC-CTF expression as compared to WT littermates using Flotillin-2 as a loading control . Statistical analysis was performed using ANOVA Kruskal-Wallis test followed by Dunn’s post hoc multiple comparison analysis . p<0 . 05 , p<0 . 001 compared to WT , blue p<0 . 001 compared to DCC KO , and ##p<0 . 001 compared to PS1 KI . The total number of neurons ( from at least 3 independent sets of cultures ) used for quantification is shown in parentheses ( WT , n = 6 embryos; PS1 KO , n = 5 embryos; PS1 KI/+ , n = 6 embryos; PS1 KI/KI , n = 6 embryos; APP KO , n = 6 embryos; APP KO x PS1 KI , n = 6 embryos; DCC KO , n = 7 embryos; DCC KO x PS1 KI , n = 5 embryos ) . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 15645 . 00710 . 7554/eLife . 15645 . 008Figure 3—figure supplement 1 . Intramembraneous proteolysis of APP in knock-in mice expressing FAD-linked PS1 variant . FAD-linked PS1 mutations correlate with APP-CTF accumulation . ( a ) Levels of endogenous APP-full length ( APP-FL ) and accumulation of APP-CTF were evaluated by Western blot in brain lysates from WT , hetero- and homozygote PSEN1-M146V knock-in mice ( PS1 KI/+ and PS1 KI/KI , respectively ) using CTM1 antibody . APP-FL and APP-CTF band intensities were quantified using Odyssey Infrared Imaging software ( Li-Cor Biosciences ) as a measure of integrated intensity per count from Western blots . Quantitative analysis of endogenous protein levels is shown as relative value from WT conditions calculated within the same gel from lysates of P10 and 6M mouse brains ( a1 and a2 , respectively ) . P values of ANOVA comparison are shown . The total number of animals used for quantification is shown in parentheses . ( b1 ) Tris-Tricine Western blot of brain lysates harvested at 6M is shown . ( b2 ) Quantitative analysis of APP-FL , APP-CTF accumulation and the ratio APP-CTF/APP-FL in brains of PS1 KI is represented . Values are reported as a relative change in the intensity of the protein as compared with the WT littermates . Statistical analysis was performed using ANOVA Kruskal-Wallis test followed by Dunn’s post hoc multiple comparison analysis . p<0 . 001 compared to WT . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 15645 . 00810 . 7554/eLife . 15645 . 009Figure 3—figure supplement 2 . No additive effect of γ-secretase inhibition on neurite outgrowth in APP-deficient neurons expressing PS1 mutation . Primary cortical neurons ( 7–8 DIV ) from WT , PS1 KI , and APP KO x PS1 KI embryonic mouse brains were transfected with YFP , treated with or without γ-secretase inhibitor Compound E ( CompE 10 nM , 24 hr ) , and immunostained with an MAP2 antibody . ( a ) Dual color images of YFP fluorescence and MAP2 staining were taken revealing axons ( green ) and dendrites ( yellow ) . ( b ) Quantitative analysis is shown as relative changes in axonal ( b1 ) and dendritic ( b2 ) outgrowth in neurons as compared to WT littermates of PS1 KI mice . Statistical analysis was performed using ANOVA Kruskal-Wallis test followed by Dunn’s post hoc multiple comparison analysis . p<0 . 001 compared to PS1 KI , ##p<0 . 001 compared to PS1 KI treated with CompE . The total number of neurons ( PS1 KI , n = 6 embryos; APP KO x PS1 KI , n = 6 embryos from at least 5 independent sets of cultures ) used for quantification is shown in parentheses . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 15645 . 00910 . 7554/eLife . 15645 . 010Figure 3—figure supplement 3 . DCC is not an essential substrate in PS1-induced neurite outgrowth . Primary cortical neurons generated from DCC KO and WT littermate mice were transfected with YFP , treated with γ-secretase inhibitor Compound E ( 10 nM , 24 hr ) , and immunostained with MAP2 . ( a ) Overlay images of YFP fluorescence and MAP2 staining of neurons at 8 DIV reveal axons ( green ) and dendrites ( yellow ) . Total axonal ( b1 ) and dendritic ( b2 ) areas were quantified and plotted relative to WT littermates . Statistical analysis was performed using ANOVA Kruskal-Wallis test followed by Dunn’s post hoc multiple comparison analysis . p<0 . 05 , p<0 . 001 compared to WT , and blue p<0 . 05 , p<0 . 001 compared to DCC KO . The total number of neurons ( WT , n = 7 embryos; DCC KO , n = 7 embryos; from at least 4 independent sets of cultures ) used for quantification is shown in parentheses . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 15645 . 01010 . 7554/eLife . 15645 . 011Figure 3—figure supplement 4 . Intramembraneous proteolysis of DCC in brains of FAD-linked PS1 mutant knock-in mice . Expression of FAD-linked PS1-M146V mutation correlates with DCC-CTF accumulation . Endogenous DCC-FL and accumulation of DCC-CTF were detected by immunoblotting in brain lysates from embryonic day 15 , postnatal day 7 , 10 and 14 of knock-in mice expressing FAD-linked PS1-M146V variant ( PS1 KI/KI ) , using G97-449 DCC antibody . Pooled results from various ages ( E15-P14 ) are shown . Quantitative analysis of DCC-FL and DCC-CTF accumulation in homozygote PS1 KI/KI and WT littermates is shown . Values are reported as a relative change in the intensity of the protein as compared with WT littermates . Statistical analysis was performed using Mann-Whitney test comparison analysis . p<0 . 001 compared to WT . The total number of animals used for quantification is shown in parentheses . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 15645 . 011 Next , we investigated the changes of γ-secretase activity on axodendritic development using primary neuronal cultures generated from PS1 KO and PS1 KI mice and their WT littermates . We observed a significant increase in both axonal and dendritic outgrowth in PS1 KI cultured neurons as compared to WT , an effect as substantial as the effect seen in neurons lacking PS1 expression ( Figure 3c and d ) . Axonal and dendritic outgrowth was increased without apparent bias toward either neuronal compartment . Strikingly , we noticed that PS1 mutation-induced neurite outgrowth was comparable in neurons expressing one or two copies of the mutated gene , which is consistent with the idea of autosomal dominant inheritance of PSEN1 mutations . Our findings also suggest that modest reduction in γ-secretase activity through an endogenous level of PS1 mutation expression is sufficient to influence drastically axodendritic outgrowth , an effect that emerges in conjunction with modest APP-CTF accumulation . To follow-up , we investigated the influence of APP expression on neurite extension . We examined whether neurons lacking APP expression would exhibit a similar level of exuberant neurite outgrowth in FAD-linked PS1-M146V variant using double crossed APP KO x PS1 KI mice . Remarkably , lack of APP expression abolished excess axonal and dendritic outgrowth found in neurons carrying the PS1-M146V mutation ( Figure 3d1; 97% and 62% reduction , respectively ) . We also subjected neuronal culture generated from PS1 KI and APP KO x PS1 KI mice to γ-secretase inhibitor Compound E . We observed that presenilin-dependent enhancement of axodendritic outgrowth was not furthermore increased in APP KO x PS1 KI cultures ( Figure 3—figure supplement 2 ) , supporting the idea that APP is the major contributing substrate in FAD-linked PS1 mutation induced neurite outgrowth . To corroborate our finding on the predominant role of APP in neurite outgrowth , we investigated whether other γ-secretase substrates may contribute to or exacerbate the aberrant neuritic processes seen in neurons expressing PS1 mutation . We and others have previously reported that γ-secretase-dependent accumulation of Deleted Colorectal Cancer C-terminal fragment ( DCC-CTF ) promotes neurite outgrowth in neuroblastoma cells ( Parent et al . , 2005 ) and motor neuron explants ( Bai et al . , 2011 ) . We used primary neurons generated from DCC KO mice , which lack DCC expression , to investigate the contribution of DCC to γ-secretase mediated axodendritic outgrowth . We observed that lack of DCC expression does not affect axonal or dendritic outgrowth in our cell culture system ( Figure 3c and d ) . Within the same experimental conditions , sets of neurons were treated with γ-secretase inhibitor Compound E to abrogate substrate proteolysis . In these groups , enhancement of axodendritic outgrowth was similar in naïve WT as compared to DCC KO neurons treated with a γ-secretase inhibitor ( Figure 3—figure supplement 3 ) . We also confirmed that lack of DCC expression did not affect axodendritic outgrowth in neurons expressing FAD-linked PS1-M146V mutation ( Figure 3c and d ) , even though DCC-CTF was increased in cultured neuron lysates of homozygous and heterozygote PS1 KI relative to WT ( Figure 3e ) . An increase of DCC-CTF detection was also significant in vivo during early brain development and the period of synaptogenesis ( E15-P14 ) in PS1 KI mice as compared to WT ( Figure 3—figure supplement 4 ) . Therefore , our results indicate that DCC is not an essential substrate for γ-secretase induced neurite outgrowth . To examine the in vivo consequence of neurite outgrowth that we see in our neuronal culture system , we measured axonal sprouting in brain slices from 6 month-old animals using GAP-43 as a marker of axonal plasticity ( reviewed by ( Benowitz and Routtenberg , 1997; Denny , 2006 ) ) . As observed by several groups ( Benowitz et al . , 1988; Goslin et al . , 1988; Oestreicher and Gispen , 1986 ) , GAP-43 staining correlates selectively with axonal development and is prominent in brain areas where axon fibers project . Consistent with our neurite outgrowth analysis , GAP-43 expression is significantly increased in hippocampal areas of PS1 KI mice , an effect that was abrogated if APP expression was absent ( see Figure 4a ) . Quantitative analysis reveals that intensity of GAP-43 immunostaining was selectively increased in the stratum lacunosum moleculare ( LMol ) and the outer and inner molecular layers of the dentate gyrus ( OMoDG and IMoDG , respectively ) in PS1 KI mice as compared to WT ( Figure 4b ) . An increase of GAP-43 immunostaining intensity levels was seen in these selective areas as shown by representative line scan projection of GAP-43 staining in 6 month-old PS1 KI mice as compared to their WT littermates , an effect that was not observed in mice lacking APP expression ( Figure 4a3 ) . We confirmed by Western blot analysis that GAP-43 expression was increased as well in the hippocampus as early as at postnatal day 10 ( P10 ) in PS1 KI as compared to APP KO x PS1 KI mice , an effect that was not observed in the cortex of these mice ( Figures 4b and c ) . Confocal microscopy corroborated the existence of stronger immunofluorescence intensity in the dentate gyrus , especially in OMoDG layer of mice expressing FAD-linked PS1 variant as compared to their counterpart APP KO x PS1 KI littermates ( Figure 4d ) . The surface render projection of zoomed areas revealed that GAP-43 immunostaining is more abundant in PS1 KI surrounding neurite-like structures , which was not seen in APP KO x PS1 KI mouse brain sections . Altogether , our results indicate that mice expressing an endogenous level of FAD-linked PS1 variant exhibit exuberant neurite outgrowth and axonal sprouting at close proximity to the perforant path , which requires APP expression . 10 . 7554/eLife . 15645 . 012Figure 4 . APP-dependent axonal sprouting in the hippocampus of knock-in mice expressing FAD-linked PS1 variant . ( a ) Immunohistochemistry of GAP-43 on coronal brain sections of 6 months-old WT , PS1 KI , and APP KO x PS1 KI mice is shown . ( a1 ) Representative pseudocolor images of brain sections present GAP-43 staining intensity in several brain areas ( Cx: cortex , alv: alveus , Or: oriens , Pyr: pyramidal layer , Rad: radiatum , LMol: lacunosum moleculare , OMoDG: outer molecular layer of the dentate gyrus , IMoDG: inner molecular layer of the dentate gyrus , GrDG: granular layer of the dentate gyrus , PoDG: polymorph layer of the dentate gyrus ) . ( a2 ) Enlarged views of the dentate gyrus areas are shown . ( a3 ) Representative line-scans ( as indicated in a1 ) show levels of overlapping-peak intensity of GAP-43 immunostaining across cortical and hippocampal areas . More intense staining is noticeable especially in hippocampal areas of PS1 KI as compared to WT and APP KO x PS1 KI . ( b ) Quantitative analysis of GAP-43 staining intensity in several brain areas is represented as relative changes compared to WT . ( c ) Western blot analysis of steady-state levels of GAP-43 is examined using cortex and hippocampus lysates from WT , PS1 KI and APP KO x PS1 KI mouse brains taken at postnatal day 10 ( P10 ) . Flotillin-2 was used as loading control . ( d ) Confocal images of GAP-43 staining in 6 months-old mice showing the axonal sprouting in OMoDG layer of PS1 KI as compared to APP KO x PS1 KI mice . The boxed regions are shown as enlarged insets . Inverted images at the bottom show the surface rendered zoomed areas generated with Huygens software . Statistical analysis was performed using ANOVA Kruskal-Wallis test followed by Dunn’s post hoc multiple comparison analysis . p<0 . 05 , p<0 . 001 compared to WT littermates , and #p<0 . 05 , ##p<0 . 001 compared to PS1 KI littermates . The total number of animals used for quantification is shown in parentheses . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 15645 . 012 We previously reported that accumulation of APP-CTF caused by overexpression of membrane-tethered APP intracellular domain ( mAICD ) favors axodendritic arborization as a result of direct coupling with GαS and subsequent activation of adenylate cyclase ( Deyts et al . , 2012 ) . Consequentially , cAMP/PKA cascade is initiated and leads to phosphorylation of CREB . As described in our previous studies ( Deyts et al . , 2012 ) and shown in Figure 5 , we observed that concomitant expression of APP-FL and inhibition of γ-secretase activity favors neurite outgrowth ( Figure 5a ) and associated increase of phosphorylated CREB ( pCREB , Figure 5b ) in cortical neuronal cultures . In support of this finding , we also observed that neurons overexpressing APP-C99 ( the APP β-secretase cleaved CTF by-product ) exhibit increased axodendritic outgrowth that was prevented by the expression of a C99 construct lacking the GαS-protein interacting site where three alanine substitutions were introduced to replace the basic residues of the BBXXB motif in APP-CTF ( named C99mutAAA; see Figure 5a ) . As shown in Figure 5b , we observed that neurons expressing APP-C99 or mAICD constructs exhibit a strong increase of CREB signaling illustrated by the enhancement of pCREB immunostaining . These outcomes were abolished in neurons overexpressing mutant lacking GαS-protein interacting site C99mutAAA and mAICDmutAAA , respectively ( Figure 5b1 and b2 ) . As expected for CREB signaling , these effects are more noticeable in the somatic/nuclear area as compared to dendritic projections ( Figure 5—figure supplement 1 ) . To further establish the importance of γ-secretase dependent APP-CTF accumulation in this process , we overexpressed membrane-targeted mAICD and APP-NTF by-products in primary mouse embryonic fibroblasts ( MEF ) or neurons generated from APP KO ( Figure 5c1 and c2 , respectively ) . We observed that neither sAPPα nor sAPPβ overexpression influenced CREB signaling in HEK293 cells or in neurons lacking APP expression ( Figure 5—figure supplement 2a and Figure 5c1 , respectively ) . We confirmed this observation by Western blots of MEF lysates generated from APP KO mice ( Figure 5c2 ) and lysates of HEK293 cells ( Figure 5—figure supplement 2b ) . Altogether , our results point out that increase in APP-CTF accumulation is causing changes in neurite outgrowth and CREB signaling in PS1 loss-of-function neurons , therefore supporting the idea that APP-CTF accumulation is the main contributing factor to PS1-dependent modulation of CREB signaling . 10 . 7554/eLife . 15645 . 013Figure 5 . Loss of γ-secretase activity is associated with a gain of APP-CTF signaling . ( a1 ) Representative inverted images of cortical neurons ( 8 DIV ) coexpressing YFP and APP-FL , C99 , membrane-tethered APP intracellular domain ( mAICD ) or mutants lacking GαS-protein interacting site ( C99mutAAA and mAICDmutAAA ) are shown . ( a2 ) Analysis of total neurite area is presented as relative to the empty vector ( EV ) transfected control group . ( b1 ) Representative pseudocolor images of phosphorylated CREB ( pCREB ) immunofluorescence are shown for WT neurons ( 14 DIV ) expressing APP-FL ( treated or not with γ-secretase inhibitor Compound E; 10 nM , 24 hr ) or various APP-CTF constructs using polyclonal phospho- ( Ser133 ) CREB antibody . ( b2 ) Quantitative analysis of pCREB staining intensity is represented as relative changes as compared to EV . ( c1 ) pCREB immunofluorescence intensity levels were analyzed in 14 DIV neurons expressing EV , soluble APPα ( sAPPα ) , soluble APPβ ( sAPPβ ) or mAICD that were generated from APP KO mice . Quantitative analysis of pCREB staining intensity is represented as relative changes compared with EV transfected neurons . ( c2 ) Western blot analysis of pCREB accumulation is shown in primary mouse embryonic fibroblasts ( MEF ) expressing EV , sAPPα , sAPPβ or mAICD that were generated from APP KO mice . The level of sAPP was examined in the conditioned media using the NTH-452 antibody that recognized the N-terminal fragment of APP . The expression level of pCREB , Flotillin-2 , and mAICD ( using CTM1 antibody ) were examined in cell lysates . Statistical analysis was performed using ANOVA Kruskal-Wallis test followed by Dunn’s post hoc multiple comparison analysis . p<0 . 001 compared to EV , and #p<0 . 05 and ##p<0 . 001 compared to the overexpressing condition within the same group . The total number of neurons ( from at least 3 independent sets of cultures ) used for quantification is shown in parentheses . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 15645 . 01310 . 7554/eLife . 15645 . 014Figure 5—figure supplement 1 . Accumulation of APP-CTF is associated with a larger increase in CREB signaling in the somatic area . Mouse cortical neurons were transfected at 11 DIV . Cells were treated with Compound E ( 10 nM ) 24 hr before fixation ( 14 DIV ) . Quantitative analysis of somatic ( a ) and dendritic ( b ) staining intensity of pCREB is shown under basal condition ( WT neurons ) and in conditions that accumulate APP-CTF through γ-secretase inhibition ( CompE 10 nM , 24 hr ) with endogenous or overexpression of APP-FL , or overexpression of APP-C99 ( C99 ) , or overexpression of membrane-tethered APP intracellular domain ( mAICD ) . Quantitative analysis of pCREB staining intensity is represented as relative change as compared to WT . Statistical analysis was performed using ANOVA Kruskal-Wallis test followed by Dunn’s post hoc multiple comparison analysis . p<0 . 05 , p<0 . 001 compared to WT basal condition . The total number of quantified cells is shown in parentheses ( from at least 3 independent sets of cultures ) . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 15645 . 01410 . 7554/eLife . 15645 . 015Figure 5—figure supplement 2 . Soluble APP does not affect CREB signaling in HEK293 transfected cell lines . Steady-state levels of pCREB were examined by immunostaining ( a ) and Western blot ( b ) in HEK293 cells transfected with empty vector ( EV ) treated with forskolin ( FSK , 1 μM ) , and in overexpressing conditions using soluble APPα ( sAPPα ) , soluble APPβ ( sAPPβ ) or mAICD plasmids . ( a ) Quantification of total phosphorylated CREB ( pCREB ) levels is shown as changes in relative intensity as compared to EV . ( b ) pCREB accumulation and mAICD expression were examined in cell lysates using phospho- ( Ser133 ) CREB and APP CTM1 antibodies , respectively . The level of sAPP was examined in the media collected from transfected HEK293 cells using the NTH-452 antibody that recognized the N-terminal fragment of APP . Flotillin-2 was used as loading control . Statistical analysis was performed using ANOVA Kruskal-Wallis test followed by Dunn’s post hoc multiple comparison analysis . p<0 . 001 compared to EV . The total number of quantified cells is shown in parentheses . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 15645 . 015 As a read-out of APP-CTF downstream signaling , we first assessed if the expression of PS1 mutation could alter cAMP/PKA-dependent signaling cascades using antibodies that detect phosphorylated PKA substrates ( pPKA; Figure 6—figure supplement 1 ) . We observed that primary cortical neurons generated from PS1 KI mice exhibit a significant overall increase of phosphorylated PKA substrates ( Figure 6—figure supplement 1 ) . Associated PKA-dependent CREB signaling was substantiated by an increase in pCREB immunostaining as compared to their WT littermates ( Figure 6a1 and a2 ) . As we have described previously ( Barnes et al . , 2008 ) , we reiterate here that neurons lacking PS1 function exhibit increased pCREB levels ( Figure 6a2 ) . As expected for CREB signaling , these effects are more noticeable in the somatic/nuclear area as compared to dendritic projections ( Figure 6—figure supplement 2b ) . We confirmed by Western blot analysis that pCREB was significantly increased in neuron lysates generated from PS1 KI mice ( Figure 6b1 and b3 ) , an effect that was also reported in neuroblastoma cell lines and mouse brains expressing FAD-linked PS1 mutations ( Muller et al . , 2011 ) . In line with PS1 loss-of-function phenotype , Western blot analysis of neuronal lysates confirmed increases of pCREB levels in neurons treated with γ-secretase inhibitor or neurons generated from PS1 KO mouse brains ( Figure 6b2 and b3 ) . Next , we examined whether or not CREB signaling was influenced by PS1 mutations in cells lacking APP expression . In support of our neurite outgrowth studies presented in Figure 1 , we likewise observed that neurons generated from APP KO present a significant decrease in pCREB level ( Figure 6a2 ) . More interestingly , we found that neurons generated from PS1 KI mice that lack APP expression ( APP KO x PS1 KI ) do not show increased pCREB as observed in PS1 KI littermates , an effect that was not seen in neurons lacking DCC expression ( Figure 6a1 , a2 , b2 , and b3 ) . Therefore , we conclude that APP expression is associated with the increase in CREB signaling in PS1 KI neurons . Our results confirm that APP is sufficient and necessary to induce CREB signaling in the loss-of-function PS1 mutant ( see summary Table 1 ) . 10 . 7554/eLife . 15645 . 016Figure 6 . APP-dependent enhancement of CREB signaling in neurons expressing FAD-linked PS1 variant . ( a ) Phosphorylated CREB ( pCREB ) immunofluorescence staining was performed in neurons at 14 DIV using polyclonal phospho- ( Ser133 ) CREB antibody . ( a1 ) Representative pseudocolor images of pCREB immunostaining levels at steady-state are shown in neurons generated from PS1 KI , APP KO , APP KO x PS1 KI , DCC KO , and DCC KO X PS1 KI embryonic brains , and treated with γ-secretase inhibitor Compound E ( 10 nM , 24 hr ) . ( a2 ) Quantitative analysis of pCREB staining intensity is represented as relative changes as compared to WT . ( b ) Steady-state levels of pCREB were examined in neuronal lysates at 14 DIV cortical neurons by Western blot analysis in PS1 KI neurons ( b1 ) and in γ-secretase deficient neurons ( b2 ) . Flotillin-2 was used as loading control . ( b3 ) The ratio of pCREB intensity over the intensity of Flotillin-2 as compared to WT is shown . Results were quantified from at least 2 independent cultures . Statistical analysis was performed using ANOVA Kruskal-Wallis test followed by Dunn’s post hoc multiple comparison analysis . p<0 . 001 compared to WT , blue p<0 . 001 compared to DCC KO , and ##p<0 . 001 compared to PS1 KI . The total number of quantified cells is shown in parentheses ( WT , n = 6 embryos; PS1 KI , n = 6 embryos; APP KO , n = 4; APP KO x PS1 KI , n = 4 embryos; DCC KO , n = 4; DCC KO x PS1 KI , n = 7 embryos ) . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 15645 . 01610 . 7554/eLife . 15645 . 017Figure 6—figure supplement 1 . Increase in cAMP/PKA signaling in neurons expressing FAD-linked PS1 mutant . Cortical cultures ( 14 DIV ) , generated from PSEN1-M146V knock-in mouse ( PS1 KI/+ and PS1 KI/KI ) and WT littermates , were immunostained with a polyclonal antibody directed against phosphorylated Ser/Thr PKA substrates ( pPKA ) . ( a ) Relative levels of total pPKA staining intensity are shown in somatic ( b ) and dendritic ( c ) areas . Statistical analysis was performed using ANOVA Kruskal-Wallis test followed by Dunn’s post hoc multiple comparison analysis . p<0 . 001 compared to WT . The total number of neurons used for quantification is shown in parentheses ( WT , n = 2 embryos; PS1 KI /+ , n = 2 embryos; PS1 KI/KI , n = 2 embryos ) . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 15645 . 01710 . 7554/eLife . 15645 . 018Figure 6—figure supplement 2 . The APP-dependent increases of CREB signaling in somatic and dendritic areas . Mouse cortical neurons were transfected at 11 DIV . 24 hr before fixation ( 14 DIV ) , cells were treated with Compound E ( 10 nM , 24 hr ) . pCREB staining intensity was performed in neurons generated from PS1 KI , APP KO , APP KO x PS1 KI , DCC KO and DCC KO x PS1 KI . Quantitative analysis of pCREB staining intensity in the soma ( a ) and in dendrites ( b ) is represented as relative change as compared to WT . Statistical analysis was performed using ANOVA Kruskal-Wallis test followed by Dunn’s post hoc multiple comparison analysis . p<0 . 001 compared to WT , yellow p<0 . 001 compared to APP KO , blue p<0 . 001 compared to DCC KO , and ##p<0 . 001 compared to PS1 KI . The total number of quantified cells is shown in parentheses ( WT , n = 6 embryos; PS1 KI , n = 6; APP KO , n = 4; APP KO x PS1 KI , n = 4; DCC KO , n = 4; DCC KO x PS1 KI , n = 7 embryos ) . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 15645 . 01810 . 7554/eLife . 15645 . 019Table 1 . Summary results of characterized mouse models . The mouse breeding to obtain the experimental genotyped , the effect on axodendritic outgrowth , CREB signaling , and APP-CTF accumulation are summarized . Changes are indicated as strongly increase ( ↑↑ ) , slightly increase ( ↑ ) , slightly reduce ( ↓ ) , no-change ( - ) , or not determined ( ND ) . Axodendritic components and associated CREB signaling alterations were evaluated in loss or partial loss of γ-secretase activity using γ-secretase inhibitor Compound E , PS1 KO , APPswe/PS1-ΔE9 ( Tg85Dbo ) and FAD-linked PSEN1-M146V knock-in ( PS1 KI ) mouse lines crossed to APP KO and DCC KO mice . Axonal as well as dendritic arborizations were increased in cortical neurons generated from PS1 KO , PS1 KI and Tg85Dbo mice . Augmentation of neurite extension correlates with the increase in CREB signaling and APP-CTF accumulation . Lack of APP expression in PS1 KI mice selectively alters axonal outgrowth that parallels a reduction in CREB signaling , as compared to their PS1 KI littermates . Lack of DCC expression in PS1 KI mice does not affect neurite outgrowth and CREB signaling , as compared to their PS1 KI littermates , therefore supporting the idea that APP is an essential substrate , but not DCC , in γ-secretase-mediated axodendritic plasticity . DOI: http://dx . doi . org/10 . 7554/eLife . 15645 . 019Mouse modelsParental strainsAxonal outgrowthDendritic outgrowthCREB signalingAPP-CTFCompound EWT↑↑↑↑↑↑↑PS1 KOPS1+/- x PS1+/-↑↑↑↑↑↑↑↑PS1 KI/KIPS1 KI/+ x PS1 KI/+↑↑↑↑↑↑↑Tg85DboAPPswe x PS1-ΔE9↑↑↑↑ND↑↑APP KOAPP+/- x APP+/- ↓ −↓noneAPP KO x PS1 KIAPP+/- x APP+/- PS1 KI/KI ↓ − −noneDCC KODCC+/- x DCC+/- − − −NDDCC KO x PS1 KIDCC+/- x DCC+/- PS1 KI/KI↑↑↑↑↑↑ND CREB signaling possesses a variety of effectors including direct upstream activation of adenylate cyclase and subsequent production of cAMP . We previously reported that APP-CTF induced neurite outgrowth was prohibited in a condition where adenylate cyclase was inhibited ( Deyts et al . , 2012 ) . We documented that APP-CTF initiates stimulation of adenylate cyclase through trimeric interaction with GαS ( Deyts et al . , 2012 ) . Therefore , to determine if the increase of CREB signaling seen in PS1 loss-of-function mutants is entirely achieved through the accumulation of APP-CTF mediating adenylate cyclase activation , we treated neurons with adenylate cyclase inhibitor MDL-12 , 330A ( Figure 7a ) . We observed that adenylate cyclase inhibition profoundly reduced CREB activity in neurons generated from PS1 KI . We also confirmed that enhanced neurite outgrowth in neurons expressing PS1 mutants was dependent on adenylate cyclase activity ( Figure 7b ) . Moreover , we examined if the FAD-linked mutation in APP could modify the propensity of PS1 mutation to induce neurite outgrowth . We generated cortical neuronal cultures from transgenic mouse embryos coexpressing APPswe and PS1-ΔE9 mutations ( Tg85Dbo ) , a mouse model associated with accumulation of Aβ burden ( Jankowsky et al . , 2005 ) . We observed a significant increase of axodendritic outgrowth in Tg85Dbo as compared to their non-transgenic littermates , at the same level seen in neurons generated from PS1 KO and PS1 KI ( Figure 7b2 ) . This effect was completely alleviated by a 24 hr treatment with adenylate cyclase inhibitor . Altogether , our results confirm the requirement of cAMP/PKA-dependent signaling to induce neurite outgrowth in neurons that exhibit increased APP-CTF ( see summary Table 1 ) . 10 . 7554/eLife . 15645 . 020Figure 7 . PS1-induced neurite outgrowth and associated CREB signaling require adenylate cyclase activation . ( a1 ) Representative pseudocolor images of phosphorylated CREB ( pCREB ) immunofluorescence is shown in 14 DIV neurons generated from PS1 KI mice treated with adenylate cyclase inhibitor MDL-12 , 330A ( 100 nM , 30 min ) using polyclonal phospho- ( Ser133 ) CREB antibody . ( a2 ) Quantitative analysis of pCREB staining intensity is represented as relative changes as compared to untreated PS1 KI neurons . p<0 . 001 compared to PS1 KI . ( b1 ) Representative inverted images of YFP fluorescence in cortical neurons ( 8 DIV ) generated from PS1 KO , PS1 KI , andTg85Dbo mice are shown after 24 hr treatment with MDL-12 , 330A ( 10 nM ) . ( b2 ) Analysis of neurite extension is represented as relative changes in total neurite area as compared to WT littermates . p<0 . 001 compared to WT , ##p<0 . 001 compared to untreated condition for each group . The total number of neurons ( from at least 3 independent sets of cultures ) used for quantification is shown in parentheses ( WT , n = 5 embryos; PS1 KO , n = 3 embryos; PS1 KI , n = 3 embryos; Tg85Dbo , n = 3 embryos ) . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 15645 . 020 For the past two decades , researchers have focused on the neurotoxicity associated with Aβ peptide production and accumulation . Our study explored the APP molecule as a whole cellular component that could affect neuronal development and perhaps have consequence on the course of AD , thus filling an important gap in AD research . Our results provide a meaningful understanding of how attenuation of γ-secretase activity and gain of APP function relate to each other , and its significance in health and disease . We took advantage of KI mice expressing FAD-linked PS1 variant , an AD mouse model that does not overexpress APP and PS1 mutations , to demonstrate the necessity of APP and its intracellular metabolites in the development of AD , before the emergence of overt cerebral Aβ burden . We show striking neuronal influences , mechanistic explanations , and in vivo evidence on nerve growth associated with PS1 mutations . Altogether , our findings clearly indicate that APP is an essential substrate in mediating neurite outgrowth in conditions where γ-secretase activity is inhibited or altered . Our results reveal that selective accumulation of membrane-anchor APP-CTF through its coupling to adenylate cyclase is sufficient to induce these changes . We observed that primary cortical neurons generated from PS1 KO and PS1 KI mice harboring the FAD-linked PS1-M146V variant exhibit exuberant axodendritic outgrowth ( see summarized results in Table 1 ) . These results correlate with a large and more moderate increase of APP-CTF in brain lysates prepared from PS1 KO and PS1 KI mice , respectively . This supports the notion of partial loss-of-function of γ-secretase activity associated with FAD-linked PS1 mutation . Strikingly , the lack of APP expression in cortical neurons expressing the PS1-M146V variant led to a recovery of basal axodendritic outgrowth , an effect that was not seen in neurons lacking DCC expression ( another γ-secretase substrate ) . Treatment with the γ-secretase inhibitor did not induce additional axodendritic morphological changes in APP KO x PS1 KI , again supporting the necessity of APP and APP-CTF accumulation in PS1-induced neurite outgrowth . Reduced γ-secretase activity was accompanied by an increase of CREB signaling , an effect that was abolished in APP-deficient neurons and prohibited by adenylate cyclase inhibition . Interestingly , we observed that adult PS1 KI mice exhibit aberrant axonal sprouting , especially in hippocampal areas . More importantly , lack of APP expression in the FAD-linked PS1 mutant mouse model eliminated that effect , providing evidence that APP expression is causally implicated with the pathological feature associated with loss-of-function PS1 mutation . Our investigations point out that APP is not an important contributing factor to neurite formation under normal physiological condition . In fact , loss of APP expression only slightly affects axonal outgrowth , at least in cell culture system . Therefore , our observations reinforce the notion that APP must be tightly regulated , likely through secretase-dependent cleavages and degradation process , which would render the molecule less vital for global cellular function . In support of this idea , knocking down APP in mice does not lead to severe phenotype , therefore questioning also the significance of the holoprotein ( Deyts et al . , 2016; Muller and Zheng , 2012 ) . Previous studies emphasize this interpretation [review by ( Deyts et al . , 2016; Haass et al . , 2012; Muller and Zheng , 2012; Thinakaran and Koo , 2008 ) ] . APP KO mice are viable ( Zheng et al . , 1995 ) , and present subtle , age-dependent decline in memory tasks ( Ring et al . , 2007; Schrenk-Siemens et al . , 2008 ) . Nonetheless , our study clearly indicates that γ-secretase-dependent inhibition of APP processing , and subsequent APP-CTF accumulation , mediates a striking impact on axodendritic growth . Indeed , we observed that axodendritic outgrowth and associated signaling are reversed in the absence of APP expression in cell culture systems . It has been proposed that ectodomain shedding of cell-surface receptors by γ-secretase complex activity may serve to prevent further cell-surface signaling by a variety of substrates ( Kopan and Ilagan , 2004; McCarthy et al . , 2009 ) . Our study clearly demonstrated the importance of this process and the essential contribution of APP-CTF in conditions when PS1 is mutated or γ-secretase activity is reduced . It appears that among all known ( and unknown ) γ-secretase substrates , APP is the key contributing substrate in observed dysregulation of neurite outgrowth . It is even more remarkable that APP expression would be so critical in controlling neurite development in selective brain areas associated with memory formation . Therefore , APP and its CTF accumulation could play a more critical role in disease states . Noteworthy , the exuberant axonal sprouting observed in adult PS1 KImice selectively affects the hippocampal areas that are connected with entorhinal cortex/perforant path projections ( Forster et al . , 2006 ) , which is reminiscent of the aberrant sprouting described in AD patients ( Arendt , 2001; Geddes et al . , 1985; Masliah et al . , 1991; Rekart et al . , 2004 ) . Higher intensity levels of GAP-43 correlated with the hippocampal sprouting in the molecular layer seen in mouse model overexpressing the FAD-linked APP variant ( Chin et al . , 2004; Phinney et al . , 1999 ) . Mice expressing the FAD-linked PS1 mutation exhibit similitude in that respect . We would consider these sequences of events as an early indication of AD development since Aβ accumulation and deposition have not occurred yet in this model . We corroborated changes seen in our culture systems with that in mouse brains expressing the FAD-linked PS1-M146V variant , reiterating the importance of APP expression as a cause of exuberant neurite outgrowth and sprouting observed in a PS1 KI mouse model . Clearly , it has been reported that lack of APP expression in rodent or Drosophila models could lead to inaccuracies in axon guidance in some parts of the brain , including the commissural projections ( Magara et al . , 1999; Rama et al . , 2012 ) and mushroom body development , a Drosophila brain structure involved in learning and memory ( Soldano et al . , 2013 ) [see also ( Muller and Zheng , 2012; Nicolas and Hassan , 2014 ) for review] . More interestingly , Drosophila lacking the APP homolog expression requires the APP-CTF component for accurate axonal outgrowth ( Soldano et al . , 2013 ) , indicating the significance of APP-CTF for proper axonal guidance and memory function in Drosophila . One important question that arises from our findings is whether APP/APP-CTF accumulation and associated signaling sequences that impact neurite outgrowth are contributing factors to the disease , or perhaps considered as a compensatory epiphenomenon ? One could argue that PS1 loss-of-function mutations favor memory formation since APP-CTF accumulation leads to CREB signaling . It is widely accepted that CREB signaling is important for memory consolidation and cellular events that preserve cell integrity [review by ( Abel and Nguyen , 2008; Alberini and Kandel , 2015; Lonze and Ginty , 2002 ) ] . Conversely , ablation of CREB is associated with axonal and dendritic outgrowth defects ( Lonze et al . , 2002; Redmond et al . , 2002 ) . Therefore , an increase in neurite outgrowth along with CREB signaling would support a prominent role of APP in learning and memory . Even though long-term synaptic plasticity is enhanced in several mouse models expressing FAD-linked PS1 variants ( Parent et al . , 1999; Parent and Thinakaran , 2010 ) , which is consistent with an increase in CREB signaling , a number of these mice performed poorly in a variety of memory tasks ( Webster et al . , 2014; Xia et al . , 2015 ) . One alternative to consider is the detrimental consequence of over-active CREB signaling . An increase of CREB signaling , in cell lines and brain lysates generated from mouse lines carrying PS1 mutations ( including the PS1M146V mutation ) , can be reversed or normalized by interfering with ER Ca2+ stores and the IP3-dependent signaling pathway , an effect that eliminated Aβ toxicity and cell death normally observed in these cells ( Muller et al . , 2011; Shilling et al . , 2014 ) [see also ( Guo et al . , 1999 ) ] . Therefore , we stipulate that excessive activation of the CREB pathway may render the cell more vulnerable to injury . Excitatory synaptic plasticity is increased in neurons expressing FAD-linked PS1 variants ( Parent et al . , 1999; Parent and Thinakaran , 2010 ) ; an effect that causes exaggeration of Aβ production ( Cirrito et al . , 2005; Kamenetz et al . , 2003 ) . Recent studies demonstrated that chronic entorhinal cortex hyperexcitability in an AD mouse model harboring APP mutations could contribute to a consequential increase in Aβ burden ( Yamamoto et al . , 2015 ) , opening up speculation about the role of APP in the synaptically-driven perforant path cascade of events ( Buxbaum et al . , 1998 ) . Indeed , Aβ burden is reduced in the Tg85Dbo AD mouse model where the perforant path projections are disrupted ( Lazarov et al . , 2002; Sheng et al . , 2002 ) . It remains to be determined if APP-CTF accumulation would affect any of these outcomes . γ-secretase dependent accumulation of APP by-products is associated with AD and is considered to be detrimental to neuronal function [review by ( Hardy and Selkoe , 2002; Musiek and Holtzman , 2015 ) ] . Indeed , it has been well documented that the expression of genes encoding several FAD-linked PS1 mutations favors the accumulation of Aβ and potential disease-related toxicity especially in combination with overexpression of APP , at least in animal models ( http://www . alzforum . org/research-models ) . However , a link between AD , aberrant neuronal sprouting in the hippocampal area , partial loss-of-function of γ-secretase activity , and accumulation of APP-CTF has not been described so far . Interestingly , it has been reported that APP-C99 accumulation also takes place selectively in the hippocampal area of 3xTgAD mice ( a more aggressive model harboring PS1-M146V/APPswe/Tau-P301L mutations ) , even prior to accumulation of Aβ ( Lauritzen et al . , 2012 ) . APP-CTF accumulation is found surrounding dystrophic neurites in KI mice expressing the FAD-linked PS1-L166P mutation and two copies of APP-WT ( Vidal et al . , 2012 ) . We also found that neurons generated from transgenic mice coexpressing APPswe and PS1-ΔE9 mutations ( Tg89Dbo ) exhibit exuberant axodendritic outgrowth that was prohibited by adenylate cyclase inhibition , despite APP-CTF accumulation . Accordingly , we argue that the AD-like aberrant sprouting phenotype depends on signaling downstream of APP-CTF and is preventable through the inhibition of adenylate cyclase activity . Altogether , we demonstrated that APP expression and accumulation of its intracellular fragment are required for exuberant neurite outgrowth associated with pathological PS1 loss-of-function mutations before the emergence of amyloid burden . Our discovery of APP-dependent axonal sprouting in AD mouse models is certainly novel and of great interest for the understanding of how AD process is initiated . Therapeutic inhibition of γ-secretase and the resulting accumulation of APP-CTF could have significant consequences for AD treatment . Mice with targeted deletion of Psen1 alleles ( PS1 KO ) , targeted deletion of App alleles ( APP KO ) , and targeted deletion of Dcc alleles ( DCC KO ) were generated by intercrossing heterozygote PS1+/- ( Wong et al . , 1997 ) , APP+/- ( Zheng et al . , 1995 ) , and DCC+/- ( Fazeli et al . , 1997 ) , respectively . All these mice were maintained in C57Bl/6J x C3H/HeJ F2 background . Heterozygote transgenic APPswe/PSEN1-ΔE9 ( Tg85Dbo ) mouse pairs were purchased from The Jackson Laboratory ( Bar Harbor , ME ) and maintained in C57Bl/6J background ( Jankowsky et al . , 2004 ) . Homozygote knock-in ( KI ) mice expressing mutant human PSEN1-M146V ( PS1 KI/KI ) were obtained by intercrossing PSEN1-M146V heterozygote ( named PS1 KI/+ ) ( Guo et al . , 1999 ) and maintained in C57Bl/6J x C3H/HeJ F2 background . Double mutants APP KO x PS1 KI and DCC KO x PS1 KI mice were obtained by crossing heterozygote APP+/- and DCC+/- respectively with heterozygote PS1 KI/+ mice . Animal parental strains are summarized in Table 1 . Mouse handling procedures were performed in accordance with National Institutes of Health guidelines . The laboratories of Drs . Susan Ackerman ( The Jackson Laboratory ) , Sangram Sisodia ( University of Chicago ) , and Hui Zheng ( Baylor University ) kindly provided DCC KO , PS1 KO , APP KO , and PS1 KI original mice for colony expansion . APP-CTM1 , APP-CT11 , PS1NT , Flotilin-2 and APPNTH-452 homemade rabbit polyclonal antibodies were generously provided by Dr . Gopal Thinakaran ( University of Chicago , Chicago , IL ) as described previously ( Deyts et al . , 2012; Vetrivel et al . , 2009 ) . Monoclonal anti-MAP2 , GAP-43 ( clone 7B10 ) and GAPDH were purchased from Sigma-Aldrich ( St . Louis , MO ) . Monoclonal DCC ( clone G97-449 ) antibody was purchased from BD Biosciences ( San Diego , CA ) . Polyclonal phospho- ( Ser/Thr ) PKA substrate antibody and phospho-CREB ( Ser133 ) antibodies were purchased from Cell Signaling Technology ( Danvers , MA ) and EMD Millipore ( Billerica , MA ) , respectively . Monoclonal Alexa-647 and Alexa-555 , and polyclonal Alexa-555 secondary antibodies were purchased from Invitrogen ( Carlsbad , CA ) . IRDye 680 and IRDye 800CW-conjugated secondary antibodies were purchased from LI-COR Biosciences ( Lincoln , NE ) . γ-secretase inhibitor Compound E was generously provided by Dr . Todd E . Golde ( University of Florida , Gainesville , FL ) ( Seiffert et al . , 2000 ) . Cis-N- ( 2-phenylcyclopentyl ) azacyclotridec-1-en-2-amine ( MDL , 12-330A ) was obtained from Enzo Life Science ( Farmingdale , NY ) . Tetrodotoxin was purchased from Tocris Bioscience ( distributed by Fisher Scientific , Pittsurgh , PA ) . Unless indicated , all other reagents were purchased from Sigma-Aldrich . Plasmids encoding empty vector ( EV ) , APP-FL , C99 , mAICD , and mAICDmutAAA have been described previously ( Deyts et al . , 2012 ) . Dr . Gopal Thinakaran ( University of Chicago , Chicago , IL ) generously provided plasmids encoding human PS1-WT , PS1 mutations ( PS1-D385A , PS1-M146L , and PS1-ΔE9 ) , and APP mutations ( APPswe , APP-F615P , and APP-M596V ) . cDNAs encoding soluble APP ( sAPPα and sAPPβ ) and C99mutAAA ( RHLSK residues in APP-CTF were mutated to AALSA ) were generated by PCR and were cloned into pMX puro retroviral vector . PCR-amplified regions were verified by sequencing . Human embryonic kidney ( HEK293 ) , and African monkey kidney ( COS-7 ) cells were originally purchased from the ATCC ( Manassas , VA ) . Stable HEK293 cells overexpressing APP695 were provided from Dr . Sangram Sisodia laboratory . Primary mouse embryonic fibroblasts ( MEF ) , HEK293 , and COS-7 cells were cultured in Dulbecco’s modified Eagle’s medium ( DMEM , Invitrogen ) supplemented with 10% fetal bovine serum , 1% glutamine and 1% penicillin-streptomycin ( Gibco ) . To avoid mycoplasma contamination , treatment with ciprofloxacin ( 10 µg/ml ) was added once a month for culture maintenance purposes . Cells were used after less than 3–4 passages . Short tandem repeat ( STR ) profiling was not tested on these well-established cell lines . MEF and HEK293 cells were transiently transfected with Lipofectamine 2000 ( Invitrogen ) or LipoD293 ( SignaGen Laboratories , Rockville , MD ) , respectively , and according to the manufacturer’s protocol . HEK293 cells stably expressing PS1-WT , PS1-D385A , PS1-M146L and PS1-ΔE9 were generated by retroviral infection as described previously ( Deyts et al . , 2012 ) . Briefly , retroviral supernatants collected 48 hr after transfection of Phoenix cells were used to infect HEK293 cells in the presence of 4 µg/ml polybrene . Stably transduced cells were selected in the presence of 1 µg/ml puromycin and pooled for further analysis . Primary mouse cortical neuron cultures were prepared from embryonic E16 mice as previously described ( Parent et al . , 2005 ) and maintained at 37°C in Minimal Essential Medium ( Invitrogen ) supplemented with 1% glutamine , 5% horse serum , 0 . 5% D-glucose , 0 . 15% HCO3 , and nutrients , in a humidified 10% CO2 incubator . Neurons were cultured in 0 . 1% polyethylenimine-coated 18 mm glass coverslip for immunostaining and poly-L-Lysine-precoated 35 mm dishes for Western blot . Transient transfection of 7 DIV neurons was carried out using Lipofectamine 2000 ( Invitrogen ) in Neurobasal medium ( Invitrogen ) . After 3 hr , the transfection medium was replaced by 50% original medium and 50% supplemented Minimal Essential Medium without serum . In some cases , cells were treated with Compound E ( 10 nM ) for 24 hr or MDL-12 , 330A ( acute treatment: 100 nM , 30 min; chronic treatment: 10 nM , 24 hr ) before being lysed for Western blotting or fixed for fluorescence imaging . Primary neurons or HEK293 cells were lysed in buffer containing 150 mM NaCl , 50 mM Tris-HCl , pH 7 . 4 , 0 . 5% NP-40 , 0 . 5% sodium deoxycholate , 5 mM EDTA , 0 . 25% SDS , 0 . 25 mM phenylmethylsulfonyl fluoride and protease inhibitor mixture ( 1:200 , Sigma-Aldrich ) , and briefly sonicated on ice . Mouse brains were lysed in 50 mM NaCl , 25 mM Tris-HCl , pH 7 . 4 , 250 mM sucrose , 1 mM dithiothreitol DTT , 0 . 25 mM phenylmethylsulfonyl fluoride , protease inhibitor mixture and a phosphatase inhibitor mixture containing the following reagents: 2 mM EGTA , 50 mM NaF , 10 mM Na-pyrophosphate , 20 mM β-glycerophosphate , 1 mM para-nitrophenylphosphate and 0 . 1 mM ammonium molybdate . Equal amounts of proteins were resolved on SDS-PAGE gels and transferred to a PVDF Immobilon-FL membrane ( Millipore , distributed by Fisher Scientific ) . Endogenous or overexpressed APP full-length ( APP-FL ) and APP C-terminal fragments ( APP-CTF ) were detected by immunoblotting with CTM1 antibody . APP-CTFα and APP-CTFβ were separated on 16 . 5% Tris-Tricine gels . Identification of APP-CTF species was assessed using revelation of APP-CTF from APP-M596V ( APP β-site cleavage mutant ) , APP-F615P ( APP α-site cleavage mutant ) , and APPswe mutant expressing constructs loaded within the same gels as previously described ( Deyts et al . , 2012; Vetrivel et al . , 2011 ) . Soluble APPs were fractionated on 4–20% SDS gels and revealed using APPNTH-452 polyclonal antibody on Western blots made from conditioned media lysates collected 24 hr after transfection ( Vetrivel et al . , 2009 ) . Endogenous DCC full-length ( DCC-FL ) and DCC endoproteolytic C-terminal fragment ( DCC-CTF ) were detected using G97-449 DCC antibody . PS1 N-terminal fragment ( PS1-NTF ) was detected using PS1NT antiserum ( Deyts et al . , 2012 ) . Detection of GAPDH or Flotilin-2 proteins with selective antibodies was used as loading control . Endogenous phosphorylated CREB was detected using pCREB ( Ser133 ) antibody . To increase accuracy and sensitivity , Western blots were quantified by fluorescence using Odyssey infrared imaging system ( LI-COR Biosciences , Lincoln , NE ) . A fixed size box was drawn surrounded the band of interest and quantified within the same gel . Using Odyssey Infrared Imaging software ( Li-Cor Biosciences ) , the band quantification method feature with average background subtraction from top and bottom was employed to determine the level of integrated fluorescence intensity . Cells were fixed with 4% paraformaldehyde/4% sucrose in phosphate buffer saline ( PBS ) for 30 min at 4°C and permeabilized with 0 . 2% Triton X-100 in PBS for 8 min on ice . After blocking in 10% BSA for 1 hr , neurons were incubated overnight at 4°C with monoclonal MAP2 ( 1:5000 ) diluted in 5% BSA in PBS . Then , after washes in PBS , cells were incubated with monoclonal Alexa-647 or polyclonal Alexa-555 conjugated secondary antibodies ( 1:400 ) for 1 hr at room temperature and mounted using Permafluor mounting solution ( Fisher Scientific ) . To examine pCREB signaling , cells were serum deprived for 3h in HEPES buffer supplemented with tetrodotoxin ( 1 µm ) as previously described ( Deyts et al . , 2012 ) . Labeled neurons were imaged using a motorized Nikon TE 2000 microscope system and Cascade II:512 CCD camera ( Photometrics , Tucson , AZ ) . Using 20X or 60X objective , images were acquired as 200 nm z-stacks and processed using MetaMorph software ( Molecular Devices Corporation , Sunnyvale , CA ) . Total neurite area was evaluated using morphology filter and thresholding features of MetaMorph software to isolate the cell of interest , followed by somatic area removal from YFP expressing cells . The total axonal area was determined by subtraction of total dendritic area ( as seen as MAP2 labeling ) from total neurite area represented as total pixel number ( as seen as YFP fluorescence ) . To quantify the intensity level of pCREB immunofluorescence , raw images were first set to the same threshold level to eliminate nonspecific fluorescence , and then the gray intensity level was determined and divided by the number of pixels per area ( calculated from fluorescence images of YFP transfected cells ) . In order to evaluate fold of changes between conditions , quantification of area was represented as relative value to the control group . All images were acquired with identical parameters and photomultiplicator values as previously described ( Deyts et al . , 2012 ) . Labeled neurons were imaged using a motorized Nikon TE 2000 microscope system and Cascade II:512 CCD camera ( Photometrics , Tucson , AZ ) . Using 20X or 60X objective , images were acquired as 200 nm z-stacks and processed using MetaMorph software ( Molecular Devices Corporation ) . Total neurite area was evaluated using thresholding feature of MetaMorph software to isolate the cell of interest , followed by somatic area removal from YFP expressing cells . The total axonal area was determined by subtraction of total dendritic area ( as seen as MAP2 labeling ) from total neurite area ( as seen as YFP fluorescence ) . To quantify the intensity level of pCREB immunofluorescence , raw images were first set to the same threshold level to eliminate nonspecific fluorescence , and then the gray intensity level was determined and divided by the number of pixels per area ( calculated from fluorescence images of YFP transfected cells ) . All images were acquired with identical parameters and photomultiplicator values as previously described ( Deyts et al . , 2012 ) . Animals were anesthetized with isoflurane and then transcardially perfused with 4% paraformaldehyde/4% sucrose at pH 7 . 4 dissolved in PBS . Brains were extracted and post-fixed 4-6h in the same solution at 4°C , equilibrated in PBS/30% sucrose for 24 hr , embedded into Tissue-Tek O . C . T . compound ( Electron Microscopy Sciences , Hatfield , PA ) and then snap-frozen and stored at −80°C until further processing . Coronal sections ( 40 µm ) were performed using a cryostat ( Leica CM3050 ) and stored in a solution containing 30% ethylene glycol , 30% glycerol , and 0 . 1 M phosphate buffer at −20°C until processing for immunohistochemistry . GAP-43 staining was performed according to the following procedure: brain sections were washed in Tris-buffered saline ( TBS , 3 times , 10 min ) , then blocked for 2 hr in TBS containing 5% horse serum and 0 . 25% Triton X-100 followed by the incubation with monoclonal GAP-43 ( 1:500 ) primary antibody overnight at 4°C . After washes in TBS , sections were then incubated for 2 hr at room temperature with the Alexa-555 secondary antibody ( 1:400 , Invitrogen ) . Sections were then washed again in TBS before mounting with Vectashield mounting medium ( Vector Laboratories , Burlingame , CA ) . For the quantification , several ﬁelds of view per section ( every sixth section ) were acquired using a motorized Nikon TE 2000 microscope using 4X objective . The immunoreactivity was quantified with Metamorph software by measuring the integrated intensity level divided by the number of pixels per thresholded area . The results were expressed as arbitrary units . Confocal 300 nm z-stack images were acquired using Leica SP5 II STED-CW Super-resolution Laser Scanning Confocal using 20X and 63X objectives . Deconvolved z-stack images were generated using Huygens software ( Scientific Volume Imaging , The Netherlands ) and processed using MetaMorph software . Data are presented as mean ± SEM . Statistical significance was determinedby ANOVA Kruskal-Wallis test with independent post hoc Dunn’s multiple comparisonanalysis using GraphPad prism software ( San Diego , CA ) . Each experiment was performed using , at least , three independent sets of cultures and more than five animals . If statistical significance was not reached among group conditions , a post-hoc power analysis was performed to determine needed sample sizes . p<0 . 05 and *p<0 . 001 as compared to empty vector ( EV ) transfected cells or WT , and #p<0 . 05 and ##p<0 . 001 as compared to control group within the same experimental conditions .	One of the hallmarks of Alzheimer’s disease is the accumulation within the brain of sticky deposits called plaques . These plaques form from clumps of molecules called amyloid-beta peptide . An enzyme called gamma-secretase generates the amyloid-beta peptide , by cutting it from a membrane-associated protein called APP . This enzyme consists of multiple subunits , and a mutation in one of these – presenilin-1 – causes a particularly severe form of Alzheimer’s disease . For decades , research into Alzheimer’s disease has focused on the harmful effects of amyloid-beta peptides and plaques . However , Deyts et al . now argue that the protein that gives rise to amyloid-beta peptides has a more direct role in Alzheimer’s disease than previously thought . Specifically , APP may contribute to the harmful effects of the presenilin-1 mutations . By studying genetically modified mice carrying a human presenilin-1 mutation , Deyts et al . show that some of these animals’ nerve cells grow abnormally . Their cell bodies sprout too many branches , while their nerve fibers – which carry electrical signals away from the cell body – become too long . These abnormalities resemble changes seen in the brain in Alzheimer’s disease . Unexpectedly , however , deleting the gene for APP in the presenilin-1 mutant mice prevents the changes from occurring . This suggests that APP must be present for the presenilin-1 mutation to exert this unwanted effect . An increase in APP-driven signaling within cells seems to trigger the observed abnormalities in nerve cells . The presenilin-1 mutation modifies how gamma-secretase cuts APP at the cell membrane to produce amyloid-beta peptides . This frees up the APP to instead interact with signaling cascades inside the cell . Given that gamma-secretase is a key therapeutic target in Alzheimer’s disease , further work is needed to explore the implications of these protein interactions for potential treatments .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Loss of presenilin function is associated with a selective gain of APP function
Cellular actin assembly is controlled at the barbed ends of actin filaments , where capping protein ( CP ) limits polymerization . Twinfilin is a conserved in vivo binding partner of CP , yet the significance of this interaction has remained a mystery . Here , we discover that the C-terminal tail of Twinfilin harbors a CP-interacting ( CPI ) motif , identifying it as a novel CPI-motif protein . Twinfilin and the CPI-motif protein CARMIL have overlapping binding sites on CP . Further , Twinfilin binds competitively with CARMIL to CP , protecting CP from barbed-end displacement by CARMIL . Twinfilin also accelerates dissociation of the CP inhibitor V-1 , restoring CP to an active capping state . Knockdowns of Twinfilin and CP each cause similar defects in cell morphology , and elevated Twinfilin expression rescues defects caused by CARMIL hyperactivity . Together , these observations define Twinfilin as the first ‘pro-capping’ ligand of CP and lead us to propose important revisions to our understanding of the CP regulatory cycle . Assembly of cellular actin structures with distinct architectural and dynamic properties requires the convergence and coordination of numerous actin assembly , stabilization , and disassembly mechanisms . Although our understanding of the functions and mechanisms of individual actin-binding proteins has grown tremendously , there is a need to consider more deeply how seemingly disparate and sometimes competing factors work together in vivo and take on new mechanistic roles within more complex mixtures . One particularly enigmatic example is the interaction of Twinfilin with Capping Protein ( CP ) . These two conserved proteins directly interact with high-affinity , and yet have seemingly opposite effects on the barbed ends of actin filaments . Twinfilin is one of five proteins in the Actin Depolymerization Factor-Homology ( ADF-H ) domain family , of which ADF/Cofilin is the founding member ( Poukkula et al . , 2011 ) . Twinfilin is unique among the members of this family in containing two ADF-H domains , which are joined by a small linker region and followed by a short C-terminal tail . Initial biochemical studies categorized Twinfilin as an actin monomer sequestering factor because of its high affinity for ADP-bound G-actin and ability to inhibit subunit addition to either end of the filament ( Goode et al . , 1998; Vartiainen et al . , 2000; Wahlström et al . , 2001 ) . However , mouse Twinfilin was later shown to interact directly with the barbed ends of actin filaments ( Helfer et al . , 2006; Paavilainen et al . , 2007 ) , and more recently yeast Twinfilin was shown to accelerate depolymerization at actin filament ends ( Johnston et al . , 2015 ) . Alone , yeast Twinfilin enhanced barbed end depolymerization by 3-fold through a processive filament end-attachment mechanism . Further , in conjunction with Srv2/CAP ( cyclase-associated protein ) , yeast Twinfilin increased the rate of pointed-end depolymerization by over 15-fold ( Johnston et al . , 2015 ) . More recently , it was shown that mouse Twinfilin isoforms accelerate barbed end depolymerization , similar to yeast Twinfilin , but do not induce robust pointed end depolymerization in conjunction with Srv2/CAP ( Hilton et al . , 2018 ) . Collectively , these studies highlight the biological significance of Twinfilin . The conserved barbed-end effects of Twinfilin are particularly interesting given that both yeast and mammalian Twinfilins bind to CP ( Falck et al . , 2004; Palmgren et al . , 2001 ) . Further , a barbed-end regulatory role for Twinfilin is suggested by its localization to the tips of stereocilia and filopodia , and to the barbed ends of Drosophila actin bristles ( Peng et al . , 2009; Rzadzinska et al . , 2009; Wahlström et al . , 2001 ) . In addition , Twinfilin localizes to endocytic actin patches in yeast , and to lamellipodia and cell-cell junctions in animal cells ( Goode et al . , 1998; Vartiainen et al . , 2000 ) . Twinfilin’s localization to cortical actin patches in yeast is dependent on its interaction with CP ( Palmgren et al . , 2001 ) . In both yeast and mammals , this interaction is mediated by conserved sequences in the C-terminal tail region of Twinfilin ( Falck et al . , 2004 ) . Despite the high affinity of the Twinfilin-CP interaction ( Kd ~10 nM for the yeast homologs [Poukkula et al . , 2011] ) , studies have revealed no significant effects of Twinfilin on the barbed end capping activity of CP in vitro , and reciprocally , no obvious effect of CP on Twinfilin interactions with ADP-actin monomers ( Falck et al . , 2004 ) . Thus , the functional significance of the Twinfilin-CP interaction has remained highly enigmatic . CP is an obligate heterodimer , consisting of alpha and beta subunits , and binds stably to the barbed ends of actin filaments to block subunit addition and loss . CP is ubiquitous and highly conserved across eukaryotes , and has universal roles in controlling the assembly of actin networks that drive cell morphogenesis and cell motility ( Cooper and Sept , 2008; Hart and Cooper , 1999; Mejillano et al . , 2004; Schafer et al . , 1994; Schafer et al . , 1995 ) . In vitro , CP binds to the barbed ends of actin filaments with sub-nanomolar affinity , and dissociates from barbed ends very slowly ( half-life of ~30 min ) ( Schafer et al . , 1996 ) . Given the relatively high abundance of CP in the cytosol ( 1–3 µM ) and the strength of its interactions with barbed ends ( Cooper and Sept , 2008 ) , it is not surprising that cells have evolved a number of regulatory mechanisms to spatiotemporally restrict CP activity . Cellular protein inhibitors of CP broadly fall into two classes: steric inhibitors and allosteric inhibitors . Steric inhibitors , which include V-1/myotrophin , bind to CP in a manner that physically obstructs its association with barbed ends ( Bhattacharya et al . , 2006; Kim et al . , 2007; Schafer et al . , 1996 ) . V-1 is a highly abundant 13 kDa protein that binds CP with a Kd ~40 nM and sterically blocks its ability to bind barbed ends ( Bhattacharya et al . , 2006; Taoka et al . , 2003 ) . Notably , however , V-1 does not catalyze dissociation of CP from barbed ends ( Bhattacharya et al . , 2006 ) . In contrast , allosteric inhibitors induce conformational changes in CP that catalyze its dissociation from barbed ends ( ‘uncapping’ or ‘displacing’ CP ) , and also decrease but do not abolish its ability to bind barbed ends . The major class of allosteric inhibitors is the capping protein interaction ( CPI ) motif family of proteins ( Edwards et al . , 2014 ) . The founding and best characterized member of the CPI family is CARMIL ( Capping Protein , ARP2/3 and Myosin I linker ) , which is conserved across metazoans ( Stark et al . , 2017 ) . CARMIL catalyzes CP dissociation from barbed ends , reducing CP’s affinity for barbed ends by ~100 fold , transforming it into a transient capper ( Fujiwara et al . , 2014; Stark et al . , 2017; Uruno et al . , 2006; Yang et al . , 2005 ) . CARMIL localizes to the leading-edge plasma membrane , where it promotes cell migration through direct interactions with CP ( Fujiwara et al . , 2014; Liang et al . , 2009; Stark et al . , 2017; Yang et al . , 2005 ) . Other proteins with CPI motifs include CD2AP , CKIP-1 , CapZIP , CIN85 , and WASHCAP ( FAM21 ) ; their roles in regulating CP are less well understood . CPI-motif proteins share a common mode of interaction with CP , but are otherwise unrelated to each other ( Edwards et al . , 2014; Hernandez-Valladares et al . , 2010 ) . To date , binding partners of CP that antagonize its inhibitors , and thus function as ‘pro-capping’ factors , have not been reported . Here , we uncover a novel role for Twinfilin in protecting CP from the negative regulatory effects of V-1 and CARMIL , and thus promoting actin filament capping . These and other data lead us to propose important revisions to current models for the CP regulatory cycle . Because CP binding proteins have been studied predominantly in mammalian systems , we focused our investigation on mouse rather than yeast CP and Twinfilin . Mutagenesis on the yeast Twinfilin tail previously identified a mutant , twf1-11 , that targets a cluster of positively charged residues ( R328A , K329A , R330A , R331A ) necessary for binding CP ( Falck et al . , 2004 ) . While truncations of the C-terminal tail in mouse Twinfilin ( mTwf1 ) also disrupt CP binding , the residues involved have not yet been defined . We therefore first sought to generate a specific mutant in mTwf1 that disrupts the interaction , analogous to yeast twf1-11 . An alignment of the three mouse and three human Twinfilin isoforms , along with the single Twinfilin genes expressed in S . cerevisiae and D . melanogaster ( Figure 1A ) , revealed a region that includes two of the basic residues mutated in the yeast twf1-11 mutant . We mutated these two residues in mTwf1 , changing them to alanines , to produce mTwf1-11 ( K332A , R333A ) . To quantify binding of mTwf1 to CP , we performed fluorescence anisotropy assays using a mTwf1 tail peptide ( 317-350 ) labeled at its N-terminus with HiLyte488 . The mTwf1 tail peptide displayed high affinity , concentration-dependent binding to CPα1β2 , a major non-muscle isoform of CP in mammalian cells ( Figure 1B ) . Moreover , full-length mTwf1 protein ( unlabeled ) competed with the labeled mTwf1 tail for CP binding , whereas full-length mTwf1-11 ( unlabeled ) did not ( Figure 1C ) . Thus , the mTwf1-11 mutant effectively uncouples mTwf1 binding to CP . Using mTwf1-11 , we addressed how CP binding affects Twinfilin’s actin depolymerization activities in total internal reflection fluorescence ( TIRF ) microscopy assays , by directly observing depolymerization at actin filament barbed ends in real time . In agreement with previous observations using yeast and mouse Twinfilin ( Hilton et al . , 2018; Johnston et al . , 2015 ) , 1 µM mTwf1 accelerated barbed end depolymerization by 2–3 fold compared to control reactions ( Figure 1D ) , and the addition of CP blocked this effect ( Figure 1E ) . Further , mTwf1-11 exhibited a similar rate ( Figure 1D ) , indicating that this mutant has wild type depolymerization activity , and thus separates Twinfilin’s ability to bind CP from its ability to promote barbed-end depolymerization . Interestingly , the addition of CP was still able to block barbed-end depolymerization by mTwf1-11 ( Figure 1E ) . These observations suggest that CP sterically blocks mTwf1 access to barbed ends , independent of its direct interaction with mTwf1 . However , this left open the question of whether CP binding to mTwf1 might alter its mechanism of depolymerization independent of blocking the barbed end . To address this possibility , we utilized a CP mutant , CPαΔ28 , which truncates the C-terminal tentacle of the alpha subunit , severely inhibiting capping activity ( Kim et al . , 2010 ) . Importantly , in binding assays the mTwf1 tail interacted equally well with wild-type CP and CPαΔ28 , demonstrating that this mutant binds normally to mTwf1 ( Figure 1B ) . In TIRF assays , equimolar amounts of CPαΔ28 did not significantly alter mTwf1 depolymerization activity ( Figure 1F; Figure 1—video 1; also summarized in Figure 1G ) , suggesting that while CP blocks Twinfilin access to barbed ends , Twinfilin-CP direct interaction does not alter Twinfilin depolymerization activity . Given that CP binding does not affect Twinfilin’s depolymerization activity , or other known activities of Twinfilin ( Falck et al . , 2004; Johnston et al . , 2015; Palmgren et al . , 2001 ) , we next considered whether Twinfilin binding might influence CP functions in the presence of known regulators of CP . We were particularly interested in how Twinfilin might impact the regulation of CP by CPI-motif proteins such as CARMIL , since we noticed that the C-terminal tail regions of evolutionarily diverse Twinfilins share sequence homology with the CPI motifs of several CPI family proteins ( Figure 2A ) . The consensus CPI motif is 17-amino acids long , with some additional contacts contributed from outside this motif , and tolerates significant divergence across the CPI-motif family ( Edwards et al . , 2014; Hernandez-Valladares et al . , 2010 ) . As an initial test , we used a mutant of CP , CP ( RY ) , which alters two surface residues on the beta subunit ( R15A , Y79A ) that make essential contacts with CPI-motif proteins ( Edwards et al . , 2014; Hernandez-Valladares et al . , 2010 ) . The CP ( RY ) mutant is insensitive to inhibition and uncapping by CARMIL and disrupts binding with at least two other CPI-motif proteins , CD2AP and WASHCAP ( FAM21 ) ( Edwards et al . , 2015 ) . In fluorescence anisotropy binding assays , we observed that the CP ( RY ) mutant has approximately 20-fold reduced affinity for mTwf1 tail compared to wild type CP ( Figure 2B ) . These data are consistent with mTwf1 and CPI-motif proteins sharing at least partially overlapping binding sites on CP . In addition , we asked whether introducing a mutation in the mTwf1 tail peptide at a conserved residue in CPI consensus sequences would alter binding to CP ( Lys 325 in mTwf1; see red asterisk , sequence alignment in Figure 2A ) . In fluorescence anisotropy binding assays , we compared the abilities of wild-type and mutant ( K325A ) mTwf1 tail peptides to compete with labeled mTwf1 tail peptide for CP binding . This analysis revealed an ~30 fold reduction in binding affinity for the mutant ( K325A ) mTwf1 tail peptide compared to wild type peptide ( Figure 2C ) . We next asked whether the CP-binding region ( CBR ) of CARMIL1 ( residues 964 – 1078 ) competes with mTwf1 tail for binding to CP . We observed that unlabeled CBR peptide competed with the fluorescent mTwf1 tail probe for CP binding ( Figure 2D ) . These results indicate that CARMIL and mTwf1 directly compete for binding CP . Next , we more narrowly defined the region of CARMIL that competes with mTwf1 by using peptides that divide the CBR into its two conserved components , the CPI motif ( 969 – 1005 ) and the CARMIL-specific interaction ( CSI ) motif ( 1019 – 1037 ) . The CSI makes additional contacts with CP , but is found only in CARMIL family members , and not in other CPI-motif proteins ( Edwards et al . , 2014 ) . As expected based on Twinfilin’s sequence similarity to CPI motifs , only the CPI-motif peptide and not the CSI peptide competed with mTwf1 tail for CP binding ( Figure 2D ) . Together , these results suggest that Twinfilin is a divergent CPI-motif protein and has important implications for CP regulation in cells ( see Discussion ) . Given that CARMIL and Twinfilin compete for binding to CP , we asked whether mTwf1 affects CARMIL’s ability to displace CP from barbed ends . We addressed this question in pyrene actin assembly assays , where actin polymerization was initiated at time zero in the presence of CP and increasing concentrations of mTwf1 , and after 400 s CARMIL1 CBR was spiked into the reaction . CARMIL1 alone ( no mTwf1 ) strongly induced uncapping , leading to the rapid polymerization of previously-capped filament seeds ( Figure 3A ) . However , increasing concentrations of mTwf1 attenuated CARMIL’s uncapping effects ( Figure 3A ) . These results are consistent with mTwf1 competing with CARMIL for binding CP , and thereby blocking uncapping . To more directly observe mTwf1 effects on CARMIL-induced uncapping of barbed ends , we used TIRF microscopy . In these experiments , we used fluorescently labeled SNAP-tagged CP ( SNAP-649-CP; 100% labeled ) to monitor lifetimes of CP molecules on filament barbed ends ( Bombardier et al . , 2015 ) . Filaments were first polymerized to a desired length ( ~10 µm ) and then capped by flowing in SNAP-649-CP . Free CP was washed out , and then proteins of interest ( or control buffer ) were flowed in . Capped filaments were identified in the field of view prior to flow-in , and then monitored after flow-in to measure the dwell time of SNAP-649-CP . As expected , in the absence of other factors , SNAP-649-CP had a long dwell time , remaining on barbed ends for tens of minutes ( Figure 3B and C ) . However , when CARMIL1 CBR was introduced , this led to the rapid displacement of SNAP-649-CP , with complete loss of CP from barbed ends by 100 s ( Figure 3B and C ) . The addition of mTwf1 with CARMIL1 CBR attenuated the uncapping effects in a concentration-dependent manner ( Figure 3B and C ) . Further , this attenuation required direct interactions between Twinfilin and CP , as mTwf1-11 failed to protect CP from CARMIL uncapping ( Figure 3D and E and Figure 3—figure supplement 1 ) . Similar effects were observed for the other major isoform of mouse Twinfilin that is expressed in non-muscle cells , mTwf2a ( Figure 3—figure supplement 1 ) ( Nevalainen et al . , 2011; Vartiainen et al . , 2003 ) . We next considered whether Twinfilin binding to CP might affect the activities of CP inhibitor V-1/myotrophin , which is distinct from CPI-motif proteins in its mode of CP interaction . Unlike CARMIL , V-1 does not displace CP from barbed ends; instead , it sequesters CP and blocks it from binding filament ends ( Bhattacharya et al . , 2006; Jung et al . , 2016; Taoka et al . , 2003 ) . In contrast to the CARMIL binding site on CP , which partially encircles the ‘stalk’ of the CP heterodimer ( Hernandez-Valladares et al . , 2010; Johnson et al . , 2018; Zwolak et al . , 2010 ) , V-1 interacts with CP on the opposite face , sterically blocking binding to the filament end ( Johnson et al . , 2018; Takeda et al . , 2010; Zwolak et al . , 2010 ) . To test how Twinfilin might affect the interaction of CP with V-1 , we used pyrene-actin seeded elongation assays ( Figure 4A ) . As expected , filament seeds pre-incubated with CP and then mixed with pyrene-actin monomers displayed minimal growth , whereas the addition of V-1 restored actin assembly to uncapped levels . Somewhat to our surprise , the further addition of mTwf1 suppressed V-1’s effects , restoring capping activity , while mTwf1-11 had no effect ( Figure 4A and B ) . These effects were unexpected given the above-mentioned differences in Twinfilin’s predicted and V-1’s known binding sites on CP , and our observation that even high concentrations of V-1 ( 1000-fold excess to mTwf1 tail probe ) fail to compete with mTwf1 for CP binding in anisotropy assays ( Figure 4C ) . These results suggest that mTwf1 attenuates V-1 effects on CP via an allosteric mechanism , distinct from a simple steric binding competition . In probing the mechanism further , we drew inspiration from a study by Fujiwara and colleagues , showing that CARMIL forms a transient ternary complex with V-1 and CP , leading to accelerated dissociation of V-1 from CP ( Fujiwara et al . , 2014 ) . We asked whether mTwf1 might similarly catalyze the dissociation of V-1 from CP . In stopped-flow fluorescence assays , fluorescently labeled V-1 ( TAMRA-V-1 ) was first allowed to bind CP , and then mixed at time zero with an excess of unlabeled V-1 . The resulting decrease in fluorescence reflects the spontaneous dissociation of TAMRA-V-1 from CP ( Figure 4D ) . The rate of V-1 dissociation from CP increased in the presence of increasing concentrations of mTwf1 , pointing to the possible formation of a transient ternary complex that destabilizes V-1 interactions with CP ( Figure 4D and E ) . Importantly , mTwf1-11 failed to enhance V-1 dissociation ( Figure 4E ) , showing that this effect depends on direct interactions between mTwf1 tail and CP . These results demonstrate that CARMIL and Twinfilin share a common function in catalyzing the dissociation of V-1 from CP using their CPI motifs to bind CP , despite having different effects on the displacement of CP from barbed ends . Given the observed competition between mTwf1 tail peptide and the CPI motif of CARMIL for binding to CP , and the similarity between mTwf1 and CARMIL in catalyzing V-1 dissociation from CP , we sought structural evidence for the nature of the interaction between mTwf1 and CP . We hypothesized that the binding sites for mTwf1 and the CPI motif were likely overlapping . To test this hypothesis , we used hydrogen-deuterium exchange with mass spectrometry ( HDX-MS ) to interrogate the conformational dynamics and solvent accessibility of the backbone and sidechains of CP , free and in complex with Twf1 . Further , we compared our results to those in our recent study on the interactions of CARMIL with CP using the same approach ( Johnson et al . , 2018 ) . We tested three different forms of mTwf1: a short tail peptide ( residues 317–350 ) , a longer tail peptide ( residues 305–350 ) , and full-length mTwf1 . These constructs were added to CP , either full-length alpha/beta heterodimer , or full-length alpha subunit with a beta subunit truncated at its C-terminus , removing the actin-binding beta tentacle . The results were essentially the same in each case . The presence of mTwf1 resulted in protection from H-D exchange at the N-terminal stalk of CP ( Figure 5A , Figure 5—figure supplements 1 and 2 ) . Similar effects to H-D exchange were observed upon CARMIL binding to CP ( Johnson et al . , 2018 ) ; also shown here in Figure 5B ) , which correspond well with the CPI-motif binding site defined by X-ray crystallography and solution NMR studies ( Hernandez-Valladares et al . , 2010; Takeda et al . , 2010; Zwolak et al . , 2010 ) . For mTwf1 , we also observed H-D exchange protection of a small region on CP corresponding to the V-1 binding site ( Figure 5A and B , Figure 5—figure supplements 1 and 2 ) , consistent with our results described above for the effects of mTwf1 in promoting V-1 dissociation from CP . These structural effects are also consistent with our previous results for CARMIL , which alters the V-1 binding site ( Johnson et al . , 2018 ) . However , it is worth noting that mTwf1-induced changes in CP conformation at the actin-binding interface were not as extensive as those induced by CARMIL , which is consistent with CARMIL , but not mTwf1 , weakening CP binding to actin at the barbed ends . To investigate the functional relationship between Twinfilin and CP in cells , we started by asking whether mTwf1 and CP colocalize . While Twinfilin and CP have been localized individually , and are each reported to be enriched at the tips of filopodia and stereocilia , endocytic actin patches , lamellipodia , and Drosophila bristles ( Avenarius et al . , 2017; Falck et al . , 2004; Goode et al . , 1998; Nevalainen et al . , 2011; Peng et al . , 2009; Rzadzinska et al . , 2009; Sinnar et al . , 2014; Vartiainen et al . , 2000 ) , to our knowledge they have never been co-imaged in vertebrate cells . To address this , we performed immunofluorescence on CP and Twinfilin in mouse B16F10 melanoma cells , co-staining the cells with Alexa 568-phalloidin to visualize F-actin . We observed strong colocalization of Twinfilin and CP throughout the cell and a co-enrichment at the actin-rich leading and trailing edges ( Figure 6A and B ) . Further , quantitative western blotting showed that Twinfilin and CP are present at ~1:2 molar ratio in B16F10 cells ( Figure 6C , Figure 6—figure supplement 1 ) . Previous studies reported the concentration of CP in B16F10 cells to be ~1 µM ( Fujiwara et al . , 2014; Pollard and Borisy , 2003 ) , suggesting that mTwf1 is present at ~0 . 5 µM . Given the high affinity of the Twinfilin-CP interaction ( Kd = 50 nM ) , these observations are consistent with mTwf1 being associated with a substantial fraction of the CP in cells . The ability of Twinfilin to function as a ‘pro-capping’ factor in vitro , by antagonizing the inhibitory effects of V-1 on CP , predicted that genetic loss of mTwf1 might at least partially phenocopy loss of CP . While a number of studies have examined how Twinfilin mutations affect whole animal development and physiology ( Iwasa and Mullins , 2007; Meacham et al . , 2009; Nevalainen et al . , 2011; Wahlström et al . , 2001; Wang et al . , 2010; Yamada et al . , 2007 ) , we are unaware of any studies that have investigated how loss of Twf1 affects the morphology and actin organization of cultured mammalian cells . Using RNAi silencing in B16F10 cells , we separately depleted endogenous mTwf1 and CP , which was verified by both western blotting ( Figure 6E and F ) and immunostaining ( Figure 6—figure supplement 1 ) . Knockdown of either mTwf1 or CP led to a similar , marked increase in the density of peripheral protrusions or microspikes with a concomitant loss of lamellipodial surfaces ( Figure 6F and G ) . Similar phenotypes have been reported for CP depletion in multiple cell lines ( Edwards et al . , 2013; Edwards et al . , 2015; Mejillano et al . , 2004; Sinnar et al . , 2014 ) . Expression of an RNAi-refractive mTwf1 construct , but not mTwf1-11 , rescued the defects caused by depletion of endogenous mTwf1 ( Figure 6F and G; Figure 6—figure supplement 1 ) , demonstrating that these cellular functions of mTwf1 critically depend on its interaction with CP . We also made the unexpected observation that knockdown of CP was accompanied by a dramatic reduction in Twinfilin levels in cells , as seen by both western blotting ( Figure 6D ) and immunofluorescence ( Figure 6—figure supplement 1 ) . This effect was confirmed using a second RNAi oligonucleotide that targets a different region of CP ( siCP2 , Figure 6D ) . Further , it was observed in additional cell lines besides B16F10 , including Neuro-2A and NIH-3T3 cells ( Figure 6—figure supplement 1 ) . These observations support the closely intertwined relationship of CP and Twinfilin in vivo . Our results above also call into question whether the full extent of the phenotype caused by knockdown of CP ( Figure 6G ) is due to loss of CP , or instead is partly due to the accompanying loss of Twinfilin . To address this , we restored mTwf1 levels in cells depleted of CP by driving mTwf1 expression from a rescue plasmid , which was confirmed by western blotting and immunofluorescence ( Figure 6—figure supplement 1 ) . Forced expression of mTwf1 partially rescued the defects associated with CP depletion , indicating that a portion of the original defects observed after CP knockdown were likely due to the accompanying loss of mTwf1 . These observations also suggest that many previously reported phenotypes arising from CP knockouts and knockdowns should be revisited or reinterpreted with the potential loss of Twinfilin in mind . Finally , we tested the prediction of our biochemical observations that loss of capping activity in cells caused by overexpressed CARMIL1 should be restored by co-overexpression of Twf1 . B16F10 cells ectopically expressing CARMIL1 showed morphological defects similar to loss of CP , and ectopic mTwf1 expression rescued the defects ( Figure 7A and B ) . Importantly , ectopic expression of mTwf1 alone caused no significant change in cell morphology . These results support our biochemical observations , and suggest that Twf1 promotes capping in vivo , at least in part by competing with CARMIL for CP binding and antagonizing the uncapping effects of CARMIL . Twinfilin and CP have been inextricably linked as interacting partners in yeast and animal cells for over 15 years ( Palmgren et al . , 2001 ) , yet until now it has remained a mystery what function their interaction serves . Here we discovered that Twinfilin binds to CP using an orphan CPI-like sequence in its C-terminal tail region , and through this interaction protects CP from inhibition and/or barbed end displacement by CARMIL and V-1 . We found that Twinfilin binds to CP in a competitive manner with the CPI motif of CARMIL , interacts with a site on CP similar to that of CARMIL , and attenuates CARMIL-mediated uncapping of actin filaments . Separately , Twinfilin binding to CP also accelerates V-1 dissociation from CP , despite Twinfilin and V-1 having non-overlapping binding sites on CP . This might be achieved by an allosteric mechanism , given that CARMIL uses its CPI motif to induce V-1 dissociation from CP through allosteric changes ( Fujiwara et al . , 2014; Johnson et al . , 2018 ) . Thus , we have demonstrated that Twinfilin promotes capping by protecting CP from interactions with V-1 and CARMIL . This functional role for Twinfilin is further supported in vivo by our observations of: ( i ) strong colocalization of Twinfilin and CP , ( ii ) knockdowns of mTwf1 and CP that each give rise to similar defects in cell morphology , and ( iii ) over-expression of mTwf1 suppressing defects caused by CARMIL hyperactivity . Taken together , these results reveal that Twinfilin is a new member of the CPI-motif family of proteins , and the first within this group to show the ability to bind CP without reducing CP affinity for barbed ends , and antagonize the negative regulatory effects of another CPI protein . These functions of Twinfilin provide important new insights into the CP regulatory cycle . The best working model to date has been the Fujiwara model ( Fujiwara et al . , 2014 ) ( depicted here as ‘Earlier Model’; Figure 7C ) . It posits that the majority of CP in the cytosol is bound to V-1 , in an inactive state , which then can be locally ‘activated’ by CARMIL at the leading edge . However , a caveat to this model is that it suggests CP-CARMIL complexes are the dominant capping species in the cell , despite this complex having ~100 fold reduced affinity for barbed ends compared to free CP . While this could potentially explain dynamic capping and uncapping near the plasma membrane , consistent with GFP-CP single molecule speckle analysis ( Miyoshi et al . , 2006 ) , it does not explain how cells maintain a pool of ‘capping competent’ CP further back from the leading edge , where CP is needed to cap barbed ends in stress fibers and other actin networks , and may cap barbed ends generated by severing to promote filament disassembly . This model goes on to suggest that an unknown factor or mechanism dissociates the CP-CARMIL complex , allowing V-1 to rebind CP , restoring it to an inactive state . In light of our results , we propose several additions and revisions to the Fujiwara model ( see ‘Revised Model’; Figure 7D ) . First , we suggest that Twinfilin’s protective effects on CP , in particular against V-1 , allow cells to maintain a larger pool of fully active CP ( Twinfilin-CP complexes ) in the cytosol than was previously thought . This view is supported by the relatively high abundance of Twinfilin in cells ( ~0 . 5 µM , compared to ~1 µM CP ) its high affinity for CP ( Kd = 50 nM ) , and its ability to increase the rate of dissociation of V-1 from CP . Given these observations , we propose that a substantial fraction of CP is available in a fully active state , as Twinfilin-CP complexes , even in the presence of a high concentration of V-1 in the cytosol ( ~3 µM ) ( Fujiwara et al . , 2014; Pollard and Borisy , 2003 ) . Second , we propose that V-1 functions to maintain a cytosolic reservoir of inactive CP , mobilized by Twinfilin and/or CARMIL dissociating V-1 to generate ‘stable capping’ ( Twinfilin-CP ) in the cytosol and possibly ‘transient capping’ ( CARMIL-CP ) complexes at the plasma membrane , respectively . CARMIL-CP complexes at the plasma membrane could facilitate actin network growth to drive leading edge protrusion . In contrast , Twinfilin-CP complexes in the cytosol may facilitate stable capping of barbed ends to limit network growth and promote filament disassembly and turnover . Third , we propose that the association of Twinfilin-CP complexes with barbed ends primes filaments for disassembly . Our data show that CARMIL , and/or other CPI proteins , compete with Twinfilin for binding CP . These interactions may competitively remove CP from barbed ends , leaving Twinfilin at the barbed end to processively depolymerize filaments , either alone or in combination with Srv2/CAP ( as depicted in Figure 6D ) ( Hilton et al . , 2018; Johnston et al . , 2015 ) . In this manner , the interaction of Twinfilin with CP could serve not only to initially promote capping , and thus limit network growth , but also to position Twinfilin at barbed ends for subsequently catalyzing the disassembly of filaments . In summary , our results show that functions of mammalian Twinfilin and CP are closely intertwined . This functional relationship is likely to extend to other species given CPI motif sequence conservation in the Twinfilin tail region ( Figure 2A ) and the conserved nature of the Twinfilin-CP interaction . Indeed , S . cerevisiae Aim21 was recently identified as the first yeast CPI motif-containing protein , and was shown to regulate CP function at cortical actin patches ( Farrell et al . , 2017; Shin et al . , 2018 ) . We generated a structural model to explore the possible ternary complex formed by Twinfilin , CP , and the barbed end of an actin filament ( Figure 7—figure supplement 1 ) . In this model , the Twinfilin tail is long enough to allow for simultaneous binding of Twinfilin’s CPI motif to CP and Twinfilin’s C-terminal ADFH domain to an actin subunit at the barbed end . Further , there are no clashes in binding between CP and Twinfilin on actin . It is worth noting that CP and Twinfilin appear to be able to associate with barbed ends individually or as a CP-Twinfilin complex , but with distinct consequences for the function and dynamics of actin networks . CP alone stably caps barbed ends , blocking subunit addition or loss , and our results suggest that CP-Twinfilin complexes may do the same . However , when Twinfilin alone associates with barbed ends , it drives processive depolymerization , while blocking new assembly ( Hilton et al . , 2018; Johnston et al . , 2015 ) . Thus , despite key differences in the nature of their associations with barbed ends , CP and Twinfilin each inhibit filament growth , likely explaining why Twinfilin can replace CP in reconstituted actin motility assays in vitro ( Helfer et al . , 2006 ) . Finally , our data add to a broader emerging view that actin dynamics in vivo are controlled by a complex set of barbed end-associated factors , many of which interact with each other and/or stimulate each other’s dissociation from barbed ends . These multi-component mechanisms may allow cells to control rapid transitions at filament ends through different functional states , including ( i ) formin-bound elongation , ( ii ) paused growth by formin-CP ‘decision complexes’ , ( iii ) stable or transiently capped states by CP alone or CP-Twinfilin complexes ( Bombardier et al . , 2015; Shekhar et al . , 2015 ) , and ( iv ) depolymerization by Twinfilin , Cofilin , and/or Srv2/CAP . These molecular mechanisms for regulating barbed end growth are vastly more elaborate and dynamic than once thought , and help explain the exquisite spatiotemporal control that cells have in tuning actin network dynamics . Plasmids used for expressing the following proteins were previously described: chicken CPα1β1 ( Soeno et al . , 1998 ) , chicken SNAP- CPα1β1 ( Bombardier et al . , 2015 ) , mouse CPα1β2 ( Kim et al . , 2012 ) , mouse CPα1Δ28 ( Kim et al . , 2012 ) , mouse CP α1β2 R15A/Y79A ( Edwards et al . , 2015 ) , human CARMIL1 CBR115 ( 964–1078 ) ( Kim et al . , 2012 ) , human V-1 ( Edwards et al . , 2015 ) . The plasmid for over-expressing CARMIL1 in mammalian cells has been described ( Edwards et al . , 2013 ) . To generate plasmids for expressing mouse Twinfilin isoforms as glutathione-S-transferase ( GST ) -fusions in E . coli , ORFs were PCR amplified from pHAT2-mTwf1 and pHAT2-mTwf2a kindly provided by Pekka Lappalainen ( Univ . of Helsinki ) ( Nevalainen et al . , 2009 ) , and subcloned into the EcoRI and NotI sites of pGEX-6p-1 , yielding pGEX-6p-1-mTwf1 and pGEX-6p-1-mTwf2a . pGEX-6p-1-mTwf1-11 ( K332A , R333A ) was generated by site-directed mutagenesis of pGEX-6p-1-mTwf1 . For V-1 fluorescence experiments , we used a previously demonstrated strategy of removing two surface cysteine residues to allow direct labeling on the single remaining cysteine ( Fujiwara et al . , 2014 ) ; this was achieved by performing site-directed mutagenesis on wild type pGEX-GST-V-1 plasmid to introduce two mutations ( C45S , C83S ) . To generate an RNAi-refractive construct of mTwf1 for expression in cultured cells , the ORF of mTwf1 was PCR amplified from pGEX-6p-1 and subcloned into the HindIII and SacI sites of pEGFP-C1 ( Clontech , Mountain View , CA ) . Then , site-directed mutagenesis was used to introduce silent mutations at specific nucleotides of the ORF ( 703 , 709 , 711 , 715 ) , and the RNAi-refractive ORF was subcloned into the EcoRI and NotI sites of pCMV-M1 , a gift from Linda Wordeman ( Stumpff et al . , 2008 ) ( Addgene plasmid # 23007 ) , yielding pCMV-myc-mTwf1 . Site-directed mutagenesis was performed on pCMV-myc-mTwf1 to generate mutant pCMV-myc-mTwf1-11 ( K332A , R333A ) . All constructs were verified by DNA sequencing . Rabbit skeletal muscle actin ( RMA ) ( Spudich and Watt , 1971 ) , was purified from acetone powder generated from frozen ground hind leg muscle tissue of young rabbits ( PelFreez , Rogers , AR ) . Lyophilized acetone powder stored at −80°C was mechanically sheared in a coffee grinder , resuspended in G-buffer ( 5 mM Tris-HCl pH 7 . 5 , 0 . 5 mM DTT , 0 . 2 mM ATP , 0 . 1 mM CaCl2 ) , and then cleared by centrifugation for 20 min at 50 , 000 × g . Actin was polymerized by the addition of 2 mM MgCl2 and 50 mM NaCl and incubated overnight at 4°C . F-actin was pelleted by centrifugation for 150 min at 361 , 000 × g , and the pellet solubilized by Dounce homogenization and dialyzed against G-buffer for 48 hr at 4°C . Monomeric actin was then precleared at 435 , 000 × g , and loaded onto a S200 ( 16/60 ) gel filtration column ( GE healthcare , Marlborough , MA ) equilibrated in G-Buffer . Peak fractions containing actin were stored at 4°C . For labeling actin with biotin ( Breitsprecher et al . , 2012 ) or Oregon Green ( OG ) ( Kuhn and Pollard , 2005 ) , the F-actin pellet described above was Dounced and dialyzed against G-buffer lacking DTT . Monomeric actin was then polymerized by adding an equal volume of 2X labeling buffer ( 50 mM Imidazole pH 7 . 5 , 200 mM KCl , 0 . 3 mM ATP , 4 mM MgCl2 ) . After 5 min , the actin was mixed with a 5-fold molar excess of NHS-XX-Biotin ( Merck KGaA , Darmstadt , Germany ) or Oregon-Green-488 iodoacetamide ( Invitrogen , Carlsbad , CA ) resuspended in anhydrous DMF , and incubated in the dark for 15 hr at 4°C . Labeled F-actin was pelleted as above , and the pellet was rinsed briefly with G-buffer , then depolymerized by Dounce homogenization , and dialyzed against G-buffer for 48 hr at 4°C . Labeled , monomeric actin was purified further on an S200 ( 16/60 ) gel filtration column as above . Aliquots of biotin-conjugated actin were snap frozen in liquid nitrogen and stored at −80°C . OG-488-actin was dialyzed for 15 hr against G-buffer with 50% glycerol and stored at −20°C . For bulk actin assembly assays , RMA was fluorescently labeled with pyrenyl-iodoacetamide on cysteine 374 ( Pollard and Cooper , 1984; Graziano et al . , 2013 ) . An RMA pellet stored at 4°C ( prepared as described above ) was dialyzed against pyrene buffer ( 25 mM Tris-HCl , pH 7 . 5 , 100 mM KCl , 0 . 02% NaN3 , 0 . 3 mM ATP , and 2 mM MgSO4 ) for 3–4 hr and then diluted with pyrene buffer to 1 mg/ml ( 23 . 8 μM ) . A sevenfold molar excess of pyrenyl-iodoacetamide was added , the actin solution was incubated overnight at 4°C , and aggregates were cleared by low-speed centrifugation . The supernatant ( containing F-actin ) was centrifuged for 3 hr at 4°C at 45 , 000 rpm in a Ti70 rotor ( Beckman Coulter , Indianapolis , IN ) to pellet F-actin . The actin pellets were disrupted by Douncing , dialyzed against G-buffer for 1–2 d , and gel filtered on a 16/60 S200 column . Peak fractions were pooled , aliquoted , snap frozen , and stored at −80°C . Mouse non-muscle CPα1β2 was purified as described ( Graziano et al . , 2014 ) . Briefly , the expression vector ( Soeno et al . , 1998 ) was transformed into E . coli strain BL21 pLysS . Cells were grown in LB to log phase , then expression was induced for 3 hr at 37°C by addition of 0 . 4 mM isopropyl-β-D-thiogalactopyranoside ( IPTG ) . Cells were collected by centrifugation , washed with 25 ml water , and resuspended in lysis buffer ( 20 mM Tris pH 8 . 0 , 1 mM EDTA , 0 . 1% Triton X-100 , protease inhibitors ) and lysed by lysozyme treatment and sonication . The cell lysate was clarified by centrifugation at 12 , 500 x g for 30 min at 4°C . Supernatants were loaded onto a 1 ml Q-HiTrap column ( GE Healthcare ) and eluted with a 45 ml salt gradient ( 0–500 mM KCl ) in 20 mM Tris , pH 8 . 0 . Peak fractions were pooled , concentrated using a centrifugal filter ( Centiprep , MWCO 10 kDa; Millipore ) to 3 ml , and loaded onto a 26/60 Superdex 75 gel filtration column ( GE Healthcare ) equilibrated in 50 mM KCl , 20 mM Tris , pH 8 . 0 . Peak fractions were pooled and loaded onto a 5 ml Mono Q column ( GE Healthcare ) and eluted with a 30 ml salt gradient ( 0–500 mM KCl ) in 20 mM Tris , pH 8 . 0 . Peak fractions were pooled , dialyzed overnight at 4°C into HEK buffer ( 20 mM HEPES , pH 7 . 4 , 1 mM EDTA , 50 mM KCl ) , aliquoted , snap-frozen in liquid N2 , and stored at −80°C . SNAP-649-CP ( CPα1β1 ) was purified and labeled as described ( Bombardier et al . , 2015 ) . SNAP-CP was expressed E . coli strain BL21 pLysS . Cells were grown to log phase at 37°C , and then expression was induced for 8 hr at 37°C by addition of 0 . 4 mM isopropyl-β-D-thiogalactopyranoside ( IPTG ) . Cells were collected by centrifugation , and resuspended in 20 mM Tris pH 8 . 0 , 1 mM EDTA , 0 . 1% Triton X-100 , protease inhibitors and lysed by lysozyme treatment and sonication . The cell lysate was centrifuged for 80 min at 60 , 000 rpm , 4°C in a Ti70 rotor ( Beckman/Coulter , Fullerton , CA ) . The supernatant was rotated with 0 . 75 ml of Ni2+-NTA-agarose beads ( Qiagen , Valencia , CA ) . SNAP-CP was fluorescently labelled using 9 μM ( ∼4-fold excess ) dye adduct for 2 hr at room temperature , yielding SNAP-649-CP . To remove free dye , beads were washed three times with 20 mM imidazole ( pH 8 . 0 ) , 1X PBS , 1 mM DTT , 200 mM NaCl . Labeled SNAP-649-CP was eluted with 0 . 5 ml of 300 mM imidazole pH 8 . 0 , 50 mM Tris pH 8 . 0 , 100 mM NaCl , 1 mM DTT , 5% glycerol , then purified by gel filtration on a Superose six column ( GE Healthcare ) equilibrated in 20 mM Hepes ( pH 7 . 5 ) , 1 mM EDTA , 150 mM KCl , 5% glycerol . Peak fractions were pooled , concentrated , aliquoted , snap-frozen in liquid N2 , and stored at −80°C . For stopped-flow kinetics , fluorescence anisotropy binding and HDX-MS experiments , His-tagged-α1 and β2 subunits of mouse CP ( pRSFDuet-1 , pBJ 2041 ) were co-expressed in E . coli BL21 ( DE3 ) pRIL and purified as described ( Johnson et al . , 2018 ) . For CP lacking the β tentacle , a premature stop codon was introduced , so that the C-terminal residue of the mouse β2 subunit was L243 instead of C272 ( pBJ 1891 ) . Twinfilin polypeptides were expressed as GST-fusions in E . coli strain BL21 pRARE . Cells were grown to log phase at 37°C , and then expression was induced for 16 hr at 18°C by addition of 0 . 4 mM isopropyl-β-D-thiogalactopyranoside ( IPTG ) . Cells were collected by centrifugation , washed with 25 ml water , and resuspended in 10 ml of PBS supplemented freshly with 0 . 5 mM dithiothreitol ( DTT ) , 1 mM phenylmethylsulphonyl fluoride ( PMSF ) , and a standard mixture of protease inhibitors . Cells were incubated with lysozyme ( 0 . 5 mg ml−1 ) on ice for 15 min and then sonicated . The cell lysate was clarified by centrifugation at 12 , 500 g for 20 min and incubated at 4°C ( rotating ) for at least 2 hr with 0 . 5 ml glutathione–agarose beads ( Sigma-Aldrich; St . Louis , MO ) . Beads were washed three times in PBS supplemented with 1M NaCl and then washed two times in PBS . Twinfilin was cleaved from GST by incubation with PreScission Protease ( GE Healthcare; Marlborough , MA ) overnight at 4°C ( rotating ) . Beads were pelleted , and the supernatant was concentrated to 0 . 3 ml , and then purified further by size-exclusion chromatography on a Superose12 column ( GE Healthcare ) equilibrated in HEK buffer ( 20 mM Hepes pH 7 . 5 , 1 mM EDTA , 50 mM KCl , 0 . 5 mM DTT ) . Peak fractions were pooled , concentrated , aliquoted , snap-frozen in liquid N2 , and stored at −80°C . CARMIL CBR115 and V-1 were purified from E . coli as above for mTwf1 proteins , except the GST tag was removed from V-1 by digestion with thrombin instead of PreScission protease . To purify and label V-1 ( generating TAMRA-V-1 ) for fluorescence experiments , BL21 E . coli expressing pGEX-GST-V-1 ( C45S , C83S ) was lysed in a Microfluidizer ( Microfluidics Corp . ; Westwood , MA ) . Fusion protein was isolated on Glutathione Superflow Agarose ( Thermo Fisher Scientific; Waltham , MA ) . The GST tag was cleaved by digestion with bovine thrombin ( MP Biomedicals; Santa Ana , CA ) overnight at 4°C , then separated from V-1 on a Sephacryl S-200 HR 16/60 column ( GE Healthcare ) equilibrated in 25 mM HEPES pH 7 . 0 , 1 mM TCEP , 100 mM KCl , 1 mM NaN3 . Residual GST was removed by re-incubating peak fractions with Glutathione Superflow Agarose . Purified V-1 ( C45S , C83S ) was then labeled with tetramethylrhodamine ( TAMRA ) −5-maleimide ( Invitrogen ) overnight at 4°C . Excess TAMRA was removed by dialysis against 20 mM 3- ( N-morpholino ) propanesulfonic acid ( MOPS ) pH 7 . 2 , 1 . 0 mM TCEP , 100 mM KCl , 1 mM NaN3 . TAMRA-V-1 was stored at −70°C The mTwf1 tail peptides used for anisotropy were sourced as follows: N-terminal HiLyte488 labeled mTwf1 ( H317-D350 ) was purchased from Anaspec ( Fremont , CA ) ; unlabeled CARMIL1 CPI ( G969-A1005 ) , CARMIL1 CSI ( M1019-M1037 ) , mTwf1 ( A305-D350 ) and mTwf1 ( A305-D350 , K325A ) , as well as N-terminal TAMRA labeled mTwf1 ( A305-D350 ) , were purchased from WatsonBio Sciences ( Houston , TX ) . Pyrene actin assembly assays were performed as previously described ( Chesarone-Cataldo et al . , 2011 ) , with slight modifications for monitoring uncapping . Reactions containing 2 μM G-actin ( 5% pyrene labeled ) , 25 nM CapZ , and variable concentrations of mTwf1 were mixed to a volume of 52 µl followed by addition of 3 µl of initiation mix ( 40 mM MgCl2 , 10 mM ATP , 1 M KCl ) . Fluorescence was monitored at excitation and emission wavelengths of 365 and 407 nm , respectively , in a fluorescence spectrophotometer ( Photon Technology International; Lawrenceville , NJ ) . Acquisition was paused at 400 s , and 5 µl of CARMIL CBR ( final concentration 250 nM ) was spiked into the reaction , mixed rapidly by pipetting , and measurement was resumed . For pyrene actin elongation assays ( as in Figure 4A and B ) , 5 µl of freshly mechanically sheared F-actin ( 10 µM ) was added to a mixture of the indicated proteins or control buffers , and then immediately mixed with 0 . 5 µM monomeric actin ( 10% pyrene labeled ) in 60 µl reactions and monitored in a plate reader ( Infinite M200; Tecan , Männedorf , Switzerland ) at excitation and emission wavelengths of 365 and 407 nm , respectively . The following anisotropy experiments were performed in HEK buffer ( 20 mM HEPES pH 7 . 5 , 1 mM EDTA , 50 mM KCl , 0 . 5 mM DTT ) . Reactions were incubated at room temperature for 15 min , and anisotropy was determined by measuring polarized emission intensities at 525 nm when excited at 497 nm using a fluorescence spectrophotometer ( Photon Technology International ) . To compare mTwf1-tail binding to wild type and mutant CP ( Figure 1B ) , HiLyte-488-mTwf1 tail peptide ( 100 nM ) was mixed with different concentrations of wild-type or mutant CP . To compare the abilities of full-length wild type mTwf1 and mutant mTwf1-11 polypeptides to compete with labeled mTwf1-tail for binding CP ( Figure 1C ) , HiLyte-488-mTwf1 tail peptide ( 100 nM ) was mixed with 1 µM CP and variable concentrations of full-length mTwf1 polypeptides . The following anisotropy experiments were performed in the indicated buffer , incubated at room temperature for 2 min , and anisotropy was determined by measuring polarized emission intensities at 525 nm when excited at 497 nm for HiLyte-488 , or at 582 nm when excited at 552 nm for TAMRA . To compare mTwf1 tail peptide binding to wild type CP and mutant CP ( RY ) ( Figure 2B ) , HiLyte-488-mTwf1 tail peptide ( 60 nM ) was mixed with different concentrations of CP or CP ( RY ) in HEK buffer containing 0 . 005% TWEEN 20 . To compare the abilities of unlabeled wild type and mutant mTwf1 tail peptides to compete with labeled mTwf1 tail peptide for binding to CP ( Figure 2C ) , TAMRA-mTwf1 tail peptide ( A305-D350 , 40 nM ) was mixed with 1 µM CP and varying concentrations of the unlabeled tail peptides ( mTwf1 A305-D350 or mTwf1 A305-D350 , K325A ) in 20 mM MOPS ( pH 7 . 2 ) , 1 mM TCEP , 100 mM KCl , 1 mM NaN3 , 0 . 005% TWEEN 20 . To test the abilities of different fragments of CARMIL to compete with mTwf1 tail peptide for binding CP , HiLyte-488-mTwf1 tail peptide ( 60 nM ) was mixed with 240 nM CP and different concentrations of mouse CARMIL1 CBR ( 964–1078 ) , CPI ( 969–1005 ) , or CSI ( 1019–1037 ) in HEK buffer containing 0 . 005% TWEEN20 . For kinetic dissociation experiments ( as in Figure 4D and E ) , an SX . 18MV stopped flow instrument with Pro-Data SX software V2 . 2 . 27 ( Applied Photophysics Ltd . , Leatherhead , UK ) was used . 100 nM TAMRA-V-1 was preincubated with 2 µM CPα1β2 . At time zero , TAMRA-V-1:CP complex was rapidly mixed via stopped-flow with an equal volume of a solution containing 5 µM unlabeled V-1 , along with varied concentrations of mTwf1 or mTwf1-11 . Experiments were performed at 25°C in HEK buffer containing 0 . 005% TWEEN20 . Excitation occurred at 505 nm , with emission detected using a 570 + nm band-pass filter . All concentrations of mTwf were performed in replicates of 5–10 , and traces were averaged . Apparent dissociation rates were determined by fitting the averaged data ( 5 ms . - 120 s . ) to a single exponential model using Pro-Data Viewer software V4 . 2 . 27 ( Applied Photophysics Ltd . ) . For all experiments , 24 × 60 mm coverslips ( Fisher Scientific; Pittsburg , PA ) were cleaned by successive sonications as follows: 60 min in detergent , 20 min in 1 M KOH , 20 min in 1 M HCl min , and 60 min in ethanol . Coverslips were then washed extensively with ddH2O and dried in an N2-stream . A solution of 80% ethanol pH 2 . 0 , 2 mg/ml methoxy-poly ( ethylene glycol ) -silane and 2 µg/ml biotin-poly ( ethylene glycol ) -silane ( Laysan Bio Inc . ; Arab , AL ) was prepared and layered on the cleaned coverslips ( 200 µl per coverslip ) . The coverslips were incubated for 16 hr at 70°C . To assemble flow cells , PEG-coated coverslips were rinsed extensively with ddH2O and dried in an N2-stream , then attached to a prepared flow chamber ( Ibidi; Martinsried , German ) with double sided tape ( 2 . 5 cm x 2 mm x 120 µm ) and five min epoxy resin . Flow cells were prepared immediately before use by sequential incubations as follows: 3 min in HEK-BSA ( 20 mM Hepes pH 7 . 5 , 1 mM EDTA , 50 mM KCl , 1% BSA ) , 30 s in Streptavidin ( 0 . 1 mg/ml in PBS ) , a fast rinse in HEK-BSA , and then equilibration in 1X TIRF buffer , pH 7 . 5 ( 10 mM imidazole , 50 mM KCl , 1 mM MgCl2 , 1 mM EGTA , 0 . 2 mM ATP , 10 mM DTT , 15 mM glucose , 20 µg/ml catalase , 100 µg/ml glucose oxidase , and 0 . 5% methylcellulose ( 4000 cP ) ) . To initiate reactions , actin monomers ( 10% OG-labeled , 0 . 5% biotinylated ) were diluted to 1 µM in TIRF buffer , and immediately transferred to a flow chamber . After several minutes , once the actin filaments reached an appropriate length ( approximately 10 µm ) , the reaction mixture was replaced by flow-in . For depolymerization experiments , the solution was replaced with TIRF buffer lacking actin monomers , with or without Twinfilin and/or CP polypeptides . For uncapping experiments , the solution was replaced with TIRF buffer lacking actin monomers , with 3 nM SNAP-649-CP ( 100% labeled ) , and filaments were allowed to be capped for 3 min . Subsequently , the solution was again replaced with TIRF buffer lacking actin monomers , with or without 50 nM CARMIL CBR and/or variable concentration of Twinfilin polypeptides . Time-lapse TIRF microscopy was performed using a Nikon-Ti200 inverted microscope equipped with a 150 mW Ar-Laser ( Mellot Griot; Carlsbad , CA ) , a 60X TIRF-objective with a N . A . of 1 . 49 ( Nikon Instruments Inc . ; New York , NY ) , and an EMCCD camera ( Andor Ixon; Belfast , Northern Ireland ) . During recordings , optimal focus was maintained using the perfect focus system ( Nikon Instruments Inc ) . Images were captured every 5 s . The pixel size corresponded to 0 . 27 µm . Filament depolymerization rates were determined by tracing filaments in ImageJ ( http://rsbweb . nih . gov/ij ) and measuring the change in length of individual filaments for 15–20 min after flow-in , or until filaments disappeared . Differences in fluorescence intensity along the length of the filament provided fiduciary marks that allowed us to distinguish barbed- and pointed-ends . Filament uncapping was measured by monitoring the as the amount of time that SNAP-649-CP puncta remained associated with the barbed end of a filament after the addition of CARMIL to the reaction ( with or without Twinfilin ) and expressing it as a fraction of filaments that remained capped at a given time point . All results shown are data from at least two independent TIRF experiments . Mouse B16-F10 ( CRL-6475 ) , Neuro-2a ( CCL-131 ) , and NIH/3T3 ( CRL-1658 ) cells obtained directly from ATCC ( American Type Culture Collection; Manassas , VA ) , where their identities were authenticated by short tandem repeat DNA profiling and where they were tested for mycoplasma contamination . Cells were used for experiments within one year . All cells were grown in DMEM ( Gibco BRL Life Technologies; Carlsbad , CA ) supplemented with 10% fetal bovine serum ( FBS; Sigma ) and 200 mM L-glutamine ( Thermo Fisher Scientific ) at 37°C and 5% CO2 . All cell culture experiments were carried out in 6-well dishes that were initially seeded with 100 , 000 cells . To knockdown Twinfilin-1 or Capping Protein cells were transfected 24 hr after seeding with 30 pmol siRNA oligo using Lipofectamine RNAiMAX ( Thermo Fisher Scientific ) according to the manufacturer’s instructions . RNAi oligos directed against the mouse Twinfilin-1 coding region targeting ( siTwf1 ) 5’- CGUUACCAUUUCUUUCUGUUU −3’; and against the Capping Protein β subunit coding region targeting ( siCP1 ) 5’- CCUCAGCGAUCUGAUCGACUU-3’ , or ( siCP2 ) 5’- GCACGCUGAAUGAGAUCUA-3’ . Cells were transfected in parallel with control RNAi oligos ( Invitrogen ) . For over expression experiments cultured cells were transfected using Lipofectamine 3000 ( Thermo Fisher Scientific ) according to the manufacturer’s instructions 24 hr after seeding . For CARMIL over expression experiments , 5 μG of DNA was transfected , and for Twinfilin over expression experiments 1 μG of DNA was transfected . The rabbit anti-Twinfilin was a generous gift from Pekka Lappalainen ( Univ . Helsinki ) and used a dilution of 1:1000 for western blot detection and 1:100 in cultured cells . A mouse anti-Capping Protein ( Development Studies Hybridoma Bank; Iowa City , IA ) was used at a dilution of 1:2000 for western blot detection and 1:50 in cultured cells . Mouse anti-Flag ( F3165 , Sigma ) and rabbit anti-Myc ( GTX29106 , GeneTex; Irvine , CA ) was used at 1:5000 for western blot detection and 1:500 in cultured cells . Mouse and Rabbit horseradish peroxidase conjugated secondary antibodies ( GE Healthcare ) were used at a dilution of 1:10 , 000 for western blot detection . Secondary antibodies for immunofluorescence ( Alexa Fluor 488 or 647 ) and Alexa Fluor 568-phalloidin ( ThermoFisher ) were used at a dilution of 1:1000 . For cell-staining experiments , 48 hr post transfection , the cells were re-plated on 3 × 1×1 mm glass coverslip ( VWR International ) that had been acid washed and coated with Laminin ( Invitrogen ) and allowed to adhere for 3–6 hr . Cells were fixed for 15 min with 4% paraformaldehyde in PBS at room temperature and then permeabilized for 15 min in permeabilization solution ( 0 . 5% Triton X-100 and 0 . 3 M glycine in PBS ) at room temperature . Slips were then blocked in 3% BSA dissolved in PBST ( 1X PBS and 0 . 1% TWEEN 20 ) for 1 hr at room temperature , then incubated in primary antibody ( in PBST ) for 12 hr at 4°C . Coverslips were then washed three times with 1X PBST and incubated with secondary antibodies ( in PBST ) for 1 hr at room temperature . Slips were washed three times with PBST and two times with PBS , and subsequently mounted on to slides with AquaMount ( Thermo Fisher Scientific ) . Cells were imaged on a Nikon i-E upright confocal microscope equipped with a CSU-W1 spinning disk head ( Yokogawa , Tokyo , Japan ) , 60x oil objective ( NA 1 . 4; Nikon Instruments ) , and an Ixon 897 Ultra-CCD camera ( Andor Technology ) controlled by NIS-Elements software . Maximum intensity projections and raw fluorescence values were measured using Fiji . To measure protein levels in cells after silencing and rescue , cells were harvest 48 hr after initial oligo transfection and incubated for 10 min at 4°C in RIPA buffer ( 50 mM Tris , pH 7 . 5 , 150 mM NaCl , 1% NP-40 , 0 . 5% Na-deoxycholate , 0 . 1% SDS , 2 mM EDTA , 50 mM NaF ) . Samples were incubated on ice for 30 min , vortexed every 10 min , then precleared by centrifugation at 20 , 800 x g for 15 min at 4°C , quantified by Bradford assay , and immunoblotted . Proteins were detected using a Pierce ECL Western Blotting Substrate detection kit ( Thermo Fisher Scientific ) . Bands were quantified using ImageLab ( Biorad ) . HDX-MS was performed as described ( Johnson et al . , 2018 ) . CP and Twf1 samples were buffer-exchanged with 1X phosphate saline buffer ( PBS ) , pH 7 . 4 . HDX was initiated by diluting samples ( 25 μM , 2 μL ) 10-fold with 1XPBS prepared in D2O buffer , or 1XPBS H2O buffer for samples measured for no-deuterium control . At different time intervals ( 10 , 30 , 60 , 120 , 360 , 900 , 3600 , and 14400 s ) , the labeling reaction was quenched by rapidly decreasing the pH to 2 . 5 with 30 μL of quench buffer ( 3 M urea , 1% trifluoroacetic acid , H2O ) at 4°C . The protein mixture was immediately injected into a custom-built HDX sample-handling device that enabled digestion with a column containing immobilized pepsin ( 2 mm ×20 mm ) at a flow rate of 100 μL/min in 0 . 1% formic acid . The resulting peptic peptides were captured on a ZORBAX Eclipse XDB C8 column ( 2 . 1 mm ×15 mm , Agilent ) for desalting ( 3 min ) . The C8 column was then switched in-line with a Hypersil Gold C18 column ( 2 . 1 mm ×50 mm , Thermo Fisher ) , and a linear gradient ( 4–40% acetonitrile , 0 . 1% formic acid , 50 μL/min flow rate , over 5 min ) was used to separate the peptides and direct them to an LTQ-FTICR mass spectrometer ( Thermo Fisher ) equipped with an electrospray ionization source . Valves , columns , and tubing for protein digestion and peptide separation were immersed in an ice-water bath to minimize back-exchange . To map the peptic peptides , the digest , in the absence of HDX , was submitted to accurate mass analysis by LC–MS/MS with the LTQ-FTICR , and the peptic peptides identified using Mascot ( Matrix Science ) . For samples that underwent HDX , raw mass spectra and peptide sets were submitted to HDX Workbench ( Pascal et al . , 2012 ) for calculation and data visualization in a fully automated fashion . Peptides for each run were assessed based on relative representation and statistical validation as implemented within HDX Workbench . Appropriate approach to determine statistical significance between these data is by using Tukey’s multiple comparison test . A representative time point was manually selected , replicate data points from multiple samples at this time point used to conduct a one-way analysis of variance ( ANOVA ) the divergence between the means of the experiments were assessed . In instances with large differences , Tukey method was used to determine statistical significance if the resulting P value is less than 0 . 05 . In the case where there was a comparison between two experiments , a t-test was used . Only the top six peptides from each MS scan were used in the final analysis . The extent of HDX at each time point was calculated by subtracting the centroid of the isotopic distribution of the nondeuterated peptide from that of the deuterated peptide . The relative deuterium uptake was plotted versus the labeling time to yield kinetic curves ( %D vs time ) . Error bars represent the results of t-tests between samples are shown above each time point to illustrate statistical significance . For comparison between apo states and the complexes , differences in HDX for all time points were calculated . Absolute differences in perturbation values larger than 5% D were considered significant . HDX values at 15 min time point were mapped onto the protein three-dimensional ( 3D ) structure for data visualization . Peptide digestions were optimized under HDX assay conditions , and the mass calculations included accommodation for back exchange with solvent .	Plant and animal cells are supported by skeleton-like structures that can grow and shrink beneath the cell membrane , pushing and pulling on the edges of the cell . This scaffolding network – known as the cytoskeleton – contains long strands , or filaments , made from many identical copies of a protein called actin . The shape of the actin proteins allows them to slot together , end-to-end , and allows the strands to grow and shrink on-demand . When the strands are the correct length , the cell caps the growing ends with a protein known as Capping Protein . This helps to stabilize the cell’s skeleton , preventing the strands from getting any longer , or any shorter . Proteins that interfere with the activity of Capping Protein allow the actin strands to grow or shrink . Some , like a protein called V-1 , attach to Capping Protein and get in the way so that it cannot sit on the ends of the actin strands . Others , like CARMIL , bind to Capping Protein and change its shape , making it more likely to fall off the strands . So far , no one had found a partner that helps Capping Protein limit the growth of the actin cytoskeleton . A protein called Twinfilin often appears alongside Capping Protein , but the two proteins seemed to have no influence on each other , and had what appeared to be different roles . Whilst Capping Protein blocks growth and stabilizes actin strands , Twinfilin speeds up their disassembly at their ends . But Johnston , Hilton et al . now reveal that the two proteins actually work together . Twinfilin helps Capping Protein resist the effects of CARMIL and V-1 , and Capping Protein puts Twinfilin at the end of the strand . Thus , when Capping Protein is finally removed by CARMIL , Twinfilin carries on with disassembling the actin strands . The tail of the Twinfilin protein looks like part of the CARMIL protein , suggesting that they might interact with Capping Protein in the same way . Attaching a fluorescent tag to the Twinfilin tail revealed that the two proteins compete to attach to the same part of the Capping Protein . When mouse cells produced extra Twinfilin , it blocked the effects of CARMIL , helping to grow the actin strands . V-1 attaches to Capping Protein in a different place , but Twinfilin was also able to interfere with its activity . When Twinfilin attached to the CARMIL binding site , it did not directly block V-1 binding , but it made the protein more likely to fall off . Understanding how the actin cytoskeleton moves is a key question in cell biology , but it also has applications in medicine . Twinfilin plays a role in the spread of certain blood cancer cells , and in the formation of elaborate structures in the inner ear that help us hear . Understanding how Twinfilin and Capping Protein interact could open paths to new therapies for a range of medical conditions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2018	A novel mode of capping protein-regulation by twinfilin
CRISPR-Cas9 genome engineering has revolutionised high-throughput functional genomic screens . However , recent work has raised concerns regarding the performance of CRISPR-Cas9 screens using TP53 wild-type human cells due to a p53-mediated DNA damage response ( DDR ) limiting the efficiency of generating viable edited cells . To directly assess the impact of cellular p53 status on CRISPR-Cas9 screen performance , we carried out parallel CRISPR-Cas9 screens in wild-type and TP53 knockout human retinal pigment epithelial cells using a focused dual guide RNA library targeting 852 DDR-associated genes . Our work demonstrates that although functional p53 status negatively affects identification of significantly depleted genes , optimal screen design can nevertheless enable robust screen performance . Through analysis of our own and published screen data , we highlight key factors for successful screens in both wild-type and p53-deficient cells . CRISPR-Cas9 genome engineering technologies have transformed cell biology , particularly high throughput functional genomic screens ( Wang et al . , 2015; Shalem et al . , 2014; Shalem et al . , 2015; Smith et al . , 2017 ) . Pooled CRISPR-Cas9 cell viability screens have been successfully employed in determining gene essentiality ( Hart et al . , 2015 ) , identifying genetic interactions ( Chan et al . , 2019 ) and assessing drug sensitivities across various genetic backgrounds ( Han et al . , 2017 ) . A number of factors influence CRISPR-Cas9 screen performance , including cellular background . In particular , recent reports concerning technical difficulties in CRISPR-Cas9 genome editing in p53-proficient cells , have brought into question the suitability of p53-proficient cell lines for high throughput CRISPR-Cas9 genetic screens ( Haapaniemi et al . , 2018; Ihry et al . , 2018 ) . TP53 , encoding p53 , acts as a master regulator of cell-cycle checkpoint activation ( Kastan et al . , 1991 ) , cellular senescence ( Shay et al . , 1991 ) and induction of apoptosis in response to DNA damage ( Clarke et al . , 1993; Lowe et al . , 1993; Lakin and Jackson , 1999 ) . TP53 is arguably the most important tumour suppressor gene , with loss of function mutations in up to 50% of human cancers ( Bouaoun et al . , 2016 ) . Consequently , the p53 status of a cell line , either wild-type ( proficient ) or mutant ( deficient ) , can be an important factor in determining the suitability of a cellular model , and hence is an important consideration in design of high throughput genetic screens . Generation of DNA double strand breaks ( DSBs ) induces p53-dependent cell-cycle arrest in normal fibroblasts ( Di Leonardo et al . , 1994 ) , and most CRISPR-Cas9 genome editing approaches rely on DSB generation to achieve efficient editing ( Jinek et al . , 2012 ) . Recent work has shown that CRISPR-Cas9-associated DSBs in hPSCs ( human pluripotent stem cells ) induce a p53-mediated apoptotic response , leading to high levels of toxicity and reduced editing efficiency in this background ( Ihry et al . , 2018 ) . Furthermore , a similar p53-mediated DSB response in wild-type retinal pigment epithelial ( RPE-1 ) cells reportedly severely impaired identification of essential genes in a CRISPR-Cas9 screen when compared to RPE-1 TP53 knockout ( TP53KO ) cells ( Haapaniemi et al . , 2018 ) . In contrast , analysis of data from a small number of additional screens in p53 wild-type RPE-1 cells has shown that performance of successful CRISPR screens , as determined by essential gene identification and enrichment of expected targets , is possible in this cellular background ( Brown et al . , 2019 ) . This controversy is confounded by the complexity of variation in experimental design between screens with a lack of controlled parallel experiments . To provide more definitive insights into this important debate , we performed parallel CRISPR-Cas9 screens in paired wild-type and TP53KO cell lines , thereby minimising additional confounding factors that can preclude accurate screen comparisons . In summary , we present data from parallel screens in TP53 wild-type and TP53KO RPE-1 cells , which demonstrate that a p53-mediated response does negatively impact the sensitivity of CRISPR-Cas9 screens . The extent of the impact of TP53 status on CRISPR-Cas9 screens might vary depending on the cell type being studied , including those with loss-of-function mutations in TP53 without being fully TP53 null . It remains to be established precisely how and to what extent different TP53 mutations , including ‘hotspot’ mutations , might influence CRISPR-Cas9 screen performance . However , we anticipate that most or all cell lines with an intact TP53 pathway and proper cell-cycle checkpoint activation would likely recapitulate our findings . Other important factors impacting sensitivity include the guide RNA library used , the magnitude of guide effects , adequate gRNA representation and sufficient sequencing depth . Selection of high-editing efficiency Cas9-expressing cells is also highly recommended and use of biological replicates enables identification of clonal variation . Considering these factors in screen design and execution allows successful CRISPR-Cas9 screens to be carried out in both p53-proficient and p53-deficient cells , thereby fostering new biological insights . A custom dual-sgRNA library was designed to target 852 genes related to the DNA damage response , 112 olfactory-receptor genes , and 14 sequence scrambled negative controls with a total of 3404 dual-sgRNAs . The genes targeted by this library include a total of 95 core essential genes . The sgRNA sequences and pairwise scores were determined using the Croatan scoring algorithm ( Erard et al . , 2017 ) . Transomic Technologies selected the top pairs of sgRNAs for each gene and assigned a distinct barcode to each pair , cloned them into the pCLIP-dual-SFFV-ZsGreen vector , and packaged them into lentiviral particles ready for transduction . For pooled screening , the viral titre was determined by exposing cells to a 6-point dose response of the lentiviral stock . The optimal concentration of virus to achieve a multiplicity of infection ( MOI ) of 0 . 3 was determined by linear regression analysis . CRISPR-Cas9 screens were performed using the custom dual-sgRNA DNA damage response library outlined above . Biological duplicates ( two independently isolated Cas9-expressing clones ) of wild-type and TP53KO RPE-1 cells were transduced at a MOI of 0 . 3 and >1 , 000 fold coverage of the library . The following day , cells were cultured with puromycin to select for the transductants for 12 additional days . Surviving cells from each biological replicate were harvested prior to puromycin selection ( day 3 ) , and at day 15 and day 19 after initial transduction . Subsequently , the genomic DNA ( gDNA ) was isolated using TAIL buffer ( 17 mM Tris pH 7 . 5 , 17 mM EDTA , 170 mM NaCl , 0 . 85% SDS , and 1 mg/mL Proteinase K ) and subjected to 24 PCR reactions with custom indexed primers designed to amplify the barcode within the lentiviral backbone and append Illumina adapter sequences . Finally , the PCR products were purified ( QIAquick PCR Purification kit , Qiagen ) , multiplexed , and sequenced on an Illumina HiSeq1500 system . Genes enriched or depleted in the day 15 and day 19 samples compared to the day 3 samples were determined using MAGeCK v0 . 5 . 9 . 2 ( Li et al . , 2014 ) . RPE-1 TP53 wild-type and TP53KO cells were cultured in DMEM/F-12 media ( Dulbecco’s Modified Eagle Medium: Nutrient Mixture Ham’s F-12 , Sigma-Aldrich ) supplemented with 17 mL of 7 . 5% NaHCO3 ( Sigma-Aldrich ) per 500 mL , 10% ( v/v ) foetal bovine serum ( FBS , BioSera ) , 100 U/mL penicillin , 100 µg/mL streptomycin ( Sigma-Aldrich ) , 2 mM L-glutamine , and 10 μg/mL blasticidin ( Sigma-Aldrich ) to select for Cas9 expressing cells . Cells were additionally cultured with 1 . 5 µg/mL puromycin during selection of the transductants . RPE-1 TP53 wild-type and TP53KO cells were harvested in 100–200 uL of Laemmli buffer ( 120 mM Tris 6 . 8 pH , 4%SDS , 20% glycerol ) . Protein concentrations were determined using a NanoDrop spectrophotometer ( Thermo Scientific ) at A280 nm . SDS-PAGE was performed with 35 µg of protein lysates , the proteins were resolved on a precast NuPAGE Novex 4–12% Bis/Tris gradient gel ( Invitrogen ) . Resolved proteins were transferred to a nitrocellulose membrane ( GE Healthcare ) and immunoblotted with the following antibodies at a 1/1 , 000 dilution: p53 ( #554293 , BD Biosciences ) and GAPDH ( #MAB374 , Merck Millipore ) . RPE-1 wild-type cells were originally obtained from the ATCC cell repository by Professor Jonathon Pines . They were routinely tested for mycoplasma and were authenticated using Affymetrix SNP6 copy number analysis . RPE-1 TP53KO cells were generated as described previously ( Chiang et al . , 2016 ) . The TP53 wild-type and TP53KO RPE-1 cells were transduced with a lentiviral vector encoding Cas9 and a blasticidin resistance cassette to facilitate the isolation of Cas9-expressing clones . Limiting dilution of the transduced population enabled isolation of monoclonal cell lines . Cas9 expression was validated by western blot and Cas9 editing efficiency was assayed by transducing clones with a lentiviral vector encoding GFP , BFP , and a sgRNA for GFP ( obtained from Dr Emmanouil Metzakopian , UK Dementia Research Institute , Cambridge , UK ) . Transduced and non-transduced cells were subjected to FACS sorting using an LSRFortessa ( BD Biosciences ) flow cytometer . The Cas9 editing efficiency for each clone was calculated by comparing the percentage of BFP+ ( i . e . edited ) cells to the GFP/BFP+ cells ( i . e . total transduced population ) using FlowJo . Statistical analyses were performed in Python ( 3 . 7 . 5 ) , using the following packages in particular: Data files containing guide abundances were downloaded from https://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE128210 . Supplementary file 5 lists the origins of the data . Where multiple timepoints were available , the day 18 timepoint was used . Guides targeting genes not present in our DDR library were removed from the abundance tables , and MAGeCK ( 0 . 5 . 9 . 2 ) was used to obtain significance values for depletion and enrichment of genes . The command line arguments remove-zero-threshold=10 and remove-zero=control were used . LFCs were normalised by subtracting the mean of the olfactory receptor ( OR ) genes from all values , and then dividing all values by the SD of the OR genes . To simulate smaller sequencing runs , guide abundances were resampled by N random draws using the initial abundances as weights . N was set to yield expected median abundances ranging between 10 and 1000 . MAGeCK was used to obtain significance values as above . five replicate draws were performed per sample . Genes within the library were annotated according to KEGG ( Kyoto Encyclopedia of Genes and Genomes ) pathway . Selection of relevant pathways within the library was based on classifications by Pearl et al . ( 2015 ) . The enrichment of genes with p<0 . 05 in these pathways was evaluated using Fisher’s exact test . Genes that were depleted over time , or enriched , were tested separately .	The invention of CRISPR-Cas9 genome editing has unlocked a greater understanding of the human genome . Researchers can use this system to make targeted cuts in any gene in the genome , forcing the cell to perform a rapid repair at the cut site . These repairs often introduce mutations into the damaged area , adding or removing DNA letters and disrupting the gene . This allows researchers to study what happens to cells when specific genes are missing , which can help to uncover what each gene is for . One of the most comprehensive ways to use this technique is to perform a CRISPR-Cas9 screen , which disrupts each gene in the genome one by one . For a CRISPR-Cas9 screen to work well , a cell needs to survive the cuts to its genome . But there is a crucial gene that can stop this happening . Often described as the 'guardian of the genome' , this gene codes for a protein called p53 , a tumour suppressor that helps to stop a cell turning cancerous when its DNA becomes damaged . This protein activates when the cell senses a cut in its genetic material and can kill the cell if it fails to make a successful repair . Recent work has shown that the presence of a working copy of the gene for the p53 protein might limit the ability of CRISPR-Cas9 to edit genes . But the evidence was inconclusive . So , Bowden , Morales-Juarez et al . performed two parallel CRISPR-Cas9 screens in human cells with and without p53 to find out more . This revealed that CRISPR-Cas9 can inactivate genes in both normal cells and cells lacking the p53 protein , but that it works better in cells without p53 . This was because , when p53 was active , the cells initiated a protective response against the CRISPR-Cas9 cuts . This changed the patterns of genes successfully inactivated by the screen , but it did not make the results unusable . Careful experimental design and thorough data analysis made it possible to get useful results even in cells with functional p53 protein . The gene for p53 has mutations in around half of human cancers . So , understanding how it affects CRISPR-Cas9 screens could influence the design of future experiments . It is possible that the effects of the p53 protein could vary from cell type to cell type , and with different p53 mutations . Comparisons like the one performed here could help to further unpick how the cell's DNA repair systems might interfere with future CRISPR experiments .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "short", "report", "genetics", "and", "genomics" ]	2020	Parallel CRISPR-Cas9 screens clarify impacts of p53 on screen performance
Many eukaryotic protein kinases are activated by phosphorylation on a specific conserved residue in the regulatory activation loop , a post-translational modification thought to stabilize the active DFG-In state of the catalytic domain . Here we use a battery of spectroscopic methods that track different catalytic elements of the kinase domain to show that the ~100 fold activation of the mitotic kinase Aurora A ( AurA ) by phosphorylation occurs without a population shift from the DFG-Out to the DFG-In state , and that the activation loop of the activated kinase remains highly dynamic . Instead , molecular dynamics simulations and electron paramagnetic resonance experiments show that phosphorylation triggers a switch within the DFG-In subpopulation from an autoinhibited DFG-In substate to an active DFG-In substate , leading to catalytic activation . This mechanism raises new questions about the functional role of the DFG-Out state in protein kinases . Stringent regulatory control of protein kinases is critically important for the integrity of cellular signal transduction . The catalytic activity of protein kinases is regulated by finely-tuned allosteric mechanisms that reversibly switch the kinase domain between active and inactive conformational states ( Huse and Kuriyan , 2002 ) . Disruption of these mechanisms , leading to constitutive kinase activity , is a major cause of human cancer , and small molecules that inhibit specific disease-associated kinases are an increasingly important component of many modern cancer therapies ( Zhang et al . , 2009 ) . Phosphorylation on a specific site in the activation loop of the kinase domain is the most widely conserved regulatory mechanism in kinases ( Johnson et al . , 1996 ) . X-ray structures show that ionic interactions between the phosphate moiety and a surrounding pocket of basic residues stabilize the activation loop in a conserved active conformation ( Knighton et al . , 1991; Yamaguchi and Hendrickson , 1996; Steichen et al . , 2012 ) . In this active state , a catalytic asp-phe-gly ( DFG ) motif at the N-terminal end of the activation loop adopts an active ‘DFG-In’ conformation , with the aspartate residue of the DFG motif pointing into the active site to coordinate Mg-ATP , and the C-terminal segment of the activation loop positioned to bind peptide substrates . The DFG-In state is stabilized by the assembly of a network of hydrophobic residues , termed the regulatory spine , that lock together the N-terminal lobe , the αC-helix , the DFG motif and the C-terminal lobe of the kinase ( Kornev et al . , 2006; Kornev et al . , 2008 ) . In the absence of phosphorylation on the activation loop , kinase activity is usually restrained by rearrangements of the activation loop and DFG motif into specific autoinhibited conformations . An important autoinhibited state , called ‘DFG-Out’ , involves a flip in the backbone torsion angles of the DFG motif , reorienting the DFG aspartate out of the active site to prevent magnesium coordination , repositioning the activation loop to block peptide binding , and disassembling the regulatory spine ( Hubbard et al . , 1994; Nagar et al . , 2003; Mol et al . , 2004 ) . This conformational change dramatically alters the chemical makeup of the active site , and selective recognition of the DFG-Out state underlies the mechanism of action of many small-molecule kinase inhibitors ( Liu and Gray , 2006 ) . The serine/threonine kinase Aurora A ( AurA ) is an essential mitotic protein that controls a variety of cellular processes including mitotic spindle assembly , centrosome maturation , and mitotic entry ( Glover et al . , 1995; Hannak et al . , 2001; Berdnik and Knoblich , 2002; Macůrek et al . , 2008; Seki et al . , 2008 ) . These functions of AurA are driven by two distinct activation mechanisms of the kinase operating in different spatiotemporal contexts . At the centrosome , AurA must first be activated by autophosphorylation on the activation loop threonine residue T288 in order to carry out its centrosomal functions . At the mitotic spindle , AurA is instead activated by binding to the spindle assembly factor Tpx2 ( Kufer et al . , 2002 ) . This spindle-associated pool of AurA must be maintained in the unphosphorylated state by the phosphatase PP6 in order for spindle assembly to proceed faithfully ( Zeng et al . , 2010; Toya et al . , 2011 ) . Extensive in vitro studies have confirmed that Tpx2 and phosphorylation can act independently to increase AurA kinase activity by up to several hundred-fold ( Zorba et al . , 2014; Dodson and Bayliss , 2012 ) . We recently showed that activation of AurA by Tpx2 is driven by a population shift from a DFG-Out to the DFG-In state ( Cyphers et al . , 2017 ) . Since crystal structures of phosphorylated AurA bound to Tpx2 show the T288 phosphothreonine residue forming the canonical ionic interactions thought to stabilize the DFG-In state ( Bayliss et al . , 2003; Zhao et al . , 2008; Clark et al . , 2009 ) , it has been assumed that phosphorylation also triggers a transition from the DFG-Out to the DFG-In state . In this paper , we show that phosphorylation on T288 in fact activates AurA through a completely different mechanism than Tpx2 . Three complementary spectroscopic methods , infrared spectroscopy , Förster resonance energy transfer , and double electron-electron resonance , all show that phosphorylation does not trigger a switch to the DFG-In state , and that phosphorylated AurA continually samples both DFG-In and DFG-Out conformational states . Instead , phosphorylation triggers a conformational switch from a previously unknown inactive DFG-In substate to a fully activated DFG-In substate , enhancing the catalytic activity of the DFG-In subpopulation within a dynamic conformational ensemble . We set out to explain how phosphorylation of AurA on T288 leads to a ~100 fold increase in catalytic activity ( Figure 1—figure supplement 1a ) ( Zorba et al . , 2014; Dodson and Bayliss , 2012 ) . We previously used an infrared ( IR ) probe that tracks the DFG motif of AurA to show that Tpx2 binding triggers a conformational change from the DFG-Out to the DFG-In state ( Cyphers et al . , 2017 ) , resulting in the assembly of the active site and the regulatory spine ( Figure 1—figure supplement 2 ) . In this method , a cysteine residue is introduced at position Q185 at the back of the active site of AurA , and chemical labeling is used to introduce a nitrile infrared probe at this position ( Fafarman et al . , 2006 ) . To test whether phosphorylation of AurA also causes a conformational shift of the DFG motif , we prepared samples of AurA Q185C phosphorylated on T288 . Homogeneous phosphorylation and nitrile labeling were verified by western blotting and mass spectrometry ( Figure 1—figure supplement 1 ) . IR spectra of nitrile-labeled phosphorylated AurA showed predominantly a single absorbance band centered at 2158 cm−1 ( Figure 1a , solid black line ) . We previously assigned this peak in IR spectra of unphosphorylated AurA to the DFG-Out form of the kinase , in which the nitrile probe is buried in a hydrophobic pocket ( Figure 1b , lower panel ) ( Cyphers et al . , 2017 ) . Addition of saturating amounts of Tpx2 peptide ( residues 1–43 of human Tpx2 ) to the IR samples caused a dramatic spectral change wherein the central peak at 2158 cm−1 is largely replaced by two new peaks at 2149 cm−1 and 2164 cm−1 ( Figure 1a , dashed black line ) . These changes are indicative of a shift to the DFG-In state , in which water molecules coordinated to the DFG motif form hydrogen bonds to the nitrile probe , causing pronounced spectral shifts ( Figure 1b , upper right panel ) ( Cyphers et al . , 2017 ) . To confirm that the peak at 2158 cm−1 arises from the DFG-Out state , we mutated residue W277 , which is positioned directly against the IR probe in the DFG-Out state , but is displaced away from it in the DFG-in state , to alanine ( Figure 1b ) . IR spectra of the W277A mutant showed a clear spectral shift of the 2158 cm−1 peak ( Figure 1c ) , consistent with this peak arising from the DFG-Out state . The addition of ADP to apo AurA resulted in the appearance of a DFG-In subpopulation , apparent in the IR spectra as small shoulders on either side of the main 2158 cm−1 peak . Experiments performed over a range of temperatures showed that this DFG-In subpopulation increases at higher temperature ( Figure 1a , colored lines ) , but does not reach the level observed in the presence of Tpx2 . A similar DFG-In subpopulation was also detected in unphosphorylated AurA bound to ADP ( Cyphers et al . , 2017 ) ( Figure 1a , inset ) , highlighting that although nucleotide binding shifts the DFG equilibrium towards the DFG-In state , phosphorylation does not seem to enhance this effect . These IR results suggested that phosphorylation alone does not substantially change the DFG-In/Out equilibrium of AurA , unlike Tpx2 binding . However , as replacement of the Q185 residue with the nitrile probe was found to alter the activation properties of AurA ( Cyphers et al . , 2017 ) , it was necessary to confirm this interpretation using an alternative method . We used intramolecular FRET to track movements of the activation loop of AurA with and without phosphorylation on T288 , using a construct with a native Q185 residue ( Cyphers et al . , 2017 ) . Donor ( D ) and acceptor ( A ) fluorophores ( Alexa 488 and Alexa 568 , respectively ) were incorporated on the activation loop ( S284C ) and αD helix ( L225C ) using maleimide chemistry ( Cyphers et al . , 2017 ) . These labeling positions were chosen to track the movement of the activation loop across the active-site cleft as the kinase switches from the DFG-Out to the DFG-In state , with the dyes predicted to be further apart in the DFG-In state ( Figure 2a ) . Phosphorylation of the protein on T288 was confirmed by tryptic mass spectrometry ( Figure 2—figure supplement 1 ) , and the labeled phosphorylated sample exhibited robust catalytic activity in the absence of Tpx2 , and was further activated ~4 fold by the addition of Tpx2 ( Figure 2—figure supplement 1c ) ( Zorba et al . , 2014; Dodson and Bayliss , 2012 ) . Steady-state fluorescence emission spectra were measured for D- and D + A labeled forms of both unphosphorylated and phosphorylated AurA . In either phosphorylation state , titrating ADP or Tpx2 onto the kinase resulted in enhanced fluorescence emission from the donor dye and reduced emission from the acceptor , indicating a decrease in FRET efficiency ( Figure 2—figure supplement 2 ) consistent with a shift towards the DFG-In state . To gain more insight into the conformation of the activation loop and how it is altered by phosphorylation and ligand binding , we performed time-resolved ( TR ) FRET experiments to quantify energy transfer through its effect on the fluorescence lifetime of the donor dye . TR fluorescence decays were recorded using time-correlated single-photon counting ( TCSPC ) ( Figure 2b , top panel ) , and were then fit to a structural model consisting of a Gaussian distribution of inter-fluorophore distances ( Muretta et al . , 2013; Agafonov et al . , 2009; Nesmelov et al . , 2011 ) to represent the ensemble of conformations sampled in solution ( Figure 2b , bottom panels ) . The distance distributions measured for the phosphorylated and unphosphorylated kinase in the absence of ligands are strikingly similar ( Figure 2b , bottom panels ) . In both cases , a broad distribution centered around ~30 angstroms is observed for apo AurA , indicating that the activation loop is highly flexible regardless of the phosphorylation state ( Figure 2b , black ) . This broad distribution is consistent with the DFG-Out state , in which the C-terminal half of the activation loop lacks contacts with the rest of the kinase domain , and is typically disordered in X-ray structures ( Wu et al . , 2013; Coumar et al . , 2009; Fancelli et al . , 2006 ) ( Figure 2a , top panel ) . In contrast , when the phosphorylated and unphosphorylated samples were saturated with both Tpx2 peptide and nucleotide ( either ADP or the non-hydrolysable ATP-analog AMPPNP ) , narrow distributions were observed that were shifted to ~54 angstroms ( Figure 2b , blue ) . This indicates adoption of a well-defined structure consistent with the DFG-In state , in which the segment of the loop containing the labeling site is anchored to the C-terminal lobe of the kinase by flanking β-sheet interactions ( Bayliss et al . , 2003 ) ( Figure 2a , bottom panel ) . In the presence of ADP or AMPPNP alone the observed distance distributions were intermediate in both distance and width between the other samples , consistent with nucleotide binding driving unphosphorylated and phosphorylated AurA into a similar equilibrium between DFG-Out and DFG-In states ( Figure 2b , red ) , as was observed in the IR experiments . We used steady-state fluorescence to measure the equilibrium dissociation constants of ADP and Tpx2 for unphosphorylated and phosphorylated AurA ( Figure 2c ) . Importantly , ADP bound to phosphorylated and unphosphorylated AurA with similar affinities ( Figure 2c , top left panel ) , indicating that the interaction of the kinase with nucleotide is not substantially affected by phosphorylation on the activation loop . This is consistent with our IR and TR-FRET experiments , which show that phosphorylation by itself fails to trigger the long-range conformational change presumably required to couple the phosphorylation site on the activation loop to the distant ATP-binding site . In contrast , Tpx2 , which does trigger a conformational change from the DFG-Out to the DFG-In state in both unphosphorylated and phosphorylated AurA , also enhances the binding affinity of nucleotides in both cases ( Figure 2c , compare top panels ) . While phosphorylation does not affect nucleotide binding to apo AurA , we found that it does substantially enhance the binding of Tpx2 to the exterior surface of the kinase , increasing the affinity by a factor of ~20 ( Figure 2c , bottom panels ) . This is remarkable considering that phosphorylation does not appear to stabilize the DFG-In state , and that there are no direct contacts between the T288 residue and Tpx2 ( Bayliss et al . , 2003 ) . In addition , once Tpx2 is bound , phosphorylation does lead to an enhancement of nucleotide affinity ( Figure 2c , top right panel , compare red and blue ) , indicating that allosteric coupling between the phosphorylation site and the active site , missing in apo AurA , is established in the AurA:Tpx2 complex . These trends in the affinity data are in good agreement with previous enzyme kinetics measurements ( Dodson and Bayliss , 2012 ) . Interestingly , the synergy observed between Tpx2 and phosphorylation is also reflected in our TR-FRET experiments ( Figure 2b ) . A comparison between the unphosphorylated and phosphorylated samples bound to Tpx2 shows that while the unphosphorylated sample requires nucleotide to fully shift to the active state , Tpx2 alone is sufficient to achieve this in phosphorylated AurA , and the further addition of nucleotide has little effect ( Figure 2b , compare yellow and blue ) . The same trend was observed in steady-state FRET experiments ( Figure 2—figure supplement 2c , double-headed arrows ) . Together these data suggest a model in which the allosteric effects of phosphorylation are somehow masked in apo AurA , and only become apparent when Tpx2 switches the kinase to the DFG-In state , at which point phosphorylation further stabilizes this state . While our results reveal synergy between phosphorylation and Tpx2 , they do not answer the key question of how phosphorylation itself activates AurA . Indeed , the IR and FRET data clearly show that phosphorylation on T288 by itself does not cause a substantial shift towards the DFG-In state , and that the phosphorylated kinase , like the unphosphorylated enzyme , instead samples a range of different conformations spanning the DFG-In and DFG-Out states . We hypothesized that phosphorylation must instead drive catalytic activation of AurA by altering the structure and dynamics of the DFG-In subpopulation , presumably allowing it to populate catalytically competent geometries . To provide insight into how phosphorylation alters the structure and dynamics of the DFG-In state , we performed molecular dynamics simulations of the wild-type kinase . Simulations were initiated from the X-ray structure of DFG-In AurA bound to ADP and Tpx2 ( PDB ID: 1OL5 ) ( Bayliss et al . , 2003 ) , and were run in the presence and absence of Tpx2 and with and without phosphorylation on T288 . For each of these four biochemical states , 250 trajectories up to 500 nanoseconds in length were obtained on the distributed computing platform Folding@home , for a total of over 100 microseconds of aggregate simulation time for each biochemical state . Analysis of the DFG conformation revealed that the simulations remained predominantly in their initial DFG-In state ( Figure 3—figure supplement 1 ) , suggesting that the simulation time was insufficient to capture the slow conformational change to the DFG-Out state . The simulations can thus be regarded as probing the conformational dynamics of the DFG-In kinase . The T288 phosphorylation site lies in the C-terminal segment of the activation loop , the correct positioning of which is essential for the binding of peptide substrates ( Figure 3a ) . In the crystal structure used to initiate the simulations , this segment of the loop appears to be stabilized by interactions between the pT288-phosphate moiety and three arginine residues: R180 from the αC helix , R286 from the activation loop , and the highly conserved R255 from the catalytic loop ‘HRD motif’ ( Figure 3a ) ( Bayliss et al . , 2003 ) . To probe the integrity of these interactions in the simulations , and to investigate loop dynamics in their absence , we examined the distribution of distances between the Cζ atoms of either R180 or R255 and the Cα atoms of T288 following equilibration within the DFG-In state ( Figure 3—figure supplement 1b ) . We also tracked the distance between the L225 and S284 Cα atoms ( the sites used for incorporating spectroscopic probes ) to capture movements of the activation loop along a roughly orthogonal axis across the active site cleft . We plotted the simulated L225-S284 distances as a function of the R255-T288 distance to assess how Tpx2 and phosphorylation affect the conformation of the activation loop ( Figure 3b ) . As expected , the simulations of unphosphorylated AurA without Tpx2 show that the activation loop is highly dynamic , reflected as relatively broad distributions of L225-S284 and R255-T288 distances ( Figure 3b , bottom right panel ) . The N-terminal lobe of the kinase was particularly dynamic in these simulations , and local unfolding occurred within the αC-helix in many of the trajectories , as seen previously in simulations of the epidermal growth factor receptor ( Shan et al . , 2012 ) as well as in X-ray structures of the related AGC-family kinase Akt in the unphosphorylated state ( Yang et al . , 2002a ) . In striking contrast , the simulations showed that phosphorylated AurA is locked into a single conformation with a long L225-S284 distance ( 42 Å , cf . 41 Å in the 1OL5 x-ray structure ) and short R255-T288 distance ( ~5 Å , cf . 5 . 5 Å in 1OL5 ) , indicative of a stable active state in which the loop is fully ordered and the phosphothreonine residue forms ion-pairing interactions with R255 and R180 ( Figure 3b , bottom left panel , and Figure 3—figure supplement 1b ) . Interestingly , phosphorylation alone is almost as effective at constraining the loop in the active state as phosphorylation and Tpx2 together ( Figure 3b , left panels ) . In contrast , the simulations show that the activation loop of unphosphorylated AurA bound to Tpx2 remains somewhat dynamic ( Figure 3b , top right panel ) , and additional phosphorylation significantly stabilizes the loop . Although unphosphorylated AurA is highly dynamic in the absence of Tpx2 , the activation loop is not in fact disordered in the simulations . Instead , two discrete subpopulations are observed: one subpopulation corresponding to the active-like state with a long L225-S284 distance ( ~43 Å ) , and another with a much shorter distance ( ~38 Å ) , representing a DFG-In conformation in which the activation loop is not correctly positioned for catalytic function ( Figure 3b , lower right panel ) . Manual inspection of the trajectories revealed that in this subpopulation the tip of the activation loop folds into a short helical turn spanning residues P282-R286 , with the P282 proline residue serving as the N-terminal capping residue ( Kumar and Bansal , 1998 ) in most of the trajectories ( Figure 3c ) . Calculating the pseudodihedral angle for the Cα atoms of S283-R286 across all trajectories confirmed that the inactive subpopulation possessed well-defined helical pseudodihedral values of 50–75° ( Figure 3d ) . Although this conformation has not been observed in X-ray structures of AurA , the formation of short helices in the activation loop is a common feature of the inactive states of other protein kinases ( Sicheri et al . , 1997; Wood et al . , 2004; Lee et al . , 2010; De Bondt et al . , 1993 ) . An interesting feature of the autoinhibited DFG-In substate observed in the simulations is that the T288 residue , which immediately follows the helical segment in the protein sequence , is positioned close to the C-terminal end of the helix in almost all of the trajectories ( Figure 3c ) , with the sidechain hydroxyl forming hydrogen bonds to the backbone carbonyls of residues R285 and R286 in many of the simulation snapshots . We reasoned that upon phosphorylation of T288 , the proximity of the phosphate group to the negatively-charged end of the helix dipole ( Hol et al . , 1978 ) would destabilize this autoinhibited substate , promoting the refolding of the activation loop to the active conformation . We wondered why the helical conformation of the activation loop has not been observed in X-ray structures of AurA . In fact , the activation loop adopts the active conformation in only a small subset of AurA structures , specifically those determined either in the presence of Tpx2 ( Bayliss et al . , 2003; Zhao et al . , 2008; Clark et al . , 2009 ) or other protein factors that stabilize the active state ( Richards et al . , 2016 ) . Instead , almost all of the structures of AurA in the DFG-In state ( 76 structures out of 138 total structures of AurA in the PDB ) were determined in the same hexagonal crystal form in which the kinase adopts an inactive conformation with the activation loop misaligned and the peptide binding site disassembled . Upon examination of the crystal lattice we noticed that this conformational state of the activation loop appears to be induced by a crystal contact between the peptide binding site and a neighboring molecule in the lattice ( Figure 3—figure supplement 2 ) . This apparent crystallographic artifact may have prevented previous observation of the helical autoinhibited DFG-In substate visualized in our simulations , which model the kinase in solution rather than in the crystallographic context . Our MD simulations , which represent over a millisecond of simulation data , predict that phosphorylation has profound effects on the activation-loop conformation of AurA within the DFG-In state , both disrupting an autoinhibited substate and promoting an active substate that is primed for catalytic function . In an attempt to confirm the simulation result that the T288 phosphothreonine residue of phosphorylated AurA is correctly coordinated in the active state , we mutated its ion-pairing partner R180 to an alanine residue and measured the effect on kinase activity . The R180A mutant possessed 4-fold lower activity in the absence of Tpx2 , whereas the activity in the presence of Tpx2 was only modestly affected ( Figure 3e ) . This is consistent with the catalytic activity of phosphorylated AurA arising from a population of molecules adopting the canonical active state in which the phosphothreonine residue is correctly ion-paired . To assess the effects of phosphorylation on the structure of the DFG-In state predicted by our simulations , we used double electron-electron resonance ( DEER ) , a pulsed electron paramagnetic resonance ( EPR ) spectroscopy technique ( Jeschke , 2012 ) . DEER experiments measure the dipole-dipole interactions of unpaired electron spins to provide high-resolution information about the distribution of spin-spin distances in the sample . Two MTSL spin labels were incorporated into AurA at the same positions used for FRET experiments ( L225C and S284C ) . Labeling and phosphorylation were confirmed by mass spectrometry , and MTSL-labeled samples retained close to full kinase activity ( Figure 4—figure supplement 1 ) . Samples were flash frozen in the presence of saturating concentrations of nucleotides and/or Tpx2 , and DEER experiments were performed at 65 K . We first sought to confirm our above results that the phosphorylated kinase still samples both the DFG-Out and DFG-In states . DEER spectra ( background-corrected dipolar evolution data ) take the form of a damped oscillating signal in which the mean distance between the spin labels is encoded in the cube root of the oscillation period , and the width of the distance distribution is encoded in the degree of damping . DEER spectra acquired for the phosphorylated kinase bound to ADP or AMPPNP showed a rapidly decaying and heavily damped signal , consistent with the activation loop adopting multiple conformations ( Figure 4a and Figure 4—figure supplement 2a , blue lines ) . Extraction of spin-spin distances from the DEER spectra using Tikhonov regularization ( Chiang et al . , 2005 ) confirmed the presence of a broad distribution of distances , spanning ~35 to~55 angstroms ( Figure 4b , blue ) . Control experiments with the model peptide substrate kemptide , which conforms to the consensus phosphorylation site sequence for AurA ( Ferrari et al . , 2005 ) , showed no effect on the conformational ensemble of the phosphorylated kinase ( Figure 4—figure supplement 2b ) . However , DEER spectra of phosphorylated AurA bound to Tpx2 decayed much less rapidly , indicating increased spin-spin distance , and exhibited pronounced oscillations , indicating a high degree of structural order ( Figure 4a , yellow ) . The corresponding distance distribution displayed a single dominant peak at 52 angstroms ( Figure 4b , yellow ) , representing a ~5 angstrom longer mean spin-spin distance than seen in the samples lacking Tpx2 , consistent with the activation loop now adopting the extended DFG-In conformation ( see Figure 2a ) . To bolster these DEER experiments we performed molecular dynamics simulations of MTSL-labeled phosphorylated AurA in either the DFG-Out state ( PDB ID: 5L8K ) or the fully-active DFG-In state ( PDB ID: 1OL5 , AurA bound to Tpx2 ) , totaling 75–110 microseconds of aggregate simulation data for each state . In these simulations , sampling of different spin label rotamers , combined with motion of the protein , gives rise to a range of predicted spin-spin distances . Although the simulated distributions for the DFG-In and DFG-Out states overlap , the DFG-In distribution is skewed to longer distances: distances beyond 45 angstroms are more populated in the DFG-In simulations , and distances beyond 50 angstroms are almost exclusively associated with the DFG-In state ( Figure 4c ) . Prominent peaks are present in the DFG-In distribution at ~44 , ~48 and~52 angstroms ( Figure 4c ) . The 44- and 52-angstrom predicted spin-spin distances are also observed in the DEER experiment performed with phosphorylated AurA bound to Tpx2 ( Figure 4b , yellow ) , whereas the 48-angstrom peak was not observed experimentally; presumably the corresponding rotamer state is too sparsely populated to be detected at the low temperature of the DEER experiment . The simulations nonetheless confirm that the 52-angstrom spin-spin distance , which is so prominent in the experiment performed in the presence of Tpx2 , almost certainly arises from the DFG-In state , and underscore that in the absence of Tpx2 phosphorylated AurA occupies a mixture of DFG-In and DFG-Out states , consistent with the IR and FRET experiments above . We next turned to assessing how phosphorylation alters the structure of the DFG-In state . DEER experiments performed on unphosphorylated AurA bound to ADP gave qualitatively similar results to those obtained with the phosphorylated kinase bound to ADP , with a broad distribution of spin-spin distances consistent with similar populations of DFG-In and DFG-Out states ( Figure 4b , compare red and blue ) . However , close inspection of the DEER spectra revealed subtle differences between the unphosphorylated and phosphorylated samples ( Figure 4a inset ) , and the corresponding distributions indicated that distances beyond 50 angstroms were more populated in the phosphorylated sample ( Figure 4b , dark blue shading ) . The same trend was observed in experiments performed with AMPPNP instead of ADP ( Figure 4—figure supplement 2a ) . We wondered whether the presence of a substantial DFG-Out subpopulation in these samples might be obscuring a more dramatic structural change occurring in the DFG-In subpopulation . To test this , we used the ATP-competitive AurA inhibitor SNS-314 , which preferentially binds to the DFG-In state of AurA ( Oslob et al . , 2008 ) ( Figure 4d ) , to induce a homogeneous population of DFG-In kinase . Strikingly , DEER spectra measured on unphosphorylated and phosphorylated AurA bound to SNS-314 showed pronounced oscillations with differing periods , indicating that both samples adopt well-defined structures , but that the major spin-spin distances are different in the two cases ( Figure 4e ) . Indeed , the Tikhonov distributions showed sharp peaks at 49 and 53 angstroms for the unphosphorylated and phosphorylated kinase , respectively ( Figure 4e , inset ) . Contrasting these results with those obtained with ADP ( Figure 4b ) highlights that the binding of SNS-314 stringently enforces adoption of the DFG-In state and dramatically simplifies the conformational ensemble . Importantly , the resulting unobstructed view of the DFG-In state shows that phosphorylation does indeed lead to a pronounced structural change of the activation loop , reflected in the increase in spin-spin distance from 49 to 53 angstroms shown in Figure 4e . The distance distribution obtained for phosphorylated AurA bound to SNS-314 ( Figure 4e , blue ) is very similar to that observed in the presence of Tpx2 ( Figure 4b , yellow ) , confirming that these longer spin-spin distances ( 52–53 Å ) arise from the catalytically active DFG-In conformation , whereas the shorter spin-spin distance observed with unphosphorylated AurA bound to SNS-314 ( ~49 Å ) corresponds to a structurally-distinct DFG-In conformation . This phosphorylation-driven structural change was also detected in a separate set of DEER experiments in which the S284 labeling site was moved to S283 , and SNS-314 again used to isolate the DFG-In state ( Figure 4—figure supplement 3 ) . Interestingly , the 4-angstrom increase in spin-spin distance observed in the DEER experiments performed with the L225/S284 labeling sites ( Figure 4e inset ) is similar to the difference in the L225-S284 Cα distances predicted by the MD simulations for the autoinhibited and active DFG-In substates ( Figure 3b , bottom right panel ) . The DEER experiments thus support the model that phosphorylation triggers a switch from an autoinhibited DFG-In substate to the active DFG-In substate ( see Figure 3c ) . We conclude that while phosphorylated AurA samples the DFG-Out and DFG-In states to a similar extent as the unphosphorylated kinase , the structure and dynamics of the DFG-In subpopulation are profoundly altered by phosphorylation , leading to catalytic activation . Our proposed model for the autoinhibited DFG-In substate of AurA provides an explanation for several puzzling observations . Firstly , the presence of a third state explains how phosphorylation can promote the active state without triggering a shift in the DFG equilibrium , the central result of this paper . By eliminating the autoinhibited DFG-In substate , phosphorylation redistributes the ensemble between the DFG-Out and active DFG-In substates ( Figure 4f ) . The fact that this does not substantially change the DFG equilibrium suggests that the stabilizing ionic interactions between the phosphothreonine and the arginine residues ( see Figure 3a ) are offset by an energetic penalty associated with refolding the activation loop into the active configuration . Secondly , our model accounts for the synergy between phosphorylation and Tpx2 , observed in our fluorescence experiments , and reported previously ( Dodson and Bayliss , 2012; McIntyre et al . , 2017 ) . By unfolding the activation loop from the autoinhibited DFG-In state , phosphorylation promotes the formation of the Tpx2-interaction surface by the N-terminal segment of the activation loop , resulting in dramatically enhanced binding affinity . By binding tightly to this surface , Tpx2 in turn compensates for the energetic cost of reconfiguring the activation loop , allowing the effects of phosphorylation to be manifested as a further stabilization of the active DFG-In state , as we observed in our fluorescence experiments in terms of an additional conformational shift and enhanced nucleotide binding . While definitive confirmation of the structural model for the autoinhibited DFG-In state awaits further experiments , our DEER data are fully consistent with this model , which provides a mechanism for the activation of AurA by phosphorylation that accounts for the available structural , dynamic and biochemical data . The majority of eukaryotic protein kinases are activated by phosphorylation on the activation loop at a site equivalent to T288 in AurA ( Johnson et al . , 1996 ) . X-ray structures have suggested that the functional role of this phosphorylation is to trap the kinase in an active DFG-In state and rigidify the flexible activation loop in a specific configuration that promotes catalysis and substrate binding ( Knighton et al . , 1991; Yamaguchi and Hendrickson , 1996; Steichen et al . , 2012 ) . Our results show that phosphorylation can drive catalytic activation of a protein kinase without restraining the protein in the DFG-In state , providing a contrasting and highly dynamic view of an activated kinase in which major conformational changes of catalytic elements may occur continuously during the catalytic cycle . Of note , a recent single-molecule fluorescence study also reported that phosphorylated AurA dynamically transitions between multiple structural states ( Gilburt et al . , 2017 ) . Binding of Tpx2 to unphosphorylated AurA causes a pronounced population shift from the DFG-Out to the DFG-In state ( Cyphers et al . , 2017 ) , in striking contrast with the phosphorylation-mediated activation mechanism described here . Our simulation data also reveal differences in how phosphorylation and Tpx2 affect the DFG-In subpopulation , with Tpx2 less effective at constraining the C-terminal segment of the activation loop . Thus it appears that phosphorylation and Tpx2 activate AurA through quite different - albeit complementary - mechanisms , with phosphorylation triggering a structural switch within the DFG-In subpopulation , and Tpx2 instead promoting a DFG flip to increase the population of the DFG-In state . Although we have not explicitly tracked the dynamics of the important αC-helix in this work , our previous work showed that the binding of Tpx2 substantially stabilized the αC-helix and the associated regulatory spine ( Cyphers et al . , 2017 ) . Given that the R180 residue on the αC-helix directly recognizes the pT288 phosphothreonine residue , and that the R180A mutation interferes with activity of the phosphorylated enzyme , phosphorylation is presumably also coupled to the αC-helix . However , while this coupling likely contributes to the switch from the autoinhibited DFG-In substate to the active DFG-In substate , our spectroscopic experiments show , surprisingly , that these interactions are not sufficiently stabilizing to substantially increase the overall population of the DFG-In state . Thus , phosphorylation can be thought of as tuning the free energy surface for the DFG-In state - favoring the active substate over the autoinhibited substate - as opposed to changing the relative free energies of the DFG-In and DFG-Out states . It is interesting to consider the dual activation mechanisms of AurA in the light of the closely-related AGC-family kinases . The AGC kinases possess a C-terminal hydrophobic motif that docks in cis onto a pocket above the αC-helix in a manner that closely resembles the interaction of Tpx2 with AurA ( Frödin et al . , 2002; Yang et al . , 2002b ) . Activation of the AGC kinases by activation loop phosphorylation and hydrophobic motif engagement are tightly coupled events ( Alessi et al . , 1996 ) and are both important for activation ( Batkin et al . , 2000 ) . The contrasting ability of AurA to be independently activated by phosphorylation or Tpx2 is dependent upon a unique active-site water network that strengthens the regulatory spine of the kinase relative to that of the AGC kinases ( Cyphers et al . , 2017 ) . Interestingly , when Tpx2 and phosphorylation act together , they are capable of overriding the deleterious effects of mutations in the regulatory spine designed to disrupt the water network and render the spine more like that of an AGC kinase ( Cyphers et al . , 2017 ) . Presumably the synergistic effects of Tpx2 and phosphorylation on the conformational dynamics of the kinase , in which the αC-helix is stabilized by Tpx2 engagement and the activation loop is locked into the active DFG-In conformation , renders the additional stabilization provided by the water network redundant . It is likely that the fully-activated AGC kinases , with their hydrophobic motifs engaged on the αC-helix and their activation loops phosphorylated , closely resemble this conformationally-rigid form of AurA . In contrast , the highly dynamic nature of AurA activated only by phosphorylation may be a unique property of the Aurora kinases that reflects the loss of the hydrophobic motif in this lineage . These motions may facilitate further regulation of AurA by additional cellular factors , allowing for graded levels of catalytic activity . For instance , phosphorylated AurA has been shown to interact with Cep-192 ( Joukov et al . , 2014 ) , Bora ( Macůrek et al . , 2008; Seki et al . , 2008 ) , and Ajuba ( Hirota et al . , 2003 ) at the centrosome , and these interactions can further regulate AurA activity towards specific substrates . Although a model peptide substrate did not modulate AurA dynamics in our experiments , in the context of the higher-order signaling complexes found at the centrosome substrates may be presented at sufficiently high local concentration to further stabilize the activation loop in the active conformation and promote phosphoryl transfer . In this context it is noteworthy that the oncogenic transcription factor N-Myc binds to AurA through a pseudosubstrate interaction , and does indeed appear to stabilize the phosphorylated kinase in the active conformation ( Richards et al . , 2016 ) . These observations are consistent with the overarching view that protein kinases evolve under strong selective pressure to optimize their responsiveness to regulatory inputs , and not for maximal catalytic efficiency . It is unclear whether activation of AurA by both phosphorylation and Tpx2 occurs in normal cells , where the Tpx2-bound spindle pool of AurA is thought to be predominantly unphosphorylated ( Zeng et al . , 2010; Toya et al . , 2011 ) , and the centrosomal pool is not bound to Tpx2 ( Kufer et al . , 2002 ) . The doubly-activated form of the kinase is prominent , however , in the tumors of ~10% of melanoma patients , where mutational inactivation of the PP6 phosphatase leads to accumulation of phosphorylated AurA bound to Tpx2 on the mitotic spindle , resulting in chromosome instability and DNA damage that can be partially reversed by AurA inhibitors ( Hammond et al . , 2013; Hodis et al . , 2012; Gold et al . , 2014 ) . The distinct conformational dynamics of doubly-activated AurA might provide opportunities for the development of improved inhibitors that selectively target this form of the kinase in melanoma cells . The DFG flip has long been considered one of the key regulatory mechanisms used by nature to control the catalytic activity of protein kinases ( Hubbard et al . , 1994; Nagar et al . , 2003 ) , but the difficulty of directly observing this structural change in solution and correlating it with activity has hampered efforts to conclusively demonstrate its regulatory role . Although AurA does adopt the DFG-Out state , our results show that activation of the enzyme by phosphorylation is not mediated by a DFG flip , but rather by inducing activating conformational changes within the DFG-In state . It is noteworthy that many other protein kinases , including non-receptor and receptor tyrosine kinases ( Wood et al . , 2004; Xu et al . , 1997 ) and the cyclin-dependent kinases ( De Bondt et al . , 1993 ) , employ an autoinhibitory DFG-In state , as opposed to the DFG-Out state , for regulatory control . A major remaining question is whether the substantial DFG-Out subpopulation of phosphorylated AurA performs an important biological function , and whether this is a unique feature of AurA itself , or a more general property of activated protein kinases . Aurora A kinase domain constructs were expressed in E . coli and purified as previously described ( Cyphers et al . , 2017 ) . We used a Cys-lite form of AurA ( C290A C393S ) that has been previously shown to possess robust kinase activity within 2-fold of WT AurA , and can be purified from bacteria in homogeneously T288-phosphorylated form ( Burgess and Bayliss , 2015; Rowan et al . , 2013 ) . For IR and EPR experiments it was necessary to mutate an additional cysteine ( C247 ) to alanine to avoid non-specific labeling . Site-directed mutagenesis was performed using the QuikChange Lightning kit ( Agilent ) . Phosphorylation was verified by mass spectrometry and activity assays . Phosphorylated Q185C AurA protein samples ( human AurA residues 122–403 containing an N-terminal hexahistidine tag ) were prepared using a cysteine-light form of the kinase in which all endogenous solvent-accessible cysteines were removed by mutagenesis ( Q185C , C247A , C290A , C393S ) . After Nickel-affinity purification , repeated rounds of cation exchange chromatography were used to isolate the homogeneous singly phosphorylated species , with enrichment of the phosphorylated species tracked during purification by western blotting and activity assays ( Figure 1—figure supplement 1 ) . Nitrile labeling of the purified protein was performed using a 1 . 5:1 molar ratio of DTNB ( Ellman’s reagent ) , followed by 50 mM KCN , and excess labeling reagents were removed using a fast desalting column ( GE Healthcare ) . Incorporation of a single nitrile label was confirmed by whole-protein mass spectrometry ( Figure 1—figure supplement 1 ) . Samples for IR spectroscopy were prepared by concentrating labeled protein ( 50–100 μM ) in the presence or absence of 4 mM ADP and 8 mM MgCl2 , and/or excess Tpx2 peptide ( ~150 μM , residues 1–43 of human Tpx2 , Selleckchem ) in FTIR buffer ( 20 mM Hepes , pH 7 . 5 , 300 mM NaCl , 20% glycerol ) . Samples were concentrated to ~1 mM and loaded into a calcium fluoride sample cell mounted in a temperature-controlled housing ( Biotools ) for IR experiments . IR spectra were recorded on a Vertex 70 FTIR spectrometer ( Bruker ) equipped with a liquid nitrogen cooled indium antimonide detector with 2 cm−1 spectral resolution . Spectra were averaged across 256 scans , background subtracted using absorbance spectra of the buffer flow-through from sample concentration , and baselined using the polynomial method in the OPUS software ( Bruker ) . Kinase activity was measured using the ADP Quest coupled kinase assay ( DiscoverX ) in a fluorescence plate reader ( Tecan Infinite M1000 PRO ) as described previously ( Cyphers et al . , 2017 ) . Assays were performed using 2 , 5 , 10 , 100 , or 200 nM kinase ( depending on the protein variant ) , 1 mM kemptide peptide substrate ( Anaspec ) , 10 μM Tpx2 residues 1–43 ( Selleckchem ) , and 50 μM ATP ( Sigma Aldrich ) . Activity was determined by fitting the fluorescence intensity as a function of time using linear regression . Background ATPase activity was determined using samples lacking peptide substrate , and was subtracted from the activity in the presence of substrate peptide . The average fluorescence for several ADP concentrations in the dynamic range of the assay was used to construct a calibration curve , and convert the background-corrected activity to ATP turnover numbers . Activities given are the average of three experiments , where error bars are the standard deviation of the replicates . For FRET experiments , the AurA variant C290S/A C393S L225C S284C was expressed , purified and labeled with donor ( Alexa 488 , Invitrogen ) and acceptor ( Alexa 568 ) using cysteine-maleimide chemistry , as previously described ( Cyphers et al . , 2017 ) . Labeled samples were validated using activity assays and mass spectrometry and retained close to full kinase activity ( Figure 2—figure supplement 1 ) . Ligand titration FRET experiments were performed in a Fluoromax 4 Spectrofluorometer ( Horiba ) at 22°C . Assays were performed in 15 mM HEPES pH 7 . 5 , 20 mM NaCl , 1 mM EGTA , 0 . 02% Tween-20 , 10 mM MgCl2 , 1% DMSO and 0 . 1 mg/mL bovine-γ-globulins at AurA concentrations of 5–50 nM . Bulk FRET efficiency and inter-fluorophore distance were calculated from the ratios of the donor fluorescence in the presence and absence of acceptor , assuming a value of 62 angstroms for the Förster radius , as previously described ( Cyphers et al . , 2017 ) . The steady-state anisotropy was below 0 . 2 for all samples and was similar across biochemical conditions ( Figure 2—figure supplement 2b ) . Dissociation constants KD were determined using the spectra obtained with the D + A labeled sample . The ratio FD/FA ( where FD is the donor fluorescence maximum and FA is the acceptor fluorescence maximum ) , a highly sensitive measure of ligand binding , was fit as a function of ligand concentration . For calculation of the KD for Tpx2 with saturating ADP bound , KD is near the concentration of fluorescent protein . Ligand depletion was accounted for by fitting the raw data to determine the plateau FD/FA value ( representing ligand saturation ) , calculating the percent saturation for each total ligand concentration , and then back calculating the concentration of free ligand . FD/FA was then re-fit as a function of free ligand concentration . The time-correlated single photon counting instrument used to collect time-resolved fluorescence data has been previously described ( Agafonov et al . , 2009 ) . The instrument response function was obtained with the emission polarization set at vertical , while fluorescence data were collected with the emission polarization set at 54 . 7o , with a GFP band pass filter in place ( Semrock ) . Experimental buffer contained 15 mM HEPES pH 7 . 5 , 20 mM NaCl , 1 mM EGTA , 0 . 02% Tween-20 , 10 mM MgCl2 . Experiments were performed at 100–200 nM unphosphorylated or phosphorylated FRET-labeled AurA , in the presence and absence of 100–125 μM Tpx2 and 0 . 2–1 mM ADP or AMPPNP; one phosphorylated AurA experiment also contained 1 mM DTT . For both unphosphorylated and phosphorylated AurA , three independent experiments were performed and analyzed . Data fitting was performed as previously described ( Muretta et al . , 2013 ) . Briefly , time-resolved fluorescence waveforms were fit using custom software designed for analysis of time-resolved fluorescence ( Muretta et al . , 2010 ) . The instrument response function and the model of the fluorescence decay were convolved to fit the measured time-resolved fluorescence waveform . Donor-only fluorescence waveforms were modeled using a multiexponential decay function , which accounts for the intrinsic lifetimes of the dye , with two exponentials required to fit the Alexa 488 fluorescence decay . Donor +acceptor ( D + A ) waveforms were modeled from the amplitudes and lifetimes present in the matched donor-only sample and modified so that a distance-dependent resonance energy transfer term , corresponding to a Gaussian distribution of inter-probe distances , describes the decrease in fluorescence lifetime relative to the donor-only control . The mean distance and full-width half maximum of the Gaussian functions were fit individually for each D + A and D-only pairing , while the parameters that described general conditions of the experiment common among all samples , such as the fraction of a given D + A sample lacking acceptor dye , were globally linked . DEER samples were prepared using the Cys-lite mutant construct of AurA C290S/A C393S C247A L225C S284C ( or S283C ) , purified as described above . AurA was labeled with MTSL ( Santa Cruz Biotechnology ) , purified by cation exchange chromatography , and concentrated . Labeling was verified using mass spectrometry , and MTSL-labeled samples retained close to full activity of unlabeled AurA in the presence and absence of Tpx2 ( Figure 4—figure supplement 1 ) . The protein was then buffer exchanged into the experimental buffer , which was 20 mM HEPES pH 7 . 5 , 300 mM NaCl , 10% deuterated glycerol , 2% v/v H2O in D2O . For DEER experiments , samples containing 30–60 μM MTSL-labeled AurA were prepared in the presence and absence of 100–200 μM Tpx2 and 300 μM ADP ( 8 mM MgCl2 was added to samples containing ADP ) . Final samples varied in v/v H2O concentration from 2–14%; however , no significant differences were observed in Tikhonov distributions derived from experiments performed in protonated and deuterated buffers . Samples were flash-frozen in an isopropanol dry ice bath followed by liquid nitrogen . Data shown are from one of two replicate experiments . DEER spectra were detected at 65 K using an Elexsys E580 Q-Band spectrometer ( Bruker Biospin ) equipped with an ER 5107D2 resonator ( Bruker Biospin ) using the standard 4-pulse pulse sequence with π/2 and π pulses ( including ELDOR ) set to 16 and 32 ns respectively . The pump frequency was set to the central resonance position of the nitroxide echo-detected field swept spectrum while the observe position was set 24 G up-field to avoid excitation bandwidth overlap ( Agafonov et al . , 2009 ) . Data were analyzed using custom software written in Mathematica which was based heavily on DeerAnalysis 2017 ( Jeschke et al . , 2006 ) . The raw spectra were phase and background corrected assuming a homogeneous background model to produce the DEER waveform . Distance distributions were determined using Tikhonov regularization , with an optimal smoothing parameter chosen using a combination of the l-curve and leave one out cross validation ( LOOCV ) techniques ( Edwards and Stoll , 2016 ) . After choice of smoothing parameter , a range of background fits were performed to identify stable populations in the distance distributions , with highly unstable , long distance populations being largely attributable to errors in the background fit and model choice ( Jeschke , 2011 ) .	The transfer of phosphate groups onto proteins ( protein phosphorylation ) is one of the most important methods used to send signals inside cells . The enzymes that catalyze this process , called protein kinases , are themselves controlled by the phosphorylation of a flexible region called the activation loop . For many years it had been thought that the purpose of activation loop phosphorylation was to clamp the otherwise flexible activation loop in an active state that allows molecules that need to be phosphorylated to bind to the kinase . This assumption was based on static pictures of protein kinases obtained by X-ray crystallography , in which individual states are trapped and visualized in a crystal lattice . However , new methods and approaches now mean it is possible to visualize how the position of the activation loop changes as it moves in solution . By applying these techniques , Ruff et al . show that the static model is incorrect in a protein kinase called Aurora A . In this enzyme , the phosphorylated activation loop continues to switch back and forth between active and inactive states . Phosphorylation instead enhances the catalytic activity of the active state . Aurora A regulates several important steps in cell division , and plays important roles in several kinds of cancer . The discovery that activated forms of Aurora A can have different dynamic properties raises the possibility that inhibitor molecules could be designed to exploit these differences and block specific activities of Aurora A in cancer cells . To realize this goal we need to better understand how a kinase switching between active and inactive states affects the ability of inhibitors to interact with it .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "computational", "and", "systems", "biology" ]	2018	A dynamic mechanism for allosteric activation of Aurora kinase A by activation loop phosphorylation
We hypothesized that human genes and disease-associated alleles might be systematically functionally annotated using morphological profiling of cDNA constructs , via a microscopy-based Cell Painting assay . Indeed , 50% of the 220 tested genes yielded detectable morphological profiles , which grouped into biologically meaningful gene clusters consistent with known functional annotation ( e . g . , the RAS-RAF-MEK-ERK cascade ) . We used novel subpopulation-based visualization methods to interpret the morphological changes for specific clusters . This unbiased morphologic map of gene function revealed TRAF2/c-REL negative regulation of YAP1/WWTR1-responsive pathways . We confirmed this discovery of functional connectivity between the NF-κB pathway and Hippo pathway effectors at the transcriptional level , thereby expanding knowledge of these two signaling pathways that critically regulate tumor initiation and progression . We make the images and raw data publicly available , providing an initial morphological map of major biological pathways for future study . The dramatic increase in human genome sequence data has created a significant bottleneck . The number of genes and variants known to be associated with most human diseases has increased dramatically ( Amberger et al . , 2015 ) . Unfortunately , the next step - understanding the function of each gene and the mechanism of each allele in the disease - typically remains non-systematic and labor-intensive . Most commonly , researchers painstakingly design , develop , and apply a disease-specific or biological process-specific assay . Over 30% of genes in the human genome are of unknown function ( Leonetti et al . , 2016 ) and even annotated genes have additional functions yet to be uncovered . Furthermore , even when a gene’s normal functions are known , methods are lacking to predict the functional impact of the millions of genetic variants found in patients . These gaps must be filled in order to convert the promise of human genome sequence data into clinical treatments . Therefore , there is a widespread need for systematic approaches to functionally annotate genes and variants therein , regardless of the biological process or disease of interest . One general approach depends on guilt-by-association , linking unannotated genes to annotated ones based on properties such as protein-protein interaction data , sequence similarity , or , most convincingly , functional similarity ( Shehu et al . , 2016 ) . In the latter category are profiling techniques , where dozens to hundreds of measurements are made for each gene perturbation and the resulting profile is compared against profiles for annotated genes . Various data sources can be used for profiling; gene expression is one that can be performed in relatively high-throughput and it has been proven useful in predicting gene function ( Lamb et al . , 2006 ) . In fact , high-throughput mRNA profiles were recently used to cluster alleles found in lung adenocarcinoma based on their functional impact , a precursor to therapeutic strategy for variants of previously unknown significance ( Berger et al . , 2016 ) . Images are a less mature data source for profiling but show tremendous promise . Morphological profiling data is complementary to transcriptional profiling data ( Wawer et al . , 2014 ) and is less expensive . Morphological profiling has succeeded across several applications , including grouping small-molecule perturbations based on their mechanism of action ( Caicedo et al . , 2016; Bougen-Zhukov et al . , 2017 ) , and grouping genes based on morphological profiles derived from cells perturbed by RNA interference ( RNAi ) ( Mukherji et al . , 2006; Boutros and Ahringer , 2008; Fuchs et al . , 2010; Pau et al . , 2013 ) . One limitation of RNAi for morphological profiling is that the number of measurements must be limited or else the resulting profiles are dominated by off-target effects , especially seed effects ( Singh et al . , 2015 ) . Some computational solutions have shown some promise in overcoming this problem for gene expression profiling ( Schmich et al . , 2015 ) , but their utility is unproven for image-based profiling , and regardless RNAi does not permit analysis of gene variants , only knockdown . Modification of genes via CRISPR will require new libraries of reagents and is as yet untested in morphological profiling . In the proof-of-concept work presented here , we tested morphological profiling using overexpression in human cells as a general approach to annotate gene and allele function . We profiled a reference series of well-known genes , and a small number of variants thereof , by Cell Painting . In particular , we wondered whether the information content of this strategy would outweigh potential limitations ( e . g . , due to cellular context or expression level ) . We found that the approach successfully clustered genes and alleles based on functional similarity , revealed specific morphological changes even when present in only a subpopulation of heterogeneous cells , and uncovered novel functional connections between important biological pathways . To profile each exogenously expressed gene ( or allele therein ) , we used our previously developed image-based profiling assay , called Cell Painting ( Gustafsdottir et al . , 2013; Bray et al . , 2016 ) . This microscopy-based assay consists of six stains imaged in five channels and revealing eight cellular components: DNA , mitochondria , endoplasmic reticulum , Golgi , cytoplasmic RNA , nucleoli , actin , and plasma membrane ( Figure 1A ) . In five replicates in 384-well plate format , we infected U-2 OS cells ( human bone osteosarcoma cells ) , chosen for their flat morphology and previous validation in the assay , with an arrayed ‘reference’ expression library of 323 open reading frame ( ORF ) constructs of partially characterized functions ( Supplementary file 1A ) , a subset of which have been previously described ( Kim et al . , 2016 ) . Of these , we prioritized analysis of the 220 constructs that were most closely representative of the annotated full length transcripts ( see Materials and methods ) . Morphological profiles were extracted using CellProfiler for image processing , yielding 1384 morphological features per cell , and Python/R scripts for data processing , including feature selection and dimensionality reduction ( Figure 1B , and see Materials and methods ) . This computational pipeline yielded a 158-dimensional profile for each of 5 replicates for each gene or allele tested . Not all genes are likely to impact cellular morphology given the limitation of our experiment; using a single cell line at a single time point under a single set of conditions and stained with six fluorescent labels . We therefore first asked what fraction of these ORFs impacted morphology . Surprisingly , we found that 50% ( 110/220 ) of these ORF constructs induced reproducible morphological profiles distinct from negative control profiles ( Figure 2A , and see Materials and methods ) . Next , we ruled out the possibility that position artifacts may have artificially inflated this result by taking an alternative pessimistic null distribution which takes well position into account ( Figure 2—figure supplement 1 ) . Therefore , we conclude that a single ‘generic’ morphological profiling assay can detect signal from a substantial proportion of genes in our reference set . We next turned to testing whether those signals are biologically meaningful and can lead to novel , unbiased discoveries about gene function . 10 . 7554/eLife . 24060 . 003Figure 1 . Morphological profiling by Cell Painting . ( A ) Example Cell Painting images from each of the five channels for a negative control sample ( no gene introduced ) . ( B ) From left to right: Cell and nucleus outlines found by segmentation in CellProfiler; raw profiles ( 2769 dimensional ) containing median and median absolute deviation of each of 1384 measurements over all the cells in a sample , plus cell count; processed profiles which are made less redundant by feature selection and Principal Component Analysis; dendrogram constructed based on the processed profiles ( see Figure 3 ) . Replicates are merged to produce a profile for each gene which is then compared against others in the experiment to look for similarities and differences . DOI: http://dx . doi . org/10 . 7554/eLife . 24060 . 00310 . 7554/eLife . 24060 . 004Figure 2 . Morphological profiles are sensitive and reproducible , and show expected relationships . ( A ) 50% of the gene overexpression constructs produced a detectable phenotype by image-based profiling . Constructs yielding a reproducible phenotype ought to have a median correlation among replicates that is higher than the 95th percentile of correlations seen for pairs of different constructs; this is true for 51% ( 112 out of 220 ) of the constructs ( as shown ) . Additionally , we removed two constructs that passed that filter but whose profiles were highly similar to negative control profiles ( not shown ) , leaving 110 constructs ( 50% ) for further analysis . ( B ) Of wild-type ORF pairs that both yielded a distinguishable phenotype , 96% showed significant correlation to each other . Correlations between the 23 pairs of constructs that are clones of the same gene ( although with potential sequence variation or possibly different isoforms ) were almost always much higher than correlations between pairs of constructs related to different genes . The threshold , shown as the dashed line , is set to 95th percentile of profile correlation for pairs of different genes . Profile correlation of these 23 pairs lie above the threshold . ( C ) Genes in pathways thought to regulate morphology were more likely to yield detectable phenotypes vs . the remainder of genes in the experiment . The same cutoff as in ( A ) is used to identify percentage of genes with a detectable phenotype . This percentage is 87% for the genes hypothesized to change morphology , while it is 48% for the other genes . DOI: http://dx . doi . org/10 . 7554/eLife . 24060 . 00410 . 7554/eLife . 24060 . 005Figure 2—figure supplement 1 . Position artifacts do not contribute to the hit rate seen in the experiment . We were concerned that position artifacts may result in overestimating the replicate correlations because replicates of the same treatment are assigned to the same well location across different plates in the experiment ( that is , it was infeasible to scramble well locations ) . We ruled out this possibility by taking an alternative pessimistic null distribution which takes well position into account . In contrast to Figure 2A , which shows a 51% hit rate , a more pessimistic alternative null distribution is shown here ( left ) , calculated based on the replicate correlation of pairs of negative controls in the same position only . We consider this less reliable because the number of such pairs is small ( 26 ) and we excluded edge wells; nevertheless the hit rate increases slightly , to 60% . DOI: http://dx . doi . org/10 . 7554/eLife . 24060 . 00510 . 7554/eLife . 24060 . 006Figure 2—figure supplement 2 . Strength of morphological phenotypes , according to annotated pathway . Morphological phenotype strength is calculated as the average replicate correlation for genes that experts manually annotated genes as belonging to each pathway . The number of genes tested in each category is shown in parentheses after the pathway name ( two wild-type clones of the same gene are only counted once , but a mutant allele is counted separately ) . The red line shows the threshold beyond which an individual gene’s profile would be considered to yield a distinguishable phenotype . Error bars indicate the deviation in the replicate correlation among the genes associated with the pathway . DOI: http://dx . doi . org/10 . 7554/eLife . 24060 . 006 Given that technical replicates produce similar morphological profiles , we next evaluated whether similarities between profiles induced by different constructs are meaningful . We began with the simplest case: for a subset of genes in the experiment , a ‘wild-type’ sequence ( see Materials and methods for important definitions ) was captured in more than one ORF construct ( 23 pairs ) . These pairs either correspond to different physical cloning events and preparations but with highly similar full-length sequence ( as defined in Methods; category a: nine pairs ) , or a substantive difference in their nucleotide sequence , for example , isoforms ( category b: 14 pairs ) . We found that , as expected , the phenotypes of over-expressed wild-type ORFs of the same gene were more similar to each other , on average , than to randomly selected genes . Of the 23 pairs for which both wild-type ORFs yielded a phenotype distinguishable from negative controls , 22 ( ~96% ) of the pairs’ profiles were correlated more than expected by chance ( Figure 2B , the one pair not meeting that threshold was in category b ) , confirming that different constructs with biological similarity indeed produce similar morphological profiles . This result also confirms that the sequence differences seen in separately cloned wild-type constructs do not generally have a major functional impact , but we caution that any individual construct of interest may have an impactful mutation; thus the raw sequence data should be examined and testing alternate constructs for a gene may be recommended . Note that if , for example , only 50% of wild-type pairs showed high profile correlation , it would remain ambiguous whether it was caused by poor assay quality or by constructs’ sequence mismatches . But in this particular case the mentioned near perfect consistency rules out either of the two possibilities . We also note that the 23 pairs analyzed here are located in different well locations on each plate; this result therefore also rules out widespread artifacts , such as plate position effects or metadata errors . We suspected that the small number of engineered constitutively activating alleles for certain genes would , on average , yield a stronger phenotype than their wild-type counterparts . We indeed found that correlations between replicates of the constitutively activating allele were typically higher than correlations between replicates of the wild-type version of a gene ( Supplementary file 1B; p-value=0 . 012 , one-sided paired t-test ) . We hypothesized that genes in pathways known to affect cellular morphology ( RAC1 , KRAS , CDC42 , RHOA , PAK1 , and genes related to the Hippo pathway ) would be more likely to yield a morphological phenotype distinguishable from negative controls than other genes in the analysis . Indeed , we found this to be true ( Fisher’s test p-value=3 . 7 × 10−3 ) ( Figure 2C ) . Reassured by this validation , we were curious which pathways would be most and least likely to yield detectable morphological phenotypes , recognizing that ‘pathways’ are neither separate nor well-defined entities . We found genes manually annotated as being in the Hippo , Hedgehog , cytoskeletal reorganization , and Mitogen-activated protein kinases ( MAPK ) pathways were more likely to result in a phenotype , whereas genes annotated as belonging to the JAK/STAT , hypoxia , and BMP pathways were among the least likely to yield a phenotype under the conditions tested ( Figure 2—figure supplement 2 and Supplementary file 1C ) . Nevertheless , the majority of pathways could be interrogated by morphological profiling . Given the caveats and limitations of overexpressing genes ( see Discussion ) , we next tested whether image-based profiling of expression constructs could capture relationships among genes known to be functionally related . Because a reliable and complete map of all gene-gene connections is not available , we evaluated the accuracy of our results via two approaches . First , we compared our data to protein-protein interaction data from BioGRID ( Stark et al . , 2006 ) . This is imperfect ground truth for judging our predictions because two proteins might physically interact without producing the same morphological phenotype when overexpressed , and genes in the same pathway might regulate the same phenotype without any physical interaction . Nevertheless , we expect that the corresponding proteins of gene pairs with highest profile similarity are more likely than average to physically interact . Indeed , looking at wild-type versions of genes showing a detectable phenotype ( the 73 genes represented in the 110 constructs ) , the ratio of verified gene connections among the top 5% correlated gene pairs ( 9% , 13 verified out of 143 possible combinations ) is significantly higher than that of other gene pairs ( 5% , 128 verified out of 2485 possible; Fisher’s test p-value=0 . 04; Supplementary file 1D ) . Second , we manually annotated each gene for the pathway with which it is associated . This approach is based on expert opinion and thus imperfect knowledge of all genes’ function; furthermore many pathways interrelate , and genes in the same pathway are not expected to have identical phenotypes given that their functions are rarely identical ( most notably , overexpression of some may activate while others suppress a biological pathway or process ) . Nonetheless , we expect pairs of genes whose morphological profiles correlate highly to be more likely than average to be annotated in the same pathway vs . different pathways . Using the same 73 genes as in the previous analysis , the ratio of gene connections with the same-pathway annotation in the top 5% most-correlated gene pairs was 20% ( 29 pairs out of 143 ) , significantly higher than the ratio for the remaining pairs ( 6% , 139 pairs out of 2485; Fisher’s test p-value = 7 . 53×10−9; Supplementary file 1E ) . Having quantitatively established that morphological profiling is sensitive , robust , and captures known gene-gene relationships , we explored these relationships in a correlation matrix ( Figure 3 bottom left and Figure 3—figure supplement 1 ) . The overall structure , with multiple groupings along the diagonal , is consistent with the fact that the 110 constructs ( 73 unique genes ) that showed a phenotype had been annotated as representing 19 different pathways . That is , we did not see large , homogeneous clusters , as would be expected if morphological profiling was sensitive to perturbation but not highly specific . This rules out uniform toxicity induced by a large number of genes , for example . Neither did we see only signal along the diagonal , which would have indicated no strong similarity between any gene pairs . 10 . 7554/eLife . 24060 . 007Figure 3 . Morphological relationships among overexpressed genes/alleles , determined by Cell Painting . Correlations between pairs of genes/alleles were calculated and displayed in a correlation matrix ( bottom left inset , full resolution is available as Figure 3—figure supplement 1 ) . Only the 110 genes/alleles with a detectable morphological phenotype were included . The rows and columns are ordered based on a hierarchical clustering algorithm such that each blue submatrix on the diagonal shows a cluster of genes resulting in similar phenotypes . The correlations were then used to create a dendrogram ( main panel ) where the radius of the subtree containing a cluster shows the strength of correlation . The 25 clusters containing at least two constructs are printed on the dendrogram in arbitrary colored fonts , while gene names colored gray and marked by asterisks are those that do not correlate as strongly with their nearest neighbors ( i . e . , they are singletons or fall below the threshold used to cut the dendrogram for clustering ) . Each colored arc corresponds to a cell subpopulation as noted in the legend . Line thickness indicates the strength of enrichment of the subpopulation in the cluster samples compared to the negative control . Solid vs . dashed lines indicate the over- vs . under-representation of the corresponding subpopulation in a cluster , respectively . Note that the number next to each cluster in the dendrogram is referenced in the main text and corresponds to the numbered supplemental data file for each cluster . DOI: http://dx . doi . org/10 . 7554/eLife . 24060 . 00710 . 7554/eLife . 24060 . 008Figure 3—figure supplement 1 . Correlation among the 110 genes/alleles with a detectable morphological phenotype . The rows and columns are ordered based on a hierarchical clustering algorithm such that each blue submatrix on the diagonal shows a cluster of genes resulting in similar phenotypes . The scale bar depicts Pearson correlation . DOI: http://dx . doi . org/10 . 7554/eLife . 24060 . 00810 . 7554/eLife . 24060 . 009Figure 3—figure supplement 2 . Smoothed stability score across different cutoffs , in order to choose a threshold for cutting the dendrogram to form clusters . The maximum occurs at threshold = 0 . 522 . Smoothing is done by taking the moving average of order 0 . 02 . The stability score is defined as the proportion of treatments whose clusters are not affected if the cutoff is increased or decreased by a small amount ( ϵ= . 002 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24060 . 00910 . 7554/eLife . 24060 . 010Figure 3—figure supplement 3 . Common cell subpopulations seen across more than one cluster . These names are used to annotate clusters of genes in Figure 3 . Example images shown are taken from individual clusters . Scale bar is 63 μm and image intensities are log normalized . References to size and shape in the subpopulation legends refer to both the nucleus and cell borders , unless otherwise noted . DOI: http://dx . doi . org/10 . 7554/eLife . 24060 . 010 We next created a dendrogram ( Figure 3 ) and defined 25 clusters ( see Materials and methods and Figure 3—figure supplement 2 ) to explore the similarities among genes . Pairs of wild-type ORFs almost always clustered adjacently , consistent with our quantitative analysis described above ( Figure 2B ) . After retaining only one copy of replicate ORFs , we found that the majority of clusters ( 19 out of the 22 clusters containing more than one gene ) were enriched for one or more Gene Ontology terms ( Supplementary file 1F ) , indicating shared biological functions within each cluster . Using this dendrogram , we began by interrogating three clusters that conformed well to prior biological knowledge . First , we analyzed Cluster 20 , containing the two canonical Hippo pathway members YAP1 and WWTR1 ( more detail in Supplementary file 2 [PDFs A2–A20 and B2–B20 ] , and in a later section of the text ) . Both are known to encode core transcriptional effectors of the Hippo pathway ( Johnson and Halder , 2014 ) , and a negative regulator of these proteins , STK3 ( also known as MST2 ) , is the strongest anti-correlating gene for the cluster ( Supplementary file 2 [PDF A20] , panel c1 ) . Second , we noted Cluster 21 is comprised of the two phosphatidylinositol 3-kinase signaling/Akt ( PI3K ) regulating genes , PIK3R1 and PTEN , both frequently mutated across 12 cancer types in The Cancer Genome Atlas ( TCGA ) ( Kandoth et al . , 2013 ) . These results are consistent with previous observations that certain isoforms of PIK3R1 reduce levels of activated Akt , a dominant negative effect ( Abell et al . , 2005 ) . AKT3 is in a cluster anti-correlated to the Cluster 21 ( ( Supplementary file 2 [PDF A21 , panel b1] ) . Third , we examined three clusters ( 19 , 6 and 3 ) that included many MAPK-related genes . Cluster 19 is the largest example of a tight cluster of genes already known to be associated; it includes four activators in the RAS-RAF-MEK-ERK cascade: KRAS , RAF1 ( CRAF ) , BRAF , and MOS . Notably , two constitutively active alleles of these genes , BRAFV600E ( Davies et al . , 2002 ) and RAF1L613V ( Wu et al . , 2011 ) , form a separate cluster ( Cluster 6 ) adjacent to their wild-type counterparts . Furthermore , the constitutively active RAS alleles HRASG12V and KRASG12V ( McCoy et al . , 1984 ) are in the next-closest cluster ( Cluster 3 ) , which also contains MAP2K4 and MAP2K3 ( known to be activated by Ras [Shin et al . , 2005] ) , as well as CDKN1A ( Jalili et al . , 2012 ) . By contrast , MAPKs that are known to be unrelated to the RAS-RAF-MEK-ERK cascade , such as MAPK14 in Cluster 5 , are far away in the dendrogram . Overall , these results support the notion that connections between genes can be efficiently discovered using our approach . We hypothesized that the specific morphologic features that segregated each of the clusters would provide insight into gene function . Examining images ( Supplementary file 2 [PDF -19A , panel 3] ) or rank-ordered lists of features that distinguish individual profiles or clusters ( Supplementary file 1G ) is tedious and lacks sensitivity for all but the most obvious of phenotypes , confirming that quantitative morphological profiling is more sensitive than the human visual system . We therefore devised several strategies to enhance biological interpretability from these experiments and applied these in combination . First , we grouped features into meta-features based on their type of measurement , i . e . , shape , texture , intensity , etc . , and the cell constituents to which they are related , to create a Feature Grid ( Figure 4A ) . Second , we performed unsupervised grouping of features by mapping the top 20 most-distinguishing features for each cluster onto a plane , creating a Feature Map ( Figure 4B ) , in which highly correlated features are mapped nearby each other ( see ‘Feature Interpretation’ in Methods for an explanation of individual feature names ) . In certain cases , these visualizations revealed the nature of the morphological phenotype ( e . g . , nuclear shape abnormalities Supplementary file 2 [PDF 7A] ) , but for others these approaches did not suffice to yield an obvious phenotypic conclusion ( e . g . , for Cluster 19 , Figure 4A and B ) . 10 . 7554/eLife . 24060 . 011Figure 4 . Visualizations used to interpret morphology of Cluster 19 ( for other clusters , see Supplementary file 2 [PDFs 1A–25A] ) . ( A ) Feature Grid . RNA and AGP ( actin , Golgi , plasma membrane ) intensity contribute most to distinguishing the genes in Cluster 19 ( KRAS , RAF1 , BRAF , and MOS ) . Dark blue colors indicate higher median z-score of the relevant measurements for genes in the cluster relative to negative controls . As ‘RadialDistribution’ features do not exist for the DNA channel , it is colored in black . ( B ) Feature Map . The feature names showing the greatest difference between the cluster and negative controls are shown , based on largest absolute value of z-scores ( full resolution version is available in Cluster 19A PDF ) . They are mapped in 2D space such that features that are highly correlated with each other across all genes’ profiles are placed close together and thus can be interpreted together . Blue/red colored names indicate positive/negative sign of the z-score ( i . e . , blue indicates that the cluster shows higher values than controls ) . According to this map , the average intensity of AGP , RNA and Mito shows high variation for cells within samples in Cluster 19 ( e . g . , large mad_Cytoplasm_Intensity_MeanIntensity_AGP , where the prefix ‘mad’ refers to median absolute deviation , a robust form of standard deviation ) . ( C ) Sample images of a subpopulation of cells enriched and de-enriched for all genes in Cluster 19 . Cells with asymmetric organelle distribution are highly over-represented for genes in the cluster , and cells with more even distribution of organelles are less abundant . Note that the exemplar cells are shown at the center of the patches . This explains the duplications observed in some patches . Scale bars are 39 . 36 μm long . Pixel intensities are multiplied by five for display . DOI: http://dx . doi . org/10 . 7554/eLife . 24060 . 011 Third , we hypothesized that leveraging the single-cell resolution of image-based profiling might be highly sensitive in enhancing interpretation , particularly for cases where only a subset of cells is distinctive from negative controls . To test this , for each given cluster of genes together with negative controls we identified 20 subpopulations using k-means clustering on single cell data . We calculated the abundance of cells in each of the 20 subpopulations to determine which are over/under-represented relative to controls for the given cluster ( corresponding images are shown; Supplementary file 2 [PDFs 1B–25B] ) . For example , the MAPK pathway activators in Cluster 19 show increased prevalence of a subpopulation of cells with strongly asymmetric ER , mitochondria , and Golgi staining , indicating a cell polarization phenotype ( Figure 4C , and Supplementary file 2 [PDF 19B] , Categories one and two ) , for which there is evidence in the literature ( Samaj et al . , 2004; Elsum et al . , 2013; Godde et al . , 2014 ) . This phenotype was not captured by manual inspection nor the first two approaches ( e . g . , Supplementary file 2 [PDF 19A , panels a2 and b2] ) . Encouraged by this , we supplemented the morphological map by compiling these and other visualizations into PDF files for each cluster , summarized in Figure 5 and provided in full as Supplementary file 2 . We also noticed that certain subpopulations were similar across several clusters ( Figure 3—figure supplement 3 shows sample cell images of each such subpopulation ) ; we annotated their enrichment/de-enrichment on the dendrogram ( Figure 3 ) . 10 . 7554/eLife . 24060 . 012Figure 5 . Data and visualizations supporting the morphological map for each cluster . For all 25 clusters , there are two corresponding Supplemental PDF files . Left: Supplementary file 2 ( type A PDFs , e . g . , ‘1A . pdf’ ) provide an overview of data about the cluster . Panel a1 lists the genes/alleles in the cluster as well as expert annotations regarding related pathways and the cell count ( as a z-score ) for each gene/allele . Panel b1 contains the average correlation of the cluster to other clusters , indicating uniqueness of the cluster’s morphological phenotype . Panel c1 lists the top five negatively correlated gene/alleles to the cluster . Panel a2 shows the Feature Grid summarizing categories of morphological features distinguishing the cluster from the negative control . Panel b2 shows the Feature Map displaying the names of the top 20 morphological features distinguishing the cluster from the negative control , positioned based on similarity . Explanations for feature names can be found in the Methods section . Panel c2 shows a correlation matrix for just those genes/alleles in the cluster . Panel 3 contains sample images of fields of view of cells expressing each gene/allele in the cluster , along with images of the control for comparison . Right: Supplementary file 2 ( type B PDFs ) contain multiple plots aiming to illustrate the phenotype based on single-cell data , including cell subpopulation enrichment/suppression in the cluster . First , a histogram of single-cell DNA content is shown for all cells from all genes/allele treatments in the cluster , indicating the overall cell cycle distribution . Next , bar plots show ( for the cluster overall and for each gene in the cluster ) which of 20 subpopulations of cells are enriched and suppressed relative to negative controls . Finally , each subsequent page of the PDF is devoted to the subpopulations whose representation differs from negative controls in a statistically significant way , whether enriched or suppressed ( subpopulations which are very small in both the cluster and negative control samples are omitted ) . For each subpopulation , a bar plot shows the top 10 most-distinguishing feature names ( versus negative control cells ) . Then , sample images are shown of individual representative cells from each subpopulation . DOI: http://dx . doi . org/10 . 7554/eLife . 24060 . 012 Using these visualizations , we began by interrogating three adjacent and correlating clusters ( Clusters 4 , 7 , and 11 ) contain wild-type and mutant alleles of CDC42 , a gene encoding a Rho family GTPase with diverse roles in cell polarity , morphology , and migration ( Melendez et al . , 2011; Martin , 2015 ) . Cluster 4 contains the constitutively active mutant CDC42 Q61L ( Nobes and Hall , 1999 ) as well as MAP3K2 and MAP3K9 . The highly similar Cluster 7 contains the dominant negative alleles CDC42 T17N ( Nobes and Hall , 1999 ) and RAC1 T17N ( Zhang et al . , 1995 ) , a related RAS superfamily member . That activating and inhibiting alleles would yield similar phenotypes when overexpressed is not surprising for CDC42 ( Melendez et al . , 2011 ) . Cluster 7 also contains isoforms and alleles of AKT: specifically , AKT3 and the constitutively active E17K alleles of both AKT1 and AKT3 ( Kim et al . , 2008; Davies et al . , 2008 ) . Akt is known to be essential for certain Cdc42-regulated functions ( Higuchi et al . , 2001 ) and vice versa ( Stengel and Zheng , 2012 ) . Finally , the nearby Cluster 11 ( which is discussed in more detail later ) contains the wild-type form of CDC42 as well as TRAF2 , a canonical NF-κB activator; these two are known to interact and share functions in actin remodeling ( Marivin et al . , 2014 ) . We also note that anti-correlating genes to these clusters ( generally in Clusters 13 and 21 ) are consistent with existing knowledge , including ( a ) AKT family member AKT1S1 ( a Proline rich AKT substrate , PRAS40 ( Kovacina et al . , 2003; Wiza et al . , 2014 ) , Supplementary file 2 [PDF 7A , panels b1 and c1] ) ( b ) CDK2 ( a known target of Akt [Maddika et al . , 2008] ) , ( c ) PIK3R1 and PTEN in Cluster 21 , described previously , which have known interactions with AKT ( Cheung and Mills , 2016; Hemmings and Restuccia , 2015 ) . Thus , all of these connections have previously been identified . Subpopulation visualization revealed that Clusters 4 , 7 , and 11 are enriched in cells that are huge and binucleate ( Figure 3 , example images shown in Supplementary file 2 [PDF 4B] ) . Genes in all three clusters also show irregularities in DNA content , namely , an enrichment in cells with sub-2N DNA content , a decrease in cells with 2N DNA content , and , for most genes , a decrease in cells with S phase and 4N DNA content , indicating a significant amount of DNA fragmentation and thus apoptosis ( DNA histograms in Supplementary file 2 [PDFs 4B , 7B , and 11B] ) . These phenotypes are consistent with these genes’ known role in the cell cycle and cell polarity ( Chircop , 2014 ) . As a second test case , we examined Cluster 8 , which contains PRKACA ( the catalytic subunit α of protein kinase A , PKA ) and two of its known substrates: GLI1 ( a transcription factor mediating Hedgehog signaling ) ( Asaoka , 2012 ) , and RHOAQ63L ( a Ras homolog gene family member ) ( Lang et al . , 1996; Rolli-Derkinderen et al . , 2005 ) . The highly similar Cluster 10 contains the wild-type RHOA , as well as ELK1 which is also linked to the Rho GTPase family and PKA ( Bachmann et al . , 2013; Murai and Treisman , 2002 ) . We investigated the morphological changes causing these genes to cluster . RhoA is a known regulator of cell morphology and cell rounding is a known related phenotype ( Oishi et al . , 2012 ) . We found that indeed all members of Clusters 8 and 10 significantly induce cell rounding ( Supplementary file 1H ) . Although cell count is lower for genes Clusters 8 and 10 , the degree varies greatly ( from z-score −0 . 67 to −3 . 02 , Supplementary file 2 [PDFs 8A and 10A , panel a1] ) , ruling out that simple sparseness of cells explains their high similarity in the assay . As well , the overall DNA content distribution of the cell populations appears relatively normal ( Supplementary file 2 [PDFs 8B and 10B] ) . Subpopulation extraction provides a satisfying biological explanation for these clusters’ distinctive phenotype: the increased roundness and strong variation in intensity levels ( per the Feature Grid ) across the population stems from an increased proportion of telophase , anaphase , and apoptotic cells ( Figure 3 and Supplementary file 2 [PDFs 8B and 10B] ) . We therefore conclude that the morphological map can link related genes to each other and that the morphological data can provide insight into their functions , particularly with the help of subpopulation visualization . We wondered whether novel relationships might emerge from our unbiased classification of gene and allele function based on morphologic profiling . We noticed that the known regulator of NF-κB signaling , TRAF2 ( in Cluster 11 , together with CDC42 ) ( Grech et al . , 2004; Tada et al . , 2001 ) , yields a signature strongly anti-correlated to YAP1/WWTR1 ( Cluster 20 ) , which encode the transcriptional effectors of the Hippo pathway , YAP ( Yes-associated protein ) and TAZ ( Transcriptional co-activator with a PDZ-domain ) . The Hippo pathway and NF-κB signaling are critical regulators of cell survival and differentiation , and dysregulation of these pathways is implicated in a number of cancers ( Varelas , 2014; Hoesel and Schmid , 2013; Tornatore et al . , 2012 ) , but we found no evidence in the literature ( in particular through BioGRID ) of physical interaction between the proteins encoded by Cluster 11 genes and Cluster 20 genes . Confirming our approach , a functional connection between CDC42 ( Cluster 11 ) and YAP1 ( Cluster 20 ) has been identified: deletion of CDC42 phenocopies the loss of YAP1 in kidney-specific conditional knockouts in mice ( Reginensi et al . , 2013 ) . Still , the NF-κB pathway ( and in particular the Cluster 11 member TRAF2 ) , has not been closely tied to YAP and TAZ in human cells ( see Discussion ) . We first wanted to characterize Clusters 11 and 20 to confirm that relationships within each cluster are supported in the literature . Indeed we found evidence for most of the within-cluster connections . CDC42 and TRAF2 ( Cluster 11 ) physically interact and share functions in actin remodeling ( Marivin et al . , 2014 ) . As described in a prior section YAP/TAZ ( Cluster 20 ) are known to share functional similarities in the Hippo pathway , being regulated by , and also regulating , cytoskeletal dynamics . Consistent with these known functions , we found that a core effector of the Hippo pathway which functions to restrict YAP/TAZ nuclear activity , STK3 ( which encodes the Mst2 kinase ) ( Meng et al . , 2016 ) , has a morphological signature strongly anti-correlated to YAP1/WWTR1 ( Supplementary file 2 [PDF 20A , panel c1] ) . We note that although STK3 and TRAF2 are both moderately anti-correlated with YAP/TAZ ( Cluster 20 ) , STK3 and TRAF2 are not themselves highly correlated , indicating each has a different subset of phenotypes that anti-correlate to YAP/TAZ . We also note that two clones that express another regulator of YAP activity , STK11 , form Cluster 22 which falls nearby YAP1/WWTR1; a connection between STK11 and YAP has been identified ( albeit with opposite directionality , identified via knockdown of STK11 [Mohseni et al . , 2014] ) . Further , YAP1 is among the highest anti-correlating genes to REL ( data not shown; REL is a singleton in the dendrogram and thus not in a cluster ) , whose protein product , c-Rel , has a known connection to TRAF2 ( Jin et al . , 2015 ) . These results reaffirm that the Cell Painting-based morphological signatures are a useful reporter of biologically meaningful connections among genes in these pathways . Given the striking inverse correlation between YAP1/WWTR1 and TRAF2 , we sought to confirm a negative regulatory relationship between the Hippo and NF-κB pathways by multiple orthogonal methods . First , we explored the observed inverse morphological impact using the Cell Painting data . The morphological impact of genes in Cluster 11 and 20 is quite strong ( median replicate correlation is at the 74th and 81st percentile , and average within-group correlations are 0 . 66 and 0 . 73 ) . Subpopulation analysis showed that Cluster 20 ( YAP1 , WWTR1 ) is enriched for cells that are slightly large , slightly elongated , and have disjoint , bright mitochondria patterns , whereas Cluster 11 ( TRAF2 , CDC42 ) is de-enriched for those subpopulations and instead enriched for binucleate cells , very large cells , and small cells with asymmetric organelles ( Figures 3 , 6A and B ) . 10 . 7554/eLife . 24060 . 013Figure 6 . Morphological and transcriptional cross-talk between the Hippo pathway and regulators of NF-κB signaling . ( A ) . The TRAF2/CDC42 cluster ( Cluster 11 ) is enriched for bi-nucleate cells , small cells with asymmetric organelles , and huge cells . Note that exemplar images shown are not labeled as to the actual gene they are associated with . Rather they are only supposed to provide a visual insight of the cell morphologies which are enriched in the gene cluster . ( B ) The YAP1/WWTR1 cluster ( Cluster 20 ) is enriched for cells with bright disjoint mitochondria patterns , slightly large cells , and slightly elongated cells . Scale bars are 39 . 36 μm long . Pixel intensities are multiplied by five for display . ( C ) Gene Set Enrichment Analysis ( GSEA ) reveals that gene overexpression leading to down-regulation of YAP1 targets ( CTGF , CYR61 , and BIRC5 ) are enriched for regulators of the NF-κB pathway ( Enrichment Score p-value = 8 . 19×10−5 ) . The horizontal axis gives the index of ORFs sorted based on the average amount of down-regulation of the YAP1 targets . Each blue hash mark on this axis indicates an NF-κB pathway member . The running enrichment score , which can range from −1 to 1 , is plotted on the vertical axis and quantifies the accumulation of NF-κB pathways members on the sorted list of ORFs . ( D ) TRAF2 and REL suppress YAP and TAZ transcriptional activity . REL and TRAF2 suppress the ability of wild-type ( D1 ) YAP and ( D2 ) TAZ to drive the expression of a TEAD-regulated luciferase reporter . Activity of nuclear active mutants of ( D3 ) YAP ( 5SA ) and ( D4 ) TAZ ( 4SA ) are similarly suppressed . Luciferase reporter activity was measured in HEK293T cells co-transfected with expression constructs as indicated and a TEAD luciferase reporter was used to measure YAP-directed transcription . ( * p-value<0 . 05 , p-value=0 . 001 , * p-value<0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24060 . 01310 . 7554/eLife . 24060 . 014Figure 6—figure supplement 1 . Gene Set Enrichment Analysis ( GSEA ) reveals that overexpression constructs sorted based on their similarity to YAP1/WWTR1 overexpression ( in terms of impact on particular mRNA targets ) , are enriched for regulators of the NF-κB pathway ( Enrichment Score p-value=0 . 0019 ) . mRNA targets common to both YAP1 and WWTR overexpression include INPP4B , MAP7 , LAMA3 , STMN1 , and TRAM2 , which are positively regulated , and SPP1 , IER3 , RAB31 , and GPR56 , which are negatively regulated . DOI: http://dx . doi . org/10 . 7554/eLife . 24060 . 01410 . 7554/eLife . 24060 . 015Figure 6—figure supplement 2 . Gene Set Enrichment Analysis ( GSEA ) reveals that overexpression constructs sorted based on their similarity to TRAF2/REL overexpression ( in terms of impact on particular mRNA targets ) , are weakly enriched for regulators of the Hippo pathway ( Enrichment Score p-value=0 . 024 ) . mRNA targets common to both TRAF2 and REL overexpression include NFKBIA , IKBKE , AKAP8 , and BIRC2 , which are positively regulated and RPA3 which is negatively regulated . As compared to Figure 6C and Figure 6—figure supplement 1 , this is a weaker/lower-confidence enrichment - note the lower maximum height ( ~0 . 44 compared to >0 . 6 ) and higher p-value ( 0 . 024 compared to <0 . 002 ) . Still , we note that WWTR1 and PPP1CA are the top two matches among those annotated as related to the Hippo pathway in KEGG; PPP1CA ( also known as PP-1A ) activates TAZ ( Liu et al . , 2011 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24060 . 015 Second , given that YAP/TAZ are transcriptional regulators , we analyzed gene expression data . Using the same constructs as in our Cell Painting experiment , we found an anti-correlated relationship at the mRNA level , consistent with the anti-correlation we had seen in morphological space . To do this , we used Gene Set Enrichment Analysis ( Subramanian et al . , 2005 ) and publicly available data , which includes data from four to nine different cell lines at one to four time points ( https://clue . io ) . Time point refers to the duration of treating the cells with over-expression constructs until the time gene expression readouts are made . This analysis revealed that the NF-κB pathway is the pathway most enriched among genes whose overexpression results in down-regulation of known YAP1 targets , CTGF , CYR61 , and BIRC5 ( Zhao et al . , 2008 ) ( Benjamini and Hochberg ( BH ) adjusted p-value = 2 ×10−8 in Supplementary file 1I , and Figure 6C ) , with TRAF2 being among the genes contributing to this enrichment ( Supplementary file 1I ) . We also saw enrichment of NF-κB pathway members when testing a data-driven set of targets of YAP1/TAZ ( Figure 6—figure supplement 1 , see Materials and methods ) . In the inverse analysis , genes that alter the levels of TRAF2/REL common targets are weakly enriched in Hippo pathway members ( Figure 6—figure supplement 2 , see Materials and methods ) . This is consistent with the hypothesis that NF-κB members can downregulate YAP/TAZ targets but not strongly vice versa . Finally , we more directly confirmed negative crosstalk between NF-κB effectors and YAP/TAZ using a synthetic TEAD luciferase reporter that is YAP/TAZ responsive ( Dupont et al . , 2011 ) . Importantly , these confirmatory experiments used different cellular contexts and perturbation constructs versus the original Cell Painting data . Co-expression of the NF-κB pathway effectors TRAF2 or C-REL with YAP or TAZ led to significantly lower reporter activity than expression of YAP or TAZ alone ( Figure 6D1 and D2 ) . Intriguingly , mutants of YAP or TAZ that are insensitive to negative regulation by the Hippo pathway ( YAP-5SA and TAZ-4SA; [Zhao et al . , 2008] ) remained sensitive to suppression of transcriptional activity by TRAF2 and C-REL , indicating that the negative relationship we identified may be independent of canonical upstream Hippo pathway signals ( Figure 6D3 and D4 ) . We conclude that connections among genes can be profitably analyzed using morphological profiling of overexpressed genes via the Cell Painting assay . In a single inexpensive experiment , we were able to rediscover a remarkable number of known biological connections among the genes tested . Further , we found that morphological data from the Cell Painting assay , together with novel subpopulation visualization methods , can be used to flesh out the functionality of particular genes and/or clusters of interest . By adopting a two-pronged approach , merging this Cell Painting morphological analysis with transcriptional data , we were able to identify an unexpected relationship in human cells between two major signaling pathways , Hippo and NF-κB , both under intense study recently for their involvement in cancer . Through validation of these clustered genes , we have identified that YAP/TAZ-directed transcription is negatively regulated by NF-κB pathway effectors and our data suggests a novel regulatory mechanism that is independent of upstream Hippo kinases . To date , there has been little evidence of the intersection between these important signaling pathways . Recent work examining osteoclast-osteoblast differentiation has suggested that Hippo pathway kinases , such as Mst2 , may affect the NF-κB pathway through phosphorylation of IkB proteins , thereby promoting nuclear translocation of NF-κB transcription factors ( Lee et al . , 2015 ) . TAZ was found to be a direct target of NF-κB transcription factors and its expression is regulated via NF-κB signaling ( Cho et al . , 2010 ) . Our work , however , supports a possible additional mode of interaction , whereby regulators of NF-κB signaling directly regulate the function of Yap and Taz as transcriptional co-factors . Recent work has demonstrated , in Drosophila , that NF-κB activation via Toll receptor signaling negatively regulates the transcriptional activity of Yorkie , the homolog of YAP/TAZ , through activation of canonical hippo pathway kinases ( Liu et al . , 2016 ) . The work described here identifies , for the first time in a mammalian system , that a negative regulatory relationship exists between NF-κB activation and YAP/TAZ transcriptional function . Furthermore , we have identified that this regulation of YAP/TAZ occurs in a manner that is independent of Hippo pathway-mediated phosphorylation events on YAP/TAZ , suggesting a more direct relationship between NF-κB and YAP/TAZ signaling . In this work , we tested quantitatively and explored qualitatively the connections among genes revealed by morphological profiling . Our underlying hypothesis was that functionally similar genes would generally yield morphologically similar cells when overexpressed , and indeed we found this to be the case . Still , some discussion of this point is warranted . Most commonly , gene overexpression will result in activation of the corresponding pathway via amplification of the endogenous gene’s function . However , it is important to note that the profiling strategy to discover functional relationships does not assume or require this . For example , overexpression could also disrupt a protein complex , producing a trans-dominant negative effect that results in precisely the opposite phenotypic effect ( Veitia , 2007 ) . In still other cases , overexpression of a particular gene may not affect any of the normal functions of the gene ( producing a false negative signal ) , or trigger a stress response ( yielding a confounded profile ) , or produce a complicated response , due to feedback loops . Further , artifactual phenotypes could be seen , e . g . , if overexpression yields a non-physiological interaction among proteins or toxic aggregates . Nevertheless , despite these caveats and complications , our results indicate that valuable information could be gleaned from the similarity and dissimilarity of the morphological perturbations induced by gene overexpression . Using overexpression avoids the complications of RNAi off-target effects ( often due to seed effects ) , which were far more prevalent ( impacting 90% of constructs in our recent study [Singh et al . , 2015] ) . In addition to functionally annotating genes , as demonstrated here , one particularly appealing application enables personalized medicine: it should be feasible to use morphological profiling to predict the functional impact of various disease alleles , particularly rare variants of unknown significance . This has recently been successful using mRNA profiles ( Berger et al . , 2016 ) . Thus , an even more exciting prospect would be to combine mRNA profiles with morphological profiles to better predict groups of alleles of similar mechanism , and ultimately to predict effective therapeutics for each group of corresponding patients . We make all raw images , extracted cellular features , calculated profiles , and interpretive visualizations publicly available , providing an initial morphological map for several major signaling pathways , including several unexplored connections among genes for further study ( see Supplementary file 2 ) . Expanding this map to full genome scale could prove an enormously fruitful resource . The Reference Set of human cDNA clones utilized here has been previously described ( Kim et al . , 2016 ) ; ~90% of these constructs induce expression of the intended gene greater than two standard deviations above the control mean . Briefly , wild-type ORF constructs were obtained as Entry clones from the human ORFeome library version 8 . 1 ( http://horfdb . dfci . harvard . edu ) with additional templates generously provided by collaborating laboratories , and cloned into the pDONR223 Gateway Entry vector . In addition , here , to maximize coverage of cellular pathways , we included additional clones with minimal sequence deviations from the intended templates . Sanger sequencing of Entry clones verified the intended transcripts and , if applicable , the intended mutation . Entry constructs and associated sequencing data will be publicly available via www . addgene . org and may also be available via members of the ORFeome Collaboration ( http://www . orfeomecollaboration . org/ ) , including the Dana-Farber/Harvard Cancer Center ( DF/HCC ) DNA Resource Core DNA Repository ( http://www . dfhcc . harvard . edu/core-facilities/dna- resource/ ) and the DNASU Plasmid Repository at ASU Biodesign Institute ( http://dnasu . asu . edu/DNASU/Home . jsp ) . Clone requests must include the unique clone identifier numbers provided in the last column of Supplementary file 1A ( e . g . ccsbBroadEn_12345 as an example for a specific entry clone and ccsbBroad304_12345 as an example for a specific expression clone ) . ORFs were transferred to the pLX304 lentiviral expression vector ( Yang et al . , 2011 ) by LR ( attL x attR ) recombination . For simplicity , throughout this paper ‘wild-type’ refers to ORFs found in the original collection without a particular known mutation intentionally engineered . Due to natural human variation , and occasional cloning artifacts , there are often non-identical matches of such constructs to reference sequence; these differences are fully documented for each construct and sequence data will be publicly available through AddGene , in addition to the sequencing data for the original Entry clones for the genome-scale library ( Yang et al . , 2011 ) . U-2 OS cells ( human bone osteosarcoma cells ) , RRID:CVCL_0042 , were obtained from ATCC and propagated in the William Hahn lab; they were not additionally authenticated prior to this experiment . The cell line tested negative for mycoplasma prior to this experiment . HEK293T cells , RRID:CVCL_0063 , were obtained from ATCC . The cell line was validated by STR profiling ( Genetica DNA Laboratories ) and was negative for mycoplasma as measured by MycoAlert Mycoplasma Detection Kit ( Lonza , Walkersville , MD ) . We followed our previously described protocol ( Kim et al . , 2016; Berger et al . , 2016 ) except for durations of some steps . Briefly , cells were plated in 384-well plates and transduced with lentiviral particles carrying ORF constructs the next day . Viral particles were removed 18–24 hr post-infection and cells cultured for 48 hr until staining and imaging ( 72 hr total post-transduction ) . The experiment was conducted in five replicates , each in a different plate . The number of replicates being five was decided based on prior experiments ( Bray et al . , 2016 ) . The Cell Painting assay followed our previously published protocol ( Bray et al . , 2016 ) . Briefly , eight different cell components and organelles were stained with fluorescent dyes: nucleus ( Hoechst 33342 ) , endoplasmic reticulum ( concanavalin A/AlexaFluor488 conjugate ) , nucleoli and cytoplasmic RNA ( SYTO14 green fluorescent nucleic acid stain ) , Golgi apparatus and plasma membrane ( wheat germ agglutinin/AlexaFluor594 conjugate , WGA ) , F-actin ( phalloidin/AlexaFluor594 conjugate ) and mitochondria ( MitoTracker Deep Red ) . WGA and MitoTracker were added to living cells , with the remaining stains carried out after cell fixation with 3 . 2% formaldehyde . Images from five fluorescent channels were captured at 20x magnification on an ImageXpress Micro epifluorescent microscope ( Molecular Devices ) : DAPI ( 387/447 nm ) , GFP ( 472/520 nm ) , Cy3 ( 531/593 nm ) , Texas Red ( 562/624 nm ) , Cy5 ( 628/692 nm ) . Nine sites per well were acquired , with laser based autofocus using the DAPI channel at the first site of each well . The workflow for image processing and cellular feature extraction has been described elsewhere ( Bray et al . , 2016 ) , but we describe it briefly here . CellProfiler ( Carpenter et al . , 2006 ) software version 2 . 1 . 0 was used to correct the image channels for uneven illumination , and identify , segment , and measure the cells . An image quality workflow ( Bray et al . , 2012 ) was applied to exclude saturated and/or out-of focus wells; six wells containing blurry images were excluded , retaining 1914 plate/well combinations in the experiment . Cellular morphological , intensity , textural and adjacency statistics were then measured for the cell , nuclei and cytoplasmic sub-compartments . The 1402 cellular features thus extracted were normalized as follows: For each feature , the median and median absolute deviation were calculated across all untreated cells within a plate; feature values for all the cells in the plate were then normalized by subtracting the median and dividing by the median absolute deviation ( MAD ) times 1 . 4826 ( Chung et al . , 2008 ) . Features having MAD = 0 in any plate were excluded , retaining 1384 features in all . The image data along with the extracted morphological features at the per-cell level were made publicly available in the Image Data Repository under DOI http://dx . doi . org/10 . 17867/10000105 . The code repository for the profiling and all the subsequent analysis is publicly available at https://github . com/carpenterlab/2017_rohban_elife ( Carpenter , 2017 ) ( with a copy archived at https://github . com/elifesciences-publications/2016_rohban_submitted ) . We will next explain details of each analysis step implemented in the code . Single cell measurements in each well and plate position are summarized into the profiles by taking their median and median absolute deviation ( abbreviated as ‘MAD’ or ‘mad’ in some tables ) over all the cells . Although this method does not explicitly capture population heterogeneity , no alternate method has yet been proven more effective ( Ljosa et al . , 2013 ) . We also include the cell count in a sample as an additional feature . This results in a vector of 2769 elements describing the summarized morphology of cells in a sample . We then use the median polishing algorithm after obtaining the summarized profiles , to remove and correct for any plate position artifacts . For each feature , the algorithm de-trends the rows , i . e . by subtracting the row median from the corresponding feature of each profile in that particular row . Next , it de-trends the columns in a similar way using column medians . The row and column de-trending is repeated until convergence is reached in all the features . For the rest of the analysis we considered only the constructs which have more than 99% sequence identity to both the intended protein and gene transcript , to avoid testing uncharacterized mutations/truncations . Not all of the morphological features contain useful reproducible information . We first filter out features for which their replicate correlation across all samples ( except the negative controls ) is less than 0 . 30 , retaining 2200 features . Subsequently , a feature selection method is used ( Fischer et al . , 2015 ) . Briefly , starting with features ( measurements ) that we identify as essential , a new feature that contributes the most information with respect to those that have been chosen , is added to the set . The contribution of each feature to the already-selected features is measured by the replicate correlation of the residue when the feature is regressed on the already selected features . This is repeated until the incremental information added drops below a threshold . The original method proposed in ( Fischer et al . , 2015 ) overfits in its regression step when the original data is very high dimensional . As a remedy , in the regression step we only use features that have a Pearson correlation of more than 0 . 50 with the selected features thus far . This prevents overfitting of regression when the dimensionality of selected features grows . We stop feature selection when the maximum replicate correlation of residue is less than 0 . 30 . The feature selection method greatly removes redundancy , but because of the non-optimal ‘greedy’ strategy , some redundancy remains . Principal component analysis is then applied to keep 99% of variance in data , resulting in 158 principal components being selected . The features measured using CellProfiler follow a standard naming convention . Each feature name is made up of several tokens separated by underscores , in the following order: Our method to identify which genes produce a discernable profile involves first normalizing each profile to the negative controls , such that a treatment’s median replicate correlation becomes a surrogate for phenotype strength . In the case that a treatment does not show a phenotype different from the negative control , its replicates would center around the origin in the feature space . This would consequently decrease the median replicate correlation . On the other hand , a phenotype which is consistently observed in the replicates and is significantly different from the controls results in the replicates to concentrate in a region far from the origin in the feature space , and hence a high median replicate correlation value . The cutoff for ‘discernible’ is set based on the top fifth percentile of a null distribution . The null distribution is defined based on the correlations between non-replicates ( that is , different constructs ) in the experiment . Treatments whose replicate correlations are greater than the 95th percentile of the null distribution are considered as ‘hits’ that have a morphological phenotype that is highly reproducible ( Figure 2A ) . At this point , for strong treatments , all profiles of the replicates are collapsed by taking the average of individual features . 110 out of the 112 selected ORFs were significantly different from the untreated profiles in the feature space . That is , their average Euclidean distances to the untreated profiles were higher than 95th percentile of untreated profile distances to themselves . This shows these two alternative notions of phenotype strength–replicate reproducibility and distance to negative control–are consistent . We restrict all the remaining analyses to the 110 ORFs . In this analysis , mutant alleles were removed and we considered only one wild-type allele for each gene with a detectable phenotype , retaining 73 genes . We calculated a threshold to identify significantly correlated gene pairs . We picked the threshold to minimize the probability of error in classifying wild-type clone pairs versus different-gene pairs . To do so , we found the value at which the probability density functions of the two groups intersect; this value ( here , 0 . 43 ) can be proved to have the desired property ( Duda et al . , 2012 ) . This approach results in about 5% of the gene pairs being categorized as highly correlated . We next formed a two by two contingency table , where the rows correspond to two groups of gene pairs , determined by whether they have high profile correlation or not . Similarly , the columns also correspond to two groups of gene pairs , determined by whether the corresponding proteins have been reported to interact in BioGRID ( or alternatively have been annotated to be in the same pathway; Supplementary file 1C and 1D ) . This table was then used to perform a one-tailed Fisher’s exact test . A dendrogram was created based on the Pearson correlation distance and average linkage , using the hclust function in R ( Figure 3 ) . Gene clusters were formed by cutting the dendrogram at a fixed correlation level , 0 . 522 , which was chosen using a stability-based measure . The measure is defined as follows: the local clustering stability is measured for a range of candidate cutoffs , from 0 . 43 ( used earlier to test consistency to protein interaction data ) to 0 . 70 . The point with highest stability was chosen ( Figure 3—figure supplement 2 ) , and the stability measure was defined as the proportion of treatments whose clusters do not change if the cutoff is slightly changed by a small amount , ϵ= . 002 . In order to extract cell categories ( subpopulations ) and subpopulation enrichment laid over the dendrogram in Figure 3 , we applied k-means clustering on the normalized single cell data for each gene cluster and the control . Data normalization was carried out on a plate-wise basis by z-scoring each feature using the control samples as reference . In order to avoid curse of dimensionality , we restricted the dataset to the features obtained from the feature selection step mentioned earlier . We set k = 20 to be the number of subpopulations . The algorithm was run for at most 5000 iterations . Each cell was assigned to the subpopulation for which it has the shortest Euclidean distance to its center . Then , the number of cells belonging to each cell subpopulation was counted and the proportion in each subpopulation for genes in the cluster was compared against that of the control . If the change in proportion of a cell category was consistent across the genes in the cluster , the cell category is shown in the Supplementary file 2 ( type B PDFs ) . To quantify this consistency , we used the inverse coefficient of variation of the change in a category proportion . If this quantity exceeded one , we called the change consistent and included the corresponding cell category in the PDFs . Images of cells which have highest similarity to the category center in the feature space are then used to interpret and give name to each cell category ( Figure 3—figure supplement 3 ) For this purpose , we used a replicate of the original experiment but with L1000 gene-expression readouts , which is provided in the supplemental data; i . e . cell line , time point , and ORF constructs are the same . This data is different from the data used in creating GSEA plots , which entails multiple cell lines and time points . The mRNA levels are all normalized with respect to the negative control . For each replicate of the overexpression construct , we sort the expression levels of landmark genes and take the list of top and bottom 50 landmark genes . Then , to find targets of the gene related to the construct , we find the landmark genes among this list which has shown up at least in p% of replicates/clones of the gene . In particular , we set p to 33% for YAP1 , 50% for WWTR1 , TRAF2 , and REL . Then , we simply take the intersection of predicted targets of YAP1 and WWTR1 ( and similarly TRAF2 and REL , separately ) to get their common targets . These targets are then used to produce Figure 6—figure supplements 1–2 . In order to produce Figure 6C , we specified the three known targets of YAP/WWTR1 ( CYR61 , CTGF , and BIRC5 ) and queried for ORFs resulting in down-regulation of these genes . This scores each ORF ( out of the 430 in the dataset ) based on the observed change in mRNA level of the specified YAP/WWTR1 targets , across between four to nine different cell lines and between one to four time points . For each ORF , we then sought the summarized score which takes the mean of 4 largest scores across time point/cell line combinations . Finally , the ORFs were sorted based on the summarized score , and top 30 ORFs were tested for enrichment in different pathways ( Supplementary file 1I ) . We used the ‘clusterProfiler’ package in R and the KEGG pathway enrichment analysis implemented in it for creating the GSEA plot ( Yu et al . , 2012 ) . Wild-type and mutant sequences of WWTR1 ( TAZ ) ( 4SA: S66A , S89A , S117A , and S311A ) and YAP1 ( 5SA: S61A , S109A , S127A , S164A , and S397A ) were previously generated and cloned into the pCMV5 backbone; these constructs are distinct from those used in the original Cell Painting data set . TRAF2 and REL were cloned from the original constructs ( using Broad ID# ccsbBroadEn_01710 and ID# ccsbBroadEn_11094 , respectively ) , into pCMV5 expression vectors . These were sequenced and confirmed to BLAST against the appropriate Broad clone ID . The empty pCMV5 backbone was used as the control condition . The Tead luciferase reporter construct , 8xGTIIC-luciferase was a gift from Stefano Piccolo ( Addgene plasmid # 34615 ) . HEK293T cells , RRID:CVCL_0063 , were transfected using Turbofect ( ThermoFisher Scientific ) according to manufacturer’s protocol . All cells were co-transfected with a β-galactosidase reporter plasmid ( pCMV-LacZ from Clontech ) as a transfection control . Cells were lysed 48 hr following transfection . Lysates were mixed with firefly luciferase ( Promega ) according to the manufacturer’s protocol and luminescence was measured using a luminometer ( BioTek ) . Lysates were mixed with o-nitrophenyl-β-D-galactoside ( ONPG ) and β-galactosidase expression was determined spectrophotometrically by measurement of absorbance at 405 nm following ONPG cleavage . All luciferase readings were normalized to β-galactosidase expression for the sample . Statistical analysis was conducted using a two tailed unpaired Student’s t test . The data shown in Figure 6D are from triplicate samples within a single experiment and is representative of replicate experiments .	Many human diseases are caused by particular changes , called mutations , in patients’ DNA . A genome is the complete DNA set of an organism , which contains all the information to build the body and keep it working . This information is stored as a code made up of four chemicals called bases . Humans have about 30 , 000 genes built from DNA , which contain specific sequences of bases . Genome sequencing can determine the exact order of these bases , and has revealed a long list of mutations in genes that could cause particular diseases . However , over 30% of genes in the human body do not have a known role . Genes can serve multiple roles , some of which are not yet discovered , and even when a gene’s purpose is known , the impact of each particular mutation in a given gene is largely uncatalogued . Therefore , new methods need to be developed to identify the biological roles of both normal and abnormal gene sequences . For hundreds of years , biologists have used microscopy to study how living cells work . Rohban et al . have now asked whether modern software that extracts data from microscopy images could create a fingerprint-like profile of a cell that would reflect how its genes affect its role and appearance . While some genes do not necessarily carry a code with instructions of what a cell should look like , they can indirectly modify the structure of the cell . The resulting changes in the shape of the cell can then be captured in images . The idea was that two cells with matching profiles would indicate that their combinations of genes had matching biological roles too . Rohban et al . tested their approach with human cells grown in the laboratory . In each sample of cells , they ‘turned on’ one of a few hundred relatively well-known human genes , some of which were known to have similar roles . The cells were then stained via a technique called ‘Cell Painting’ to reveal eight specific components of each cell , including its DNA and its surface membrane . The stained cells were imaged under a microscope and the resulting microscopy images analyzed to create a profile of each type of cell . Rohban et al . confirmed that turning on genes known to perform similar biological roles lead to similar-looking cells . The analysis also revealed a previously unknown interaction between two major pathways in the cell that control how cancer starts and develops . In the future , this approach could predict the biological roles of less-understood genes by looking for profiles that match those of well-known genes . Applying this strategy to every human gene , and mutations in genes that are linked to diseases , could help to answer many mysteries about how genes build the human body and keep it working .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "tools", "and", "resources" ]	2017	Systematic morphological profiling of human gene and allele function via Cell Painting
Deciphering how signaling enzymes operate within discrete microenvironments is fundamental to understanding biological processes . A-kinase anchoring proteins ( AKAPs ) restrict the range of action of protein kinases within intracellular compartments . We exploited the AKAP targeting concept to create genetically encoded platforms that restrain kinase inhibitor drugs at distinct subcellular locations . Local Kinase Inhibition ( LoKI ) allows us to ascribe organelle-specific functions to broad specificity kinases . Using chemical genetics , super resolution microscopy , and live-cell imaging we discover that centrosomal delivery of Polo-like kinase 1 ( Plk1 ) and Aurora A ( AurA ) inhibitors attenuates kinase activity , produces spindle defects , and prolongs mitosis . Targeted inhibition of Plk1 in zebrafish embryos illustrates how centrosomal Plk1 underlies mitotic spindle assembly . Inhibition of kinetochore-associated pools of AurA blocks phosphorylation of microtubule-kinetochore components . This versatile precision pharmacology tool enhances investigation of local kinase biology . Protein kinase inhibitor drugs are an emerging class of therapeutics for a variety of clinical indications ( Ferguson and Gray , 2018 ) . These small molecules are also powerful research tools that can be used to discover new aspects of kinase signaling ( Caunt et al . , 2015 ) . While ‘drugging’ individual kinases can establish their role in cellular events , this global approach cannot discriminate where or when these signaling enzymes operate inside the cell . Thus , designing pharmacological strategies that influence the spatial and temporal action of kinases is at the frontier of precision medicine . Polo-like kinase 1 ( Plk1 ) and Aurora A ( AurA ) are important regulators of cell division ( Barr et al . , 2004; Combes et al . , 2017; Lens et al . , 2010 ) . Accordingly , ATP-competitive drugs that block their activity , such as BI2536 and MLN8237 , ascribe functions to these kinases and are promising anticancer therapies ( Steegmaier et al . , 2007; Tang et al . , 2017; Lénárt et al . , 2007; Asteriti et al . , 2014; Manfredi et al . , 2011 ) . However , elucidating the individual spatial and temporal actions of Plk1 and AurA remains challenging as these enzymes continually change their location and activity throughout mitosis ( Bruinsma et al . , 2015; Joukov and De Nicolo , 2018; Lera et al . , 2016 ) . As a result , global drug delivery strategies mask the unique contributions of each kinase at distinct mitotic structures . Moreover , standard drug regimens that saturate dividing cells with these compounds may increase off-target effects and toxicity ( Klaeger et al . , 2017 ) . Plk1 and AurA have been implicated in the control of mitotic progression ( Barr et al . , 2004; Combes et al . , 2017; Lens et al . , 2010 ) . The anchoring protein Gravin/AKAP12 participates in this process by forming a macromolecular complex with these enzymes ( Hehnly et al . , 2015 ) . A-kinase anchoring proteins ( AKAPs ) are scaffolding proteins that limit the scope of cell signaling events at distinct cellular locations ( Scott and Pawson , 2009; Langeberg and Scott , 2015; Esseltine and Scott , 2013 ) . For example , anchored protein kinase A action is constrained to within 200–400 angstroms of the AKAP ( Smith et al . , 2013; Smith et al . , 2017 ) . This has led to the formulation of a signaling island model where catalytic activity of anchored kinases is restricted to the immediate vicinity of select substrates ( Scott and Pawson , 2009; Langeberg and Scott , 2015; Esseltine and Scott , 2013 ) . Likewise , anchoring of Plk1 and AurA occurs on a phosphorylated species of Gravin at Threonine 766 ( Canton et al . , 2012; Hehnly et al . , 2015; Colicino et al . , 2018 ) . Consequently , loss or disruption of this scaffold abrogates Plk1 and AurA organization at centrosomes and promotes mitotic delay ( Hehnly et al . , 2015 ) . Yet , an outstanding question that remains is exactly how do centrosome-localized pools of Gravin-anchored Plk1 and AurA coordinate mitotic signaling events . In the present study , we first establish that Gravin is required for localizing active pools of Plk1 and AurA at mitotic centrosomes . We then develop a novel chemical-biology tool , LoKI ( Local Kinase Inhibition ) , to probe the actions of Plk1 and AurA at defined subcellular locations . Finally , we demonstrate that local inhibition of Plk1 and AurA kinases at centrosomes and kinetochores disrupts substrate phosphorylation , spindle organization , and mitotic duration . Together , these studies decipher how activities of individual kinases at precisely defined microenvironments contribute to the global signaling events that underlie mitosis . Gravin depletion via shRNA-mediated knockdown perturbs mitotic progression ( Hehnly et al . , 2015 ) . Here , we test whether complete loss of Gravin also prolong mitosis . We used time-lapse video microscopy to monitor mouse embryonic fibroblasts ( MEFs ) from wild-type and Gravin knockout ( KO ) mice ( Figure 1A , Figure 1—figure supplement 1A ) . Live-cell imaging of wild-type MEFs expressing GFP-tagged histone 2B ( Figure 1—video 1 ) established a baseline mitotic duration ( nuclear envelope breakdown to anaphase ) as 32 . 6 min ( Figure 1B ) . In contrast , mitosis was delayed by 11 . 9 min in Gravin KO cells ( Figure 1B ) . These data further establish that Gravin promotes timely progression of cells through mitosis ( Gelman , 2010 ) . Gravin is required for organization of Plk1 and AurA at mitotic centrosomes ( Hehnly et al . , 2015 ) . Whether Gravin loss reduces active pools of Plk1 and AurA at this location remains unclear . To test this , we explored whether activity of centrosome-localized Plk1 and AurA is perturbed in cells lacking Gravin . We examined pT210-Plk1 and pT288-AurA immunofluorescence ( measures of Plk1 and AurA activity , respectively ) in HEK293 cells stably expressing a control or Gravin shRNA ( Figure 1C , D ) . Gravin depletion reduced immunofluorescence of pT210-Plk1% to 80 . 2% of control shRNA-expressing cells ( Figure 1C , E ) . Strikingly , pT288-AurA signal in Gravin-depleted cells dropped to 58 . 5% of control cell immunofluorescence ( Figure 1D , F ) . These findings reveal that Gravin is required for localizing active pools of Plk1 and AurA at mitotic centrosomes . To decipher how Gravin-associated pools of Plk1 and AurA coordinate mitotic signaling events , we needed to selectively target drugs to this location without disrupting the Gravin-Plk1-AurA macromolecular complex . During interphase Gravin is dispersed throughout the cell ( Figure 1G , Figure 1—figure supplement 1B ) . In mitosis , however , pT766-Gravin accumulates at centrosomes , a major hub of Plk1 and AurA signaling ( Figure 1G , Figure 1—figure supplement 1B ) . This provided the impetus to pharmacologically inhibit mitotic kinases at centrosomes ( Figure 1H ) . A key advance in our studies came with the development of the LoKI tool which allows us to target kinase inhibitor drugs to specific subcellular locations . We fused a pericentrin AKAP450 centrosomal-targeting ( PACT ) domain ( Gillingham and Munro , 2000 ) to a SNAP-tag moiety which can be covalently labeled with chloropyrimidine ( CLP ) -linked substrates inside cells ( Keppler et al . , 2003 ) ( Figure 2A ) . A CLP-conjugated analog of BI2536 ( CLP-BI2536 ) was generated to selectively target Plk1 ( Figure 2A , B , Figure 2—figure supplement 1A ) . In vitro kinase activity measurements demonstrated that CLP-BI2536 potently inhibits Plk1 ( IC50 = 49 ± 26 nM; Figure 2C , Figure 2—figure supplement 1B ) . To generate stable cell lines , U2OS osteosarcoma cells were infected with lentiviral constructs encoding the SNAP-PACT moieties fused to an mCherry reporter ( Figure 2—figure supplement 1C ) . Inducible protein expression was accomplished by a doxycycline-inducible promoter ( Figure 2—figure supplement 1D ) . Immunoblot detection of mCherry-SNAP-PACT persisted up to 4 hr upon removal of doxycycline ( Figure 2—figure supplement 1E ) . As anticipated , mCherry-SNAP-PACT associates with centrosomes during interphase and mitosis ( Figure 2—figure supplement 1F ) . Super-resolution structured illumination ( SIM ) imaging revealed that the SNAP-PACT construct ( magenta ) was labeled by CLP-fluorescein ( yellow ) at centrosomes ( Figure 2D , E , Figure 2—video 1 ) . Counterstaining with α-tubulin ( green ) revealed the mitotic spindle and DAPI ( blue ) detected DNA ( Figure 2D , Figure 2—video 2 ) . Collectively these results demonstrate that centrosomal targeting of SNAP-PACT creates a platform for the delivery of CLP-conjugates ( Figure 2—figure supplement 1G ) . This new drug targeting method is herein referred to as LoKI-on ( Local Kinase Inhibition-on ) . In parallel , a LoKI-off vector containing an inactivating mutation ( C144A ) in SNAP-tag was constructed ( Figure 2—figure supplement 1H ) . LoKI-off is unable to incorporate CLP-conjugates and serves as the control platform ( Figure 2—figure supplement 1H ) . Pulse-chase experiments were used to determine how efficiently CLP-BI2536 labeled SNAP-PACT . U2OS cells were treated with CLP-BI2536 ( over a range of concentrations ) to block CLP-rhodamine conjugation ( Figure 2F ) . Incubation with 250 nM CLP-BI2536 for 4 hr at 37°C was defined as the optimal drug regimen ( ~50% labeling of SNAP-PACT; Figure 2F , Figure 2—figure supplement 2A , B ) . Next , we measured the pT210-Plk1 immunofluorescence signal as an index of active kinase ( Lee and Erikson , 1997 ) ( Figure 2—figure supplement 3A , B ) . In mitotic cells expressing LoKI-off , incubation with 250 nM CLP-BI2536 reduced pT210-Plk1 immunofluorescence to 58 . 1% of DMSO-treated controls ( Figure 2G , I ) . Strikingly , the pT210-Plk1 signal was reduced to 21 . 4% in cells expressing LoKI-on ( Figure 2H , I ) . This trend persisted with lower CLP-BI2536 concentrations and even after a 1 hr washout of drug ( Figure 2I , J ) . Further validation confirmed that the reduction of pT210-Plk1 does not result from a loss in total Plk1 protein at centrosomes ( Figure 2—figure supplement 3C ) . Additional controls established that inducible expression of LoKI-on was necessary to attenuate the pT210-Plk1 signal ( Figure 2—figure supplement 3D ) . Immunoblot analyses of nocodazole-synchronized cells collected via mitotic shake-off further support these findings ( Figure 2—figure supplement 3E ) . Parallel analyses were conducted in HeLa and hTERT-immortalized RPE retinal pigment epithelial cells ( Figure 2—figure supplement 3F–J ) . We note that due to clonal cell line differences between LoKI-off and LoKI-on cells , baseline immunofluorescence signal was normalized to DMSO-treated controls ( Figure 2—figure supplement 4 ) . Collectively , these findings establish LoKI as a new pharmacological tool to selectively block Plk1 activity at centrosomes . Correct assembly of bipolar spindles ensures the fidelity of chromosome segregation into daughter cells ( Prosser and Pelletier , 2017 ) . Abrogation of Plk1 activity has been linked to mitotic spindle defects that include abnormal bipolar and monopolar structures ( Sunkel and Glover , 1988; Lane and Nigg , 1996; Sumara et al . , 2004 ) ( Figure 3A ) . Spindle classification measurements were carried out to assess if centrosomal inhibition of Plk1 induces these morphological anomalies ( Figure 3B ) . Analysis in U2OS cells revealed that application of CLP-BI2536 in LoKI-on cells increased the incidence of abnormal bipolar ( green ) and monopolar ( purple ) spindles by 24 . 6% as compared to LoKI-off controls ( 10 . 3%; Figure 3B , C , Figure 3—figure supplement 1A ) . More pronounced spindle defects were observed when local drug delivery experiments were repeated in RPE cells ( Figure 3D , Figure 3—figure supplement 1B ) . Interestingly , local delivery of CLP-BI2536 did not further exacerbate defective spindle organization in Hela cells , which naturally exhibit a high incidence of aberrant spindles ( Figure 3E , Figure 3—figure supplement 1C ) . Thus , targeting Plk1 inhibitor drugs to centrosomes promotes mitotic spindle defects in various cell types . Spindle assembly relies on γ-tubulin , a protein that interacts with α/β-tubulin polymers ( Moritz et al . , 1995; Zheng et al . , 1995 ) . Plk1 phosphorylates pericentriolar substrates that coordinate γ-tubulin accumulation at mitotic centrosomes to facilitate microtubule nucleation ( Lane and Nigg , 1996; Haren et al . , 2009; Xu and Dai , 2011 ) ( Figure 3F ) . Accordingly , we monitored centrosomal accumulation of γ-tubulin in U2OS cell after application of CLP-BI2536 for 4 hr followed by a 1 hr washout ( Figure 3G–I , Figure 3—figure supplement 1D ) . The γ-tubulin signal ( yellow ) accumulated at centrosomes in LoKI-off controls ( Figure 3G , I ) . However , γ-tubulin levels were drastically reduced when LoKI-on cells were exposed to the same drug regimen ( Figure 3H , I ) . Amalgamated data from five independent experiments are presented ( Figure 3I ) . These findings indicate that targeting CLP-BI2536 to centrosomes impairs accumulation of γ-tubulin at this location . A versatile feature of the LoKI system is the ability to compartmentalize a variety of drug analogs . The AurA inhibitor MLN8237 was a logical candidate to highlight the broad applicability of this approach . CLP-MLN8237 was synthesized ( Figure 4A , Figure 4—figure supplement 1A ) . In vitro kinase activity measurements demonstrated that CLP-MLN8237 potently inhibits AurA ( IC50 <9 . 5 nM; Figure 4B , Figure 4—figure supplement 1B ) . Cell-based characterization established that treatment with 100 nM CLP-MLN8237 for 4 hr at 37°C was sufficient to label ~50% of drug binding sites ( Figure 4C ) . Next , we used immunofluorescent detection of pT288-AurA as an index of kinase activity ( Figure 4—figure supplement 2A , B ) . In mitotic cells expressing LoKI-off , incubation with 100 nM CLP-MLN8237 reduced pT288-AurA immunofluorescence to 27 . 8% of DMSO-treated controls ( Figure 4—figure supplement 2C ) . Importantly , the pT288-AurA signal was further reduced to 14 . 2% in cells expressing LoKI-on ( Figure 4D , E ) . Thus , increasing the local concentration of MLN8237 enhances drug action by approximately 2-fold at this subcellular location . These data demonstrate the versatility of the LoKI system as a pharmacological platform to block distinct kinases that operate at mitotic centrosomes . To investigate the coordinate activities of Plk1 and AurA at the centrosome ( Asteriti et al . , 2015 ) we took advantage of another feature of the LoKI-on platform , the ability to co-localize CLP-drug combinations via a dual SNAP conjugation moiety ( Figure 4F ) . Live-cell imaging of U2OS cells expressing GFP-tagged histone 2B ( Figure 4G , Figure 4—figure supplement 2D , Figure 4—videos 1 and 2 ) was used to calculate a baseline for mitotic timing ( nuclear envelope breakdown to anaphase ) as 35 . 1 min ( Figure 4H ) . Mitosis was delayed by 19 . 4 min when CLP-BI2536 ( 250 nM ) and CLP-MLN8237 ( 100 nM ) were simultaneously applied to LoKI-off cells ( Figure 4G , H ) . However , the same combination treatment prolonged mitosis 3-fold ( 59 min delay ) when experiments were repeated in LoKI-on cells ( Figure 4G , H ) . Moreover , the mitotic duration observed in LoKI-on cells after combination treatment extended beyond what was seen with either inhibitor alone ( Figure 4—figure supplement 3A , B ) . These data accentuate the utility of LoKI-on as a means to direct drug combinations to defined cellular locations in space and time . Zebrafish provide an excellent model organism to test local drug action using the LoKI system because their transparency simplifies imaging analysis ( Zon and Peterson , 2005 ) . Zebrafish embryos were microinjected with mCherry-LoKI-on mRNA and allowed to develop for 5 hr until they reached ~50% epiboly ( Figure 5A , B ) . Detection of mCherry fluorescence confirmed expression of the local drug-targeting construct ( Figure 5B ) . Higher resolution imaging of fixed embryos confirmed accumulation of LoKI-on at centrosomes during interphase and mitosis ( Figure 5—figure supplement 1A ) . Co-distribution of the SNAP moiety ( magenta ) with CLP-647 dye ( yellow ) confirmed assembly of the drug-targeting platform at centrosomes ( Figure 5—figure supplement 1B ) . Microinjection of the Plk1 inhibitor adduct CLP-BI2536 ( 250 nM ) permitted local drug delivery . Live-cell imaging 5 hr post injection exposed a range of adverse mitotic phenotypes . Mitotic spindles were visualized using a microtubule binding protein , EMTB-3xGFP ( Figure 5C ) . Multipolar spindles , spindle orientation defects , and prolonged mitoses were evident in drug-treated embryos expressing LoKI-on ( Figure 5C , Figure 5—video 1 ) . Fixed-cell imaging of whole embryos revealed intact microtubule organization and few mitotic cells in LoKI-off embryos treated with CLP-BI2536 ( Figure 5D ) . In contrast , drug-treated LoKI-on embryos exhibited microtubule abnormalities and a higher incidence of mitotic cells ( Figure 5E ) . Fluorescent detection of the SNAP moiety confirmed centrosomal targeting of LoKI platforms ( Figure 5D , E ) . Additional analyses correlated centrosomal inhibition of Plk1 with increased mitotic indices in LoKI-on embryos ( Figure 5F ) . Conversely , control experiments in LoKI-off embryos showed that CLP-BI2536 had minimal effect , as indicated by a significantly higher proportion of interphase cells ( Figure 5F ) . Thus , targeted delivery of kinase inhibitor drugs to mitotic centrosomes induces a range of adverse mitotic phenotypes in developing embryos relative to global drug application . Kinetochores are proteinaceous structures that ensure the proper attachment of spindle microtubules to the centromeric region of condensed chromatin ( Hinshaw and Harrison , 2018 ) ( Figure 6A ) . To further demonstrate the versatility of the LoKI system , we utilized a targeting domain from the kinetochore protein Mis12 ( Goshima et al . , 2003 ) ( Figure 6A ) . Inducible expression of the mCherry-tagged fusion was accomplished by a doxycycline-inducible promoter ( Figure 6B , Figure 6—figure supplement 1A ) . Immunoblot detection of Mis12-LoKI-on persisted up to 4 hr upon removal of doxycycline ( Figure 6—figure supplement 1B ) . Immunofluorescent staining revealed that mCherry-tagged Mis12-LoKI-on co-localized with centromeric DNA ( anti-centromere antibodies ( ACA ) , cyan ) at kinetochores during mitosis ( Figure 6C , Figure 6—video 1 ) . Counterstaining with α-tubulin antibodies ( green ) revealed the mitotic spindle ( Figure 6C ) . SIM imaging revealed that a CLP-dye ( CLP-647 , yellow ) accumulated with the SNAP moiety ( magenta ) at kinetochores of Mis12-LoKI-on cells ( Figure 6D ) . In contrast , recruitment of CLP-647 was not evident in Mis12-LoKI-off cells ( Figure 6D ) . Line plot analyses of the CLP-647 signal in selected kinetochores emphasizes this result ( Figure 6E ) . Pulse-chase experiments established that incubation with 100–250 nM of the AurA inhibitor adduct CLP-MLN8237 for 4 hr at 37°C was the optimal drug regimen ( ~50% of drug binding sites occupied; Figure 6F ) . Parallel , validation studies were performed with CLP-BI2536 ( Figure 6—figure supplement 1C ) . Roles for AurA at centrosomes and mitotic spindles are well documented ( Nikonova et al . , 2013 ) . However , recent reports have implicated AurA as a modulator of microtubule attachment to kinetochores ( Chmátal et al . , 2015; Ye et al . , 2015 ) . At this location AurA phosphorylates serine 69 in Hec1 , a subunit of the NDC80 complex ( DeLuca et al . , 2018 ) . This local phosphorylation stabilizes microtubule-kinetochore interaction to safeguard chromosome segregation ( DeLuca et al . , 2018 ) . However , the proximity of centrosomes to kinetochores in early mitosis has hampered attempts to resolve the contribution of discrete AurA pools . Therefore , immunofluorescent detection of pS69-Hec1 served as an index for local AurA kinase activity at kinetochores ( Figure 6G , H ) . As before , counterstaining for α-tubulin ( green ) and DNA ( blue ) revealed the mitotic spindle ( Figure 6G , H ) . In cells expressing Mis12-LoKI-off , incubation with 100 nM CLP-MLN8237 caused a negligible decrease in pS69-Hec1 signal as compared to DMSO-treated controls ( Figure 6G , I ) . Conversely , in Mis12-LoKI-on cells the pS69-Hec1 signal was reduced to 59 . 8% ( Figure 6H , I ) . Representative heat maps further illustrate this phenomenon ( Figure 6G , H ) . Importantly , the pS69-Hec1 signal at centrosomes was unaffected by drug treatments ( Figure 6G , H , Figure 6—figure supplement 2A ) . Amalgamated data from three independent experiments reveal that this trend persisted even at higher concentrations ( Figure 6I ) . Immunoblot analyses of nocodazole-synchronized cells collected via mitotic shake-off further support our findings ( Figure 6—figure supplement 2B ) . Finally , when SNAP-PACT LoKI-expressing cells were used to sequester CLP-MLN8237 at centrosomes , we no longer observed a reduction in pS69-Hec1 signal at kinetochores ( Figure 6—figure supplement 2C ) . These data further support a role for AurA at the kinetochore . Ultimately , our findings illustrate how LoKI platforms can be adapted to pharmacologically investigate kinase signaling at distinct subcellular locations within the mitotic cell . Cells have evolved a highly organized architecture that is segregated into functionally distinct microenvironments ( Figure 6J ) . However , traditional methods of drug delivery do not account for this exquisite degree of molecular organization . Conventional approaches flood cells with bioactive compounds , masking the unique contributions of individual kinases at distinct subcellular locations . Although it is well established that Plk1 and AurA coordinate various aspects of cell division , current drug-targeting strategies limit our ability to decode the spatiotemporal regulation of these events ( Barr et al . , 2004; Combes et al . , 2017; Lens et al . , 2010 ) . Studying molecular scaffolds that form complexes with these key mitotic enzymes provides important mechanistic insight into how these processes are coordinated ( Figure 1; Hehnly et al . , 2015 ) . Moreover , designing pharmacological tools that restrict the spatial and temporal action of kinase inhibitor drugs is paramount to deciphering local kinase action . In this study , we discovered that the anchoring protein Gravin is required for organizing active pools of Plk1 and AurA at centrosomes ( Figure 1C , D ) . These data support previous findings in which Gravin loss led to increased Plk1 mobility and aberrant CEP215 phosphorylation ( Canton et al . , 2012; Hehnly et al . , 2015; Colicino et al . , 2018 ) . Thus , we suggest that Gravin constrains enzymes in a signaling island to provide spatiotemporal control of kinase activity . Conversely , depletion of this anchoring protein alters Gravin-Plk1 and Gravin-AurA protein-protein interactions which underlie healthy cellular function . This further emphasizes a need for designing strategies that inhibit kinase activity locally . For this reason , we developed a novel chemical-biology tool , LoKI , to more precisely probe the actions of Plk1 and AurA at centrosomes and kinetochores . Previous work from Gower and colleagues utilized antibody mimetics to target promiscuous inhibitor drugs to specific kinases ( Gower et al . , 2016 ) . Our strategy advances this technology by combining AKAP-targeting domains with SNAP-tagging technologies . Additionally , we direct selective ATP-competitive inhibitors to specific subcellular locations to achieve local kinase inhibition . By combining biochemical approaches , quantitative imaging , and live-cell microscopy we reveal that local targeting of Plk1 and AurA kinase inhibitor drugs disrupts substrate phosphorylation , spindle organization , and mitotic duration more profoundly than global drug distribution . Thus organellar targeting of drugs offers a new means to advance the investigation of broad-spectrum kinases at precise locations . Previous studies suggest that Plk1 phosphorylates pericentriolar substrates that coordinate γ-tubulin accumulation at mitotic centrosomes to facilitate microtubule nucleation ( Lane and Nigg , 1996; Haren et al . , 2009; Xu and Dai , 2011 ) . We advance this concept and extend these findings by demonstrating that centrosomal inhibition of Plk1 prevents accumulation of γ-tubulin and correct organization of bipolar mitotic spindles ( Figure 3 ) . Thus , by using the LoKI system we are able to definitively establish that Plk1 activity at mitotic centrosomes is a driver in these processes . Furthermore , the utility of LoKI drug targeting was underscored by our in vivo studies using zebrafish embryos . We provide evidence that embryos treated with centrosome-targeted Plk1 inhibitors have more microtubule abnormalities than those treated with a non-localized inhibitor ( Figure 5C–E ) . These data implicate centrosome-localized pools of Plk1 in coordinating mitotic events such as spindle organization and mitotic progression during early zebrafish development . In a broader context , we show that local targeting of Plk1 inhibitors in developing organisms offers an innovative precision technique to probe local drug action . Another key advance in our studies came with the discovery that AurA-mediated Hec1 phosphorylation is a spatially-coordinated event that occurs at kinetochores . We provide quantitative imaging ( Figure 6G–I , Figure 6—figure supplement 2C ) and biochemical ( Figure 6—figure supplement 2B ) data that implicates AurA activity at this mitotic substructure . Although this kinase was originally thought to reside exclusively at centrosomes and mitotic spindles , our findings extend recent evidence for the existence of an AurA pool at kinetochores ( DeLuca , 2017 ) . We reveal that kinetochore-targeted MLN8237 reduces pS69-Hec1 signal more drastically than globally distributed drug ( Figure 6H–I ) . Our data suggest that even during prometaphase , when kinetochores may encounter centrosome-associated AurA , this phosphorylation is solely a kinetochore-associated event . Furthermore , when we target the AurA inhibitor to centrosomes and measure pS69-Hec1 at kinetochores we no longer see a loss of pS69-Hec1 signal ( Figure 6—figure supplement 2C ) . Thus , we show that centrosome-associated AurA is not likely responsible for this phosphorylation event as has been previously suggested ( Chmátal et al . , 2015; Ye et al . , 2015 ) . More importantly , these findings uncover that S69-Hec1 phosphorylation is a local event that depends on AurA activity at kinetochores . This allows us to postulate that isolated pockets of AurA may act independently and concurrently to orchestrate complex cellular events . The versatility of this new chemical-biology platform is demonstrated in three ways . First , this approach works in a variety of cell types and microinjection of LoKI mRNA into live zebrafish embryos permits local drug targeting in vivo ( Figure 5 ) . We foresee that LoKI platforms will be adapted to acutely probe local signaling in other genetically tractable organisms . Second , while derivatized Plk1 and AurA drugs delineate roles for each mitotic kinase , conjugation of chloropyrimidine ( CLP ) to other ATP analogs offers a general method to synthesize localizable inhibitors for additional members of the kinome ( Gower et al . , 2016 ) . However , it is worth noting that the reduced cell permeability of certain CLP-drug conjugates , including CLP-BI2536 , may necessitate their use at approximately 10-fold higher concentrations than the unmodified drugs ( Figure 2I versus Figure 2—figure supplement 3B , Figure 4E versus Figure 4—figure supplement 2B ) . Additionally , derivatization of certain inhibitors may sterically hinder their access to the ATP-binding pockets of some kinases or , as is the case of the PKA antagonist H89 , the lack of a functional group prevents CLP derivatization . Third , plasma membrane and mitochondrial targeting domains from AKAP79 and dAKAP1 expand the repertoire of subcellular compartments reached by LoKI platforms ( Figure 6J ) . Although our strategy uncovers local contributions of anchored kinase pools within the cell , certain limitations to our current approach exist . For example , we treat cells for 4 hr with CLP-conjugated drug adducts to achieve sufficient inhibitor targeting ( ~50% of drug binding sites occupied; Figure 2F , Figure 6F ) . This is a relatively long time period in the context of measuring cell-cycle events . Likewise , in cells we find that high concentrations of CLP-drug ( 100 nM ) are required to produce equivalent effects as 10 nM of non-derivatized MLN8237 ( Figure 4E versus Figure 4—figure supplement 2B ) . We postulate that reduced cell permeability of certain CLP-drug conjugates may account for both of the aforementioned findings . As a result , this necessitates long incubation periods and application of higher concentrations of drug . Finally , it is possible that kinase inhibitor drugs directed to centrosomes have off-target effects at nearby structures such as spindle microtubules . We envision that future advancements of this platform would include photo-caged inhibitors that are inert until they are ready to be released at the site of desired inhibition ( Ellis-Davies , 2007 ) . Employing this strategy would more strictly define the range of inhibitor action , provide another level of control to the LoKI system , and allow us to better delineate the effects of global versus local inhibition . One exciting feature of our LoKI platform is the ability to co-localize CLP-drug combinations via a dual SNAP conjugation moiety ( Figure 4F ) . Although in our study this allowed combined inhibition of Plk1 and AurA at centrosomes , we hope that future work will advance on our strategy and provide a system that utilizes multiple self-labeling enzymes to deliver distinct inhibitors to the same location . Employing orthogonal tagging systems such as CLIP-tag or Halo-tag in tandem with SNAP-tag could eliminate the possibility that CLP-BI2536 and CLP-MLN8237 compete for conjugation to the same targeting moiety . Nonetheless , by exploiting our knowledge of how AKAPs compartmentalize signaling enzymes we have developed tools that define the local kinase terrain at the angstrom level . This will allow investigators to probe local signaling events at a level of precision that has not been possible before . SNAP , mCherry , eGFP , PACT , Mis12 , AKAP79 , and dAKAP1 components and were individually PCR amplified with overlapping ends and/or Gateway ‘att’ sites and assembled using Gibson Cloning . Gateway cloning was carried out to subclone SNAP constructs into pLIX402 ( a gift from David Root; Addgene plasmid #41394 ) for PACT and Mis12 studies or pcDNA3 . 1+ ( Life Technologies ) for AKAP79 and dAKAP1 . To generate mutant SNAP , site-directed mutagenesis was performed with a QuikChange II XL kit ( Aligent ) . GFP-H2B and EMTB-3xGFP constructs were used for live-cell imaging studies . Constructs were verified by Sanger sequencing . Cells used to generate stable cell lines in this study originated as follows: U2OS ( purchased from ATCC ) , HeLa ( from L . Wordeman lab and maintained in-house ) , and hTERT-RPE ( gift from P . Jallepalli lab and maintained in-house ) . HeLa and hTERT-RPE cells were tested by STR at ATCC . Chang Liver cells , a HeLa contaminant , were detected in the HeLa line while hTERT-RPE cells were an exact match to ATCC cell line CRL-4000 ( hTERT-RPE-1 ) . U2OS , HeLa , and hTERT-RPE cells tested negative for mycoplasma contamination as assessed by the Universal Mycoplasma Detection Kit ( ATCC 30-1012K ) . U2OS , HeLa , Control and Gravin shRNA HEK293 ( Canton et al . , 2012 ) , and immortalized MEF ( generated as described in Hehnly et al . ( 2015 ) and maintained in-house ) cells were maintained in DMEM , high glucose and hTERT-RPE cells were maintained in DMEM/F-12 , Hepes ( Life Technologies ) at 37°C and 5% CO2 . All media was supplemented with 10% FBS . Infections for generation of stable SNAP cells were performed using lentiviral particles created in-house . In brief , SNAP pLIX402 vectors were transfected alongside pMD2 . G and psPAX2 plasmids ( gifts from Didier Trono; Addgene plasmid #12259 [RRID:Addgene_12259] and plasmid #12260 [RRID:Addgene_12260] ) ) into HEK293 cells using Lipofectamine 2000 reagent ( Invitrogen ) in Opti-MEM ( Life Technologies ) media . Virus-containing supernatant was collected , passed through a . 45 µm filter , and transduced into cells in the presence of 1 µg/ul Polybrene ( Santa Cruz ) . Cells were selected and maintained in supplemented media with 4 µg/mL Puromycin dihydrochloride ( Santa Cruz ) . Single clones were isolated using Scienceware cloning discs ( Sigma-Aldrich ) . Infections for generation of stable knockdown in HEK293 cells were performed with shRNA lentiviral particles ( Santa Cruz Biotech ) as described in Canton et al . ( 2012 ) . For expression of AKAP79 , dAKAP1 , and Gravin constructs in U2OS cells , transient transfections were performed using TransIT-LT1 reagent ( Mirus ) in Opti-MEM ( Life Technologies ) media according to manufacturer’s instructions . 1 Eq 1d ( 0 . 2 M ) and 1 . 1 Eq of 2a ( Chem Scene ) were dissolved in DMF at RT . The reaction was placed on ice . While stirring , 1 . 3 Eq HOAt and 3 Eq DIEA were added . After 5 min on ice , 1 . 3 Eq of EDCI was added . The reaction was allowed to stir for 24 hr ( letting the ice melt and the reaction slowly come to RT ) . DMF was removed and 2b ( BI2536-CLP ) was purified with HPLC . Identity was verified with MS . [M+H]+ = 817 . 7 m/z . 1 Eq 1d ( 0 . 2 M ) and 1 . 1 Eq of 3a ( MLN8237 ) were dissolved in DMF at RT . The reaction was placed on ice . While stirring , 2 Eq HOAt and 3 Eq DIEA were added . After 5 min on ice , 1 . 2 Eq of EDCI was added . The reaction was allowed to stir for 24 hr ( letting the ice melt and the reaction slowly come to RT ) . DMF was removed and 3b ( MLN8237-CLP ) was purified with HPLC . Identity was verified with MS . [M+H]+ = 911 . 0 m/z . CLP-rhodamine Preparation: 1 Eq 1a ( 0 . 2 M ) and 1 Eq 5 ( 6 ) -carboxytetramethylrhodamine N-succinimidyl ester ( Thermo Fisher ) were dissolved in DMF at RT . While stirring , 3 Eq DIEA were added . The reaction was allowed to stir for 24 hr . DMF was removed and product was purified with HPLC . Identity was verified with MS . [M+H]+ = 676 . 2 m/z . His6-SNAP-tag in pMCSG7 ( Addgene ) was expressed in Escherichia coli BL21 ( DE3 ) cells in 250 mL LB Miller broth . The evening prior to expression , 5 mL LB Miller broth , containing 50 μg/mL Ampicillin , was inoculated with transformed cells , and they were grown at 37°C overnight . The following day , the starter culture was used to seed 250 mL LB Miller broth in a 500 mL baffle flask . Cells were grown to OD600 ~0 . 3 and the temperature was then reduced to 20°C . Cells were allowed to grow to OD600 ~0 . 8 , and then induced with 500 μM isopropyl β-D-thiogalactopyranoside . Induced cells were grown at 20°C overnight . Subsequent purification steps were carried out at 4°C . Cells were spun down at 6500 g , suspended in 10 mL of wash/lysis buffer [50 mM HEPES ( pH 7 . 5 ) , 300 mM NaCl , 20 mM imidazole , and 1 mM phenylmethanesulfonyl fluoride] , and lysed via sonication . The lysate was centrifuged at 10000 g for 20 min , and the supernatant was allowed to batch bind with 0 . 7 mL of Ni-NTA ( Ni2+-nitrilotriacetate ) beads for 60 min . The resin was collected by centrifugation at 500 g for 5 min and washed with 10 mL of wash/lysis buffer . The wash step was repeated three times . The Ni-NTA/His6-SNAP-tag was added to a BioRad purification column , and washing was continued until the wash showed no remaining protein by Bradford . The protein was eluted using ∼ 5 mL of elution buffer [50 mM HEPES ( pH 7 . 5 ) , 300 mM NaCl , 200 mM imidazole] . The eluate was dialyzed against 50 mM HEPES ( pH 7 . 5 ) , 200 mM NaCl , 5% glycerol , and 1 mM fresh dithiothreitol ( DTT ) . Protein was aliquoted , flash-frozen in liquid N2 , and stored at − 80°C . For induction of SNAP expression cells were treated for 48–72 hr in FBS-supplemented DMEM with 1 µg/mL ( for SNAP-PACT ) or 4 µg/mL ( for SNAP-Mis12 ) doxycycline hyclate ( Sigma-Aldrich ) prior to inhibitor treatments . For degradation assays , cells were dox-induced for 72 hr after which doxycycline was washed out ( cells were incubated in normal media ) . At selected time point plates were collected , cells were washed once with PBS , plates were dried quickly , and frozen at −80°C until lysis . For nocodazole synchronization experiments , dox-induced cells were treated for 16 hr with nocodazole and 4 hr with nocodazole plus DMSO , 250 nM CLP-BI2536 , or 100 nM CLP-MLN8237 . Cells were washed once with PBS , collected via mitotic shake-off , and spun at 2000 rpm for 5 min at 4°C . Supernatants were discarded and pellets were kept for lysis . All lysates were prepared as described under ‘immunoblotting’ . For fixed cell experiments , both dox-induced and non-induced cells were grown on 1 . 5 poly-D-lysine coated coverslips ( neuVitro ) for at least 16 hr in complete DMEM and then treated with DMSO or CLP-compounds in serum-free DMEM for 1–4 hr . For washout experiments ( pT210-Plk1 1 hr washout and γ-tubulin data ) , cells were incubated in serum-free DMEM without inhibitors for an additional 1 hr . Cells were washed once with PBS prior to fixation . For live-imaging experiments , cells were treated with CLP-compounds for 18 hr ( see ‘microscopy’ for more details ) . Zebrafish were bred and embryos were collected . Embryos were injected with EMTB-3xGFP ( 100 pg mRNA or 20 pg pCS2 plasmid construct ) , SNAP-Cell Fluorescein ( 300 µM final embryo concentration ) , and/or CLP-BI2536 ( 250 nM final embryo concentration ) , and/or SNAP-PACT active or dead ( 200 pg mRNA ) at the 2 cell stage using a microinjector ( Warner Instruments PLI-100A ) with a Kanatec magnetic base ( MB-B ) , and a micromanipulator ( Marzhauser Wetzlar MM33 ) . Embryos were raised at 30°C until ~50% epiboly , at which point they were imaged on stereoscope , mounted in 2% agarose for live confocal imaging , or fixed using 4% paraformaldehyde + 0 . 5% Triton-X overnight at 4°C . Embryos were dechorionated in PBST ( phosphate buffered saline + 0 . 1% Tween-20 ) and incubated in DAPI solution ( 1 µg/mL in PBS ) for 2 hr at room temperature . Fixed embryos were then mounted in 2% agarose and imaged on confocal microscope . Cells were lysed in RIPA lysis buffer ( 50 mM Tris HCl pH 7 . 4 , 1% Triton X-100 , 0 . 5% Sodium Deoxycholate , 0 . 1% SDS , 50 mM NaF , 120 mM NaCl , 5 mM β-glycerophosphate ) supplemented with protease and phosphatase inhibitors ( 1 mM benzamidine , 1 mM AEBSF , 2 µg/mL leupeptin , 100 nM microcystin-LR ) . Lysed samples were boiled for 5 min at 95°C in NuPAGE LDS Sample Buffer 4X ( Thermo Fisher ) + 5% BME ( Sigma-Aldrich ) . Protein concentration was determined using a Pierce BCA Protein Assay Kit ( Thermo Fisher ) . Samples were resolved on Bolt 4–12% Bis-Tris Plus Gels ( Invitrogen ) or AnykD Criterion TGX Precast Midi Protein Gel ( Biorad ) . Proteins were transferred to nitrocellulose for immunoblotting and probed with Anti-SNAP-tag rabbit antibody ( NEB ) and Anti-GAPDH−HRP mouse mAb , ( Novus ) . Detection was achieved with a HRP-conjugated rabbit secondary antibody ( GE Healthcare ) followed by enhanced chemiluminescence with SuperSignal West Dura Extended Duration Substrate ( Thermo Fisher ) . Densitometry was performed using NIH ImageJ ( Fiji ) software . Cells grown on 1 . 5 poly-D-lysine coated coverslips ( neuVitro ) for at least 16 hr were fixed for 10 min in 4% paraformaldehyde in PBS or in ice-cold methanol . Cells were permeabilized and blocked in PBS with 0 . 5% Triton X-100% and 1% BSA ( PBSAT ) for 30 min . Primary antibodies were diluted in PBSAT and cells were stained for 1 hr . Secondary antibodies conjugated to Alexa Fluor dyes ( Invitrogen ) were diluted in PBSAT and applied for 1 hr . Staining with FITC-tubulin antibodies and/or DAPI staining always followed secondary incubation step and was carried out for 10–45 min in PBSAT . Washes ( quick on and off ) with PBSAT were carried out 10X between antibody and/or dye incubation steps and prior to mounting . Coverslips were mounted on slides using ProLong Diamond Antifade Mountant ( Life Technologies ) . To de-identify cell type and dox-treatment conditions and allow for blinded analysis of mitotic spindle differences , all identifying information on microscope slides was masked by a third-party individual . All mitotic cells within a coverslip were classified as either having normal bipolar , abnormal bipolar , or monopolar spindles based on morphology of the DNA and microtubules . Maximum intensity projections from z-stack images were generated using SoftWoRx ( GE Healthcare ) or NIH ImageJ ( Fiji ) software . All immunofluorescence signal measures were carried out using Fiji software . Sum slice 32-bit Tiff projections were generated from z-stack images for analysis of immunofluorescence at centrosomes . For kinetochore measurements the ImageJ ‘SubtractMeasuredBackground’ macro was first applied and sum slice 32-bit Tiff projections were generated . For centrosomes measurements , the oval selection tool in Fiji was used to draw a circle ( ROI ) around the centrosome in the 568 ( SNAP-PACT ) channel . The area of the circle remained consistent for all measurements and all replicates of an experiment . Measurements were taken in the 647 channel ( which contained pT210-Plk1 , Total Plk1 , pT288-AurA , or γ-tubulin ) using the predefined centrosome ROI . Using the measure function in Fiji , with ‘Area’ and ‘Raw Integrated Density’ predefined as measurements , values were determined for each centrosome and for an arbitrarily selected background region . The raw integrated density was recorded for each centrosome and the background . The average raw integrated density for the centrosomes was determined by adding together the raw integrated densities for each centrosome in a cell and dividing that value by 2 . The integrated density for the background was subtracted from the average centrosome integrated density to yield a background-subtracted average integrated density signal for a centrosome . If the signal value was negative ( signal at centrosome was lower than at background ) the value was replaced with a 0 . For kinetochore measurements , the selection tool in Fiji was used to draw an arbitrary region ( ROI ) around the kinetochore in the 405 ( ACA , centromeric DNA ) channel or in the 568 ( SNAP-Mis12 ) channel . Measurements were taken in the 647 channel ( which contained pS69-Hec1 ) using the predefined kinetochore ROI . Using the measure function in Fiji , with ‘Area’ and ‘Raw Integrated Density’ predefined as measurements , values were determined for each kinetochore . The raw integrated density was recorded for each kinetochore . For both centrosome and kinetochore experiments and average was calculated for each control and experimental condition . To do this the normalized average integrated densities were added together and divided by the total number of cells for that condition . This yielded a value that represents the background-normalized average integrated density at a centrosome or at the kinetochore for a particular condition . Values for drug-treated cells were then normalized to their respective DMSO-treated control . Integrated intensity surface plots were generated from sum-slice 32-bit Tiff projections of representative images using the 3D Surface Plot function in Fiji software . Maximum intensity heat maps were generated from maximum projection representative images using the 3D Surface Plot function with Fire LUT in Fiji software . Zebrafish three-dimensional renderings: Three-dimensional renderings were created using Imaris software ( Bitplane ) . Individual mitotic cells were isolated and assigned a new color channel using the ‘Surfaces’ function to create a surface rendering . Surface renderings were created through the use of the Isoline function , where regions of individual mitotic cells were isolated based on intensity . Completed surface renderings were then merged and masked to create a channel that encompassed the mitotic cells of each embryo . Statistics were performed using an unpaired two-tailed Student's t-test in GraphPad Prism software . All values are reported as mean ± standard error of the mean ( s . e . m ) with p-values less than 0 . 05 considered statistically significant . Number of independent experiments ( N ) and number of individual points over several experiments ( n ) are presented . For γ-tubulin experiments a ROUT ( Q = 1% ) outlier test was performed and two values were removed prior to performing an unpaired Student’s t-test . The sample size was not statistically determined . Where applicable , n > 15 independent measurements were conducted across N ≥ 3 independent experiments . For doxycycline removal experiments ( Figure 2—figure supplement 1E , Figure 6—figure supplement 1B ) at least 2–3 independent experiments were conducted per time point .	In order for an animal cell to divide it needs to duplicate its DNA and split this genetic material equally between its daughter cells . This process , also known as mitosis , requires a number of different proteins that work together to coordinate this vital aspect of the cell’s lifecycle . For example , two enzymes known as Polo-like kinase 1 ( Plk1 ) and Aurora A ( AurA ) accumulate at specialized structures within the cell called centrosomes , where they receive signals from other parts of the cell to promote this process . Plk1 and AurA also operate at other locations in the cell , but it remains unclear whether these proteins have different activities at each individual location . Current methods for measuring protein activities use drugs or genetic approaches that switch off the target protein’s activities everywhere in the cell . Now , Bucko et al . have developed a new tool named LoKI that is able to direct drugs to specific locations in animal cells to switch off a target protein’s activity only at these sites . The experiments showed that inhibiting the activities of Plk1 and AurA at the centrosomes led to defects that prolonged mitosis . The new tool developed by Bucko et al . provides a means to more precisely study local events that occur in healthy and diseased cells . This will help us to understand how cancer and other diseases develop and may ultimately lead to new treatments for human patients with these conditions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2019	Subcellular drug targeting illuminates local kinase action
Cytidine deaminases are single stranded DNA mutators diversifying antibodies and restricting viral infection . Improper access to the genome leads to translocations and mutations in B cells and contributes to the mutation landscape in cancer , such as kataegis . It remains unclear how deaminases access double stranded genomes and whether off-target mutations favor certain loci , although transcription and opportunistic access during DNA repair are thought to play a role . In yeast , AID and the catalytic domain of APOBEC3G preferentially mutate transcriptionally active genes within narrow regions , 110 base pairs in width , fixed at RNA polymerase initiation sites . Unlike APOBEC3G , AID shows enhanced mutational preference for small RNA genes ( tRNAs , snoRNAs and snRNAs ) suggesting a putative role for RNA in its recruitment . We uncover the high affinity of the deaminases for the single stranded DNA exposed by initiating RNA polymerases ( a DNA configuration reproduced at stalled polymerases ) without a requirement for specific cofactors . Cytidine deaminases are a family of polynucleotide mutators that modify cytosines into uracil in viral nucleic acids as part of the innate immune defences ( Harris and Liddament , 2004 ) . Their success in restricting infection is reflected in the fact that the family has undergone a rapid expansion in primates and humans ( Jarmuz et al . , 2002 ) . The ancestral founder of the family , activation induced deaminase ( AID ) , functions in the adaptive immune system to mutate antibody genes in B cells as a fast mechanism to promote diversity of the antibody response to match the rapid evolution of pathogens during infection . The evolutionary advantages of these strategies are counterbalanced by the risk of exposing the host genome to active mutagenesis , a frequent cause of oncogenic transformation in leukaemia and lymphomas of B cell origin . All members of the AID/APOBEC family are selective in the sequence context of the deaminated cytosine , with the two preceding nucleotides identifying the signature of individual deaminases ( Beale et al . , 2004 ) . This mutation context signature has identified the human APOBEC3A and 3B proteins as the source of many of the somatic mutations accumulated by cancer genomes ( Nik-Zainal et al . , 2012; Burns et al . , 2013; Roberts et al . , 2013 , Taylor et al . , 2013 ) . The combined mutational landscape observed in mammalian genomes is complicated by the contribution from multiple cellular processes in addition to enzymatic deamination , such as metabolic oxidation , methyl-CpG deamination and aging , thus elucidating the precise contribution of the APOBECs is far from straightforward ( Alexandrov et al . , 2013; Lawrence et al . , 2013 ) . However the peculiar clustering of same strand mutations at TpC dinucleotides observed in kataegic mutations in breast cancers constitutes a hallmark of the APOBEC3A and 3B deaminases that can be experimentally induced . Repair of double stranded DNA breaks can expose long patches of single stranded DNA with multiple deaminations leading to the mutation clusters observed in association with genomic rearrangements in breast cancer genomes ( Nik-Zainal et al . , 2012; Roberts et al . , 2012; Taylor et al . , 2013 ) . Physiologically , the activity of such mutators is targeted to specific substrates and restricted from the rest of the genome to limit genomic instability . In the case of AID , expressed upon activation in only a fraction of B cells , by limiting access to the nuclear compartment and preferential recruitment to the immunoglobulin genes; in the case of APOBEC3G , expressed preferentially in lymphoid cells , by its localisation in the cytosol and binding to the viral genome and capsid . The mechanism that preferentially directs AID to the immunoglobulin genes is not fully understood , but active transcription has been repeatedly invoked as a requirement ( Reviewed in Storb , 2014 ) and many of the proteins found to be associated with AID are also involved in transcription and mRNA processing ( Pavri et al . , 2010; Basu et al . , 2011; Okazaki et al . , 2011; Willmann et al . , 2012 ) . Access of AID to off-target loci are documented not only by the anecdotal occurrence of mutations in oncogenes and chromosomal break points bearing the signature of the deaminase ( Bcl6 , MYC [Pasqualucci et al . , 2001] ) but also by AID dependent chromosome-break-capture and direct ChIP , where widespread off-target presence of AID is experimentally detected outside the immunoglobulin locus in mouse B cells ( Chiarle et al . , 2011; Klein et al . , 2011; Yamane et al . , 2011 ) . In addition to the sporadic off-target mutations induced by AID in B cells , APOBEC3A and 3B are thought to be responsible for many of the non-clustered/non-kataegic mutations at TpC dinucleotides observed not only in breast cancers but in other tumour types where the kataegic signature is not obviously present ( Kuong and Loeb , 2013 ) . As with sporadic AID mutations , the circumstances that promote or grant access of the APOBECs to single stranded DNA substrates of the host are not known . We have shown that overexpresison of deaminases in yeast faithfully recapitulates the mutation signatures observed in mammalian genomes . Here we have attempted to identify genomic features that promote or are permissive for enzymatic deamination by footprinting mutator activity on multiple genomes . Our results indeed reveal a preferential targeting of the deaminases to defined regions of the genome that is not dependent on cofactors but is rather based on accessibility , with structural features of the DNA at the promoter of actively transcribed genes being the key determinant . We also uncover a potential mechanistic explanation for the targeting and off-target preferences of the antibody diversification mutator AID . Overexpression of cytidine deaminases in yeast leads to the accumulation of genome wide mutations , which can be monitored by the number of cells resistant to the arginine analogue L-Canavinine through inactivation of the arginine permease CAN1 gene ( Figure 1A ) . We have previously shown that such overexpression leads to an uracil-DNA glycosylase ( UNG ) dependent enrichment of kataegic mutations through deamination of cytosines on single stranded DNA intermediates during the repair of double strand breaks ( Taylor et al . , 2013 ) . To assess the distribution of isolated mutations , we obtained a dataset largely devoid of kataegic mutations by expressing the deaminases in ungΔ cells . Overexpression of AID* ( an AID hyperactive mutant [Wang et al . , 2009; Taylor et al . , 2013] ) in haploid cells results in highly elevated frequency of Canavinine resistant colonies ( 164 × 10−6 ) , but relatively few mutations , averaging 61 single nucleotide variations ( SNVs ) per genome ( Figure 1A , B ) . Diploid cells can overcome this limit as they avoid the reduction in fitness costs caused by accumulated mutation ( Waters and Parry , 1973; Lada et al . , 2013 ) . Our experimental setting confirms this effect; whereas the mutation frequency is reduced almost 40-fold due to the requirement to inactivate both CAN1 alleles , the genome wide SNV increase over 10-fold , averaging 796 SNVs per genome for AID* and 592 SNVs for transformants expressing sA3G* ( a hyperactive mutant of the catalytic domain of human APOBEC3G [Wang et al . , 2009]; Figure 1A , B ) . For comparison , a database of mutations at C•G pairs was generated using the alkylating agent ethyl methane sulfonate ( EMS ) . Alkylation of guanosines promotes base pairing with thymine , thereby causing G > A transitions during replication . Overnight exposure of diploid cells to 0 . 2% EMS resulted in increased mutation frequency and SNV load per genome similar to that elicited by the deaminases ( Figure 1 ) . 10 . 7554/eLife . 03553 . 003Figure 1 . Genome wide distribution and signature of unclustered deaminase induced mutations in ung1Δ diploid yeast . ( A ) Mutation frequency ( expressed as the number of canavinine resistant colonies per 106 ) at the CAN1 locus in ung1Δ haploid yeast ( data in part from Taylor et al . , 2013 ) and ung1Δ/ung1Δ diploid yeast transformants expressing AID/APOBEC proteins or upon treatment with 0 . 2% EMS . Red bars indicate the median mutation frequency ( n = 12–126 colonies ) . ( B ) Genome wide SNV number in ung1Δ haploid and ung1Δ/Δ diploid yeast transformants expressing AID/APOBEC proteins or with EMS treatment . Red bars indicate the median mutation per genome ( n = 25–50 independent clones ) . ( C ) Sequence context of mutations at G•C pairs in diploid yeast genomes ( indicated as mutations at cytosines ) exposed to AID* , sA3G* or EMS mutagenesis . The numbers indicate total mutations per dataset , with the height of colour bars proportional to the frequency of each base found in the vicinity of a mutation . ( D ) Distribution of mutations per diploid yeast chromosome expressed as the number of mutations per chromosome in each independent genome against the chromosome length . The bars represent the projected linear trend for mutations at C ( in black ) or G ( in red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03553 . 003 When interrogating the mutations ( 99 . 8% of which occur at C:G pairs; A:T mutations were excluded from further analysis; all detected mutations are given in Supplementary file 1 ) , the expected flanking sequence context of WRC was found for AID* and YCC for sA3G* ( Figure 1C ) . In stark contrast , no consensus motif was observed in the EMS data , highlighting the random nature of this mutagenesis . In all three datasets SNVs appeared distributed throughout the genome , with all chromosomes displaying similar overall mutation that is strongly correlated with chromosome length , ruling out major biases in the targeting of mutations ( Spearman's correlation coefficient for AID: ρ > 0 . 65; for sA3G: ρ > 0 . 55; for EMS: ρ > 0 . 68; Figure 1D ) . Whilst mutations are equally distributed amongst chromosomes , they are not uniformly arranged along the chromosome . By combining the SNVs from independent transformants , regions can be observed in AID* and sA3G* genomes which show pronounced mutational peaks ( Figure 2A ) . Only one such region of high mutation density is seen in the EMS treated clones , that of the CAN1 gene . The presence of multiple loci with high mutation density is therefore a deaminase specific process . 10 . 7554/eLife . 03553 . 004Figure 2 . Mutation enriched loci ( MELs ) identified by focussed deaminase-induced mutation . ( A ) Radial histograms depict the density ( Z-score ) of pooled mutations for each dataset in 2 kb overlapping genomic segments along each chromosome . The CAN1 locus is highlighted in red . The peak highlighted in cyan is further enlarged in panels ( B ) , ( C ) and ( D ) . ( B ) Mutation densities along ChrII in AID* ( red ) , sA3G* ( black ) and EMS ( blue ) treated genomes , expressed as the Z-score of mutation density per dataset ( y-axis ) along chromosome II ( x-axis; 200 bp bin size ) . The region shadowed in cyan is magnified in ( C ) . ( C ) Regions of high mutation density identify narrow mutation enriched regions ( MELs ) , shown as green boxes for AID* and purple boxes for sA3G* in the bottom panel . Horizontal lines represent a single genome with each non-clonal mutation at C or G indicated by a dot ( black or red respectively ) . Regions in Chr II and Chr X containing mutation enriched loci shown at the same scale , with the genomic coordinates indicated . ( D ) Mutations in the pronounced MEL on ChrII ( highlighted cyan in panels ( A ) , ( B ) and ( C ) shown in green for AID* and purple for sA3G* . Coordinates are indicated . ( E ) Overlap of detected MELs in AID* , sA3G* and EMS datasets . ( F ) Distribution of MELs width with the median indicated for AID* and sA3G* mutated genomes . ( G ) Fraction of the total deaminase mutations in MELs ( black boxes ) relative to genomic coverage of MELs . ( H ) Distribution of distances between AID and A3G mutable motifs within MELs vs genome wide mutable motif distances . DOI: http://dx . doi . org/10 . 7554/eLife . 03553 . 00410 . 7554/eLife . 03553 . 005Figure 2—figure supplement 1 . Overlap between Haploid and Diploid MELs . DOI: http://dx . doi . org/10 . 7554/eLife . 03553 . 00510 . 7554/eLife . 03553 . 006Figure 2—figure supplement 2 . Strand bias in deaminase induced mutations calculated as fraction of mutations at C ( +strand ) or G ( - strand ) within each MEL . ( A ) Strand distribution of mutations within AID* and sA3G* MEL regions . MELs comprising a single base were excluded . ( B ) Strand distribution of mutations within MEL regions in relation to the direction of transcription of the associated gene . ( C ) Strand distribution of WRC and YCC deaminase motifs within MEL regions and their flanking 50 base pairs . DOI: http://dx . doi . org/10 . 7554/eLife . 03553 . 006 A more detailed look at regions with high density of mutations reveals narrow peaks of accumulated mutation that are in many cases common to both deaminases ( Figure 2B ) , with the most prominent peaks resulting from the proximity of several regions of densely targeted loci . These peaks represent high mutation densities within a bin size of 150 base pairs but surprisingly reflect the accumulation of mutations focussed to very narrow intervals within targeted loci ( Figure 2C , D ) . To further delineate mutation favoured loci , we defined regions of high mutation density by identifying overlapping 150 base pair fragments containing higher than expected mutation loads ( minimum of six mutations per fragment , originating from three independent transformants ) . We identify 1227 and 568 such mutation-enriched loci ( MELs ) in the AID* and sA3G* treated genomes , in contrast to just 1 obtained for EMS treatment ( overlapping the body of the CAN1 gene and hence due to canavinine selection ) . On average 35 such MELs would be expected for simulated datasets of equivalent mutation loads ( Figure 2E and Supplementary file 2 ) . MELs span remarkably narrow regions , with a window width averaging 110 bp for AID* and 71bp for sA3G* ( Figure 2F ) , and with almost 41% of all AID* and 22% of all sA3G* induced mutations localised to these regions ( Table 1 and Supplementary file 2 ) . In total , 25 , 618 of the combined 72 , 196 AID* and sA3G* mutations are occurring in MELs which account for just 1 . 5% of the genome ( Figure 2G ) . 10 . 7554/eLife . 03553 . 007Table 1 . Deaminase induced Mutation Enriched Loci ( MEL ) in yeast genomesDOI: http://dx . doi . org/10 . 7554/eLife . 03553 . 007ObservedSimulatedAIDsA3GEMSAIDsA3GEMSMELs1227568150213% MEL mutation40 . 721 . 60 . 240 . 750 . 390 . 14 Both AID and APOBEC3G target cytosines for deamination within a specific sequence context , leading to the mutation hotspots associated with antibody diversification and the recurrent mutations at CCC trinucleotides observed in HIV-1 genomes during the evolution of viral clades and which accumulate in viral genomes from infected individual ( Kijak et al . , 2008 ) . We therefore analysed the distribution of AID and APOBEC3G preferred sequence context in the yeast genome and find that the densities of AID and APOBEC3G motifs ( WRC and YCC respectively ) show no enrichment within the highly targeted regions compared to the remaining genome ( Figure 2H ) . Therefore , the accumulation of mutations in MELs is not a consequence of localised clustering of mutable motifs . Reinforcing the notion that MELs are highly favoured targets for mutations , we find these areas are frequency mutated on both alleles: 48% of AID* genomes and 56% of sA3G* genomes have mutations within MELs occurring on both chromosome alleles , compared to just 2–3% predicted for random fragments of equivalent size and mutation loads . MELs also contain most of the homozygous mutations detected ( 82% of AID* and 78% of sA3G* ) . Targeting of both alleles in MELs suggests they represent highly mutable regions within the genome , with the deaminases returning repeatedly to the same sites ( albeit on a second chromosome ) to mutate . Re-analysis of deaminase mutations we previously reported in haploid yeast ( Taylor et al . , 2013 ) identified 39 MELs which overlap with hypermutated MELs in diploid yeast , thus the focusing of mutations to MELs is seemingly unaffected by ploidy , suggesting the skewing of mutations due to selective pressures , such as fitness , is negligible ( Figure 2—figure supplement 1 ) . Equally , we observe no significant strand bias in the hypermutated hotspots associated with AID* MELs suggesting that both strands are targeted in a similar fashion . A broader distribution of sA3G* strand bias more likely reflects the partial skewing in the presence of YCC motifs at MELs ( Figure 2—figure supplement 2 ) . In conclusion , deaminases preferentially target narrow focussed regions throughout the genome independent of the sequence density of deaminase targets . There is a well-recognised relationship between AID induced mutations and transcription both in B cells at immunoglobulin genes , and for off-target loci , with mutations preferentially accumulating towards the promoter proximal region of the transcription unit ( Pasqualucci et al . , 2001; Rada and Milstein , 2001 ) . The transcription link is interpreted as a mechanism that facilitates access of AID due to the generation of single stranded DNA intermediates ( Chaudhuri et al . , 2003 ) . We therefore wondered whether AID induced MELs would be found associated with transcription . Contrary to expectation , enrichment analysis reveals that both AID* and sA3G* MELs are depleted within the body of RNA polymerase II ( RNAP II ) transcribed mRNA genes and rather that the deaminase induced mutations are preferentially associated with promoter regions , with over 76% of deaminase targeted hotspots found at promoters , compared to just 24% for simulated fragments ( Figure 3A ) . 10 . 7554/eLife . 03553 . 008Figure 3 . Deaminase mutation footprints are focussed to the pre-initiation complex region of active promoters . ( A ) Proportion of promoters , gene bodies , intergenic regions and replication origins ( ARS ) harbouring a MEL ( green ) or not ( grey ) for AID* and sA3G* datasets vs the expected distribution ( sim . AIDsA3G ) determined by Monte Carlo simulation of equivalent sized fragments for each MEL dataset distributed randomly across the genome . ( B ) Density of mutations in relation to their distance to the nearest transcription start site ( TSS ) of mRNA ( RNAP II ) transcripts compared to the density relative to transcription termination sites ( TTS ) . Data includes all mutations in addition to MELs . ( C ) Deaminase mutations relative to the TATA or TATA-like element for each RNAP II promoters ( Rhee and Pugh , 2012 ) compared to the mutation distance distribution aligned to the transcription start site ( TSS ) . ( D ) Proportion of AID* or sA3G* mutable motifs within RNAP II promoter regions , centred on the TATA-elements ( Rhee and Pugh , 2012 ) . Total number of mutations for each dataset is shown at each position ( black line ) . ( E ) Relative transcription rates ( see methods ) at RNAP II promoters targeted by MELs compared to relative transcription rates for all RNAP II genes in gal induced conditions ( García-Martínez et al . , 2004 ) . ( F ) Relative enrichment of RNAP II and RNAP II CTD phosphorylation ( S2P , S5P and S7P ) in promoters containing AID* ( red ) and sA3G* ( black ) MELs and all RNAP II promoters ( grey ) ranked according to transcriptional activity ( García-Martínez et al . , 2004 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03553 . 00810 . 7554/eLife . 03553 . 009Figure 3—figure supplement 1 . Paucity of deaminase mutations at replication origins is not a consequence of absence of mutable motifs . Proportion of AID* or sA3G* mutable motifs around replication origins ( ARS ) , depicted as in Figure 3D . Total number of mutations for each dataset is shown for at position ( black line , scale as in Figure 3D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03553 . 00910 . 7554/eLife . 03553 . 010Figure 3—figure supplement 2 . Density of mutations in relation to their distance to the nearest TATA box or TATA-like element . Mutations are grouped according to the TAF1 enrichment status ( data from Rhee and Pugh , 2012 ) with the line colour depicting the mutator ( AID* , red; sA3G* , black; EMS , blue ) . Data includes all mutations in addition to MELs . DOI: http://dx . doi . org/10 . 7554/eLife . 03553 . 01010 . 7554/eLife . 03553 . 011Figure 3—figure supplement 3 . Distribution of the deaminases on chromatin is unrelated to mutation preferences . Enrichment of ( A ) deaminase , ( B ) serine 5 phosphorylated RNAPII and ( C ) Histone H3 at MEL promoters , unmutated promoters and intergenic regions . Enrichment is shown relative to input chromatin ( B and C ) or further normalised to control cell lines ( A ) . Data from 2–3 independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 03553 . 01110 . 7554/eLife . 03553 . 012Figure 3—figure supplement 4 . Transcription factor binding sites compared to MEL preferences . ( A ) Frequency of each yeast transcription factor at individual promoters as described in ( Venters et al . , 2011 ) ( blue dots ) compared with the frequency that the transcription factor appears in the promoter of genes containing AID* ( red dots ) and sA3G* ( black dots ) MELs . Factors are ordered according to number of binding sites in all promoters . Basal transcription factors are the most commonly associated with deaminase targeted promoters ( labelled ) . ( B ) Transcription rates of genes grouped according to Spt16 promoter occupancy and presence of MELs . ( C ) List of transcription factors found to vary in occupancy at MEL targeted promoters vs their overall frequency at all yeast promoters ( Venters dataset ) . Transcription factors showing ±10% variation which are present in at least 25% of MELs are listed . DOI: http://dx . doi . org/10 . 7554/eLife . 03553 . 012 Initiation of replication also transiently generates single stranded DNA at defined locations . However , there is no enrichment of mutated hotspots associated with replication origins ( ARS ) ( Figure 3A ) . Although this could reflect the relative depletion of mutable motifs within ARS core consensus sequence , we find similar densities of mutable cytosines within the broader sequence context encompassing 200–300 base pairs nucleosome depleted regions associated with functional origins ( Figure 3—figure supplement 1 ) , suggesting that single strand availability provided by melting the DNA by the ORC complex might not be sufficient to efficiently target the deaminases . Mutation enrichment at promoters is not restricted to hotspots identified within MELs , which exclude 73% of the total mutations due to the threshold applied in defining enriched loci . Aligning all mutations to mRNA transcriptional starts ( TSS ) and termination sites ( TTS ) ( Xu et al . , 2009 ) , revealed a strong association of deaminase induced mutations with the TSS , with over 57% of AID* and 46% of sA3G* mutations occurring within the promoter region ( defined as 500 base pairs upstream and 50 base pairs downstream of the TSS ) , compared to only 21% of EMS mutations ( the expected frequency for randomly distributed mutations ) . Mutation accumulation is skewed upstream of the TSS ( peak at -21 bp and -38 bp for AID* and sA3G* respectively; Figure 3B ) , corresponding to the nucleosome free region where the pre-initiation RNAP complex forms before scanning for the TSS ( Rhee and Pugh , 2012 ) . Indeed , aligning SNVs to the TATA box/TATA-like element or TSS revealed that not only are the majority of promoter associated mutations occurring between these two features ( Figure 3C ) , there is also a paucity of mutations at the TATA-element suggesting this region is protected by TBP/TFIID binding ( this paucity is , at least for AID* , not due to an absence of mutable sequence motifs; Figure 3D ) . Intriguingly , the peak of AID* and sA3G* induced SNVs centred 30 base pairs from the TATA-element , the region where TBP guides TFIIB to load RNAPII for the formation of the pre-initiation complex ( PIC ) ( Rhee and Pugh , 2012 ) . The deaminase mutated hotspots thus identify the position where promoter melting occurs before the scanning polymerase encounters the TSS , suggesting a mechanistic basis for the hypothesis that the deaminases access the promoter coincidentally with the assembly of the pre-initiation complex . Consistent with the notion that initiating polymerases create transient access for the deaminases rather than specifically loading the proteins , we detect robust association of RNAP II with the promoter region of deaminase targeted promoters in yeast but negligible enhancement in the association of either AID or sA3G with mutated promoters compared to unmutated or intergenic regions ( Figure 3—figure supplement 3 ) . Additionally , while there is a correlation between the mutated strand and the direction of transcription ( Figure 2—figure supplement 2 ) , MELs are predominantly composed of mutations occurring in both strands suggesting the PIC makes both strands available during initiation . Supporting the idea that the deaminases preferentially mutate promoters due to their ability to recognize the melted DNA associated with the transcription pre-initiation complex , we observe that MELs occur in genes with above average transcriptional activity ( García-Martínez et al . , 2004 ) but targeting appears unrelated to any particular transcriptional program ( Figure 3—figure supplement 4 ) . Rather than simply transcription factor binding at the promoter , active initiation by RNAP II is important for MEL development ( Wilcox test p < 0 . 005 for all groups; Figure 3E ) . The transition of RNAP II from the pre-initiation complex to the elongation complex is associated with a shift in phosphorylation of the C-terminal domain ( CTD ) serine 5/7 to serine 2 ( Kim et al . , 2010 ) . In agreement with the transcription rate analysis , deaminase MELs are associated with both high levels of RNAP II occupancy and CTD-S5P , that parallels the association with the highest transcribed genes ( Figure 3F ) . Indeed , the recurrent association of both AID* and sA3G* MELs with regions enriched for the basal transcription machinery and in particular Spt16 -a chromatin chaperon associated with highly transcribed genes ( Formosa , 2013 ) ( Figure 3—figure supplement 4 ) , reinforces the idea that active transcription and potential pausing ( at promoters highly dependent on the FACT/Spt16 complex ) determines the deaminases targeting . In summary , cytidine deaminases mutate at specific loci through the yeast genome , predominantly within active gene promoter regions . In B cells , AID is found in association with components of the transcription machinery such as SPT5 and SPT6 , and RNAP II itself ( Nambu et al . , 2003; Pavri et al . , 2010; Okazaki et al . , 2011 ) , therefore we wondered whether the enrichment of mutations associated with promoters might be a feature restricted to RNAP II dependent genes . Analysis of mutations in highly transcribed non-RNAP II dependent transcripts , such as RNAP III dependent tRNA genes , astonishingly reveals an even more pronounced enrichment of targeted hotspots with 78% of the genomic regions corresponding to tRNAs harbouring repeated mutations . While we find that both sA3G* and AID* MELs overlap with tRNAs , AID* MELs are disproportionately overrepresented , with 228 of 275 tRNA genes being highly targeted ( Figure 4A ) . Furthermore , aligning of mutations within 250 base pairs of the TSS of tRNA genes shows that all occur within the tRNA gene body , which is also the site of RNAP III initiation ( Figure 4B ) . As in the case of mRNA promoters , the mutations in tRNAs are highly focussed to narrow hotspots that span the site where loading of the polymerase is thought to occur ( Figure 4C ) . 10 . 7554/eLife . 03553 . 013Figure 4 . AID* and sA3G* target both RNAP II and RNAP III promoters . ( A ) Number of tRNA genes harbouring ( green ) an AID* or sA3G* MEL compared with expected number from Monte Carlo simulations . ( B ) Density of mutations in relation to the transcription start site ( TSS ) of tRNA genes . Mutations within the 500 base pair interval centred at the TSS are included . ( C ) Mutation frequency in promoters of mRNA genes ( within a window 500 bp upstream and 50 bp downstream of the TSS ) compared to the frequency of mutations in the promoters of tRNA ( 550 bp window centred on the middle of the tRNA gene ) , snoRNAs and snRNA genes ( 550 bp window as for mRNA genes ) . mRNA genes are binned according to transcription rate as in Figure 3 . Both RNAP II and III driven snoRNAs are included . ( D ) Example of MELs in ChrIV and ChrXV corresponding to tRNA tI ( UAU ) D and tG ( CCC ) O , depicted as in Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 03553 . 01310 . 7554/eLife . 03553 . 014Figure 4—figure supplement 1 . Median number of mutable motifs in promoter regions . DOI: http://dx . doi . org/10 . 7554/eLife . 03553 . 01410 . 7554/eLife . 03553 . 015Figure 4—figure supplement 2 . Mutationally enriched loci are not a consequence of increased density of mutable motifs . The number of deaminase motifs for each MEL vs the number of mutations within each MEL for AID* and sA3G* datasets . DOI: http://dx . doi . org/10 . 7554/eLife . 03553 . 01510 . 7554/eLife . 03553 . 016Figure 4—figure supplement 3 . Mutations in the rDNA locus are restricted to the replication fork block ( RFB ) site . ( A ) Sequence context of low allelic frequency mutations detected in the rDNA locus , as depicted in Figure 1C . ( B ) Schematic of the rDNA repeat region . Panels show mutations detected in yeast transformants at the rDNA locus . Each line represents one clone with dots representing mutations ( mutation at C , black; at G , red; at A , green ) . Clones with no detected mutations are not depicted . DOI: http://dx . doi . org/10 . 7554/eLife . 03553 . 01610 . 7554/eLife . 03553 . 017Figure 4—figure supplement 4 . Deaminase induced mutation distribution in relation to R-loop forming potential . Tables showing the correlation between R-loops formation predicted by the QmRLFS-finder ( Wongsurawat et al . , 2012 ) or SkewR package ( Ginno et al . , 2012 ) and the presence of MELs . DOI: http://dx . doi . org/10 . 7554/eLife . 03553 . 017 The mutation frequency ( normalised number of mutations per 550 base pairs ) in AID* genomes within tRNA genes is higher than at mRNA gene promoters ( p value < 2e-16 , Wilcox non-parametric test; Figure 4C ) and much higher than that observed even in the subset of highly transcribed mRNA promoters . While the differences in mutation frequency between mRNA promoters and tRNAs is still statistically significant for A3G* ( p value < 8e-10 ) , this effect is less pronounced . Enhanced mutation is also observed in the promoters of snoRNA and snRNA genes , again particularly in the case of AID* genomes , whereas no statistically significant differences are observed between any of the promoter subsets for mutations driven by EMS . The enhanced mutation attributable to AID* is not likely a feature of RNAP III , since snoRNAs are even more targeted for mutation though all but snR52 are transcribed by RNAP II ( Moqtaderi and Struhl , 2004 ) . Targeting of tRNA , snRNA and snoRNA genes by the deaminases could be enhanced by the availability of hypermutable motifs , as there is on average one more YCC motif in the tRNA genes ( 1 . 5 more in the MEL region itself ) targeted by sA3G* than in those tRNA genes not targeted by sA3G* . We see no such difference with AID* target motifs which are present within tRNA , snRNA and snoRNA gene promoters at similar frequency as in other promoters ( average 52 to 63 motifs per 550 base pairs promoter window , Figure 4—figure supplement 1 ) . Overall , there is only weak correlation between the number of motifs within the 550 base pair promoter window and the number of mutations ( Spearman's ρ = 0 . 02 and ρ = 0 . 2 for AID* and sA3G* respectively; Figure 4—figure supplement 2 ) , confirming that motif availability is not the main determinant for targeting . Mutations at rRNA genes were poorly mapped due to the repetitive nature of the region on Chr XII ( 150–200 copies of the 9 . 1 kb unit containing the 35S pre-RNA and the 5S RNA ) . By including repeatedly mapped reads across the rDNA locus , we could detect several hundred mutations at low allele frequency all within the expected deaminase mutation context , giving confidence in their detection and location ( Figure 4—figure supplement 3A ) . Mutations were restricted to the well defined ribosomal replication fork barrier ( rRFB ) located between the 5S and 35S transcriptional units . No enhanced mutation was detected at the promoter regions ( which are transcribed in opposite directions by RNAP III and RNAP I respectively ) . However mutations clustered at the rRFB site for both deaminases ( Figure 4—figure supplement 3B ) , at a site where induced homologous recombination maintains the size of the ribosomal gene array . Although DNA double-strand breaks ( DSB ) have been detected at the site , it is likely that in vivo persistent breaks are rare in undamaged yeast ( Fritsch et al . , 2010 ) . Accordingly we did not detect kataegic like clusters in the region , but rather localised mutated hotspots . Thus it is possible that other mechanisms such as cryptic transcription ( Houseley et al . , 2007 ) might expose the site to the action of the deaminases , rather than repair of double strand breaks . While AID overexpression in yeast deficient for components of the RNA processing machinery ( THO ) have enhanced genomic instability , particularly in highly transcribed GC-rich regions prone to R-loop formation ( Gómez-González and Aguilera , 2007 ) , in wild type yeast this effect is only mild . Nonetheless we observe positive association of MELs with predicted R-loop potential genes although the paucity of these features across the genomes ( between 59–78 sites ) precludes any predictive dissociation between high density of mutation , R-loop potential and transcription rates ( Figure 4—figure supplement 4 ) . An alternative explanation for the enhanced targeting of small RNA promoters by AID* is that the RNAs themselves preferentially bind AID , thereby creating co-transcriptional enrichment of AID in the vicinity of their genes . Purified AID binds RNA , with its in vitro deamination activity enhanced by treatment with RNAse A ( Bransteitter et al . , 2003 ) , whereas the non-catalytic domain of APOBEC3G is responsible for its ability to bind RNA and form high molecular weight ribonucleic–protein complexes ( Huthoff et al . , 2009; Bélanger et al . , 2013 ) . It is not known whether binding in both cases is specific for any particular RNA species , but based on our current observations we decided to test the ability of human AID and human APOBEC3G to bind in vitro transcribed tRNA as well as polyU RNA . Whereas both Flag-tagged overexpressed human AID and full length human APOBEC3G can be recovered from cell extracts by binding to biotin labelled RNAs , the catalytic domain of APOBEC3G ( sA3G ) is not ( Figure 5A ) . Furthermore , full length APOBEC3G is efficiently recovered from extracts by the extended linear polyU RNA , a reflection of its ability to oligomerise in an RNA dependent fashion , whereas AID recovery is not enhanced by its binding to linear polyU RNA . Binding of AID to tRNA species was also found for endogenous yeast tRNAs , suggesting that the modifications found in vivo ( pseudouridylation and 2′-O-ribose methylation ) do not affect the interaction . The single domain APOBEC3A protein shows no RNA binding ability except a limited amount to doubled stranded RNA , despite sharing the preferential targeting to promoters as the rest of the deaminases ( Figure 5—figure supplement 1A ) . Taken together , this data suggest a degree of specificity in the RNA binding preferences of the deaminases , with AID preference linked to structured rather than linear RNA ( Figure 5B ) . Interestingly the catalytic activity of AID is not required for the binding or the specificity , since similar binding was observed for the inactive mutant AID-E58A ( Figure 5A ) . 10 . 7554/eLife . 03553 . 018Figure 5 . RNA binding by human AID and APOBEC3G . ( A ) Left panel shows the in vitro transcribed pre-tI ( UAU ) D tRNA used for affinity purification . Right panel shows immunoblots for transiently overexpressed AID/APOBEC3G proteins following RNA-immunoprecipitation with pre-tRNA . ( B ) Affinity purification with tl ( UAU ) D probe , total yeast tRNA , homopolymeric single stranded ( polyU ) and double stranded ( polyA:U ) RNA . Left panel shows input proteins , right panel shows immunoblots for transiently overexpressed AID/APOBEC3 proteins following RNA-immunoprecipitation . Results representative of at least 3 independent experiments . ( C ) Deaminase induced mutations in the promoter region of the YBR194W locus . Top panels: accumulated mutations in the AID* , sA3G* and EMS whole genome datasets . Bottom panels: mutations detected in Sanger sequenced yeast clones unmodified or harbouring a chimeric YBR194W-snR6 locus . Each line represents one clone with dots representing mutations ( at C , black; at G , red ) . Clones with no mutations are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 03553 . 01810 . 7554/eLife . 03553 . 019Figure 5—figure supplement 1 . Promoter mutations are driven by APOBEC3A and 3B and are a feature of cancer genomes enriched for TC mutations . ( A ) Mutation density relative to the TSS for APOBEC3A and APOBEC3B induced mutations from ungΔ haploid cells ( data from Taylor et al . , 2013 ) . The density at tRNA promoters is shown separately in red . ( B ) Mutations in breast cancer genome PD4120a and lung adenocarcenoma LUAD-S01345 . Pie charts show the contribution of mutations at TC over mutations at the remaining dinucleotides and histograms show mutation density relative to all human TSS ( Ensemble annotation ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03553 . 01910 . 7554/eLife . 03553 . 020Figure 5—figure supplement 2 . Functional comparison of the YBR194W locus in modified yeast clones . ( A ) Immunoprecipitation of chromatin associated RNAP II or ( B ) Histone H3 from unmodified or YBR194W-snR6 chimeric yeast . Black bars show enrichment relative to input in the unmodified strain with the modified strain in red . An unrelated locus , YJL105W is shown as control . Data from three independent experiments . ( C ) mRNA levels of YBR194W shown relative to ACF1 . Levels at the TAF10 gene are shown as a control . Data from three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 03553 . 020 In order to test the RNA binding properties of AID in modulating its targeting preferences we introduced a chimeric snR6 RNA into the RNAP II driven YBR194W gene , which was identified in our dataset as a transcribed but poorly targeted promoter by both deaminases ( Figure 5C top panels ) . Initiation and transcription of the modified locus remained overall unaffected ( Figure 5—figure supplement 2 ) , while comparison of the YBR194W promoter region by Sanger sequencing revealed enhanced mutation focused to the immediate vicinity of the TSS by AID* but not sA3G* . No such focussing of mutations was observed in the unmodified yeast overexpresing AID* ( Figure 5C ) . We conclude that the differential preference of AID* for tRNA , snRNAs and snoRNAs in yeast might reflect the ability of AID to preferentially associate with abundant small RNA species , in contrast to the catalytic domain of APOBEC3G ( sA3G* ) that possesses no RNA binding activity . Targeting mutations to initiating promoters is not likely a function of the size of the deaminase , as could be inferred from the results described for both AID and the single domain fragment of APOBEC3G used in our study . Similar promoter associated recurrent mutations can be elicited not only by APOBEC3A ( also a single domain deaminase ) but also by the double domain APOBEC3B ( Figure 5—figure supplement 1A ) . It is therefore not entirely unexpected to observe enrichment of mutations at TpC ( versus other dinucleotides ) in association with promoter regions in a breast cancer genome that has the highest incidence of APOBEC3 kataegic mutations ( Figure 5—figure supplement 1B ) , suggesting that the deaminases could access dsDNA at initiating or paused RNAPs also in mammalian cells . The involvement of AID and APOBEC3A and 3B in cancer suggests that enzymatic deamination of genomic targets is an infrequent but recurrent consequence of the presence of the deaminases in vertebrates . Despite subcellular compartmentalisation , specific targeting and restricted expression limiting AID off-target activity , some genomic regions other than the natural target , the immunoglobulin loci , are predisposed to mutation . BCL6 , PIM1 and MYC are recurrent off-targets of AID mutation in B cell malignancies ( Pasqualucci et al . , 2001 ) ; in the case of BCL6 it is estimated that AID induced mutations are also prevalent in non transformed B cells at just 103-fold lower frequency than at immunoglobulin genes ( Liu et al . , 2008 ) and even in the absence of the mutator phenotypes attributable to malignant transformation , normal B cells frequently show AID induced translocations at the MYC locus ( Roschke et al . , 1997; Casellas et al . , 2009 ) . In cancer genomes , the association of APOBEC mutations with genomic rearrangements suggests that replication stress , persistent DNA lesions and incomplete repair expose single stranded DNA that becomes a substrate for deaminases leading to clustered mutations . It is unclear how APOBEC3A and 3B gain access to single stranded DNA leading to the singlet isolated mutations highly prevalent in mutated cancer genomes that bear the APOBEC signature ( Taylor et al . , 2013 ) . It is therefore important to understand the genomic context that facilitates off-target activity of the deaminases in the absence of explicit DNA damage . Expression of AID and other APOBEC proteins in yeast faithfully recapitulates the signature of mutations observed in mammalian cells in a smaller genome with no background mutations due to unrelated processes , such as DNA repair ( Lada et al . , 2013; Taylor et al . , 2013 ) . In this study we demonstrate the non-random nature of the mutations induced by the deaminases , which is remarkably focussed to just 1 . 5% of the yeast genome but nonetheless overlaps more than half of the active promoters . AID is known to interact with components of the transcription machinery in mammalian cells ( reviewed in Kenter , 2012 ) . However , the overlap between highly mutated promoters by both AID and APOBEC3G suggests that rather than conservation of protein–protein interactions of the deaminases with the transcription complex , properties of the promoter itself can determine targeting . Enhanced targeting of RNAP III transcribed genes argues against active recruitment of the deaminases by conserved initiation factors , whereas the structural conservation of the DNA template conformation at the core pre-initiation complex of all polymerases ( Vannini and Cramer , 2012 ) supports the idea that the conformation of the DNA template is the common element in the recruitment of the deaminases . Indeed , the site of polymerase loading ( within the body of the tRNA genes ) rather than the TSS is the preferred target of deamination in the case of the RNAP III transcribed tRNAs in contrast to the 5′ region of the RNAP III transcribed SNR52 snoRNA , where the loading of the RNAP is fixed at the 5′ promoter region . Furthermore , the high density of mutations focussed to the small region between the TATA binding protein site ( TBP ) and the transcription start site ( TSS ) , more precisely identify the pre-initiation complex ( PIC ) as the target for the deaminases . Budding yeast RNAP II promoters show characteristic and highly regulated nucleosome exclusion . This is partly due to sequence composition , with regions enriched for poly dA•dT nucleotides that confer rigidity to the DNA and are therefore thermodynamically less favourable to wrap around nucleosomes ( Yuan et al . , 2005 ) , and partly due to the regulated and precise positioning of the +1 nucleosome relative to the TSS that includes specific histone variants ( H2A . Z and H3 . 3 ) that promote chromatin accessibility ( reviewed in Jiang and Pugh , 2009 ) . Therefore it is highly significant that other nucleosome free regions , such as ARS are not targeted by the deaminases , despite undergoing DNA melting during the initiation of replication . This reinforces our interpretation that intrinsic properties of active promoters , in particular the configuration associated with loading of the polymerase at the pre-initiation complex ( open pre-initiation complex ) ( Grünberg et al . , 2012 ) , are sufficient to generate persistent single stranded DNA accessible for deamination . Our data supports the presence of such open PICs in most yeast active promoters . Neither the preferential targeting of promoters nor the narrow focus of the MELs is due to the preferential clustering of mutable motifs . Interestingly , the nature of the mutation hotspots within MELs ( both at C and G ) , reveals that both strands of the melted DNA structure associated with active promoters are accessible . Furthermore , protection from mutation is evident at the TBP binding site while the peak of mutations ∼30 base pairs downstream identifies the site of RNAP loading and DNA melting mapped by permanganate footprinting ( Giardina and Lis , 1993 ) and high resolution ChIP ( Rhee and Pugh , 2012 ) . Our deaminase footprinting data further confirms the persistent open configuration and single stranded nature of this region potentially identifying open pre-initiation promoters . Differences in the assembly of the PIC in TATA and TATA-like promoters , do not seem to affect mutation susceptibility , although predictably , TATA box promoters show a more defined distance between the TBP protected footprint and the accessible melted DNA ( Figure 3—figure supplement 2 ) indicating that it is the structure of the single stranded DNA rather than the assembly ( SAGA or THIID dependent ) of the transcription initiation complex itself that determines targeting ( Rhee and Pugh , 2012 ) . Up to 75% of human promoters in different cell types are occupied by a pre-initiating form of RNAP II ( Guenther et al . , 2007 ) , whereas pausing and stalling are much more common in metazoan transcription compared with Saccharomyces cerevisiae . Mammalian promoters are frequently regulated by proximal pausing , with most promoters pausing within 200 base pairs of the TSS ( Adelman and Lis , 2012 ) . In the presence of a deaminase , initiating and or paused sites would become accessible for mutation , thus it is intriguing to observe promoter proximal enrichment of mutations at TpC dinucleotides in PD4120a , a breast cancer genome with dramatic accumulation of kataegis that betrays its mutagenesis by APOBEC3B ( Nik-Zainal et al . , 2012 ) . Our data favours the idea that accessibility of single stranded DNA at RNAP II stalled sites suffice to recruit APOBECs or indeed AID . This model offers explanation for the association of AID with mammalian SPT5 , which functions in modulating the pausing of RNAP II during elongation as transcription stalls , and is consistent with the recurrent targeting by AID of the promoter proximal region of MYC ( Duquette et al . , 2005 ) a well characterised promoter-proximal pausing regulated gene ( Krumm et al . , 1992; Strobl and Eick , 1992 ) . The correlation between high transcription rates and enhanced deaminase targeting reinforces the hypothesis that repeated loading of the pre-initiation complex leads to the persistence of a small region of melted DNA that is very efficiently targeted by the deaminases . Indeed the enhanced targeting of tRNA , snoRNAs and snRNA genes could reflect the high transcription rates of these essential RNAs given that RNAP I and III transcripts constitute almost 80% of the total nuclear gene expression in dividing cells ( Vannini , 2013 ) . The unexpected finding that tRNAs are disproportionally targeted for mutation by AID compared with APOBEC3G , as are the promoters of other highly structured RNAs ( snRNA or snoRNA ) , and the indication that this difference is not due to motif enrichment at those promoters , brings into focus the potential involvement of the RNA binding properties of the deaminases in promoting targeting . While APOBEC3G has been shown to bind not only HIV RNA , but cellular RNAs , including abundant 7S RNA ( Huthoff et al . , 2009 ) , this ability is dependent on the N-terminal domain . Mutation targeting of the RNAP initiation complex is not linked to the ability of the deaminases to bind RNA per se , as the catalytic C-terminal domain of APOBEC3G in this study is inert regarding RNA binding . Notably , our results show that AID binds structured RNAs in vitro ( such as tRNAs ) , and preferentially targets tRNAs and other small RNA promoters for mutation in yeast , prompting the speculation that binding to abundant RNAs sequesters AID to subnuclear localities such as nucleolar areas , where small RNAs genes also localise during transcription . Indeed nucleolar localisation of overexpressed AID has been reported in mammalian cells , although its significance under physiological levels remains to be tested ( Hu et al . , 2013 ) . Alternatively preferential recognition of particular RNA structures such as folded tRNAs could determine the recruitment of AID to genomic regions . In conclusion , our study uncovers the remarkable preference of mammalian cytidine deaminases to mutate active promoters when expressed in yeast , a preference blind to the type of RNA polymerase ( both RNAP II and III genes are targets ) and not ascribable to sequence context or targeting by specific cofactors . The precise and narrow location of the recurrent mutations pinpoints the site where the RNAP pre-initiation complex is loaded highlighting the conservation of the TBP ( TATA binding protein ) site and the formation of the pre-initiation complex , whereas exclusion of mutations from the TBP site confirms the poised nature of active yeast promoters . These results suggest that initiating polymerases create a small but persistent accessible patch of single stranded DNA in vivo , which has high affinity for deaminases and where both strands are accessible for mutation . They also strongly support the notion that AID might directly bind to single stranded DNA at the pre-initiating or stalling RNAP sites without a requirement for specific cofactors and that its targeting is modulated by its ability to interact with structured RNA species . Yeast strain BY4743 ungΔ/ungΔ was generated by crossing BY4741 ungΔ ( MATa; his3Δ1; leu2Δ0; met15Δ0; ura3Δ0 ) obtained from Euroscarf deletion collection ( Frankfurt , Germany ) with the BY4742 ungΔ strain . BY4742 ungΔ was generated by removal of the UNG1 open reading frame by homologous recombination in the parental BY4742 strain , using a PCR generated URA3 cassette flanked by a 57-bp 5′ homology and 51-bp 3′ homology arms that include adaptamers for post integration removal of the URA3 selection cassette ( Reid et al . , 2002 ) . The YBR194W-snR6 chimeric strain was generated by inserting a URA3 cassette at the 5′ end of the YBR194W gene in BY4741 ungΔ cells . Homology arms and the snR6 gene were amplified from genomic DNA using the primers ( 1 ) 5′-CCTGCCACTTTCAAAAGGCG-3′ and 5′-CGAAGGGTTACTTCGCGAACTCCTGTCCCTATTACATATTCAACC-3′ , ( 2 ) 5′-GGTTGAATATGTAATAGGGACAGGAGTTCGCGAAGTAACCCTTCG-3′ and 5′-GCCAGGCATGCTAATGGCAAAACGAAATAAATCTCTTTGTAAAAC-3′ , ( 3 ) 5′-GTTTTACAAAGAGATTTATTTCGTTTTGCCATTAGCATGCCTGGC-3′ and 5′-TGGTGGTCATATGCTCGGTG-3′ . A PCR fusion of all three fragments with the first and last primer was used to retarget the URA3 containing locus . 5-Fluoroorotic acid counter-selection was used to isolate targeted colonies that were then mated with BY4742 ungΔ to generate the final BY4743 ungΔ/ungΔ YBR194W-snR6/YBR194W strain . Correct integration of all targeting constructs was confirmed by PCR . Yeast transformation and selection , genomic DNA extraction and mutation frequency calculation were performed as described previously ( Taylor et al . , 2013 ) . Control and AID* expression vectors were as described previously ( Taylor et al . , 2013 ) . The sA3G* vector was generated by PCR amplification of the C-terminal domain of A3G* fused with a 5′ SV40 nuclear localisation sequence and FLAG tag using primers 5′-GCAAGCTTGCCACCATGCCTAAAAAGAAGCGTAAAGTCGAGATTCTCAGACACTCG-3′ and 5′-CCAGAATCAGGAAAACGGAGCAGACTACAAGGACGATGACGACAAGTAGCTCGAGGC-3′ and ligating the resultant Hind III-Xho I fragment it into pRS426-GAL1pr-tADHpolyA vector described previously ( Taylor et al . , 2013 ) . Ethyl methanesulfonate ( EMS ) mutagenesis was performed by culturing BY4743 ungΔ/ungΔ yeast overnight in YEPD with 0 . 2% EMS , after which cells were washed in 5% sodium thiosulfate and plated for viability and canavanine resistance as above . DNA libraries were generated using the multiplexing Nextera DNA Sample Prep Kit ( Illumina , Little Chesterford , UK ) according to manufactures instructions . The libraries were sequenced by BGI ( BGI , Beijing , China ) . The de-multiplexed sequence reads were aligned to the reference yeast genome ( SacCer_Apr2011/sacCer3 ) using BWA-MEM ( Li and Durbin , 2009 ) . Optical duplicates were removed using Picard ( http://picard . sourceforge . net ) and only uniquely mapped paired reads were retained . On average 43-fold sequence coverage was achieved for each yeast genome . Unprocessed sequence reads for this study have been deposited at the EMBL-EBI European Nucleotide Archive , study accession number PRJEB7456 ( http://www . ebi . ac . uk/ena/data/view/PRJEB7456 ) . RNA from 1 ml overnight cultures purified with RNAeasy plus ( Qiagen ) was used to generate cDNA using oligo-dTs and the GoScript Kit ( Promega , Southampton , UK ) followed by qPCR employing QuantiFast SYBR ( Qiagen ) all as per manufactures instructions . Primers used are TAF10; 5′-ATATTCCAGGATCAGGTCTTCCGTAGC-3′ and 5′-GTAGTCTTCTCATTCTGTTGATGTTGTTGTTG-3′ , ACT1; 5′-CTTTCAACGTTCCAGCCTTC-3′ and 5′-CCAGCGTAAATTGGAACGAC-3′ , YBR194W-snR6; 5′-CCTGCCACTTTCAAAAGGCG-3′ and 5′-CAGGGGAACTGCTGATCATCTCTG-3′ , YBR194W; 5′-GGGTCGTGAAAAAGAGAACGG-3′ and 5′-ATGTGATGGTGCAGTGCCTC-3′ . The YBR194W promoter region was amplified using the following primers; 5′-ATTGTGGCAGTTCGGCTTTG-3′ and 5′-AGGTTTCCCAGTCTGGCTTG-3′ and Sanger sequenced using the latter .	In cells , genetic information is stored within molecules of DNA , which contain sequences of four ‘bases’ arranged in different orders . Replacing one of these bases with a different base results in a mutation , which can have a positive or negative influence on the cell . Mammals use a group of enzymes called cytidine deaminases to help defend themselves against harmful invaders . These enzymes work by introducing mutations into the DNA of viruses , microbes and even the mammal itself . For example , an enzyme called APOBEC3G can mutate the DNA of viruses to prevent them spreading around the body . Another enzyme , called AID , can mutate the genes that make antibodies—proteins that attack the invading microbes—in order to make new varieties of antibodies . Unfortunately , the enzymes sometimes target other genes , which can lead to cancer and other diseases . Cytidine deaminases can only access and mutate single strands of DNA , so most of the DNA in a cell is protected because it is in a two-stranded double helix . However , there are times when the two strands are separated , such as when a section of DNA is being repaired , or when it is being transcribed to produce a molecule of RNA , which is subsequently used to make a protein . It is not clear when cytidine deaminases are able to target single stranded DNA , and whether they need help from any other components . Now , Taylor et al . have studied how these enzymes access single stranded DNA when artificially introduced into yeast . These experiments showed that AID and APOBEC3G can access single stranded DNA without the help of any extra components . The enzymes target genes that are being transcribed to make RNA , with the DNA at the start of the transcription site being the most prone to mutation . In mammal cells , most genes are normally protected from the mutations introduced by cytidine deaminases , but this protection does not appear to work in many cancer cells . The next challenge will be to develop a better understanding of how this protection works , and to work out why it sometimes goes wrong .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "immunology", "and", "inflammation" ]	2014	Active RNAP pre-initiation sites are highly mutated by cytidine deaminases in yeast, with AID targeting small RNA genes
Insertion of helix-forming segments into the membrane and their association determines the structure , function , and expression levels of all plasma membrane proteins . However , systematic and reliable quantification of membrane-protein energetics has been challenging . We developed a deep mutational scanning method to monitor the effects of hundreds of point mutations on helix insertion and self-association within the bacterial inner membrane . The assay quantifies insertion energetics for all natural amino acids at 27 positions across the membrane , revealing that the hydrophobicity of biological membranes is significantly higher than appreciated . We further quantitate the contributions to membrane-protein insertion from positively charged residues at the cytoplasm-membrane interface and reveal large and unanticipated differences among these residues . Finally , we derive comprehensive mutational landscapes in the membrane domains of Glycophorin A and the ErbB2 oncogene , and find that insertion and self-association are strongly coupled in receptor homodimers . The past four decades have seen persistent efforts to decipher the contributions to membrane-protein energetics ( Reynolds et al . , 1974; Cymer et al . , 2015 ) . Membrane-protein folding can be conceptually divided into two thermodynamic stages ( Popot and Engelman , 1990; Cymer et al . , 2015 ) , each of which affects membrane-protein structure , function , and expression levels: the insertion into the membrane of transmembrane segments as α helices , and their association to form helix bundles ( Ben-Tal et al . , 1996; Heinrich and Rapoport , 2003; Moll and Thompson , 1994; White and Wimley , 1999; Popot and Engelman , 1990 ) . While significant progress has been made in structure prediction , design , and engineering of soluble proteins ( Fleishman and Baker , 2012 ) , important but fewer successes were reported in design of membrane proteins ( Joh et al . , 2014; Li et al . , 2004 ) , largely owing to the complexity of the plasma membrane and the lack of systematic and accurate measurements of membrane-protein energetics ( Cymer et al . , 2015 ) . Recently , experimental systems that offer a realistic model for biological membranes have advanced . von Heijne and co-workers quantitated the partitioning of engineered peptides fused to the bacterial transmembrane protein , leader peptidase ( Lep ) , between membrane-inserted and translocated states , and highlighted the importance of interactions between the translocon and the nascent polypeptide chain in determining partitioning ( Hessa et al . , 2007; Öjemalm et al . , 2013 ) . The insertion energetics obtained from this assay , however , were significantly lower than expected from previous theoretical and experimental studies; for instance , the apparent atomic-solvation parameter , which quantifies the free-energy contribution from the partitioning of hydrophobic surfaces to the membrane core , was only 10 cal/mol/Å2 according to the Lep measurements ( Ojemalm et al . , 2011 ) , compared to ~30 cal/mol/Å2 from previous analyses ( Andrew Karplus , 1997; Vajda et al . , 1995 ) . Additionally , the magnitude of the insertion free energies for individual amino acids were substantially lower according to the Lep system ( Hessa et al . , 2007; Ojemalm et al . , 2011; Öjemalm et al . , 2013 ) compared to other studies ( Moon and Fleming , 2011; Shental-Bechor et al . , 2006 ) . These discrepancies led to suggestions that the Lep measurements were 'compressed' relative to others due to interactions between the engineered protein and other membrane constituents ( Johansson and Lindahl , 2009; Shental-Bechor et al . , 2006 ) . Membrane-protein energetics are governed not only by the insertion but also by the association of helices into bundles . A significant body of work has shown that association is driven by packing interactions and short sequence motifs comprising small-xxx-small residues , where small is any of the small polar residues ( Ser , Gly , or Ala ) and x is any residue ( Russ and Engelman , 2000; Senes et al . , 2004; Melnyk et al . , 2004 ) . However , while it is recognized that insertion and association both play roles in protein energetics ( Duong et al . , 2007; Finger et al . , 2006; Moll and Thompson , 1994; Ben-Tal et al . , 1996; Heinrich and Rapoport , 2003; Popot and Engelman , 1990 ) , the interplay between these two aspects has not been subjected to systematic experimental analysis . Given the remaining open questions on membrane-protein and protein-protein interactions within the membrane , we established a high-throughput assay to shed light on both factors and their coupling in a systematic and unbiased way within the bacterial plasma membrane . To overcome gaps in our understanding of membrane-protein energetics , we adapted the TOXCAT-β−lactamase ( TβL ) screen ( Lis and Blumenthal , 2006; Russ and Engelman , 1999; Langosch et al . , 1996 ) for high-throughput analysis by deep mutational scanning ( Boucher et al . , 2014; Fowler and Fields , 2014 ) ; we refer to this new method as deep-sequencing TOXCAT-β−lactamase ( dsTβL ) . TβL is a genetic screen based on a chimera , in which a membrane-spanning segment is flanked on the amino terminus by the ToxR dimerization-dependent transcriptional activator of a chloramphenicol-resistance gene and on the carboxy terminus by β-lactamase ( Figure 1a ) . In this construct , bacterial survival on ampicillin monitors membrane integration ( Broome-Smith et al . , 1990; Jaurin et al . , 1982 ) , and survival on chloramphenicol correlates with self-association of the membrane span ( Lis and Blumenthal , 2006; Russ and Engelman , 1999; Langosch et al . , 1996 ) . Furthermore , β-lactamase and ToxR function only in the periplasm and cytoplasm , respectively; therefore , unlike previous assays of membrane-protein insertion ( Hessa et al . , 2007 ) , the orientation of the inserted segment relative to the membrane is known , and only proteins inserted with their carboxy terminus located in the periplasm would be selected . Most studies using the TOXCAT screen fused maltose-binding protein ( MBP ) in the carboxy-terminal domain ( instead of β-lactamase ) and used the MBP-null E . coli strain MM39 and maltose as sole carbon source to monitor membrane integration ( Russ and Engelman , 1999; Langosch et al . , 1996; Melnyk et al . , 2004; Li et al . , 2004 ) . Unlike this conventional TOXCAT experiment , in TβL survival on ampicillin depends linearly on β−lactamase expression levels ( Broome-Smith et al . , 1990; Jaurin et al . , 1982; Li et al . , 2004; Lis and Blumenthal , 2006 ) , thus providing a more sensitive reporter for membrane insertion than MBP . Furthermore , the MM39 strain is not amenable to high-throughput transformation as required by our study; with TβL , we were able to use the high-transformation efficiency E . cloni cells ( see Materials and methods ) . 10 . 7554/eLife . 12125 . 003Figure 1 . The dsTβL assay for measuring the sequence determinants of membrane-protein energetics . ( a ) The TβL genetic construct fuses a membrane segment to two antibiotic selection markers: β-lactamase and ToxR , which report on insertion and self-association , respectively . ( b ) Libraries encoding every point mutation of a membrane segment are plated on selective and non-selective medium . Following overnight growth , the libraries are extracted and DNA segments , which encode the membrane domain , are subjected to deep-sequencing analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 12125 . 00310 . 7554/eLife . 12125 . 004Figure 1—source data 1 . Deep-sequencing read quality is high throughout the membrane-spanning segment . The deep-sequencing analysis software ( Babraham Institute , Cambridge , UK ) provides a quality control assessment , which is high ( green ) throughout the membrane span ( marked by red lines ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12125 . 00410 . 7554/eLife . 12125 . 005Figure 1—source data 2 . Deep-sequencing counts for each mutant . Amino acid positions and substitutions are represented in rows and columns , respectively . ( Top ) spectinomycin selection ( reference ) counts . ( Bottom ) spectinomycin and ampicillin selection . DOI: http://dx . doi . org/10 . 7554/eLife . 12125 . 005 Previous studies based on ToxR activity measured the effects of mutations using colony growth and enzyme-linked immunosorbant assay ( ELISA ) , which do not allow high-throughput analysis ( Mendrola et al . , 2002; Langosch et al . , 1996; Lis and Blumenthal , 2006; Russ and Engelman , 2000; Melnyk et al . , 2004 ) . Here , instead , we subject libraries encoding every amino acid substitution in the membrane domain to selection on agar plates containing either ampicillin alone or ampicillin and chloramphenicol to monitor insertion and self-association , respectively ( Figure 1b ) ; the same bacterial population is also plated on non-selective agar and serves as a reference to control for clonal-representation biases . Following overnight growth , the bacteria in each plate are pooled , plasmids encoding the TβL construct are extracted from each pool , and the variable gene segment , which encodes the membrane span , is amplified by PCR . The three resulting DNA samples are subjected to deep sequencing , which reports the relative frequency of each mutant in the selected and reference populations ( Boucher et al . , 2014 ) ( see equation 1 in Materials and methods ) . If the cytoplasmic protein fraction were perfectly constant among different mutants , the measured population frequencies could be interpreted as the relative propensities of each mutant to insert into the membrane or to self-associate in the membrane . This condition is unlikely to hold for all mutants; still , the agreement reported below with multiple lines of biophysical evidence on purified proteins suggests that the population frequencies provide a reasonable measure for changes in membrane-insertion and self-association partitioning . Hence , if we treat the population frequencies as if the mutants’ partitioning between cytosolic , membrane-inserted , and self-associated fractions were under thermodynamic control , following the Boltzmann equation we can derive , at each position i in the membrane span , apparent free energy changes for insertion or self-association due to the substitution from wild type to amino acid aa , ∆∆Gaa , iapp ( see equations 2–3 in Materials and methods ) . Although confounding factors , such as nonspecific interactions between the inserted segment and other bacterial membrane proteins , may affect the readout from the experiment , insertion and self-association are likely to dominate , since every mutant in this library differs from the wild type by only one amino acid; furthermore , all mutants are subjected to identical selection conditions , including temperature and antibiotic , thereby minimizing experimental noise ( Mackenzie and Fleming , 2008 ) . We used dsTβL to comprehensively map the sequence determinants of membrane insertion in a single-pass membrane segment ( Figure 2a ) . To minimize the effects of self-association on experimental readout , we chose the C-terminal portion of human L-Selectin ( CLS ) , which does not self-associate ( Srinivasan et al . , 2011 ) ( Figure 2—figure supplement 1 ) . Furthermore , the CLS amino acid sequence includes several polar amino acids ( Figure 2a , bottom ) ; we therefore reasoned that its membrane-expression levels would be sensitive even to point mutations . To verify that the deep-sequencing data reflected trends observed in experiments with single clones , we selected 10 single-point substitutions at the membrane-spanning segment’s amino terminus and at its core , and subjected them to experimental analysis on a clone-by-clone basis . Each clone and the wild type were grown overnight in non-selective medium , normalized to the same density , and plated in serial dilutions on ampicillin-containing agar to estimate relative viability ( Figure 2—figure supplement 2 ) . Nine of the 10 selected single-point substitutions ( all but Val302Lys ) showed qualitatively similar trends of viability in deep sequencing and single-clone analysis . The resolution of the deep-sequencing data , however , is much greater than that seen in the single-clone assays; for instance , whereas both charge mutations , Met303Glu and Ala311Arg , are highly deleterious according to deep sequencing and plate viability , the ∆∆Ginsertionapp value from deep sequencing for the former is 3 . 7 kcal/mol compared to 1 . 3 kcal/mol for the latter , emphasizing the larger dynamic range of deep sequencing compared to traditional viability screens . We next expressed these 10 mutants in non-selective conditions , isolated membrane preparations for each ( Molloy , 2008 ) , and measured membrane-localization levels relative to wild type using Western blots with an antibody targeting β-lactamase ( Figure 2—figure supplement 3 see Materials and methods ) . All clones expressed in the membrane fraction and ran at the expected size of ~55 kDa . Of the 10 tested mutations , 6 showed the expected trends , including mutations that increased ( Met303Leu , Val304Phe , and Ala311Leu ) or decreased expression ( Val302Lys , Leu310Ala , and Ala311Arg ) in agreement with the deep-sequencing and single-clone viability data ( Figure 2—figure supplement 3 ) . For instance , Ala311Arg was much less viable and showed lower membrane localization than wild type , whereas Ala311Leu was more viable and had higher membrane localization . However , three mutants to charges at the amino terminus ( Val311His , Met303Glu , and Val304Asp ) increased expression levels according to Western blots but were disruptive according to dsTβL . We attribute this difference to the fact that ampicillin viability reports not only on membrane-expression levels but also on the appropriate membrane integration , which is not captured by Western blots . 10 . 7554/eLife . 12125 . 006Figure 2 . The sequence determinants of protein expression in native bacterial membranes . ( a ) Each tile reports the apparent change in free energy ∆∆Ginsertionapp relative to wild type for every CLS point mutant ( see equation 3 in Materials and methods ) . Gray tiles mark substitutions that were eliminated from the analysis due to low counts ( <100 ) in the reference population . ( b ) Per-position insertion profiles for each amino acid residue . ( c ) Comparison of dsTβL insertion results at the plasma membrane mid-plane ( Z = 0 ) with values from the Moon scale ( Moon and Fleming , 2011 ) . ( d ) The apparent atomic-solvation parameter is the slope of the linear regression of ∆∆Ginsertionapp and computed change in solvent-accessible surface area ( SASA ) due to each mutation ( slope = -37 cal/mol/Å2 , r2 = 0 . 48 , p<0 . 0001 ) ( see Materials and methods ) . ( inset ) Inferring the atomic-solvation parameter from the relationship of ∆∆Ginsertionapp at Z = 0 for aliphatic residues and their change in SASA computed on a model poly-Ala α helix ( slope=-32 cal/mol/Å2 , p = 0 . 002 ) . CLS , C-terminal portion of human L-SelectinDOI: http://dx . doi . org/10 . 7554/eLife . 12125 . 00610 . 7554/eLife . 12125 . 007Figure 2—figure supplement 1 . Human CLS and its single-site mutation library do not self associate . By plating wild-type human CLS ( a ) and the library encoding every single-site mutation in its putative membrane span ( b ) on agar containing different antibiotic markers ( spectinomycin on the left; ampicillin in the middle; and chloramphenicol on the right ) , we show that human CLS inserts into the membrane ( ampicillin marker , middle plate ) but does not self associate ( Srinivasan et al . , 2011 ) ( right plate ) . Supplementary file 2 lists antibiotic concentrations . CLS , C-terminal portion of human L-SelectinDOI: http://dx . doi . org/10 . 7554/eLife . 12125 . 00710 . 7554/eLife . 12125 . 008Figure 2—figure supplement 2 . Deep-sequencing data reflects trends observed in clone-by-clone experiments . 10 single-point substitutions at the membrane-spanning segment’s amino terminus and at its core were grown overnight in non-selective medium , normalized to the same density , and plated in serial dilutions on agar containing 400 μg/ml ampicillin to estimate relative viability . DOI: http://dx . doi . org/10 . 7554/eLife . 12125 . 00810 . 7554/eLife . 12125 . 009Figure 2—figure supplement 3 . Western blot quantification of membrane expression of TβL mutants . Cells harboring the wild-type construct and point mutations were fractionated by ultra-centrifugation . Whole-cell extract and pelleted membrane fractions are shown with anti-β-lactamase antibody . The band intensity was analyzed by densitometry and normalized to wild type and displayed as log2 fold change . DOI: http://dx . doi . org/10 . 7554/eLife . 12125 . 00910 . 7554/eLife . 12125 . 010Figure 2—figure supplement 4 . Comparison of dsTβL insertion results at the membrane mid-plane with published hydrophobicity scales . Comparing the membrane mid-plane insertion energy ( parameter c in Supplementary file 1 with corresponding values from previously published hydrophobicity scales ( Wimley et al . , 1996; Kyte and Doolittle , 1982; Kessel and Ben-Tal , 2002; Hessa et al . , 2007; Moon and Fleming , 2011 ) . Black points represent aliphatic amino acids , blue represent polar amino acids , orange aromatic amino acids . The aliphatic insertion data in dsTβL and the Hessa scale ( Hessa et al . , 2007 ) and Moon scale ( Moon and Fleming , 2011 ) are highly correlated ( r2 = 0 . 79 and r2 = 0 . 90 , respectively ) . The slope of the correlation line for the aliphatics is close to 1 for the Moon scale , whereas it is 0 . 26 for the Hessa scale , reflecting a roughly four-fold lower change in hydrophobicity upon membrane insertion in the Hessa et al . ( 2007 ) assays compared to the dsTβL assay . DOI: http://dx . doi . org/10 . 7554/eLife . 12125 . 010 We next computed the apparent free-energy change of each substitution across the membrane relative to a substitution to Ala , and at each position i computed the running average over five neighboring positions [i-2…i+2] ( Figure 2b; Supplementary file 1 ) . The resulting profiles describe the energetics of inserting each of the twenty amino acids relative to Ala at each position across the bacterial plasma membrane ( Figure 2b ) . Although the location of the membrane mid-plane ( Z = 0 ) could not be determined unambiguously in this assay , we estimated it by aligning the hydrophobic residues’ profiles ( Leu , Ile , Met , and Phe ) , thereby locating the profiles' troughs and the presumed membrane mid-plane at CLS position Ala311 . The small and polar amino acids , Ser , Thr , and Cys have shallow , nearly neutral profiles , ranging from −0 . 1 to +0 . 8 kcal/mol . By contrast , the helix-distorting amino acids , Gly and Pro , which expose the polar protein backbone to the hydrophobic membrane environment , have a high disruptive profile , which peaks ( ~2 kcal/mol ) at the membrane mid-plane , emphasizing the strong unfavorable impact of exposing the polar protein backbone to the membrane environment . The large polar ( Asn , His , and Gln ) and charged residues ( Asp , Glu , Lys , and Arg ) are all highly disruptive in the membrane mid-plane . We note that the energetic penalties for Asp , Asn , His , Gln , Glu , and Lys cannot be determined precisely from the dsTβL assay , since the number of reads for substitutions to these residues at the membrane mid-plane in the selected population is nearly 0 , reflecting exceedingly large negative-selection pressures ( Figure 1; Supplementary file 2 ) . At the membrane mid-plane , the hydrophobic residues , Val , Ile , Leu , Met , and Phe , show the expected troughs , which are shallower for the small amino acid Val ( approximately −0 . 5 kcal/mol ) than for the large amino acids ( <−1 . 5 kcal/mol ) . We compared the dsTβL values for hydrophobic residues in the membrane mid-plane to values from five hydrophobicity scales ( Figure 2—figure supplement 4 ) . dsTβL fits well to the Moon scale ( Figure 2c , r2 = 0 . 90 , with a slope close to 1 ) , which similar to dsTβL measures substitution effects in a bacterial membrane – albeit the outer membrane ( Moon and Fleming , 2011 ) . The correspondence between dsTβL , which is based on in vivo measurement of membrane integration in a bacterial population , with biophysical assays on purified proteins , partly confirms the use of dsTβL for studying membrane-protein energetics . The dsTβL profile for Trp is similar to the profiles of the aliphatic residues , whereas Tyr makes a nearly neutral contribution to insertion in the membrane core . These profiles diverge from statistical inferences from membrane-protein structures and partitioning experiments , which show that Tyr and Trp preferentially line the membrane-water interface ( Ulmschneider et al . , 2005; Schramm et al . , 2012; Senes et al . , 2007; Nakashima and Nishikawa , 1992; Yau et al . , 1998 ) . Further experimental analysis of the role of aromatic residues in membrane-protein stability is warranted , and one possible explanation for these differences is that in the dsTβL assay aromatic residues on the membrane-spanning segment lack neighboring aromatic residues with which to form stabilizing stacking interactions; indeed , experimental stability measurements have shown that stacking makes a significant contribution to the net stabilization provided by aromatic residues in membrane proteins ( Hong et al . , 2007 , 2013 ) . Recently , controversy has surrounded the question of how hydrophobic are biological membranes ( Johansson and Lindahl , 2009 ) . On the one hand , theoretical considerations and values inferred from hydrocarbons in solution suggested that the free energy contribution due to inserting aliphatic groups into the membrane , or the atomic-solvation parameter , is ~30 ± 5 cal/mol/Å2 of nonpolar surface area ( Vajda et al . , 1995; Andrew Karplus , 1997 ) ; on the other hand , the Lep measurements suggested values of only 10 cal/mol/Å2 ( Ojemalm et al . , 2011 ) . We analyzed dsTβL data for 39 substitutions from one aliphatic residue ( Ala , Val , Ile , Leu , and Met ) to another at the core of the membrane ( −9 Å<Z<13 Å ) and inferred an apparent atomic-solvation parameter of 37 ± 6 cal/mol/Å2 ( Figure 2d ) . We additionally derived an atomic-solvation parameter of 32 ± 4 cal/mol/Å2 by analyzing the apparent insertion free energies at the membrane mid-plane ( ∆∆Gz=0app ) for each of the aliphatic residues ( Figure 2d , inset ) . The values we infer for the atomic-solvation parameter are therefore in fair agreement with values for protein cores and hydrocarbons in aqueous solution ( Vajda et al . , 1995 ) , and 3–4 times larger than the value inferred from the Lep system ( Ojemalm et al . , 2011 ) . We further note that while the ranking of apparent insertion free energies of aliphatic amino acids in dsTβL and Lep ( Hessa et al . , 2007 ) is similar ( r2 = 0 . 79 , Figure 2—figure supplement 4 ) , the magnitude of the insertion free-energy changes is nearly four times greater according to dsTβL . We conclude that our results support the view that the hydrophobicity of the plasma membrane core is similar to that of hydrocarbons and much higher than measured in the Lep system . A hallmark of plasma membrane proteins is the charge asymmetry known as the ‘positive-inside’ rule , according to which the cytoplasmic-facing side of the protein is much more positively charged than the periplasmic or extracellular-facing side ( von Heijne , 1989 ) . This asymmetry has been used to successfully predict the orientation of membrane proteins ( von Heijne , 1989 ) , but experimental quantification of the energetics of this asymmetry met with difficulty; previous studies measured only a small energy difference ( -0 . 5 kcal/mol ) between inserting Arg and Lys in the cytoplasmic relative to the extracellular-facing side of the membrane and no asymmetry for His ( Lerch-Bader et al . , 2008; Öjemalm et al . , 2013 ) . A striking feature of the dsTβL profiles , by contrast , is that they show clear and large asymmetries for Arg , Lys , and His , in agreement with the ‘positive-inside’ rule ( Figure 2b ) . The three profiles , however , are not identical: whereas Lys and Arg are favored by 2 kcal/mol near the cytoplasm compared to near the periplasm , the titratable amino acid His shows a more modest asymmetry of 1 kcal/mol; moreover , of these three amino acids , only Arg stabilizes the segment near the cytosol , whereas Lys and His are nearly neutral at the cytosol-membrane interface . This difference between Arg and Lys , which has not been noted until now , may be due to charge delocalization in the Arg sidechain and Arg’s ability to form more hydrogen bonds with lipid phosphate headgroups . We compared the relative propensity of each of the 20 amino acids at each position across the membrane ( Figure 3; equation 4 in Materials and methods ) . The results clearly reflect the asymmetric distribution of charges across the plasma membrane , with Arg as the most favored amino acid at the cytoplasmic surface , giving way to the large and hydrophobic amino acids , Leu , Ile , and Phe . Since the propensities reflect protein-lipid interactions , they could be used to engineer variants of natural membrane proteins that exhibit higher stability and expression levels by mutating membrane-facing positions to the highest-propensity identity . Furthermore , the results suggest that the insertion profiles could be used for bioinformatics prediction of the locations and orientations of membrane-spanning proteins , [manuscript in preparation ( Elazar et al . ) ] . 10 . 7554/eLife . 12125 . 011Figure 3 . Amino acid propensities across the plasma membrane . ( a ) The relative frequencies of each amino acid within the membrane ( see equation 4 in Materials and methods ) in sequence-logo format; the height of each letter corresponds to the amino acid’s propensity . DOI: http://dx . doi . org/10 . 7554/eLife . 12125 . 011 With the accumulation of membrane-protein molecular structures , it has become possible to derive knowledge-based potentials for the insertion of amino acids across the membrane from distributions of amino acids observed in structures ( Ulmschneider et al . , 2005 ) . We compared the dsTβL profiles to three knowledge-based profiles published over the past decade ( Senes et al . , 2007; Ulmschneider et al . , 2005; Schramm et al . , 2012 ) ( Figure 4 ) . The dsTβL profiles are similar to the knowledge-based ones for the weakly polar and hydrophobic residues ( Val , Ser , and Thr ) , but they diverge with respect to the more hydrophobic and polar residues: for instance , the dsTβL apparent insertion energy for Leu , Ile , and Phe at the membrane mid-plane is −2 kcal/mol and around −0 . 5kcal/mol for the other scales . Additionally , the only knowledge-based scale that attempted to derive an asymmetric insertion potential ( Schramm et al . , 2012 ) reported much smaller asymmetries for Lys and Arg , of around 0 . 5 kcal/mol , compared to 2kcal/mol in dsTβL . These differences are likely due to the fact that knowledge-based potentials reflect the frequencies of amino acids in membrane proteins and are biased by functional constraints; indeed polar residues are often found in the membrane core , where they have important roles in oligomerization , substrate binding , transport , and conformational change . Furthermore , the dsTβL experiment is based on a single-pass segment , where every position is exposed to the membrane , whereas the knowledge-based profiles are derived from all structures , including multi-pass proteins , where many residues mediate interactions with other protein segments or line water-filled cavities . 10 . 7554/eLife . 12125 . 012Figure 4 . Comparison of the dsTβL insertion profiles and knowledge-based profiles of amino acid distributions in membrane-protein structures . Equations for the knowledge-based profiles were taken from Ulmschneider et al . ( 2005 ) , Schramm et al . ( 2012 ) , and Senes et al . ( 2007 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12125 . 012 Insertion and association of membrane-spanning helices are thermodynamically coupled ( Kessel and Ben-Tal , 2002; Moll and Thompson , 1994; Popot and Engelman , 1990 ) , but except in one study ( Duong et al . , 2007 ) these two aspects were assayed separately ( Fleming et al . , 1997; Finger et al . , 2009; Hessa et al . , 2007; Mendrola et al . , 2002 ) . To test the coupling between insertion and association , we applied dsTβL to two model systems for studying membrane protein self-association: the membrane domains of the human erythrocyte sialoglycoprotein Glycophorin A ( GpA ) and the ErbB2 oncogene , and compared survival on ampicillin and chloramphenicol to survival on ampicillin alone . Some of the amino acid positions that mediate self-association in GpA and ErbB2 according to their experimentally determined structures ( Bocharov et al . , 2008; MacKenzie et al . , 1997 ) show large decreases in chloramphenicol viability upon mutation . Unexpectedly , however , many other mutations that have large effects on chloramphenicol viability are not close to the dimerization surfaces ( Figure 5—figure supplement 1 ) , suggesting that factors other than self-association dominate the chloramphenicol-viability landscape . To test whether these confounding results are due to variability in expression levels among the mutants , we subtracted from the observed effects of every mutant the expected effects due to expression-level changes according to the dsTβL insertion profiles ( see equation 5 in Materials and methods ) . The corrected self-association mutational landscapes now correctly discriminate positions that mediate self-association from those that do not ( Figure 5a ) , and clearly highlight the interaction motifs in GpA and ErbB2 . We compared the results from the systematic self-association landscape of GpA to a previous analysis of 24 GpA mutants that were individually screened for self-association using TOXCAT and corrected for differences in membrane expression using Western blots ( Figure 5b ) ( Duong et al . , 2007 ) . The results from dsTβL and the previous study are consistent ( r2 = 0 . 66; Figure 5b ) , confirming the use of the dsTβL insertion scale to correct self-association mutational landscapes in single-pass homodimers . Furthermore , by systematically probing every position across the membrane , our results highlight additional positions that are sensitive to mutation , such as GpA’s Gly86 , which was previously not subjected to mutagenesis ( Lemmon et al . , 1992; Duong et al . , 2007 ) . Moreover , it is notable that although the vast majority of mutations are either neutral or disruptive to self-association , some mutations , for instance to Met in the GpA amino terminus , may promote self-association in the context of the TβL construct ( Figure 5a ) . Our results further confirm the strong interplay between membrane-protein expression levels and association , and the importance of accounting for both in biophysical experiments on membrane proteins . 10 . 7554/eLife . 12125 . 013Figure 5 . Mutational scanning reveals strong coupling between insertion and self-association in receptor homodimers . ( a ) The mutational landscapes discriminate positions that are involved in self-association ( known associating residues are depicted in boldface ) from those that do not . ( b ) Comparison of expression-corrected apparent free energy of insertion for 24 Glycophorin A ( GpA ) mutants ( Duong et al . , 2007 ) with results from the dsTβL self-association mutation landscapes . ( c ) Dimer models that associate through positions that are sensitive to mutation ( * in panel a ) , and do not associate through positions that are insensitive to mutation ( † in panel a ) . The models ( green ) are close to the experimental structures ( blue ) for Glycophorin A ( MacKenzie et al . , 1997 ) ( 1 . 3 Å root mean square deviation ) and ErbB2 ( Bocharov et al . , 2008 ) ( 1 . 9 Å ) . Another ErbB2 model , which agrees with biochemical and computational evidence ( Endres et al . , 2013; Arkhipov et al . , 2013; Fleishman et al . , 2002 ) but has not been observed in experimental structures , is also suggested . DOI: http://dx . doi . org/10 . 7554/eLife . 12125 . 01310 . 7554/eLife . 12125 . 014Figure 5—figure supplement 1 . The mutational landscapes comparing survival on chloramphenicol and ampicillin are dominated by insertion effects . Chloramphenicol resistance in the dsTβL assay is expected to correlate with self-association ( Lis and Blumenthal , 2006; Russ and Engelman , 1999; Langosch et al . , 1996 ) . While mutations at some positions that mediate self-association disrupt viability , the majority of extreme effects observed in the mutational landscapes can be attributed to insertion rather than self-association . For instance , the charged and polar residues are highly disfavored in the membrane core , whereas the large hydrophobic residues , Leu and Phe , are favored in most positions in the membrane core . Figure 5a shows mutational landscapes where the effects of insertion were subtracted . DOI: http://dx . doi . org/10 . 7554/eLife . 12125 . 01410 . 7554/eLife . 12125 . 015Figure 5—figure supplement 2 . Alternative predicted structures for Glycophorin A and ErbB2 . In addition to the model structures of Figure 5b modeling constrained by the dsTβL , experimental data produced three models for Glycophorin A and two for ErbB2 . Improvements in the RosettaMembrane ( Yarov-Yarovoy et al . , 2006 ) energy function would be needed to eliminate these alternative models . DOI: http://dx . doi . org/10 . 7554/eLife . 12125 . 015 We also tested whether the systematic mutational landscapes generated by dsTβL could be used to provide constraints for structure modeling of receptor membrane domains ( Fleishman et al . , 2002; Kim et al . , 2003; Polyansky et al . , 2014 ) . We used the Rosetta biomolecular-modeling software ( Das and Baker , 2008; Yarov-Yarovoy et al . , 2006 ) to generate 100 , 000 structure models of GpA and ErbB2 directly from their sequences , and selected structures that self-associate through positions that are sensitive to mutation according to dsTβL but not through positions that are insensitive to mutation ( Figure 5c , Figure 5—figure supplement 2 ) . In both cases , fewer than five models passed the selection criteria and of those , some models were within 2 Å of experimentally determined structures . Despite progress in measuring protein energetics within biological membranes , significant open questions remained , among them , what is the hydrophobicity at the core of biological membranes; what is the magnitude of the bias for positively charged residues at the cytoplasm surface; and how strong is the coupling between membrane-protein insertion and association energetics ? To shed light on these fundamental questions , we established a high-throughput genetic screen and used it to generate systematic mutation landscapes of insertion and self-association in the plasma membrane of live bacteria . The apparent insertion energies in dsTβL are in line with biophysical stability measurements on outer-membrane proteins ( Moon and Fleming , 2011 ) , and the inferred atomic-solvation parameter is close to measurements in model systems and protein cores ( Andrew Karplus , 1997; Vajda et al . , 1995 ) . Our measurements , however , are three to four times larger than the corresponding ones using the Lep system ( Ojemalm et al . , 2011; 2013; Hessa et al . , 2007 ) . To be sure , we are not the first to note these large differences ( Johansson and Lindahl , 2009; Shental-Bechor et al . , 2006 ) ; yet , we find it significant that our measurements , similar to those in the Lep system , use biological membranes . The observation that the dsTβL insertion measurements for aliphatic side chains have the same ranking but are fourfold larger in magnitude compared to those from Lep ( Figure 2—figure supplement 4 ) may indicate that the Lep system measures only a part of the energy contribution to insertion . While further investigation is needed , we speculate that the reason for the large differences between dsTβL and Lep is that total membrane-protein expression levels were not quantified in the Lep system ( Hessa et al . , 2007; Ojemalm et al . , 2011; 2013 ) . We note the following two caveats regarding the dsTβL insertion profiles . First , the penalties for most polar residues at the membrane mid-plane are likely to indicate lower bounds on their insertion energies , since the number of clones counted in the deep-sequencing data for these mutants is close to 0 ( supplementary data ) . Second , statistical analyses ( Ulmschneider et al . , 2005; Schramm et al . , 2012; Senes et al . , 2007 ) and experiments ( Hessa et al . , 2007 ) demonstrated that the aromatics Tyr and Trp are preferred in the water-membrane interface rather than in the core , although dsTβL shows the reverse ( Figure 2b ) . We suggest that these results reflect the fact that dsTβL is based on a monomeric construct where the aromatics are fully exposed to the membrane environment; however , these uncertainties require further research . The TOXCAT genetic screen has made essential contributions to our understanding of self-association in the membrane ( Lindner and Langosch , 2006; Lis and Blumenthal , 2006; Russ and Engelman , 1999; Finger et al . , 2009; Mendrola et al . , 2002; Li et al . , 2004; Srinivasan et al . , 2011Reuven et al . , 2012 ) . Some early reports demonstrated that chloramphenicol survival also depends on membrane-protein expression levels ( Russ and Engelman , 1999; Duong et al . , 2007 ) . Our results strongly support this view and show that expression levels are a dominant factor in chloramphenicol survival . This dominance is perhaps not surprising in retrospect , given that a mutation’s effects on monomer concentrations are counted twice in computing its effects on homodimer concentrations , and therefore on chloramphenicol viability ( see equation ( 5 ) in Materials and methods ) . A key contribution of unbiased and systematic assays , such as dsTβL , is that they clarify such trends unambiguously . Furthermore , the dsTβL insertion profiles derived from the monomeric CLS provide a self-consistent way to factor out the contributions from insertion energetics in future assays on membrane-protein association or function in unrelated membrane proteins , thereby eschewing the need to measure the expression levels of individual mutants . Deep mutational scanning has made important inroads to analysis and optimization of diverse protein systems ( Whitehead et al . , 2012; Fowler and Fields , 2014; Boucher et al . , 2014 ) . The main strengths of deep mutational scanning are the ability to measure the effects of all point mutations without bias and that all mutants experience strictly equal experimental conditions , thereby limiting experimental noise . The structural simplicity of the model systems tested here , consisting of a single α helix or of helix homodimers , plays a further role in the ability to accurately infer energetics . Combined with structural modeling , the assay can provide essential information both on association energetics and the molecular architecture of membrane receptors . More generally the data on protein-membrane and protein-protein energetics obtained from dsTβL will be used to improve models of membrane-protein energetics and to design , screen , and engineer high-expression mutants of specific membrane proteins ( Fleishman and Baker , 2012; Joh et al . , 2014 ) . The p-Mal plasmid was generously provided by the Mark Lemmon laboratory . We replaced the maltose-binding protein domain at the open-reading frame carboxy-terminus with β-lactamase ( Lis and Blumenthal , 2006 ) . The restriction sites in multiple-cloning site 1 were changed to XhoI and SpeI . The p-Mal plasmid contains a gene for spectinomycin resistance , which is constitutively expressed , providing selection pressure for transformation . The open-reading frame encompassing the TβL construct is also constitutively expressed and is under the control of the weak ToxR promoter . The DNA coding sequence for the transmembrane constructs used in the paper: >human CLS CCGCTGTTCATCCCGGTTGCAGTTATGGTTACCGCTTTTAGTGGATTGGCGTTTATCATCTGGCTGGCT ( amino acid sequence: PLFIPVAVMVTAFSGLAFIIWLA ) >Glycophorin A CTCATTATTTTTGGGGTGATGGCTGGTGTTATTGGAACGATCCTGATC ( amino acid sequence: LIIFGVMAGVIGTILI ) >ErbB2 CTGACGTCTATCATCTCTGCGGTGGTTGGCATTCTGCTGGTCGTGGTCTTGGGCGTGGTCTTTGGCATCCTGATC ( amino acid sequence: LTSIISAVVGILLVVVLGVVFGILI ) The CLS construct was deposited in the AddGene repository [pMAL_dstβL- ( Plasmid #73805 ) ] . All experiments were conducted using the high-transformation efficiency E . cloni cells ( Lucigen Corporation , Middleton , WI ) . Customized MatLab 8 . 0 ( MathWorks , Nattick , Massachusetts ) scripts for generating primers were written ( supplementary files ) to generate forward and reverse DNA oligos of lengths 40–85 base pairs , where the central codon is replaced by the degenerate codon NNS , where N is any of the four nucleotides ( ATGC ) and S is G or C , encoding all possible natural amino acids . Resulting primers were ordered from Sigma ( Sigma-Aldrich , Rehovot , Israel ) . For example , to replace the central 302nd codon of human CLS with an NNS codon , the following two primers were ordered: >forward GCTGTTCATCCCGGTTGCAGTTNNSTGGTTACCGCTTTTAGTGGATTG >reverse CAATCCACTAAAAGCGGTAACCASNNAACTGCAACCGGGATGAACAGC Each pair of oligos was then cloned into the wild type by restriction-free ( RF ) cloning ( van den Ent and Löwe , 2006 ) . The resulting plasmids from the library-construction step above were electroporated into E . cloni and plated on agar plates containing 50 μg/ml spectinomycin . Plasmids for each position were transformed and plated separately and positions with fewer than 200 colonies were retransformed . All positions were then pooled and used to inoculate 10 ml of Luria Broth medium ( LB ) with 50 μg/ml spectinomycin and grown in a shaker at 200 rpm and 37°C over-night , diluted 1:1000 and grown to OD = 0 . 2–0 . 4 . The libraries were then diluted to OD = 0 . 1 and 200 μl of the resulting cultures were plated at different dilutions ( 1:1 , 1:10 , 1:100 , 1:1000 ) on large 12-cm petri dishes containing spectinomycin , ampicillin alone , or ampicillin and chloramphenicol . After overnight incubation at 37°C , p-Mal plasmids were extracted from the resulting colonies using a miniprep kit ( Qiagen , Valencia , California ) . Every wild-type membrane-spanning segment exhibits different sensitivity to chloramphenicol and ampicillin . To determine the concentrations that are most likely to provide maximal dynamic range , we started by cloning mutants that are predicted to reduce insertion of the membrane-spanning segment or its self association ( Mendrola et al . , 2002 ) . Results are represented in Supplementary file 1 . We next titrated the wild-type construct as well as the mutant on plates with varying concentrations of antibiotic to find the concentration that shows the largest difference in viability between the wild type and the compromising mutants . Supplementary file 2 provides the ampicillin and chloramphenicol concentrations used in each of the experiments reported in the paper . We generated a model of the CLS membrane domain by threading its sequence on a canonical α helix , and used Rosetta to singly introduce each substitution from one aliphatic identity ( Ala , Val , Ile , Met , Leu , and Phe ) to another in the membrane core . Amino acid sidechains were combinatorially repacked and the change in solvent-accessible surface area ( ΔSASA ) was computed . Four additional data points ( marked with asterisks , Figure 2d ) were extracted from Glycophorin A’s position Ala82 , which is located at the membrane center and away from the dimerization interface . To compute the atomic-solvation parameter from the insertion energies of the aliphatics at the membrane mid plane ( Figure 2d , inset ) , we compared the insertion energy at the membrane mid plane for each aliphatic residue ∆∆Gz=0app with computed ΔSASA of a change from that residue to Ala on a canonical poly-Ala α helix .	Cells are defined by a thin membrane that separates the inside of the cell from the outside . The core of this membrane is hydrophobic , meaning that it repels water . Many signals and nutrients cannot pass through the membrane itself , but can pass through the proteins that span the membrane . Membrane proteins are therefore essential for living cells; yet even after decades of research , it remains unclear how proteins interact with the membrane and which features determine a protein’s stability in a biological membrane . Since the early 1980s it was known that the bacterium E . coli could grow on a common antibiotic called ampicillin if it had enough of an antibiotic-degrading enzyme called β-lactamase anchored into its inner membrane . Now , Elazar et al . have used this enzyme to obtain detailed information on the interactions between a biological membrane and a membrane protein . First , hundreds of different mutations were introduced into the gene that encodes the enzyme to generate a population of bacteria that each had a slightly different membrane anchor . The mutant bacteria were then grown in the presence of the antibiotic , meaning that those mutants with a more stable membrane anchor were more likely to survive and grow than those with less stable anchors . Elazar et al . then collected all the surviving bacteria , sequenced their DNA and measured how common the different mutations were in the final population . This approach was less labor-intensive and more accurate than traditional methods for monitoring membrane-anchored proteins , and the resulting large dataset was used to uncover which features affect a protein’s stability in a membrane . These results also showed that a biological membrane’s core is considerably more hydrophobic than was previously thought . In addition to being hydrophobic , biological membranes have more negative charge in the side that faces into the cell . This means that membrane proteins with a positive charge in this region will be more stable , and Elazar et al . were able to use their new system to measure this effect for the first time . Finally , membrane proteins do not only span the membrane; they also bind with other membrane proteins in order to carry out their roles . Elazar et al . used their system to look at the surfaces of human membrane proteins that interact with one another , and build a detailed map of the interaction surfaces , from which they derived accurate models of the membrane proteins . Overall , these new findings could now be used to model the three-dimensional structures of membrane proteins and improve their stability . This in turn may help efforts to develop these proteins into more robust experimental tools and in the search for drugs that target membrane proteins .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2016	Mutational scanning reveals the determinants of protein insertion and association energetics in the plasma membrane
While the human brain is clearly large relative to body size , less is known about the timing of brain and brain component expansion within primates and the relative magnitude of volumetric increases . Using Bayesian phylogenetic comparative methods and data for both extant and fossil species , we identified that a distinct shift in brain-body scaling occurred as hominins diverged from other primates , and again as humans and Neanderthals diverged from other hominins . Within hominins , we detected a pattern of directional and accelerating evolution towards larger brains , consistent with a positive feedback process in the evolution of the human brain . Contrary to widespread assumptions , we found that the human neocortex is not exceptionally large relative to other brain structures . Instead , our analyses revealed a single increase in relative neocortex volume at the origin of haplorrhines , and an increase in relative cerebellar volume in apes . Primates vary almost a thousand-fold in endocranial volume – a measure which closely approximates brain size – ranging from 1 . 63 mL in mouse lemurs ( Isler et al . , 2008 ) to 1478 mL in humans ( Robson and Wood , 2008 ) . Body size is perhaps the most important statistical predictor of brain size across primates , with larger bodied species having larger brains , but substantial variation remains after accounting for the effects of body size ( Isler et al . , 2008 ) . While numerous comparative studies have sought to identify ecological , behavioral , and cognitive correlates of this variability ( Barton , 1999; MacLean et al . , 2014; DeCasien et al . , 2017; Powell et al . , 2017; Noonan et al . , 2018 ) , much less is known about the evolutionary patterns and processes that generated extant variation in brain size within the primate clade , how these differ for different components of the brain , or the degree to which the brain phenotypes of particular species , such as humans , are the result of exceptional patterns of evolutionary change . A common approach to investigating human uniqueness is to test whether humans fall ‘significantly’ far from a regression line , for example by regressing brain size on body mass ( Azevedo et al . , 2009; de Sousa et al . , 2010; Herculano-Houzel and Kaas , 2011 ) . One surprising recent result reported from such an analysis is that the mass of the human brain is only 10% greater than expected for a primate of human body mass ( Azevedo et al . , 2009 ) . However , such non-phylogenetic methods may give misleading results because they fail to incorporate trait co-variation among species that results from shared evolutionary history . Valid analysis requires methods that account for phylogeny both when estimating scaling parameters and when evaluating deviations from scaling patterns exhibited by individual species ( Garland and Ives , 2000; Ross et al . , 2004; Organ et al . , 2011 ) . An additional source of error arises if the species being investigated is included in the regression model ( e . g . Azevedo et al . , 2009 ) , particularly when , as for humans , the phenotypic trait lies at the extreme of the distribution for the other species in the analysis . This procedure would reduce the magnitude of deviations from expected trait values for lineages that have undergone exceptional change , and in the case of humans , would bias the results toward failing to detect uniqueness . Comparative methods make it possible to incorporate phylogeny into analyses and to model phenotypic evolution in ways that uncover hitherto hidden patterns . Such methods are now being applied to a wide variety of traits ( e . g . Vining and Nunn , 2016; Pagel , 1999; Orme , 2013 ) , including brain size . Pagel ( 2002 ) estimated phylogenetic scaling parameters to characterize the evolutionary trajectory of endocranial volume ( ECV ) in fossil hominins . His analyses revealed that ECV evolution accelerated towards the present . As this analysis did not account for body size , it is not clear to what extent this pattern reflects changes in brain size independent of body size . Montgomery et al . ( 2010 ) used ancestral state reconstruction with fossil data to demonstrate a directional trend in primate brain size evolution and to identify branches in the primate phylogeny along which exceptional evolutionary change occurred . They found that while the absolute change in the mass of the human brain was exceptional , the rate of change relative to body size was not . Phylogenetic methods have also been used to examine how specific brain components evolved and the extent to which the branch leading to humans exhibited unusual amounts or rates of change in the size of these components ( Barton and Venditti , 2013; Barton and Venditti , 2014 ) . Recently , Lewitus ( 2018 ) suggested that comparative analyses of neuroanatomical data can be improved by incorporating and comparing results from different evolutionary models . Here , we use phylogenetic methods to model the evolution of brain size and to identify exceptional evolutionary change along phylogenetic branches . We employ three methods: The first method models trait evolution both as a multi-optima Ornstein-Uhlenbeck ( OU ) process ( which incorporates stabilizing selection and drift ) and as a Brownian motion process ( Felsenstein , 1985 ) , and then compares the fit of the two models . In cases where the OU model is favored , exceptional patterns of trait evolution are indicated by recent shifts in adaptive optima in humans’ ( or other species’ ) evolutionary lineage . In cases where the Brownian model is favored , we apply our second method , which is a phylogenetic outlier test that uses phylogenetic generalized least squares ( PGLS ) to predict a phenotype for a species and then compares observed and predicted values . With this method , we can assess whether humans are a phylogenetic ‘outlier’ relative to expectations based on their phylogenetic position and trait covariation in other primate species . Our last method tests for directional and accelerating evolution by fitting phylogenetic scaling parameters to data on deviation from trait expectations and evolutionary time , building on previous efforts with these approaches ( Pagel , 2002 ) . Using the first two methods , we investigate the evolution of absolute brain size and brain size relative to body mass within primates . Absolute brain volume has been shown to predict cognitive ability in primates better than other metrics that account for body mass ( MacLean et al . , 2014; Deaner et al . , 2007 ) . However , brain size is highly correlated with body size ( Isler et al . , 2008 ) , and as such it is difficult to interpret the significance of brain size alone . Additionally , accounting for body mass gives more insights into the significance of brain size in life history processes , as relative brain size better approximates relative investment in cognitive ability . Accounting for body mass is also important as the relationship between this trait and brain size is associated with scaling effects that reflect conservation of neural function , such as preservation of somatosensory acuity across large surface areas ( St Wecker and Farel , 1994 ) and compensation for increased neural conduction distances in larger animals through ( i ) larger neuron and axon sizes , increased myelination , and increased white matter volume , all of which result in reduced neuron density ( Barton , 2012; Wang et al . , 2008; Collins et al . , 2013 ) and ( ii ) increased neural resources devoted to prediction-based sensorimotor control that result from escalating neural conduction delays as body size increases ( More et al . , 2010 ) . Other measures of relative brain size such as encephalization quotients , ratios , and residuals have been used in the past , but all make theoretical assumptions about the underlying relationship between brain and body size evolution that may not hold . Using relative measures can bias parameter estimates and is not recommended as a good statistical practice ( Freckleton , 2002 ) . Instead , an empirical approach is preferred in which the covariation of brain size with body size is accounted for within a statistical model that also accounts for phylogenetic history ( such as PGLS ) . We also apply the first two phylogenetic comparative methods to investigate the evolution of major brain structures involving the neocortex , cerebellum , and medulla . It is widely assumed that the neocortex expanded disproportionately relative to other brain structures during the evolution of anthropoid primates and most particularly in human evolution ( Kriegstein et al . , 2006; Geschwind and Rakic , 2013; Florio and Huttner , 2014 ) . Surprisingly however , direct tests of this hypothesis are lacking , despite the focus of much evolutionary and developmental neuroscience on the neocortex as the site of interest for understanding human uniqueness and its developmental mechanisms ( Mitchell and Silver , 2017 ) . Recent evidence suggests that the cerebellum may have contributed more to human brain evolution than previously appreciated: it underwent rapid evolutionary expansion in the great ape clade including hominins ( Barton and Venditti , 2014; Smaers et al . , 2018 ) and has been implicated in shape changes of the brain in hominin fossil endocasts ( Kochiyama et al . , 2018; Neubauer et al . , 2018 ) . Molecular evidence now corroborates the proposal that selection on cerebellar function was an important feature of hominoid and hominin brain evolution ( Sousa et al . , 2017 ) , with changes in protein-coding genes implicated in cerebellar development more likely to have evolved adaptively in apes than those implicated in neocortical development ( Harrison and Montgomery , 2017 ) . It therefore appears that the neocortex and cerebellum have had different evolutionary trajectories in primate evolutionary history . More research is needed to document and understand these patterns . We examined volumetric change in the neocortex and cerebellum relative to both body mass and the volume of the rest of the brain . As a check to establish whether changes in evolutionary patterns for relative neocortex and cerebellum size are primarily attributable to changes in those structures or to changes in the rest of the brain , we investigated the evolution of the rest of the brain relative to body mass . We also conducted analyses of the volume of the medulla relative to body mass and the volume of the rest of the brain . The relative volume of the medulla does not vary significantly across clades ( Barton , 2000 ) and as such it has not been attributed a major role in brain expansion . For the analyses of fossil species , brain component volumes are not available; thus , analyses of these lineages are restricted to overall brain size ( ECV ) . Although our main focus is on broad patterns across primate phylogeny and on the extent to which human brain evolution fits or departs from these patterns , we also examined brain evolution in other species that are considered to be unusually large-brained , such as the aye-aye ( Daubentonia ) and capuchins ( Cebinae ) ( Isler et al . , 2008; Pagel and Harvey , 1989 ) . Our analyses also help to identify other primate species that have experienced exceptional expansion or reduction of the brain or its components , generating new questions for future research on exceptional brain evolution in primates . We used our third method to characterize patterns of brain evolution in humans and extinct hominins . Pagel ( 2002 ) conducted similar analyses of raw ECV . Our analyses advance his findings in two ways . First , we incorporate body mass as a predictor . Second , we focus on the deviation from brain size expectations , based on the PGLS methods used to assess outlier status . Our findings therefore provide insights to the evolutionary trajectory of exceptional hominin ECV relative to primate-wide brain-body mass scaling relationships . We compiled ECV and female body mass data on non-human primates ( Isler et al . , 2008 ) as well as humans and fossil hominins ( Robson and Wood , 2008 , Tables 1 and 2 ) . Given that sex specific body mass estimates are available for ancient humans and extinct hominins ( Robson and Wood , 2008 ) , we used female values for body mass because female values are more tightly linked to ecological and life-history factors ( Gordon , 2006 ) and sexual selection can drive increases in male body mass unlinked to ecology , obscuring brain-body scaling relationships ( Fitzpatrick et al . , 2012 ) . We also compiled data on neocortex , cerebellum , and medulla volume ( Barton and Venditti , 2014; Stephan et al . , 1981; Bush and Allman , 2004 ) . Values used to compute predictor variables ( described below ) for analyses of brain sub-structures were taken from Isler et al . ( 2008 ) . We used several phylogenies in our analyses . For analyses of hominin ECV , we constructed a ‘hominin phylogeny’ by combining the hominin consensus tree from Organ et al . ( 2011 ) and the non-human primate consensus tree from 10kTrees version 3 ( Arnold et al . , 2010 ) . To ensure that our results in this set of analyses were not dependent upon the topology of the hominin phylogeny , we repeated them using an ‘alternate hominin phylogeny , ’ constructed in a similar manner using another hominin tree from Organ et al . ( 2011 ) . Details of the tree construction process are given in Appendix 1 . In all other analyses we used either the consensus primate phylogeny or a block of 100 primate phylogenies from 10kTrees , version 3 . To determine whether patterns of exceptional evolution represent absolute or relative changes in scaling , we included several predictor variables in our analyses . To investigate whether the volumes of structures changed relative to body size , we used body mass as a predictor variable , while we used a ‘rest-of-brain’ metric as a predictor variable to investigate whether the volumes of structures changed relative to other brain structures . For the analyses of all structures other than the medulla , the ‘rest-of-brain’ was computed as whole brain volume – ( neocortex volume +cerebellum vol ) . In analyses of the medulla , we calculated ‘rest-of-brain’ volume as brain volume - medulla volume . We also analyzed the volume of the ‘rest-of-brain’ [whole brain volume – ( neocortex volume +cerebellum vol ) ] relative to body mass . The data sets used in all analyses , along with more detailed descriptions , are given in Appendix 1 . We compared the fit of multi-optima Ornstein-Uhlenbeck ( OU ) models of evolution and Brownian models of evolution using a developmental version of the R package bayou ( Uyeda and Harmon , 2014; Uyeda , 2017 ) . OU models of evolution incorporate stabilizing selection and drift , while Brownian models only include drift . Bayou fits multi-optima OU models to a phylogeny using a Markov-Chain Monte Carlo ( MCMC ) approach . A shift in selection regime refers to a change in the parameters that determine the optimum trait value ( towards which species evolve ) at a specific location on a phylogeny . Thus , inferred changes in selective regime provide insights to how lineages differ . Shifts in selection regime along terminal branches of a tree would provide particularly strong evidence for a species’ uniqueness . Grabowski et al . ( 2016 ) proposed the following OU model to describe the evolution of a trait , y , as a function of a predictor variable , x: Equation 1:dy= − α ( y− y0 ) dt+ σ2 dB Equation 2:y0= θ+xβ In these equations , dy is the change in the trait value , α is the magnitude of the selective 'pull' towards the optimum trait value , y0 , and σ2 is the variance of the white noise process dB . The variables θ and β can be interpreted as the intercept and slope of the optimum regression line specified in Equation 2 . The optimum regression line represents the state that a species is evolving towards rather than the actual evolutionary trajectory . This model has limited utility when data for x are only available for the tips of the phylogeny because the values of x must be known along the branches of the phylogeny to infer the expected value of y for a lineage . We utilize two similar models implemented in the developmental version of bayou – the unweighted predictor model and the weighted predictor model ( corresponding to ‘immediate’ and ‘alphaweighted’ options for ‘slopechange’ in bayou ) – as these circumvent the issue of unknown phenotypes in ancestral lineages while incorporating a predictor variable into the OU model . The weighted predictor model considers the evolutionary history of the predictor variable while fitting models , and the unweighted predictor model only considers the values of the predictor variables at the tips of the phylogeny while fitting models . The details of these two models are provided in Appendix 2 . Bayou uses a MCMC to parameterize the models to fit the data by inferring the location and magnitude of concurrent shifts θ and β on a phylogeny and by inferring the values of α and σ2 , which remain constant across the phylogeny . The parameters α and σ2 are used in the calculation of the variance-covariance matrices used in evaluating model fit to the phylogeny . The phylogenetic half-life , the time needed for a trait to evolve halfway to the optimum , is computed as ln ( 2 ) / α . We present phylogenetic half-life in units of tree height . A phylogenetic half-life less than tree height indicates that the evolutionary processes can 'pull' parameter values to the optimum within the timescale in question , while a phylogenetic half-life that exceeds tree height or constitutes a large percentage of tree height indicates that evolutionary processes have a weak 'pull' and trait values are not expected to closely approach the optimum during the timescale in question . The expected variance in trait values evolving to the same optima at equilibrium ( stationary variance ) can be computed as σ22α . For each analysis , we ran the weighted and unweighted predictor models . We also ran a Brownian motion model in which the strength of stabilizing selection ( α ) was fixed at 10−6 ( resulting in a phylogenetic half-life ~9500 times greater than tree height; bayou cannot compute model likelihoods when α is 0 ) , and no shifts away from the root regime were allowed . The predictor variable is still incorporated in the Brownian motion model , but no changes in its coefficient occur on the phylogeny . We used the hominin tree for the analysis of ECV and the consensus tree of extant primates for all other analyses . All MCMCs were run for 5 , 005 , 000 time steps , sampling every 10 time steps . The priors used are given in Table 3 . For each analysis , two chains were run and checked for convergence in terms of likelihood , α , and σ2 ( see Appendix 3 for discussion of chain non-convergence issues in analyses of ECV ) . We also checked for correlation in branch-wise posterior shift probability between chains . Diagnostic plots pertaining to chain convergence are given in Source data 1 . The two chains were combined , with the first 30% of samples being discarded as burn in . We then obtained the likelihood of each model and calculated Bayes factors for each model pairing ( Kass and Raftery , 1995; Jeffreys , 1998 ) using the steppingstone algorithm in bayou , which implements the method of Fan et al . ( 2011 ) . We imposed a posterior probability cutoff of 0 . 3 for shift detection . When the multi-optima OU model was selected over the Brownian motion model , we used the location and magnitude of shifts in adaptive optima to assess changes in patterns of evolution . The inference of a shift on a terminal branch would indicate an exceptional pattern of evolution for a given species . Ho and Ané ( 2013 ) identified several potential problems with OU models , including un-identifiability of parameters and over-fitting , but acknowledged that such models may be necessary , and recommended that Bayesian models , specifically bayou , be used to overcome these problems . Several other phylogenetic OU models have been developed ( most notably Hansen , 1997 ) , but none utilized Bayesian parameter estimation . Cooper et al . ( 2016 ) echoed the concerns of Ho and Ané ( 2013 ) and again recommended using Bayesian approaches . Additionally , they recommended weighing the fit of an OU model of evolution against that of a Brownian model , which do through our model selection process . When bayou indicated that the Brownian model of trait evolution was favored over the multi-optima OU model , we conducted a phylogenetic outlier test . This was accomplished using BayesModelS , an R script that generates distributions of predicted trait values for a species or several species based on phylogenetically controlled analyses of trait covariation with predictor variables ( Nunn and Zhu , 2014 ) . BayesModelS uses a Markov-Chain Monte Carlo ( MCMC ) to fit parameters of a PGLS model and assumes a Brownian motion model of evolutionary change . The PGLS models are used to generate trait value predictions for the species of interest . Uncertainty in phylogenetic structure can be accounted for by sampling from a set of trees ( Pagel , 2002 ) . BayesModelS accounts for phylogenetic non-independence of residual trait values by incorporating branch scaling factors when fitting PGLS models . The MCMC samples between two branch length scaling factors , λ and κ , to improve the fit of the models . The parameter λ scales the internal branches of the phylogeny and measures phylogenetic signal ( Nunn , 2011 ) . Values for λ were constrained to be in the interval [0 , 1] . In the κ model phylogenetic tree branch lengths are raised to the power κ . The value of κ has previously been used to assess support for a ‘speciational’ mode of evolution ( see Pagel , 2002 ) . When predicting the value of a trait for a species ( or a group of species ) , its data were excluded from the BayesModelS analysis to avoid biasing the predictions . BayesModelS was then used to generate a posterior probability distribution of predicted values for that species , based on the predictor variable , estimated phylogenetic signal , and estimated trait co-variation with the other species in the analysis . Species were identified as outliers when their trait value was more extreme than 97 . 5% of the predicted trait values ( i . e . when trait values fell outside 95% credible interval ) . A species was identified as a positive outlier when its true value fell above the majority of predictions , and a negative outlier when the opposite was true . The analyses conducted using BayesModelS proceeded as follows . First , we investigated whether hominins follow primate brain size to body mass scaling rules by using BayesModelS to predict ECV based on body mass and phylogeny . We tested each hominin species for outlier status while excluding data on all hominins when generating predictions . When computing mean estimates for hominin ECV , we corrected for back transformation bias using the quasi-maximum likelihood estimator method described in Smith ( 1993 ) . We used the hominin phylogeny or the alternate hominin phylogeny in these analysis , and the data spanned 225 extant primate species ( including humans ) and 10 extinct hominin species . Next , we identified individual primate species that are evolutionary outliers for ECV and other brain structures ( neocortex , cerebellum , medulla , rest-of-brain ) . In these analyses , we accounted for phylogenetic uncertainty by using the block of 100 trees , which included H . sapiens and H . neanderthalensis but no other hominins . We iteratively tested each species in the data set for outlier status . Our analysis for ECV included data from 145 species , and our analyses for other brain structures structures included data from between 39 and 53 species . MCMC chains were run for 1 , 000 , 000 time steps , and the first 200 , 000 time steps were discarded as burn in . Flat priors were used for all variables being predicted . To assess whether the post-burn in results were drawn from a stable distribution , we used the ‘heidel . diag’ function in the R package coda ( Plummer et al . , 2006 ) . When post-burn-in results were not drawn from a stable distribution , we discarded an additional portion of the chain ( as indicated by ‘heidel-diag’ ) so that only results drawn from a stable distribution remained . We ensured that the effective sample sizes for the PGLS model parameters ( slope , intercept , most frequently selected phylogenetic scaling parameter ) were greater than 1000 using the ‘effectiveSize’ function in coda ( Plummer et al . , 2006 ) . Details of the MCMC diagnostics are given in supplementary materials S6 , along with detailed results concerning the posterior predicted distribution and phylogenetic scaling parameters for each species in each analysis . We investigated the evolutionary trajectory of brain-body scaling in hominins relative to other primates . We calculated the difference between observed ECV and the mean BayesModelS prediction for brain size ( generated in the first described BayesModelS analysis in which data for all hominin species was excluded while generating predictions ) for each of the hominin species . This difference , which we call ‘brain size deviation’ represents the magnitude and direction of the deviation in brain size from what would be expected under primate brain-body scaling rules . We fit four PGLS model to hominin brain size deviation to examine how brain size deviation covaried with the phylogenetic distance from the hominin-Pan split: First , we fit a ‘Brownian’ model of brain size deviation with no predictor . We fixed λ at one in this and all subsequent models . Next , we fit a ‘directional’ model of brain size deviation predicted by phylogenetic distance from the hominin-Pan split , expecting to find a positive relationship between these variables if brain volume relative to body size has increased since the split of hominins and Pan . To determine whether evolutionary rates in brain size deviation have accelerated over time , we fit an ‘acceleration’ model that included the phylogenetic scaling parameter δ ( Pagel , 2002; Pagel , 1999 ) . Values of δ greater than one are consistent with accelerating evolution , but not necessarily directional evolution . Finally , we fit a ‘directional acceleration’ model in which we fit the parameter δ and used phylogenetic distance from the hominin-Pan split as a predictor of brain size deviation . In this model , a positive relationship between brain size deviation and phylogenetic distance , along with a value of δ greater than 1 , would indicate that brain volume relative to body size has increased at an accelerating rate since the divergence of hominins from Pan . We compared these models using AICc . Analyses were conducted in the R package caper ( Orme , 2013 ) . In the bayou analysis of ECV predicted by body mass using the hominin phylogeny , the Brownian model was favored over the weighted and unweighted predictor OU models with Bayes factors greater than 22 . When we repeated this analysis using the alternate hominin phylogeny , we found that the un-weighted predictor OU model was favored over the weighted predictor OU model and the Brownian model with Bayes factors greater than 42 , despite displaying poor convergence in terms of α and σ2 . However , both chains inferred a similar set of shifts , indicating that this is likely an issue related to parameter identifiability rather than to shift identifiability . In this model , progressive shifts towards larger ECV relative to body mass were detected within the hominin clade along the human lineage ( Figure 1A , B ) . Shifts towards larger relative brain size were also detected on the terminal branch leading to D . madagascariensis and the internal branches leading to the Lemuridae and Cebinae , clades , and shifts towards smaller relative brain size were detected on the branch leading to the Alouatta clade , the branch leading to the clade containing the Aotidae and Callitrichidae families , and the branch leading to the Colobinae sub-family ( Figure 1—figure supplement 1 ) . The rejected weighted predictor OU model , as well as both OU models that were rejected in the bayou analysis using the hominin phylogeny , detected a very similar set of shifts that included shifts towards progressively larger ECV relative to body mass along the human lineage ( Source data 1 ) . Because the Brownian model was favored in the bayou analysis using the hominin phylogeny , we proceeded with BayesModels analyses using both the hominin and alternate hominin phylogenies . In the BayesModelS analysis predicting ECV based on body mass while excluding all hominin data , the observed values for H . sapiens and H . neanderthalensis exceeded the mean values predicted by BayesModelS by 7 . 63 and 6 . 96 standard deviations respectively ( Figure 2C ) . All hominin species were strongly supported positive outliers , with more than 99 . 9% of predictions falling below the observed values for ECV . The mean ECV prediction for a primate with the body mass of H . sapiens was 438 mL . Remarkably , the observed value for humans is 1478 mL , which is 238% greater than the mean of the predicted posterior distribution . A similar result was found for H . neanderthalensis; the observed ECV for this species exceeded the mean predicted value for a primate of their body mass by 952 mL , or 201% . Humans exceeded their predicted ECV by the greatest percentage , but all hominins exceeded predictions by at least 51% ( Figure 2C , Table 4 ) . We obtained similar results using the alternate hominin phylogeny ( Figure 2—figure supplement 1 , Table 5 ) . When we iteratively predicted ECV based on body mass and phylogeny for each species in the data set ( no hominins besides H . sapiens and H . neanderthalensis were included in this analysis ) and while using all data to generate predictions . We again found that humans were strongly supported positive outliers ( Figure 4A ) . H . neanderthalensis was not identified as an outlier , perhaps because these analyses included all species except for the one being predicted , and thus inclusion of H . sapiens resulted in a wide posterior distribution when predicting ECV in H . neanderthalensis . Indeed , when we excluded H . sapiens in this analysis we found that H . neanderthalensis was identified as a strongly supported positive outlier ( Source data 1 ) . We also identified several other primate species as outliers ( see Table 6 and Source data 1 ) . In the bayou analysis of ECV with no predictor variable using the hominin phylogeny , the Brownian model was selected over the un-weighted predictor OU models ( in which the influence of the predictor was set to 0 ) with a Bayes factor >10 . No weighted predictor model was run , as it would have been equivalent to the unweighted model given that no predictor variable was incorporated . An equivalent result was found when we repeated the analysis using the alternate hominin phylogeny . We then proceeded with the BayesModelS analysis , iteratively testing the outlier status of each species in the data set . We used the tree block for this analysis , and as such H . sapiens and neanderthalensis were the only hominins included . We found that neither humans nor Neanderthals were detected as an outlier ( figure 4—figure supplement 1; Source data 1 ) , indicating that without correcting for body mass , the variance in ECV across primates is great enough to prevent humans’ brains from being detected as exceptionally large . We conducted PGLS analyses of brain size deviation conducted to characterize the evolution of exceptional brain size in hominins ( data shown in Figure 3 ) . The analyses revealed evidence for both accelerated evolution of brain size deviation and directional evolution towards larger brain size deviations , as indicated by the directional acceleration model ( AICc = −23 . 38 ) being favored over the acceleration ( AICc = −21 . 93 ) , directional ( AICc = −17 . 56 ) , and Brownian ( AICc = −14 . 58 ) evolution models . In this best model , there was evidence of directional evolution towards larger brain size relative to body size ( slope = 0 . 04 ) over time , and of accelerating evolution ( δ = 8 . 36 ) . These results suggest that the exceptionality of the human brain evolved recently . We found similar results when we repeated this analysis using the alternate hominin phylogeny ( Figure 3—figure supplement 1 ) . These analyses therefore support a model of accelerating evolution towards larger brain volume relative to body mass in Homo sapiens . In the bayou analysis of neocortex volume as predicted by body mass , the Brownian motion model was strongly favored over the weighted and unweighted predictor OU models , with Bayes factors > 18 . Humans were detected as strongly supported positive outliers for neocortex volume by BayesModelS when body mass was used as the predictor variable ( Figure 4B ) . In the bayou analysis of neocortex volume with ‘rest-of-brain’ as the predictor variable , the weighted predictor model was selected over the unweighted predictor and Brownian motion models with Bayes Factors > 9 . 2 . In the weighted predictor model , different scaling patterns were detected for strepsirrhines and haplorhines , with the optimum regression line for haplorhines falling above that of strepsirrhines . The only other detected transition in scaling occurred on the terminal branch leading to Nasalis larvatus , indicating a shift towards lower relative neocortex size ( Figure 5A , B ) . In the bayou analysis of cerebellar volume predicted by body mass , the Brownian motion model was favored over the weighted predictor and unweighted predictor OU models , with Bayes factors of 11 . 96 and 22 . 79 , respectively . BayesModelS identified humans as strongly supported positive outliers for cerebellum volume when body mass was used as the predictor variable ( Figure 4C ) . In the bayou analysis of cerebellum volume relative to the rest-of-brain , the comparison between the unweighted predictor model and the Brownian motion model gave a Bayes factor of 10 . 65 , while the comparison between the unweighted and weighted predictor models gave a Bayes factor of 0 . 20 . This indicates that the OU models clearly outperform the Brownian model , but that neither OU model performs significantly better than the other . Both OU models detected a shift on the branch leading to apes associated with an increase in optimum cerebellar volume relative to the ‘rest-of-brain’ volume ( Figure 5C , D ) . In the bayou analysis of medulla volume predicted by body mass , the Brownian motion model was selected over the two OU models with Bayes factors > 7 . 4 . BayesModelS identified humans as strongly supported positive outliers for medulla volume ( Figure 4D ) . No other species were identified as exceptional in this analysis . When medulla was predicted by the ‘rest-of-brain’ volume , the Brownian motion model was again selected over the OU models , with Bayes factors > 3 . 8 . Humans were identified as strongly supported negative outliers ( Figure 4E ) . In the bayou analyses of the rest-of-brain relative to body mass , the OU models were selected over the Brownian motion model , with Bayes factors > 13 . However , the comparison between the two OU models gave a Bayes factor of 0 . 20 , indicating that neither model is supported relative to the other . No shifts were detected in either model ( Figure 5E , F ) . Our phylogenetic analyses revealed that the human brain is 238% larger than the size expected for a primate of similar body mass and phylogenetic position . The exceptional size of the human brain was achieved through progressive scaling shifts towards larger size over several million years of hominin evolution , and the evolution towards increased brain size relative to expectations based on primate scaling patterns accelerated over time . These findings add an important dimension to previous observations of gradual phyletic increases in hominin brain size . Du et al . ( 2018 ) fit six evolutionary models to within- and between-lineage change in hominin brain sizes ( random walk , gradualism , stasis , punctuated equilibrium , stasis-random walk and stasis-gradualism ) , obtaining the best fit for a gradualism model . However , their non-phylogenetic analysis did not test explicitly for accelerating directional increase . Our findings extend the results obtained by Pagel ( 2002 ) on absolute cranial volume , as the pattern of accelerating evolution is found even after accounting for body size . The pattern of accelerating brain size increase documented here is consistent with hypotheses that postulate a co-evolutionary positive feedback process driving human brain evolution , such as feedback between brain size and culture or language ( Wills , 1993; Deacon , 1998 ) or between the brain sizes of conspecifics engaged in a socio-cognitive evolutionary arms race ( Dunbar , 1998; Miller , 2011 ) . While humans clearly have the largest relative brain size among extant primates , anatomically modern humans were closely matched by H . neanderthalensis . However , even when accounting for the close phylogenetic relationship between humans and H . neanderthalensis and the exceptionally large brain of the latter , the human brain is still much larger than expected: humans were identified as strongly supported outliers when their ECV ( relative to body mass ) was predicted by phenotypic data from all primates , including H . neanderthalensis . This pattern was not reciprocal , however; H . neanderthalensis was not significantly different from other primates when H . sapiens was included in the model . Significant variation exists between estimates of ECV and body mass made from different fossil specimens of the same hominin species ( Robson and Wood , 2008 ) . Thus , using single specimens to represent a species would not be a good statistical practice . We used a dataset in which almost all mean species values were calculated from multiple fossil specimens ( Table 1 ) . Unfortunately , we could not explicitly account for intraspecific variation in our analyses , as the multi-optima OU model fitting approach and the outlier test are unable to account for variation in both a trait and predictor variable . It would therefore be worthwhile to revisit our analyses as new phylogenetic comparative methods that can account for intraspecific variation become available . Additionally , data quality will likely improve over time . More hominin fossils will be discovered , increasing sample sizes for estimated ECV and body mass . The hominin phylogeny will also likely become better resolved and more complete . We accounted for some phylogenetic uncertainty by repeating our analyses with an alternate phylogeny . The use of different phylogenies influenced outcomes of some statistical tests , as the Brownian model favored when we used the hominin phylogeny and OU model was favored when we used the alternate hominin phylogeny . However , we found that all of the OU models we fit inferred the same pattern of evolution towards larger ECV along the human lineage . The results of our outlier tests and PGLS model fitting – which assume a Brownian mode of evolution – also detected this pattern on different phylogenies . Collectively , these results indicate that our findings are likely to be robust to variations in assumed evolutionary relationships , and potentially to assumptions about the mode of evolution . It is widely assumed that primate brain size evolution in general , and the large size of the human brain in particular , reflects expansion of the neocortex relative to other brain structures ( Kriegstein et al . , 2006; Rakic , 2009 ) . Our results contradict this assumption: human neocortical volume was exceptionally large relative to body mass , but not exceptional relative the volume of the rest of the brain . We documented only one shift in neocortex size relative to the rest of the brain during primate evolution: an increase at the origin of all haplorrhines . This shift may be related to the visual specializations of haplorrhines for high-acuity photic vision , mediated by extensive cortical visual areas that make up over 50% of the cortex in these species ( Drury et al . , 1996; Barton , 1998; Barton , 2007 ) . On branches postdating the split between haplorrhines and strepsirrhines , neocortex size is largely predictable from its scaling relationship to the rest of the brain , in line with the proposed importance of cortical-subcortical connectivity in primate brain evolution ( Whiting and Barton , 2003 ) . In contrast , we found that the cerebellum increased in size relative to the rest of the brain on the branch leading to apes . This finding is consistent with the results of recent studies implicating the cerebellum , and especially the lateral cerebellum , in brain expansion in apes and some other mammalian lineages ( Barton and Venditti , 2014; Smaers et al . , 2018; MacLeod et al . , 2003 ) . Our findings also reinforce the argument that subcortical structures should be given greater consideration in studies of mammalian brain evolution and cognition ( Barton , 2012; Miller and Clark , 2018 ) . Cerebellar specialization in apes may have been initiated by the demands on motor control and route-planning imposed by arboreal below-branch locomotion and/or by complex extractive foraging ( Barton and Venditti , 2014; Barton , 2012 ) . The fact that shifts in the relative size of neocortex and cerebellum occurred on different parts of the tree supports the theory of mosaic brain evolution ( Barton and Harvey , 2000 ) and suggests that no single adaptive hypothesis is likely to be capable of accounting for primate brain evolution; rather , different selection pressures , on different information-processing capacities , likely operated at different times on different lineages . Consistent with previous studies , we found that the medulla expanded in humans ( positive outlier status for medulla volume relative to body mass ) , but to a lesser degree than other structures ( negative outlier status for medulla volume relative to the rest of the brain ) . Relative to body mass , medulla volume has been shown to be much less variable across taxa than other brain structures , particularly compared to the neocortex and cerebellum . For example , unlike neocortex and cerebellum , medulla volume does not differ significantly between insectivores , strepsirrhines and haplorrhines ( Barton , 2000 ) . Accordingly , we found that after controlling for either body mass or brain size , the evolution of the medulla was not modulated by selection towards a stationary optimum in the primate clade . These results further support mosaic brain evolution ( Barton and Harvey , 2000 ) , and also suggest that scaling constraints related to connectivity with other brain regions ( Montgomery et al . , 2016 ) was less critical for the medulla than for the neocortex and cerebellum . Several non-human primate species exhibited exceptional brain evolution in one trait or another , but only humans showed exceptional brain evolution for multiple brain components . As predicted , we detected shifts towards larger brain size on the terminal branches leading to D . madagascariensis , and on the branch leading to the Cebinae clade . Large brain size in Daubentonia and Cebinae has been attributed to extractive foraging and tool use ( Kaufman , 2005; Melin et al . , 2014; Parker , 2015 ) . Although not one of our a priori expectations , we also documented shifts towards smaller brain size on branches leading to several clades , including Alouatta . We also found that two Gorilla species exhibit a smaller brain or neocortex size relative to body mass than expected . Given the extremely large body mass of Gorilla species , these unique traits may be the byproduct of a body mass increase rather than a reduction in brain size . Also unexpectedly , two Pan troglodytes sub-species were found to have exceptionally large and small ECV relative to body mass respectively . However , because more closely related species are weighed more heavily when BayesModelS generates distributions of predicted trait values , sister taxa deviating from expectations in opposite directions could result in both taxa being identified as outliers , even if they both conform to patterns of brain-body scaling for other primates . If the trait distributions for each species overlap significantly , then accounting for intraspecific variation in future analyses could remedy this problem . The unexpected patterns that we observed amongst non-human primates raise several questions for further research . Given the well-established positive correlation between overall brain size and extended life history ( Isler and van Schaik , 2009; Sol , 2009; González-Lagos et al . , 2010 ) , what are the life history implications of mosaic shifts in the sizes of different structures , and do these support any specific interpretations of the correlation between brain size and life histories ? One hypothesis , the developmental costs hypothesis , is that large brains simply take longer to grow and mature , leading to extended periods of maternal investment and slower maturation , with other life history correlates of brain size being byproducts of developmental prolongation . Support for this hypothesis is provided by the finding that , amongst mammals , the durations of gestation and lactation have independent effects on pre- and postnatal brain growth , and once these effects are accounted for , other life history correlates are non-significant ( Barton and Capellini , 2011 ) . Despite their generally correlated evolution ( MacLeod et al . , 2003 ) , we found shifts in the relative size of neocortex and cerebellum on different parts of the phylogenetic tree . Because these two structures have different developmental trajectories , the developmental costs hypothesis predicts different life history correlates; this prediction has now received support ( Powell et al . , 2019 ) . Further work is needed to establish exactly what developmental changes allowed for the neocortex and cerebellum to rest-of-brain scaling rules to change at the origin of haplorrhines and hominoids , respectively . Another area of interest concerns the cases we found of brain or brain component size reduction . Montgomery et al . ( 2010 ) found that brain size reductions were rare during primate evolution , and that there was a general trend for brain size to increase across multiple branches of the phylogeny . This raises questions for future work concerning the causes , developmental mechanisms and functional implications of specific types of size reduction , such as those that we uncovered in brain size relative to body size in Alouatta and other clades , and in neocortex size relative to the rest of the brain in N . larvatus . Finally , a key question that has attracted considerable attention concerns the ecological and social drivers of brain size and structure across large-scale evolutionary radiations . It has become increasingly apparent that correlations between overall brain size and behavioral ecology needed to be treated with caution ( Powell et al . , 2017; Healy and Rowe , 2007; Wartel et al . , 2018 ) . However , as suggested by the hypothesis of mosaic brain evolution , correlations between ecology and individual , less functionally heterogenous brain components may be more reliable and robust ( Barton and Venditti , 2014; Barton , 2012; Barton , 2007; Whiting and Barton , 2003; Montgomery et al . , 2016; Barton et al . , 1995 ) . Our analyses focused on gross subdivisions within the brain , and we suggest that further insights could be obtained by applying the phylogenetic methods used in this paper to more fine-grained neuro-anatomical data , using this approach to tease apart the contributions of correlated and mosaic change among brain components ( Melin et al . , 2014 ) and by incorporating ecological , behavioral , and developmental predictor variables that may account for additional variation in the traits of interest . In conclusion , we provided robust evidence for directional and accelerating selection towards larger brain size over the course of human evolution , resulting in the human brain being exceptionally large for a primate of similar body mass . We also found that the sizes of human brain components – including the neocortex , cerebellum , and the rest of the brain – are not larger or smaller than expected relative to the size of the rest of the brain , but all are larger than expected for a primate of similar body mass . These results suggest that relative neocortical expansion is not a hallmark of our species . The diversity of evolutionary patterns for various brain components that we observed within primates suggests that no single factor fully explains primate brain evolution; instead , comparative research should investigate how different selection pressures influenced the evolution of different neuroanatomical components at different times on different parts of the phylogenetic tree . Additionally , future work should seek to analyze the evolution of other brain traits , including neuronal composition , using similar phylogenetic comparative methods that account for the non-independence of data from related species .	Humans have much larger brains than other primates , but it is not clear exactly when and how this difference emerged during evolution . Some scientists believe that the expansion of a part of the brain called the neocortex – which handles sight , hearing , conscious decision-making and language – drove the increase in the size of the human brain . Newer studies have challenged that idea . One way to learn more about how humans evolved bigger brains is to compare the size of the brain , and specific parts of the brain , between humans and our closest relatives: non-human primates . To make accurate comparisons , scientists must account for many factors . Closely related primates may have more similar traits because they more recently shared a common ancestor . This means the evolutionary relationships between species need to be considered . Larger animals also tend to have larger brains so it is important to consider body size , too . Now , Miller at al . show that the human brain is much larger than expected even after accounting for these factors , and that increases in brain size accelerated over the course of early human evolution . In the analyses , the brain and skull sizes of different living primate species , like chimpanzees and gorillas , and fossils of extinct primates , including Neanderthals , were compared using mathematical models . These findings suggest that larger brains provided fitness advantages that led to large brain sizes in modern humans and Neanderthals . These increases in brain size were not driven by disproportionate growth in the neocortex alone , but rather by increases in the size of many parts of the brain . Increases in the relative size of the cerebellum , which is essential for balance and movement , were also important .	[ "Abstract", "Introduction", "Materials", "and", "methods", "Results", "Discussion" ]	[ "evolutionary", "biology", "neuroscience" ]	2019	Quantitative uniqueness of human brain evolution revealed through phylogenetic comparative analysis
The cerebellum plays a crucial role in the regulation of locomotion , but how movement is represented at the synaptic level is not known . Here , we use in vivo patch-clamp recordings to show that locomotion can be directly read out from mossy fiber synaptic input and spike output in single granule cells . The increase in granule cell spiking during locomotion is enhanced by glutamate spillover currents recruited during movement . Surprisingly , the entire step sequence can be predicted from input EPSCs and output spikes of a single granule cell , suggesting that a robust gait code is present already at the cerebellar input layer and transmitted via the granule cell pathway to downstream Purkinje cells . Thus , synaptic input delivers remarkably rich information to single neurons during locomotion . In order to ensure generation of precise and reliable movements , information about movement parameters must be represented in neural circuits of the mammalian brain with high fidelity . Firing patterns directly related to specific movement parameters have been reported in single-unit recordings across several brain areas involved in movement representation and generation ( Armstrong , 1988; Beloozerova et al . , 2003 ) . Crucial to generating such accurate representations is delivery of synaptic input patterns containing rich information about the animal's movement . However , we currently lack information about the input patterns received during movement for any neuron in the mammalian brain , and thus the input–output transformations performed in these neurons during movement are only poorly understood . The cerebellum is thought to play a key role in precise limb coordination during voluntary movements ( Flourens , 1824; Vercher and Gauthier , 1988; Muller and Dichgans , 1994; Bastian et al . , 1996; Holmes , 1922 ) . Purkinje cells , the output cells of the cerebellar cortex , exhibit firing linked to the phase of the step cycle during locomotion ( Orlovsky , 1972; Armstrong and Edgley , 1984; Edgley and Lidierth , 1988 ) . How this information is represented in and transmitted by upstream neurons in the circuit , in particular in cerebellar granule cells—which form the input layer of the cerebellar cortex—remains unknown . Here we have taken advantage of the electrical compactness of granule cells , and their small number of excitatory inputs—4 on average ( Eccles et al . , 1967; Palkovits et al . , 1971; Jakab and Hamori , 1988 ) —which allows for individual synaptic inputs to be resolved in vivo ( Chadderton et al . , 2004; Jörntell and Ekerot , 2006; Rancz et al . , 2007; Arenz et al . , 2008; Chadderton et al . , 2014 ) . Moreover , since granule cell activity plays a key role in regulating locomotion , including the coordination of individual limbs ( Vinueza Veloz et al . , 2014 ) as well as in in motor learning ( Galliano et al . , 2013 ) , they represent particularly attractive targets for an electrophysiological dissection of their input–output relationships during locomotion . By recording the activity of mossy fiber boutons , EPSCs in granule cells , and granule cell output while mice are moving on a treadmill , we can thus reconstruct single cell integration of synaptic inputs in awake animals during locomotion , and identify the cellular representation of movement parameters in a defined site in the circuit . In vivo whole-cell recordings were made from mossy fiber boutons and cerebellar granule cells in lobule V of the cerebellar vermis . All recordings were performed in awake mice head-fixed on a spherical treadmill ( Figure 1A ) . Granule cells and mossy fiber boutons were identified on the basis of their distinctive electrophysiological signatures ( Chadderton et al . , 2004; Jörntell and Ekerot , 2006; Rancz et al . , 2007; Arenz et al . , 2008 ) . To study the link between voluntary movement and granule cell input and output we extracted a motion index from captured video frames ( Figure 1B , see ‘Materials and methods’ ) and aligned this to the simultaneously acquired electrophysiological data ( example recordings Figure 1C–E ) . The motion index was used to categorize the electrophysiological data recorded during quiet wakefulness ( defined as periods where the motion index remained below a threshold rate of change of 0 . 025 a . u . per frame , for at least 30 consecutive frames , see ‘Materials and methods’ ) and voluntary movement . 10 . 7554/eLife . 07290 . 003Figure 1 . Whole-cell recordings from granule cells and mossy fibers during locomotion . ( A ) Schematic of recording configuration . ( B ) Calculation of motion index . Top panel: a single frame from a video of a mouse walking on the treadmill . Bottom panel: pixel intensity variation between the frame shown in the above panel and the previous frame are highlighted in red . The pixel variation between each frame was quantified for each video to give a continuous signal relating to average motion of the mouse ( motion index calculated as described in the ‘Materials and methods’ and normalized to the maximum value in the video ) . ( C–E ) Example whole-cell recordings ( black ) from a presynaptic mossy fiber terminal ( C ) , a granule cell recorded in voltage-clamp mode ( D ) and a granule cell recorded in current clamp mode ( E ) , together with the corresponding motion index ( red ) . ( F–H ) Section of each example trace shown in C–E at a higher timescale show spontaneous input recorded during a quiet period ( left panels , orange frames in ( C–E ) indicate the location within the trace ) and typical bursts of activity during locomotion ( right panels , blue frames in ( C–E ) indicate the location within the trace ) . Summary data comparing the average instantaneous frequencies of mossy fiber spikes ( I , n = 4 in 4 mice ) , granule cell EPSCs ( J , n = 9 in 6 mice ) and granule cell spikes ( K , n = 6 , in 4 mice ) . Mean group averages are indicated with black open and closed circles , error bars indicate standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 07290 . 003 During quiet wakefulness , granule cells exhibited a resting membrane potential of −67 . 0 ± 8 . 9 mV and an input resistance of 420 ± 210 MΩ ( n = 47 , in 24 mice ) . Recordings from mossy fiber boutons ( Rancz et al . , 2007 ) demonstrated a spontaneous average firing frequency of 21 ± 20 Hz ( n = 6 , in 5 mice ) . The average frequency of spontaneous excitatory synaptic currents in granule cells was 60 ± 35 Hz ( n = 32 , in 16 mice ) . Despite this relatively high input frequency , granule cell output spiking frequency was low: the average firing rate was 0 . 12 ± 0 . 31 Hz ( n = 27 , in 13 mice ) , and 62% of granule cells showed no spiking during periods in which the mouse was sitting quietly . In some cases we were able to successfully maintain recordings of activity during periods of voluntary locomotion ( ranging from 17 s to 65 s , mean duration: 33 . 2 ± 12 . 3 s ) . This allowed to us to compare activity within single cells during quiet periods and during locomotion . Locomotion was accompanied by an increase in mossy fiber input to the granule cells , observed as an increase in firing frequency in 3 out of 4 recorded mossy fibers , although overall this was not statistically significant across the four recordings ( Figure 1F , I; average , quiet: 11 . 7 ± 11 . 6 Hz; locomotion: 46 . 4 ± 43 . 3 Hz , n = 4 in 4 mice , p = 0 . 14 , paired t-test ) and increased EPSC frequency in all recorded granule cells ( Figure 1G , J; quiet: 77 . 5 ± 37 . 6 Hz; locomotion: 144 . 0 ± 53 . 5 Hz , n = 9 in 6 mice , p < 0 . 0001 , paired t-test ) . The firing rate of the granule cells also increased dramatically during locomotion ( Figure 1H , K , average; quiet: 0 . 09 ± 0 . 2 Hz; locomotion: 5 . 3 ± 5 Hz , n = 7 in 4 mice , p < 0 . 05 , paired t-test ) . During locomotion , granule cell action potentials occurred in sparse high-frequency bursts with high instantaneous firing frequencies ( Figure 1H right panel , average instantaneous frequency: 106 ± 65 Hz; 4 out of 7 cells fired in bursts , defined as groupings of 4 or more spikes with ISIs less than 50 ms ) . These bursts consisted of an average of 11 . 9 ± 2 . 1 spikes with ISI 10 . 7 ± 3 ms , occurred with an average inter-burst interval of 1 . 88 ± 0 . 99 s , and were associated with a high coefficient of variation of the inter-spike interval ( CV = 2 . 2 ± 1 . 1 ) . The CV of the inter-spike interval was significantly higher than the CV of the EPSC inter-event interval during movement ( 1 . 1 ± 0 . 3 , n = 9 , p = 0 . 039 , unpaired t-test ) . We used the bootstrap method ( Roy , 1993 ) to calculate 95% confidence intervals on the GC EPSC/MFT spike ratio , which were 3 . 5–17 . 7 at rest , and 1 . 7 to 8 . 4 during motion , consistent with the known 4:1 anatomical convergence ( Eccles et al . , 1967; Palkovits et al . , 1971; Jakab and Hamori , 1988 ) . The sustained high-frequency barrages of EPSCs during locomotion ( Figure 1G , right panel ) were associated with large , slow inward currents , comparable to the slow spillover currents that have been observed at this synapse in vitro ( DiGregorio et al . , 2002; Xu-Friedman and Regehr , 2003; Nielsen et al . , 2004 ) . NMDA receptor activation could in theory underlie such slow currents; however the NMDA current is negligible in cerebellar granule cells at our −70 mV holding potential , due to the voltage-dependent Mg2+ block of the NMDA receptor channel ( Nieus et al . , 2006; Figure 2 ) . We confirmed that the NMDA current is negligible using a GC model ( Diwakar et al . , 2009 ) , adapted to match the NMDA/AMPA ratio of 0 . 2 found in adult mice ( Cathala et al . , 2003 ) . At −70 mV , NMDA contributes at most 0 . 4% of the total excitatory synaptic current in this model . Furthermore , it has previously been shown ( Cathala et al . , 2003 ) that in the age and species of mice from which our data is derived , the vast majority of the NMDA current is transmitted by extrasynaptic receptors . We therefore concluded that spillover transmission must be responsible for most of the slow current we measured . 10 . 7554/eLife . 07290 . 006Figure 2 . Glutamate spillover enhances transmission during locomotion . ( A ) Voltage clamp traces recorded at a holding potential of −70 mV from two granule cells . Red line indicates the spillover current obtained by subtracting the fitted single exponential decays of the phasic EPSCs . ( B ) Relationship between the EPSC frequency and the relative proportion of excitatory current carried by the phasic EPSC component ( average for all recorded cells; n = 9 in 6 mice ) . ( C ) Average cross-correlation of instantaneous EPSC frequency with spillover current ( n = 9 , dotted lines indicate standard deviation ) . Dashed line indicates zero lag . ( D ) Average EPSC burst triggered spillover current for all cells ( dotted line indicates standard deviation ) . ( E , F ) Graphs showing increasing EPSC frequency and spillover conductance occurring with motion in an example granule cell ( E ) and as an average for all cells ( F , n = 9 ) . ( G ) Overlaid voltage traces showing spike bursts ( red traces indicate the average ) . High-frequency bursting appears to be associated with greater subthreshold depolarization ( left panel ) . ( H ) Graph showing the relationship between spike burst frequency and subthreshold depolarization for individual bursts across all cells ( n = 37 bursts from 4 cells , r = 0 . 45 , p = 0 . 0055 ) . ( I ) A representative current trace showing the separation of phasic and spillover EPSCs . A granule cell model was then used to estimate the effect of spillover conductance on granule cell spike output . ( J ) Summary data showing the simulated granule cell spike frequency resulting from spillover , phasic and combined conductances , as well as combined conductances including an NMDA conductance ( n = 9; see ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07290 . 00610 . 7554/eLife . 07290 . 007Figure 2—figure supplement 1 . Synaptic charge transfer with and without spillover . The synaptic charge transfer over 100 ms as a function of EPSC frequency with and without spillover . DOI: http://dx . doi . org/10 . 7554/eLife . 07290 . 007 To isolate the putative spillover components , we fit the fast EPSCs and separated them from the underlying slow current recorded at −70 mV ( Figure 2A ) . The amplitude of the slow current was correlated with the EPSC frequency ( Figure 2E shows this for an individual cell; Figure 2F for the population of n = 9 cells ) , such that the proportion of the total current carried by the fast EPSCs vs the slow current diminished with EPSC rate ( Figure 2B ) , similar to what has been shown in vitro in response to trains of synaptic stimulation ( Sargent et al . , 2005 ) . Since spillover results from synaptic accumulation of neurotransmitter , a delay would be expected between the peak of the EPSCs and that of the putative spillover current . Indeed , a cross-correlation between the instantaneous frequency of fast EPSCs and the slow currents showed a peak with a lag of 40 ms ( Figure 2C ) , with the slow current lagging the fast events . Similarly , aligning the slow current on a burst of at least 5 EPSCs at 200 Hz or above exhibited a negative peak at positive delay ( 31 ms latency; 729 bursts from the 9 cells , Figure 2D ) . We further plotted the synaptic charge transfer over 100 ms as a function of EPSC rate with and without spillover ( Figure 2—figure supplement 1 ) . The spillover dramatically affects the slope of this curve . What is the functional role of the slow putative spillover current ? In current clamp recordings from granule cells , we observed that spiking occurs in bursts ( Figure 2G; 37 bursts recorded in 4 out of 7 cells ) , which were frequently accompanied by a slow depolarization ( Figure 2G , left panel ) . As predicted , there was a significant correlation ( r = 0 . 45 , p = 5 . 5 × 10−3; Figure 2H ) between the amplitude of the slow depolarization and the spike frequency in the burst . We hypothesized that the spillover could contribute to such spiking episodes . To study this , we used a spiking compartmental model of a granule cell ( Diwakar et al . , 2009 ) , with an excitatory synaptic conductance based on the fast EPSCs and slow spillover currents recorded in voltage clamp . The depolarization generated by the slow spillover current was able to facilitate bursts that were not observed with fast EPSCs alone ( Figure 2I ) . Consistent with this , the combination of spillover and fast currents produced significantly more spikes than either the spillover or fast current in isolation , indicating a synergistic non-linear interaction between the two currents ( spiking rate for spillover alone: 2 . 69 ± 3 . 78 Hz , for fast EPSCs: 1 . 01 ± 0 . 94 Hz , for both: 9 . 27 ± 6 . 36 Hz , Figure 2J , p = 3 . 39 × 10−6; one way ANOVA ) . Next , we assessed the possible contribution of NMDA receptor currents to spiking under these conditions . We performed simulations using an NMDA to AMPA ratio of 0 . 2 , which is physiological at this synapse ( Cathala et al . , 2003 ) . Under these conditions a spiking rate of 9 . 74 ± 6 . 51 Hz is obtained , which is only slightly higher than without the NMDA channels ( rate 9 . 37 ± 6 . 35 Hz , p = 2 . 70 × 10−8; paired t-test ) . We conclude that the slow putative spillover current contributes significantly to spike generation in awake , locomoting animals . We next wanted to study the relationship between activity in the granule cell layer and movement parameters . The mice were free to initiate locomotion at will , and were mostly quietly resting on the treadmill . We examined how EPSC activity changes at the start of these periods of locomotion , and found a sustained increase in rate from 100 ± 8 Hz to 145 ± 12 Hz ( p = 0 . 0036 , paired t-test , n = 23 periods from 9 cells comparing a 500 ms window 1 s before and immediately after locomotion onset; Figure 3—figure supplement 1 ) . Similarly , at the termination of locomotion , there was a sustained decrease in EPSC rate from 139 ± 10 Hz to 106 ± 7 Hz ( p = 0 . 0118 , paired t-test , n = 23 periods from 9 cells comparing a 500 ms window immediately before and 1 s after locomotion onset ) . Note however that even in the absence of locomotion , EPSCs occurred at high frequencies . Next , we examined the precise relationship between activity in mossy fibers and granule cells during locomotion . Positive correlations were found between the motion index and MFB spiking , GC spiking , GC EPSC rate and spillover current in most cells at low temporal resolution ( 1 . 5 s; Figure 3C ) . At higher temporal resolution significant peaks in the normalized sliding cross-correlations were observed between motion and mossy fiber bouton spiking , granule cell EPSCs and granule cell spiking ( Figure 3A , B ) . For spillover , only a minority of the cells ( 2 out of 9 ) showed a significant peak cross-correlation at high temporal resolution , consistent with it being a slow current . Together , these results indicate that overall motion during locomotion is represented in both granule cell excitatory input and spike output patterns . 10 . 7554/eLife . 07290 . 004Figure 3 . Relationship between motion and activity parameters of single granule cells . ( A ) Normalized cross-correlation between motion index and event frequency for a MFB ( green ) , EPSCs ( red ) , spillover ( orange ) , and GC spikes ( blue ) . ( B , C ) Peak normalized cross-correlation coefficients for each cell using fine time bins ( 33 . 3 ms; asterisks denote points with p < 0 . 05 ) and raw correlation coefficients using coarse time bins ( 1 . 5 s; asterisks in C denote correlation coefficients significant at the p = 0 . 05 level ) . ( D ) Spike triggered average of motion map overlaid on video of mouse . ( E , F ) PCA analysis: each video was decomposed into 50 8-frame principal components ( PCs , top row , 4 frames shown for PC 1 , 4 and 50 ) . The activity of each cell was then cross-correlated over time with the PC coefficient ( F ) . Each line represents the maximum cross-correlation of a single cell with the different PCs for the corresponding video ( green = MFBs , red = GC-EPSC , orange = GC-spillover , blue = GC-spikes ) . The gray traces represent the average cross-correlation with bootstrapped data and therefore indicate the noise level . Note that the PCs were ranked according to their cross-correlation . DOI: http://dx . doi . org/10 . 7554/eLife . 07290 . 00410 . 7554/eLife . 07290 . 005Figure 3—figure supplement 1 . Modulation of responses by locomotion onset and termination . ( A ) EPSC rates from 23 locomotion periods from 9 cells aligned to the start of locomotion ( at t = 0 ms , defined as the start of the first swing phase ) . ( B ) EPSC rates from 22 locomotion periods from 9 cells aligned to the end of locomotion ( at t = 0 ms , defined as the end of the last swing phase ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07290 . 005 To correlate specific movements with granule cell input and output , we performed activity-triggered averaging of motion maps in video images of locomoting mice ( Figure 3D ) . Anatomically coherent regions of the mouse showed strong signals ( 19 out of 20 of these maps showed significant signals in the regions corresponding to one or more of the limbs , 9 out of 20 for the head , and 11 out of 20 for the body of the mouse ) , suggesting that more specific features of the motion could be read out by granule cell activity . We therefore decomposed the videos into principal components ( PCs ) , which corresponded to motions of the mice , most often showing limb motion and/or locomotion at different speeds ( Figure 3E ) . For all the 20 cells recorded , we used a semi-automated labelling algorithm ( see ‘Materials and methods’ ) to classify the 50 highest eigenvalues and found that 738 out of 1000 showed periodic limb motion , 434 showed periodic head motion , and 370 showed periodic motion of the rest of the body . We performed a sliding cross-correlation of the electrophysiological activity with the projection of the video onto the 50 PCs with the largest eigenvalues . For all recording modalities , peak cross-correlation higher than 0 . 3 and up to 0 . 55 could be found ( Figure 3F ) ; these correlations dropped below 0 . 2 after the 10th component . Note that the correlations were higher than with surrogate data ( grey traces , Figure 3F ) , which indicates that the neural activity correlated with specific PCA components more than would be expected by chance . Next , we examined the relationship of activity parameters to the step cycle of the mouse , focusing on the two forelimbs . Animals initiated periods of locomotion spontaneously , and would often be still for several seconds between these periods . Electrophysiologically recorded parameters in mossy fiber boutons and granule cells were strikingly modulated by the step cycle ( Figure 4A–C ) , in either or both of the forelimbs . Specifically , statistically significant step cycle modulations were observed for mossy fiber bouton spiking ( 3 out of 4 cells for the right limb , 1 out of 4 cells for the left limb ) , granule cell EPSCs ( 5 out of 9 for the right limb , 4 out of 9 cells for the left limb ) , and granule cell spiking ( 0 out of 7 for the right limb , 2 out of 7 cells for the left limb ) ; while spillover was not significantly correlated with step cycle . In Figure 4—figure supplement 1 , we show why it is possible that despite high cross-correlations between spillover and EPSCs , significant step cycle modulations are not seen: a temporally filtered EPSC trace ( using a biexponential kernel with rise time = 50 ms and decay = 100 ms ) exhibits a cross-correlation with the original trace which is similar to spillover ( Figure 4—figure supplement 1B ) , however its step cycle modulations are significantly lower ( from 0 . 30 ± 0 . 03 to 0 . 13 ± 0 . 2 , p = 4 . 65 × 10−4 , paired t-test; n = 18—that is , 2 limbs for n = 9 cells ) , due to the low-pass properties of the filter . 10 . 7554/eLife . 07290 . 008Figure 4 . Decoding activity in a single granule cell can predict the step cycle . ( A ) Example step-triggered averages of activity for three different cells ( EPSC and spillover examples are from the same granule cell ) . Gray-shaded area indicates the swing phase of the step cycle . ( B , C ) Step cycle modulation index for each forelimb across all cells . ( D ) Two state Hidden Markov Model ( HMM ) used to reconstruct the step cycle from electrophysiological recordings . ( E ) Example of successful step reconstructions for an MFB ( red ) , GC EPSC recording ( green ) and GC spikes ( blue ) . The top traces represent the electrophysiological event rate in Hz , the middle traces the step transition ( with the high state being the swing phase and the low state being the stance ) predicted by the HMM , and the lower trace being the actual step of the best modulated limb . ( F ) Prediction quality for all cells . DOI: http://dx . doi . org/10 . 7554/eLife . 07290 . 00810 . 7554/eLife . 07290 . 009Figure 4—figure supplement 1 . Spillover acts like a temporal filter . ( A ) We convolved the ESPCs rates with a biexponential trace to give a filtered trace . ( B ) The cross-correlation between the filtered traced and EPSCs resembles the cross-correlation between spillover and EPSCs ( Figure 2C ) . ( C ) The step cycle modulation of the EPSCs is drastically reduced by the filtering procedure ( from 0 . 30 ± 0 . 03 to 0 . 13 ± 0 . 2; p = 4 . 65 × 10−4; n = 18 , 2 limbs for n = 9 cells ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07290 . 00910 . 7554/eLife . 07290 . 010Figure 4—figure supplement 2 . Tuning of responses to the step cycle . ( A ) Polar plot representing the step cycle modulation for each of the forelimb . Each of these plots is a representation of the step cycle for one recording ( respectively an MFT , EPSC and GC spike recording ) , 0° being the start of the swing phase , the green line is end of the swing phase . Each trace ( blue: left forelimb , red: right forelimb ) represents as the radius in the plot the modulation of activity ( see ‘Materials and methods’ ) for that phase of the step cycle ( for the MFT plot , the scale corresponds to 2 . 4 Hz for the blue trace and 2 . 7 Hz for the red trace , for the EPSC plot the scale is 120 Hz for the blue trace and 65 Hz for the red trace , and for the GC spike trace , the scale is 13 . 6 Hz for the blue trace and 9 . 4 Hz for the red trace ) . The arrows show the phase of maximal modulation . Note that for these recordings , the maximal modulations are at roughly 90° . ( B ) The phases at which maximal modulation occurs can vary widely across cells for both the left and right forelimb ( left and middle trace ) . The phase difference between these modulations for each cell is shown in the right panel . ( C ) The magnitude of modulation is highly correlated for the right and left limb across recording modalities ( green: MFTS , red: EPSCs , blue: GC spikes ) . ( D ) The direction of maximum modulation is plotted here for all cells as a z-value ( i . e . , negative values indicate a decrease in activity and positive values an increase ) for both the left and right forelimb . DOI: http://dx . doi . org/10 . 7554/eLife . 07290 . 010 To examine in greater detail the tuning of events to the step cycle , we plotted the step cycle modulation of the event rates on a polar plot for both forelimbs . Mapping the start of the step cycle to 0° of the circle , we observed that cells could show maximal modulation in a wide range of phases ( Figure 4—figure supplement 2; range from 0 to 351° ) . The phase difference between the maximal modulation for one forelimb vs the other was clustered around 90°: 102 ± 30° for MFTs , 104 ± 19 . 2 for EPSCs , 74 ± 15 for GC spikes . Not surprisingly , the magnitude of modulation was highly correlated between the two forelimbs ( n = 20 cells , R = 0 . 9 , p = 3 . 48 × 10−8 ) . However as the MFT example in Figure 4A makes clear , not all modulations of activity were in the same direction for both forelimbs . In the right panel of Figure 4—figure supplement 2C , we have plotted the maximum modulations as z-scores for one limb vs the other , which shows across the population that modulations can happen in all four directions ( up-up , up-down , down-up , down-down ) . If these step-related modulations are sufficiently robust to provide a code for downstream Purkinje cells , it should be possible to reconstruct the step sequence from the recorded activity . We therefore implemented a two-state Hidden Markov Model ( HMM; Rabiner , 1989 ) , with one state representing the stance and the other state representing the swing phase of the step cycle ( Figure 4D ) , and trained this model on the electrophysiological data . Note that the algorithm used to constrain the model is unsupervised in the sense that the learning procedure is never shown the actual step sequence . A predicted step sequence was generated for each trace by producing the most likely state sequence for the HMM trained on the trace . This prediction was then compared to the actual step sequence . For mossy fiber spikes , granule cell spikes , and granule cell EPSCs we found that in some cells ( 2 out of 4 , 2 out of 5 and 2 out of 9 cells respectively , Figure 4E ) accurate step sequence reconstructions ( i . e . , with p-values < 0 . 05 ) could be obtained ( Figure 4F ) . In contrast , step sequences could not be reconstructed from spillover currents alone , indicating that while these currents are permissive for spiking during movement , they do not represent the details of individual movements . Together , these findings indicate that rich information about locomotion parameters is available to single granule cells in the cerebellar cortex , even at the level of synaptic input . Our experiments represent the first whole-cell recordings from granule cells in awake mice . Information about granule cell activity patterns in awake animals has been extremely difficult to obtain . Extracellular recordings from the granule cell layer have suffered from the fact that the exceptionally dense packing of the granule cells has made it difficult to unambiguously assign spikes as arising from single granule cells ( Hartmann and Bower , 2001; Gao et al . , 2012 ) . Although two-photon imaging approaches should in principle help to resolve this issue , the scattering nature of the densely packed granule cell layer , and the lack of reliable single-spike sensitivity of current genetically encoded calcium sensors combined with the difficulty of separating synaptic and action potential-linked calcium signals in granule cells has made it difficult to reliably estimate granule cell firing rates using two-photon imaging ( Ozden et al . , 2012 ) . Our patch-clamp recordings allowed us to unambiguously record from single granule cells , and demonstrate that granule cells in awake mice exhibit remarkably low firing rates when the animal is at rest , similar to those observed in mormyrid fish under paralysis using similar techniques ( Sawtell , 2010 ) . Moreover , the electrical compactness of the granule cells allows us to make voltage clamp recordings to probe synaptic currents driving spiking activity . We demonstrate that the low spontaneous firing rates observed at rest in granule cells , which are surprisingly similar to those in anaesthetized ( Chadderton et al . , 2004; Duguid et al . , 2012 ) and decerebrate ( Jörntell and Ekerot , 2006 ) preparations , persist despite far higher spontaneous EPSC rates than are present under anaesthesia . This indicates that the low firing rate of granule cells is a general property of granule cells and suggests that it is under tight control by intrinsic mechanisms and by synaptic inhibition ( Duguid et al . , 2012 ) , especially in the awake animal . Despite the low firing rates observed at rest , granule cells exhibited a dramatic increase in firing at the onset of locomotion , paralleling observations using two-photon calcium imaging in the granule cell layer ( Ozden et al . , 2012 ) . This increase in granule cell firing rate was widespread across the granule cell population , indicating that granule cells switch from a sparse to a dense mode of activation during locomotion . The high firing rates of granule cells during locomotion was organized in bursts , reminiscent of the high-frequency bursts observed in granule cells with some forms of sensory stimulation ( van Beugen et al . , 2013 ) , which in turn can be transmitted reliably to the Purkinje cell ( Valera et al . , 2012 ) . Our voltage clamp experiments reveal that the locomotion-evoked spiking is driven by a significant increase in excitatory synaptic input to the granule cells , which is paralleled by an increase in spiking in mossy fibers , as shown by direct recordings from mossy fiber boutons . Some of the increase in mossy fiber input may be provided by activity in spinocerebellar pathways , which exhibit locomotion-related activity ( Arshavsky et al . , 1972a , 1972b; Orsal et al . , 1988 ) . These results indicate that the sensorimotor computations in neuronal circuits upstream from the mossy fiber pathways provide sufficiently strong activation to overcome the high threshold for output spiking in the granule cells . Future experiments are required to examine how the interaction between mossy fiber input and Golgi cell inhibition ( D'Angelo and De Zeeuw , 2009 ) jointly determine the enhanced spiking output of granule cells during locomotion . Our experiments provide the first investigation of glutamate spillover in the awake behaving animal , and demonstrate that this feature of mossy fiber-granule cell transmission ( DiGregorio et al . , 2002; Nielsen et al . , 2004 ) plays a crucial role in driving movement-related output spikes from the granule cell layer . Previous investigation of glutamate spillover currents has been restricted to brain slices , both in hippocampus ( Kullmann and Asztely , 1998 ) and cerebellum ( DiGregorio et al . , 2002; Nielsen et al . , 2004 ) , and it has been unclear whether the stimulus conditions and levels of glutamate uptake present in vitro are representative of the physiological pattern of in vivo activation . We demonstrate that spillover currents make a substantial contribution to the synaptic charge during locomotion-evoked mossy fiber input . Furthermore , using a realistic compartmental model of the granule cell , we show that these spillover currents are essential for driving the observed output firing rates of the granule cell associated with locomotion . The biophysical properties of spillover currents appear to be ideally suited to help ensure appropriate granule cell firing patterns under resting conditions and during locomotion . Specifically , spillover currents are synergistically enhanced by the activation of multiple neighbouring synapses ( Carter and Regehr , 2000; Arnth-Jensen et al . , 2002; DiGregorio et al . , 2002 ) , as is likely during the high-frequency barrage of mossy fiber inputs activated during locomotion . Thus , the requirement of the spillover current for high spiking rates provides another mechanism to both ensure sparse granule cell firing at rest , and enhance firing during locomotion , increasing the signal-to-noise ratio of granule cell transmission of locomotion information . We show that granule cells exhibit very low spike rates in awake resting mice , despite high excitatory synaptic drive . Since granule cells represent the most abundant neuronal type in the brain , and spikes are energetically expensive ( Attwell and Laughlin , 2001; Carter and Bean , 2009 ) , the metabolic cost of sustained firing in this population would be considerable ( Howarth et al . , 2012 ) . Therefore the ability to maintain sparse firing in granule cells at rest could be an important mechanism for energy conservation in the mammalian brain . During locomotion , however , spike rates can increase dramatically . We demonstrate that movement during locomotion exhibits an unexpectedly strong representation in mossy fiber input received by granule cells , as well as in the spiking output of individual granule cells . Remarkably , in some neurons it is possible to reconstruct the entire step sequence during locomotion from both input patterns and spike output , indicating that gait information is present even at the level of single granule cells . This indicates that a rich amount of information about movement parameters is already represented by individual synapses at the input layer of the cerebellar cortex . Thus , selective sampling of granule cells exhibiting tuning to a given phase of the step cycle may underlie the step cycle modulation of simple spikes observed in Purkinje cells ( Orlovsky , 1972; Armstrong and Edgley , 1984; Edgley and Lidierth , 1988 ) . Our results suggest that the synaptic representation of locomotion , rather than being based on a sparse code ( Marr , 1969; Attwell and Laughlin , 2001; Howarth et al . , 2012 ) , relies on a population of tuned neurons , which shift from sparse activity to a dense gait code during movement . All experiments were carried out in accordance with UK Home Office regulations . Adult C57BL/6J mice ( P40–60 ) were anesthetized with isoflurane ( 3–5% for induction , 0 . 5–2% for surgery , in pure oxygen ) . Throughout general anaesthesia , rectal temperature was monitored and body temperature maintained constant using a homeothermic blanket ( FHC ) . Mice were placed in a stereotaxic frame , which allowed horizontal positioning of the head and implanted with a lightweight L-shaped metal head plate and recording chamber . The recording chamber was positioned directly above lobule V of the cerebellum . This region of the cerebellum was the most accessible , requiring the least amount of muscle removal , which was an important consideration for these awake experiments . Mice were allowed to recover from head implant surgery for a minimum of 48 hr . At least 3 hr prior to performing electrophysiological recordings , mice were re-anaesthetized with isoflurane ( as described above ) . A small ( ∼200–500 μm ) craniotomy was created through the skull directly above lobule V of the cerebellar vermis roughly 0 . 5–1 mm lateral of the midline and the dura removed . Agar and a silicone elastomer ( Kwik-Cast , World Precision Instruments Ltd , Sarasota , FL ) were applied to the craniotomy to seal it . When the silicone was fully set the mouse was removed from the head-holder and placed in a warm cage to recover from anaesthesia for ∼2 hr . Following recovery from anaesthesia mice were head-fixed ∼1 hr prior to starting the recording session . Mice were placed on a 20 cm diameter polystyrene ball that was secured to an air-table directly below and slightly in front of the headstage . The ball was mounted through its center on a horizontal axle resting on bearings . Mice were placed on the ball such that they could rest comfortably on its center and walk voluntarily . With this configuration mice habituated readily to head restraint , usually sitting quietly after 30–60 min . In vivo whole-cell voltage-clamp and current-clamp recordings were obtained from cerebellar granule cells and mossy fiber boutons as previously described ( Chadderton et al . , 2004; Rancz et al . , 2007; Duguid et al . , 2012 ) . Recordings used for analysis lasted from 100 s to 950 s ( mean average: 300 ± 186 s , n = 53 , in 26 mice ) . Pipettes had resistances of 6–7 MΩ and were filled with an internal solution containing ( in mM ) : K-Methanesulphonate , 133; KCl , 7; HEPES , 10; Mg-ATP , 2; Na2-ATP , 2; Na2-GTP , 0 . 5 . EGTA , 0 . 1; pH 7 . 2 . Granule cells were identified based on their characteristically small capacitance and depth from the pial surface ( 450–600 μm ) . Mossy fiber boutons were identified by their relatively high spontaneous spike rates , lack of synaptic input and characteristic spike waveform ( Rancz et al . , 2007 ) . Electrophysiological measurements were amplified using a Multiclamp 700B amplifier ( Axon Instruments , Molecular Devices , Sunnyvale , CA ) . Data was filtered at 4–10 kHz and acquired at 20 kHz using pCLAMP software in conjunction with a Digidata 1440A acquisition system ( Axon Instruments , Molecular Devices ) . Electrophysiological data was inspected for artifacts relating to movement ( large perturbations either side of the baseline ) . Recordings that demonstrated such artifacts were excluded from further analysis . Electrophysiological data were analyzed using custom-written macros in Igor Pro 6 . Synaptic currents and potentials were detected using an amplitude threshold algorithm where the threshold for event detection was set at two times the standard deviation of the baseline noise ( typically about 10 pA ) . Detected currents and potentials were verified manually through careful inspection of all electrophysiological data . Videos were acquired using a Canon Exilim-F1 digital camera at a frame rate of 30 Hz . The electrophysiological traces were synchronized with the video by aligning the recordings with the onset of an LED timed by the electrophysiology acquisition software . Analysis of the video was performed using custom built software in Matlab ( MathWorks ) which is available as Source code 1 . The video was first cropped to contain a small area displaying the mouse and then used to calculate a motion map as follows: First a background ( corresponding to pixels that have not moved recently ) was calculated by the following formula:BGi[x , y]=α×Framei[x , y]+ ( 1−α ) ×BGi−1[x , y] , where BGi is the background at time point i , and Framei is the video frame . Empirically , we found α = 0 . 3 to perform well with our dataset , this causes the background to be a weighted average over about the last 15 frames . Next , a difference image was calculated as follows:diffimi[x , y]=\|Framei[x , y]−BGi[x , y]\| . To provide a less noisy estimate of motion , we used a time smoothed difference image as the motion map:motion_mapi[x , y]=β×diffimi[x , y]+ ( 1−β ) ×motion_mapi−1[x , y] , where β = 0 . 9 . Spike-triggered averages of this motion map were computed by binning the neural event rates in 33 . 3 ms bins ( i . e . , matching the frame rate ) and weighting the frames by the event rate in the bin . A motion index was calculated by thresholding the motion map ( with a threshold of 40 at 8 bit resolution ) and summing the pixels exceeding the threshold . The motion index was used to define periods of quiet wakefulness and periods in which the mice were moving . Quiet periods were defined as periods in which the motion index changed by a rate of less than 0 . 025 arbitrary units per frame ( with maximum rate of 1 given by the maximal pixel change ) for at least 30 consecutive frames . Conversely , periods of movement were defined as rates of change greater than 0 . 025 a . u . per frame for at least 30 consecutive frames . These definitions were used to divide the corresponding electrophysiological data into periods recorded during quiet wakefulness and during locomotion for the analysis of spike and EPSC frequencies . We correlated the binned electrophysiological activity with the motion index at either coarse ( 1 . 5 s ) or fine ( 33 ms ) time scale . The electrophysiological activity was binned as follows: For EPSCs , GC Spikes and MFT spikes , we computed the average event rate in a bin . For spillover , we computed the average spillover current ( as determined by our fitting procedure below ) in a bin . For the coarse time scale correlation , we split the motion index and electrophysiological activity into 1 . 5 s bins and averaged the data in these bins , and then calculated Pearson's r coefficient between the event rates ( for EPSCs , GC spikes , MFT spikes ) or current ( for spillover ) and the motion index . For the fine scale correlation , we made the bin size the duration of a video frame ( i . e . , 33 . 3 ms ) for ease of analysis . We performed a rolling normalized cross-correlation between the electrophysiological data and the motion index , by shifting a 3 s window across the data . For each window , the mean was subtracted from both the motion index and electrophysiological data , and a cross-correlation performed . For each cell , an average cross-correlogram was computed by averaging across shifts , and the peak in this cross-correlogram was measured . To establish whether these correlations were significant , we generated bootstrap samples for each cell by repeating the analysis with shuffled versions of the corresponding binned electrophysiological activity , which allowed us to generate a z-score for each cell . This z-score was used to look up a one sided p-value in a standard normal table . p-values were Bonferroni-corrected for multiple comparisons , and the significance level was set at p = 0 . 05 . To separate the phasic EPSCs from the underlying spillover currents we fit a model to the raw voltage clamp traces . The model consisted of a train of biexponential functions ( representing the fast events ) on top of a spline function , representing an underlying slow current . Formulating this mathematically , we fit a function g ( t ) to the data:g ( t ) =βc1 . . . cNCP ( t ) +∑i=1NEPSCsampi×ζ ( τrise , τfall , t−Ti ) , where ß is a cubic spline function with control points , Ci is the ith control point of the spline , NEPSCs is the number of fast EPSCs in the data , ampi is the amplitude for the ith fast EPSC , ζ is the biexponential function with a rise time of τrise and a decay time τfall , Ti is the onset of the of the ith fast ESPC . ζ ( τrise , τfall , t ) =[exp ( −tτrise ) −exp ( −tτfall ) ] ( τrise−τfall ) . The data was collected in continuous sweeps of 5 s , and we therefore fit the model separately to these 5 s episodes . The fitting procedure was as follows: a peak finding algorithm was used to detect the fast EPSCs in the raw voltage clamp traces . We initiated ß to be equal to zero , NEPSCs to be equal to the number of peaks found by the algorithm , ampi to the amplitudes found by the algorithms , Ti the time of the events , τrise was set to 1 ms and τfall was set to 10 ms . We then used the Matlab nlinfit function to fit the model to the raw traces , adjusting the spline parameters , the ampi and τrise and τfall . To correct for slow drifts in the holding current during a long recording , we set the maximum of the ß trace for each 5 s episode to be zero . The spline fits baseline shifts in the traces not accounted for by summation of fast events , and we used it as our measure of putative spillover . Note that since spillover currents are known to contribute to the tails of fast EPSCs ( DiGregorio et al . , 2002 ) , we are underestimating the total contribution of spillover transmission in granule cells in the awake mouse . The relative contribution of phasic and spillover transmission as a function of EPSC rate was calculated as follows: the smoothed EPSC rate was computed by convolving a causal exponential kernel ( with tau = 50 ms ) with a train of delta functions placed at the times of the fast EPSCs as found by our fitting procedures . We then iterated over all events in all cells and calculated the ratio of the fast current at the peak of the event on the spillover current at that time , treating each event as a data point . These data points were then binned by EPSC rate . For Figure 2E , F we also related the currents to the motion index , convolved with a causal exponential kernel ( with tau = 660 ms ) . To calculate the cross-correlation between EPSC rate and spillover ( Figure 2C ) , we used the smoothed EPSC rate trace and the putative spillover traces from the fitting procedure , both mentioned above , and used a 2 s sliding window over the traces . For each window , we computed a normalized , mean subtracted cross-correlation between the EPSC rate and spillover traces . We then computed the mean cross-correlation , averaged across all cells and all the sliding windows . For the burst-triggered spillover , we defined a burst as a group of 5 or more EPSCs occurring at 200 Hz or more . We averaged the spillover trace triggered by the first event of such bursts ( Figure 2D ) . We employed a published cerebellar granule cell model ( Diwakar et al . , 2009 ) to study synaptic integration using in vivo patterns of activity . This model consists of a detailed compartmental model of a spiking granule cell for the NEURON simulation environment . We only modified the model by adding a fixed tonic inhibitory conductance at the soma of 1 nS ( Erev = −70 mV ) . The model was run in current clamp mode . To inject the patterns of excitatory input we recorded in vivo , we added an AMPA synaptic conductance ( Erev = 0 mV ) which was varied dynamically to correspond to the conductance underlying our voltage clamp traces , assuming a driving force of 70 mV . Our spillover analysis described above separated our voltage clamp recordings into the contributions from the fast EPSC current , and the slow putative spillover traces , so we could feed these separately or summated into the model and record the spike output of the model cell . We also performed simulations in which an NMDA receptor conductance was added to the spillover and phasic AMPA conductances , with an NMDA:AMPA ratio of 0 . 2 ( Cathala et al . , 2003 ) , and a voltage-dependent Mg2+ block modelled according to ( Nieus et al . , 2006 ) . For the current clamp GC recordings , burst analysis was done as follows: spikes were detected in the raw traces using a voltage threshold ( −10 mV ) . A procedure then iterated through the spikes and grouped them into bursts if 3 or more spikes appeared in succession with an ISI less than 50 ms . The depolarization underlying a burst was quantified as the voltage averaged across a 5 ms interval starting 1 ms after the first spike in the burst , after subtracting a baseline voltage , defined as the average voltage over a 180 ms interval starting 20 ms before the first spike in the burst . The videos were downsampled spatially by a factor of 2 , and PCs of the video were calculated by first computing the motion map of the video as above , and then shifting an 8 frame window across this motion map . For each of the videos , we created a data matrix where the rows contained the 8 frame windows of the motion maps . The standard pca function in Matlab was then used to obtain the PCs of this matrix , as well the coefficients of each video in this eigenbasis . We kept the 50 components with the highest eigenvalues for each video . For each of these components , we then computed a sliding cross-correlation between its coefficient in the video and a binned version of the electrophysiological trace ( as described in the video analysis section ) . The sliding cross-correlation was computed as detailed above for the motion index . To establish the noise level , we generated bootstrap samples by repeating the analysis on shuffled versions of the binned electrophysiological traces . To label the PCs according to which body parts were involved in the motion , we devised a semi-automated labelling algorithm: for each video , we defined rectangular ROIs for each of three regions ( head , body , limbs ) . Each region could have more than one ROI ( e . g . , one ROI for each limb ) . For all the PCs , we looked for framewise pixel changes in the ROI's as follows:Δ=∑x , y∈ROI∑i=2n\|PC[x , y , i]−PC[x , y , i−1]\| , where PC[x , y , i] is the PC at pixel x , y and the i-th frame , and n is the total number of frames in the PC . A ROI , and therefore a body region , was deemed to be participating in the motion of the PC if Δ exceeded a threshold . Note that in these head-fixed recordings , ‘head’ movement refers primarily to movement of the whisker pad and ears . To select the threshold in a non-arbitrary manner , we also selected ROIs corresponding to background for each of the video . We then charted the distribution for the Δ's in these background ROIs . In all of the videos , Δ ranged from 0 to 2500 , and therefore we selected 3000 as our threshold . We built a custom GUI to step through the videos frame by frame and annotate limb movement . We noted in each frame whether either of the forelimbs were in swing or stance phase . The swing phase was defined as starting when the paw was lifted off the ball , and ending when contact between the paw and the ball was reinitiated . For either forelimb , we calculated the step-evoked electrophysiological activity for each cell by triggering a 40-bin ( or 1 . 32 s ) episode of the activity on the start of the stance phase , and averaged these episodes for each cell , giving an average step-evoked response vector σ1 , 2 , . . . , 40 . We computed a modulation index m for each cell by taking this vector and computing:m=maxi=140 ( σi ) −mini=140 ( σi ) maxi=140 ( σi ) +mini=140 ( σi ) . To test for significance , we generated bootstrap samples by selecting for each step a random window from the corresponding binned electrophysiological trace . For the individual cells , we considered a modulation index to be significant if its associated z-score corresponded to a p-value < 0 . 05 after Bonferroni correction for the numbers of cells within that recording modality . When measured across the population , while the mean pooled z-scores were positive , they were not significant ( mean z-scores were MFBs: 2 . 4427 ± 2 . 8788 , p = 0 . 3995 for left limb , 5 . 2776 ± 4 . 2794 for right limb . EPSCs: 3 . 0553 ± 1 . 5468 , p = 0 . 0695 for left limb , 2 . 3266 ± 1 . 0831 , 0 . 0522 for right limb . GC-spikes: 0 . 0980 ± 1 . 0836 , p = 0 . 9243 for left limb , 0 . 1723 ± 2 . 0968 , p = 0 . 9312 for right limb . Spillover: −0 . 1011 ± 0 . 3651 , p = 0 . 7764 for left limb , 0 . 0192 ± 0 . 4422 , p = 0 . 9645 for right limb ) . This is not surprising , since it reflects the fact that that only a subset of the granule cells are well-modulated by the step cycle , which is to be expected given the variety of inputs to the granule cell layer . For the polar plot in Figure 4—figure supplement 2 , we plotted the modulation to step cycle by taking the absolute deviation from the mean activity as the radius of the plot . That is , R ( θ ) =\|activity ( θ ) −∑i=0Nactivity ( i×360N ) N\| , where activity is the step-triggered activity vector as in Figure 4A mapped on 360° ( i . e . , this works out to each degree in the plot corresponding to 5 ms ) . For ease of visualization , we subtracted from R the minimum value across phases and then divided by the maximum value , so R is between 0 and 1 . We picked the phase of maximal modulation as the phase that maximized R . In Figure 4—figure supplement 2 , we plotted the modulations as z scores by subtracting the mean of the step triggered activity vector for each recording and dividing by its standard deviation . A two state Markov model was constructed using the Bayes Net Toolbox ( https://code . google . com/p/bnt/ ) . The output distribution was modelled as a mixture of four Gaussians . The observed data at time i consisted of a 360 ms window of the electrophysiological data starting at time i . For the EPSCs , the activity in each window was normalized to the mean; for the spiking data , the baseline activity ( average activity of the first 90 ms in the window ) was subtracted . All the parameters were initiated at random , except for the prior probabilities of the states and initial state transition matrix that was derived for each cell from the step data annotated manually . The Bayes Net Toolbox was then used to perform a Baum-Welch algorithm to optimize the prior probabilities , the transmission matrix , and mixture of Gaussian parameters . The Viterbi algorithm was then used to reconstruct the predicted state transition through the HMM . To score the prediction against the real stepping data , we first converted both the prediction and real data into spike trains , by placing spikes at the transitions from stance to swing . We then used a commonly utilized metric between spike trains ( Schreiber et al . , 2003 ) to compute the quality of the step reconstruction . Briefly , this measure computes a normalized , mean-subtracted cross-correlation between spike trains convolved with a Gaussian kernel ( we used a sigma of 100 ms ) , and then takes the peak of the cross-correlation as the measure of similarity between the trains . This value can vary between 0 ( no similarity ) and 1 ( identical trains ) . We multiplied this value by 100 to get a percentage score . For each limb , we took the average of the percentage score over 10 runs of the HMM algorithm to get the average prediction score . For each cell , we selected the limb that was best predicted by the activity and reported its prediction score . To test for significance for each cell , we generated bootstrap samples for each cell by repeating the above procedure on shuffled version of the electrophysiological data for the cell , and generating a z-score of the prediction score . The prediction for each cell was deemed significant if its associated one-sided p-value was <0 . 05 after Bonferroni correction .	Our voluntary movements , such as shaking hands and walking , are controlled by a region of the brain called the cerebellum . Inside this region is a layer of cells called granule cells , which are the smallest and also the most numerous type of neuron in the brains of mammals . Granule cells receive information from many other parts of the brain and respond by producing electrical signals that influence the motor system , which tells our muscles how to move . However , it is not clear how the granule cells interpret the information they receive and ensure that the right muscles are stimulated at the right time by the motor system . Powell et al . have now used ‘patch-clamp electrodes’ to measure the electrical activity of individual granule cells in the cerebellum of mice , both at rest and as they walked . This is a powerful approach as it enables the recording of both the information received by each granule cell ( input ) and the electrical signals produced by it in response ( output ) . Each mouse was placed on a treadmill with its head held still and given the choice to either rest or walk . These experiments show that when the mouse is resting , the granule cells are mostly inactive , producing only very low levels of fast electrical signals called ‘spikes’ . When the mouse starts walking , the input to the granule cells triggers a strong increase in spiking in the granule cells . Powell et al . used a computer model to understand how the granule cells represent movement . Remarkably , this model could be used to predict walking patterns of the mouse based on the activity of a single granule cell and its inputs . These findings suggest that even single neurons in the cerebellum contain rich information about the movement of the animal . The next challenge is to understand how this code interacts with the rest of the motor system to produce precisely coordinated movements . Furthermore , it will be important to determine whether a similar code is used in other parts of the brain that control movement .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "short", "report", "neuroscience" ]	2015	Synaptic representation of locomotion in single cerebellar granule cells
The spatial organization of synaptic inputs on the dendritic tree of cortical neurons plays a major role for dendritic integration and neural computations , yet , remarkably little is known about it . We mapped the spatial organization of glutamatergic synapses between layer 5 pyramidal cells by combining optogenetics and 2-photon calcium imaging in mouse neocortical slices . To mathematically characterize the organization of inputs we developed an approach based on combinatorial analysis of the likelihoods of specific synapse arrangements . We found that the synapses of intralaminar inputs form clusters on the basal dendrites of layer 5 pyramidal cells . These clusters contain 4 to 14 synapses within ≤30 µm of dendrite . According to the spatiotemporal characteristics of synaptic summation , these numbers suggest that there will be non-linear dendritic integration of synaptic inputs during synchronous activation . Computations on the level of single neurons critically depend on the location and spatial relationship between active synapses ( e . g . Branco and Häusser , 2011; Losonczy and Magee , 2006; Polsky et al . , 2004 ) . Thus knowing the connectivity of neurons with single synapse resolution is essential to understand how synaptic integration at the cellular level contributes to neural circuit function . One type of spatial organization is of particular interest . These are clusters of synaptic inputs , which can give rise to superlinear summation during synchronous activity ( e . g . DeBello et al . , 2014; Larkum and Nevian , 2008; Mel , 1993 ) . Clustering has been reported for synchronously active synapses during spontaneous activity ( Kleindienst et al . , 2011; Takahashi et al . , 2012 ) , for synapses undergoing plasticity ( Makino and Malinow , 2011; McBride et al . , 2008 ) as well as for synapses from anatomically defined populations of presynaptic neurons ( Druckmann et al . , 2014; Rah et al . , 2013 ) . However , the cluster parameters relevant for dendritic computations are not rigorously and quantitatively assessed . Furthermore , in different cortical areas the patterns of synapses activated by sensory stimulation suggest that inputs onto central neurons may not be clustered ( Chen et al . , 2011; Jia et al . , 2010; Varga et al . , 2011 ) . Here we address two issues: First , we explore the spatial arrangement of synapses of an anatomically defined intracortical connection , i . e . the intralaminar connection between layer 5 pyramidal neurons . Secondly , we analyze the spatial organization of synapses with a novel method for identifying individual synapse clusters . This allows characterizing cluster parameters , such as dendritic length and number of inputs , in order to put them into the perspective of the rules of dendritic integration and the arithmetic of synaptic summation ( e . g . Branco and Häusser , 2011; Losonczy and Magee , 2006; Polsky et al . , 2004 ) . For mapping the synapses between layer 5 ( L5 ) pyramidal cells in mouse primary visual cortex ( V1 ) we combined optogenetics and 2-photon calcium imaging ( Little and Carter , 2012; MacAskill et al . , 2012 ) , which we refer to as Channelrhodopsin ( ChR2 ) -assisted synapse identity mapping ( CASIM ) . The principle of CASIM is depicted in Figure 1A: NMDA receptor mediated Ca2+ signals identify those dendritic spines on a postsynaptic neuron , which receive input from photostimulated presynaptic neurons expressing ChR2 . CASIM has the advantage over other methods for mapping synapses by light or electron microscopy ( e . g . Druckmann et al . , 2014; Rah et al . , 2013 ) that it identifies synapses in functional rather than only in structural or molecular terms . For presynaptic expression of ChR2 , we used transgenic mice expressing ChR2-YFP under control of the THY1 promoter in L5 pyramidal cells ( line 18; Wang et al . , 2007; Figure 1B ) . Expression levels and stimulus sensitivity were different from cell to cell . High photostimulus intensities evoked electrical responses in all probed L5 pyramidal cells . Because in more sensitive cells the number of generated action potentials dropped with higher stimulus intensities , we applied an intermediate stimulus intensity for synapse mapping , which evoked action potentials in 58% of L5 pyramidal cells ( Figure 1—figure supplement 1 , 2 ) . For synapse mapping we selected layer 5 pyramidal cells with no detectable ChR2-YFP fluorescence , indicating that the levels of ChR2 expression are so low that the selected stimulus intensity is insufficient to cause significant depolarization . For recording calcium signals in spines of postsynaptic L5 pyramidal cells , we patched and filled cells with Alexa 594 ( 30 µM ) and Fluo-5F ( 1 mM ) in acute slices . In order to isolate NMDA receptor mediated calcium signals and prevent spreading of unspecific excitation , patched cells were depolarized to 10 mV above the NMDAR reversal potential in voltage clamp in extracellular solution containing NBQX ( 10 µM ) , picrotoxin ( 50 µM ) and D-serine ( 10 µM ) . Synapses between pairs of L5 pyramidal cells are reported to be located to a large extend on the proximal basal dendrite of the postsynaptic cell ( Markram et al . , 1997; Sjoström and Häusser , 2006 ) . Therefore we focused on the basal dendrites of L5 pyramidal cells to map their inputs from other L5 pyramidal cells . Basal dendritic branches were systematically scanned with overlapping rectangular tiles of 2-photon calcium imaging to probe all spines for photostimulus-evoked calcium signals . Figure 1C shows a sequence of calcium imaging frames zoomed in on two spines , where the left spine shows a clear synaptic calcium signal . Due to the stochastic nature of neurotransmitter release , it is not expected that each AP leads to transmitter release . Thus we classified spines as receiving L5 input if they displayed at least once a higher ( ≥3×standard deviation ) and earlier peak calcium signal evoked by photostimulation than observed in the parent dendrite ( Figure 1D - 1G ) . In Figure 1E L5 input positive ( L5+ ) spines are marked by white arrowheads on a dendritic stretch . In total , out of 2168 analyzed spines , 199 were L5 input positive ( 9 . 2% , n = 20 cells ) . The average peak calcium signal amplitude in successful trials ( ΔG/R = 0 . 48 ± 0 . 01 ) was significantly higher than the calcium signal in the parent dendrite ( ΔG/R=0 . 18 ± 0 . 005 , p<0 . 001 , n = 199 spines; Figure 1G ) . There was no significant difference in the morphological properties of L5+ spines and L5 input negative ( L5- ) spines ( length: L5+ , 1 . 23 ± 0 . 06 μm , n = 76; L5- , 1 . 11 ± 0 . 02 μm , n = 419; p=0 . 04; normalized volume: L5+ , 1 . 00 ± 0 . 08 , n = 77; L5- , 1 . 00 ± 0 . 03 , n = 437; p=0 . 97; Figure 1—figure supplement 3A , B ) . 10 . 7554/eLife . 09222 . 003Figure 1 . Mapping intralaminar inputs on the dendrites of L5 pyramidal neurons with channelrhodopsin-assisted synapse identity mapping ( CASIM ) . ( A ) Schematic diagram illustrating the principle of CASIM . Synapses between presynaptic L5 neurons expressing ChR2 ( green ) and a postsynaptic L5 neuron filled with calcium indicator ( red ) are identified by calcium signals in dendritic spines ( L5+ spines ) evoked by photostimulation of ChR2 . ( B ) Acute slice of primary visual cortex from a Thy1-ChR2 mouse . A L5 pyramidal cell ( arrowhead ) was patched and filled with Fluo-5F . ( C ) Example of a spine calcium signal evoked by ChR2 activation: Left , image of two spines and their parent dendrite ( from white box in panel E ) in the red channel ( Alexa 594 ) for structural imaging . Right , sequence of frames from calcium imaging in one trial showing calcium signal in the left spine . ( D ) Ca2+ signal corresponding to panel C ( ΔG/R; G , Fluo-5F; R , Alexa 594; red , spine; black , dendrite at spine base ) . ( E ) Analyzed dendritic stretch from cell in B . L5+ spines are marked by white arrowheads . Blue arrow head marks an example L5- spine . Dashed boxes indicate the size of imaging tiles . Inset , corresponding dendrogram; scale bar , 10 µm . ( F ) Ca2+ signals from all trials recorded in a selection of L5+ spines ( S1 to S3 ) , one L5- spine ( S4 ) and at their respective dendritic bases ( D1-D4 ) in E . ( G ) Peak ΔG/R of L5+ spines ( S ) and dendrites ( D ) in cell in B ( blank bars; n = 13 spines ) and in all cells ( filled bars; n = 199 spines ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09222 . 00310 . 7554/eLife . 09222 . 004Figure 1—figure supplement 1 . Characterization and calibration of optical stimulation . The photostimulus protocol was optimized to maximize the probability of synaptic transmission . 3 pulses at 30 Hz were applied to deliver the photostimulus in the duration of one frame ( 100 ms ) while the PMTs for imaging were protected with a shutter . Calibration curves were acquired at 2 ( left ) and 5 ms ( right ) pulse width on L5 pyramidal neurons selected blindly ( n = 12 ) . Number of APs was measured in cell-attached mode . The mean overall calibration curves ( blue ) did not show a uniform increase with photostimulation power , but a declining AP yield at high powers . To analyze whether this may be explained by potential adverse effects of strong photostimulation on strongly driven neurons , we analyzed the calibration curves with respect to two populations of neurons: Strongly excitable neurons ( with more than one AP on average at 1 , 2 and 4 mW power; black; dashed , individual cells; solid , mean curve; n = 4 cells at 2 ms pulse width , n = 5 cells at 5 ms pulse width ) showed saturation above 4 mW power with 2 ms pulse width and reduced activity at high illumination intensity at 5 ms pulse width . Weakly excitable neurons ( gray; dashed , individual cells; solid , mean curve; n = 8 cells at 2 ms pulse width , n = 7 cells at 5 ms pulse width ) were not affected adversely by illumination . Therefore for mapping , we chose 2 ms pulses at 4 mW as stimulus , which activates but does not overexcite weakly and strongly excitable neurons , respectively . Error bars correspond to standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 09222 . 00410 . 7554/eLife . 09222 . 005Figure 1—figure supplement 2 . Specificity of evoked presynaptic activity . AP generation in neurons outside of L5 was measured by cell attached recordings obtained during a wide-field light stimulus in L5 ( green circle ) . Left panel: marks ( + ) indicate the vertical location of the cells which were recorded in slices from Thy1-ChR2 mice . Color codes for the neurons with no evoked APs ( blue ) and evoked APs ( red ) . Right panel: example traces . In total 45 cells were recorded , of those only 3 cells in L6 exhibited light evoked activity ( 6 . 7% ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09222 . 00510 . 7554/eLife . 09222 . 006Figure 1—figure supplement 3 . Characterization of L5+ spines . ( A , B ) Cumulative fraction of spine volume ( A ) and normalized spine length ( B ) of L5+ ( red ) and L5- spines ( black ) . Spine volume and length were measured only from spines extending laterally in image stacks , since spines aligned along z could not reliably be separated from the dendrite . For spine volume , the summed intensity of pixels enclosed by the respective ROI was calculated for each z-plane after background subtraction . The highest sum represented the volume of the spine . Measured volumes were normalized to the mean spine volume within dendritic segments between branching points . Spine lengths were measured on maximum z-projections as the distance between the tip of the spine head and where it attached on the dendrite . ( C ) Number of L5+ spines versus total number of spines probed per cell ( open circles; correlation , dotted line , slope = 0 . 12 , R = 0 . 93 ) and per segment ( filled circles; correlation , continuous line , slope = 0 . 12 , R = 0 . 89 ) . ( D ) Distribution of L5+ spine distances from the soma ( red ) in the experiments and from a simulation with 1000 randomly reshuffled distributions of the same number of inputs over all recorded spine positions ( blue , individual simulations; dark blue , mean ) . For comparison also the distribution of distances of all spines from the soma is shown ( black ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09222 . 006 The number of L5+ spines scaled with the total number of probed spines ( Figure 1—figure supplement 3C ) but displayed a high degree of heterogeneity between cells ( 4% to 29% ) and individual dendritic segments ( 3% to 50% ) as has been previously reported for Schaffer collateral synapses ( Druckmann et al . , 2014 ) . The classical analysis of the distribution of pairwise nearest neighbor distances applied to L5+ spines ( e . g . Druckmann et al . , 2014; Rah et al . , 2013; Takahashi et al . , 2012 ) revealed significantly shorter recorded distances than expected from a random distribution ( Figure 2A ) , suggesting that L5 inputs are clustered . 10 . 7554/eLife . 09222 . 007Figure 2 . Spatial input organization . ( A ) Distribution of nearest neighbor distances between L5+ spines from experiments ( red ) and from 1000 times random reshuffling the recorded number of inputs over all present spines within each individual dendritic segment ( blue , dark blue mean trace ) . Median values of recorded and reshuffled data are significantly different ( data , median = 3 . 10 μm; simulations , median = 5 . 02 ± 0 . 43 μm; p<0 . 0001 ) . The plot of the difference between inter-input distances from experiments and simulations ( purple ) shows that the deviation is most pronounced for small inter-input distances below 10 µm with a peak deviation around 5 µm , suggesting a clustered arrangement of L5+ spines . ( B ) Combinatorial cluster analysis: Illustration of the principle of calculating the specific ensemble likelihood ( combinatorial cluster analysis step 2 ) . Top , example of a dendritic segment with N = 30 spines in total and n = 5 L5+ spines ( red ) . m = 4 L5+ spines are part of an input ensemble of size M = 6 spines ( gray box ) by fulfilling the maximal nearest neighbor distance criterion of △crit = 2 ( blue bar; cluster analysis step 1 ) . The ensemble is flanked by 'gaps' of L5- spines ( gray brackets ) . As illustrated below , the likelihood for observing an ensemble of M = 6 spines with m = 4 L5+ spines is determined by three factors: The number or ways the L5+ spines can be distributed inside the ensemble , the number of ways the remaining L5+ spines can be distributed outside of the ensemble and the number of ways an ensemble of the given size can be placed on the dendritic segment . Bottom left , all ( M−2m−2 ) =6 possible arrangements of m = 4 L5+ spines in an ensemble of M = 6 spines , note that the ensemble edges ( red arrowheads ) have to be occupied always to delimit the ensemble . Bottom middle , four examples of the ( N−M−2gn−m ) =22 possible assignments of the remaining n-m = 1 L5+ spine to the spines outside of ensemble and gaps . Bottom right , four examples of the N−M+1=25 possibilities to place the ensemble of size M = 6 spines into the dendritic segment with N = 30 spines in total . Note that at positions close to the beginning and end of the segment , the leading or trailing gap , respectively , are not fully realized . ( C ) Dendrograms of 10 out of 20 mapped pyramidal cells summarizing 7 observed input clusters ( L5+ spines , red; L5- spines , black; dendrograms of all other recorded cells with input clusters in Figure 2—figure supplement 1 ) . Gray boxes mark input ensembles ( cluster analysis step 1 ) , red numbers are the specific ensemble likelihood values ( cluster analysis step 2 ) , red asterisks mark ensembles identified as clusters ( specific ensemble likelihood ≤0 . 01 , inputs ≥3; cluster analysis step 3 ) , black numbers are overall cluster likelihood values for each cluster ( cluster analysis step 4 ) . Note that spines , which are located very close to each other , are not resolved at this scale and their marks appear to overlap; dashed lines indicate unmapped stretches ) . ( D ) Specific ensemble likelihood scores of all ensembles ( thin line ) and those with at least three inputs ( thick line ) . Dashed line , cluster likelihood threshold . ( E ) Number of L5+ spines in individual clusters ( red ) and all ensembles ( black ) . ( F ) Length of clusters ( red ) and all ensembles ( black ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09222 . 00710 . 7554/eLife . 09222 . 008Figure 2—figure supplement 1 . Dendrograms of all other recorded cells with input clusters not shown in Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 09222 . 00810 . 7554/eLife . 09222 . 009Figure 2—figure supplement 2 . Combinatorial cluster analysis: Example of ensemble types and translation to non-uniform inter spine distances . ( A ) Combinatorial cluster analysis step 4: Example of all ensemble types possible with a total of n = 5 inputs ( L5+ spines ) and a gap/nearest neighbor distance criterion of △crit = 2 to be considered for calculating the overall cluster likelihood . ( B ) Translation of ensemble definition from uniform to non-uniform inter spine distances: Example of a dendritic segment with irregularly spaced spines . 5 spines receive the specific input ( red , L5+ ) , 4 of which are part of an ensemble ( gray square ) by complying with the nearest neighbor distance criterion △crit ( indicated by the blue bar ) . The length of the leading and trailing edge , which flank the ensemble , corresponds to the nearest neighbor distance criterion . No spine located within these gaps receives the specific input . DOI: http://dx . doi . org/10 . 7554/eLife . 09222 . 00910 . 7554/eLife . 09222 . 010Figure 2—figure supplement 3 . Additional properties of input clusters . Cumulative fraction of input density ( A ) , input packing ratio , i . e . ratio of L5+ spines over total number of spines in cluster or ensemble ( B ) , and location as distance from soma ( C ) or distance from proximal branching point ( D ) of detected clusters ( red ) and all ensembles ( black ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09222 . 01010 . 7554/eLife . 09222 . 011Figure 2—figure supplement 4 . Comparison of the observed and expected numbers of dendritic segments containing an input cluster ( cluster analysis step 5 ) . ( A ) For dendritic segments , which contain a synapse cluster , the overall cluster likelihood OCL is plotted against the segment number sorted by increasing OCL value . For sets of segments with OCL values below an upper bound , this graph is the same as plotting the upper bound OCLmax ( c ) as function of the number c of segments with OCL values below this upper bound . ( B ) P value P ( c , OCLmax ) ( Equation 4 ) for the binomial test of the null hypothesis that the number of observed segments with an input cluster arises from a random input distribution plotted against the recorded number of segments c containing a synapse cluster . The broken line indicates the significance level of p=0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 09222 . 01110 . 7554/eLife . 09222 . 012Figure 2—figure supplement 5 . Estimation of missed L5 inputs due to slicing . ( A ) Diagrams illustrating three cases to be considered for calculating the axon density at a specific dendritic site at depth D in a slice of thickness T arising from neurons at different radial distances r from this site: Case 1 , shell volume completely contained in slice ( top panel , see Equation 6 ) ; case 2 , top cap of shell missing from slice ( middle panel , see Equation 7 . ) ; case 3 , top and bottom caps of shell missing from slice ( bottom panel , see equ . 8 ) . ( B ) Radial axon density A ( r ) of L5 pyramidal cell axons in mouse visual cortex based on the sholl analysis by Blackman et al . , 2014 ( Figure 2C , 2nd panel from left ) calculated according to Equation 5 . ( C ) Remaining fraction of axonal density within a slice of thickness T as a function of the depth D in the slice calculated according to Equation 9 . DOI: http://dx . doi . org/10 . 7554/eLife . 09222 . 012 In order to place the clustered arrangement of synapses into the framework of the known rules of dendritic integration and arithmetic of synaptic summation ( e . g . Branco and Häusser , 2011; Losonczy and Magee , 2006; Polsky et al . , 2004 ) , it is necessary to identify individual clusters so as to characterize for example their size and the number of inputs they receive . This means to go beyond the classical pairwise analysis of nearest neighbor distances . To this end , we developed a new approach for the identification of individual synapse clusters based on combinatorial analysis of the likelihood of specific synapse arrangements ( Material and methods; Figure 2B ) . We identified 12 clusters of L5+ spines on the basal dendrites of 20 cells ( Figure 2C and Figure 2—figure supplement 1 ) . These clusters extend over roughly 15 μm with , on average , approximately 8 spines ( dendritic length , 12 . 52 ± 2 . 40 μm; 7 . 17 ± 0 . 94 L5+ spines; density , 0 . 84 ± 0 . 19 L5+ spines/μm; ratio of L5+ over all spines within the cluster , 0 . 48 ± 0 . 04; Figure 2E , F and Figure 2—figure supplement 2A; n = 12 clusters ) . Almost 50% of all L5+ spines were part of a cluster ( 86 out of 199 ) . There is a possibility , that due to the slicing procedure some inputs may be missed . We estimate that the size of spine clusters could be underestimated by a factor between 1 . 4 and 2 ( see Material and methods; Figure 2—figure supplement 3 ) . This result shows that in the cortex cluster of synapses of an anatomically defined intracortical connection exist , which have the required size and spacing to be relevant for non-linear dendritic computations ( ≥4 synapses within 30 µm of dendrite; see e . g . Branco and Häusser , 2011; Losonczy and Magee , 2006; Polsky et al . , 2004 ) . Furthermore , the scale of these clusters is in line with local learning rules ( Engert and Bonhoeffer , 1999; Harvey and Svoboda , 2007 ) , which might play a role in cluster formation and furthermore in synapse interactions within the cluster during synaptic plasticity . Our results address a pressing question in the field of neural signal processing: How are the synapses of neocortical connections spatially organized on the dendrites of cortical neurons and how do the spatial patterns relate to the known rules of dendritic integration and the rules of spatiotemporal synaptic summation ( e . g . Branco and Häusser , 2011; Losonczy and Magee , 2006; Polsky et al . , 2004 ) ? To address this question we mapped the intralaminar synapses between L5 pyramidal neurons in visual cortex by combining optogenetics and 2-photon calcium imaging ( CASIM; Little and Carter , 2012; MacAskill et al . , 2012 ) . Our results show clustering of these synapses on the basal dendrites of L5 pyramidal neurons . For quantitative analysis of the spatial organization we developed a new method for the identification of individual synapse clusters and characterization of cluster parameters . Our results are in contrast to previous studies mapping sensory inputs in different neocortical areas , which reported no obvious spatial structure in input patterns and suggested that local non-linear dendritic integration may not be important for sensory processing in the cortex ( Chen et al . , 2011; Jia et al . , 2010; Varga et al . , 2011 ) . However , more recent studies provided evidence for non-linear computations during cortical processing ( Lavzin et al . , 2012; Smith et al . , 2013; Xu et al . , 2012 ) . A potential explanation for this apparent contradiction ( DeBello et al . , 2014 ) is that in the experiments described here , we stimulate a well-defined set of neurons , while inputs which are defined by the same sensory stimulus ( Chen et al . , 2011; Jia et al . , 2010; Varga et al . , 2011 ) are in all likelihood arising from a mix of genetically/anatomically defined types of neurons . Input clustering has been shown for axodendritic contacts in the barn owl auditory system ( McBride et al . , 2008 ) , and more recently , in mice , for thalamocortical ( Rah et al . , 2013 ) and hippocampal CA3-CA1 synapses ( Druckmann et al . , 2014 ) . In all of these cases however there is no quantification of cluster parameters , which is crucial if one wants to relate the findings to any kind of biophysical model of dendritic computation . In this way our data and analysis go considerably beyond these reports by identifying individual clusters and quantifying cluster parameters , such as for example the number of inputs per cluster . Our novel method for spatial cluster analysis by combinatorial statistics is not limited to the synaptic connections described here . It can be applied to characterize any distribution or pattern of synapses on the postsynaptic dendrite as obtained for example by large scale electron microscopy , GFP reconstitution across synaptic partners , array tomography or mapping of synaptic activity ( e . g Bock et al . , 2011; Chen et al . , 2011; Druckmann et al . , 2014; Kleindienst et al . , 2011; Rah et al . , 2013; Takahashi et al . , 2012 ) , as well as any other data regarding the spatial organization of structures along a one dimensional axis like for example the distribution of presynaptic boutons along an axon ( e . g . Schuemann et al . , 2013 ) . Our results show that the clusters of intracortical connections between L5 pyramidal neurons contain 4 to 14 synapses within 30 µm of basal dendrite . According to the known rules of dendritic integration and the spatiotemporal arithmetic of synaptic summation ( Branco and Häusser , 2011; Losonczy and Magee , 2006; Polsky et al . , 2004 ) such clusters would result in superlinear dendritic integration during synchronous synaptic activity . Superlinear integration can also occur without tight clustering ( Branco et al . , 2010; Losonczy and Magee , 2006 ) , but the specifics of the spatial input organization are expected to define how different temporal activity patterns determine the firing pattern of the postsynaptic cell . Our results are well explained by local learning rules , which may not only be involved in cluster formation but also support interactions among synapses in a cluster during synaptic plasticity events ( Engert and Bonhoeffer , 1999; Harvey and Svoboda , 2007 ) . They are also in line with observations of spatially clustered spontaneous activity of synapses during development ( Kleindienst et al . , 2011; Takahashi et al . , 2012 ) and clustered spine formation and synaptic potentiation during learning and experience-dependent plasticity ( Fu et al . , 2012; Makino and Malinow , 2011 ) . A further potential role of input clusters may be to overcome the stochasticity of neurotransmitter release and help to increase the reliability of synaptic transmission ( Bagnall et al . , 2011 ) . Typically connections between L5 pyramidal cells consist of 4 to 8 synapses , which are distributed over different branches of the postsynaptic dendrite ( Deuchars et al . , 1994; Markram et al . , 1997 ) . This suggests that the synapses within L5 input clusters most likely arise from a set of presynaptic neurons rather than a single one . Nevertheless , synchronous activation of L5 inputs within a cluster is expected to occur during L5 network oscillations ( e . g . Buffalo et al . , 2011; Sun and Dan , 2009 ) . In anesthetized rats , L5 pyramidal cells in visual cortex receive synchronously oscillating excitatory inputs ( Sun and Dan , 2009 ) , which are thought to arise from rhythmically firing L5 pyramidal cells ( Flint and Connors , 1996; Silva et al . , 1991 ) within the network of L5 pyramidal cells ( Markram , 1997 ) . Similar oscillations in L5 have been observed in awake animals ( e . g . Buffalo et al . , 2011 ) . In conclusion , our results , to our knowledge , provide the first characterization of synapse clusters arising from an anatomically and genetically defined input on the dendritic tree of cortical pyramidal cells . Only such quantitative analysis of the spatial organization of inputs together with the rules of dendritic integration and synaptic summation will allow working out the logical operations of neural computations at the cellular level , which are critical for understanding information processing in neural circuits . 300 µm thick acute coronal slices of primary visual cortex were prepared from THY1-ChR2 ( line 18 ) mice ( Wang et al . , 2007 ) between postnatal day 40 and 55 in choline-based ACSF ( in mM , 110 choline chloride , 2 . 5 KCl , 25 NaHCO3 , 1 . 25 NaH2PO4 , 0 . 5 CaCl2 , 7 MgCl2 , 25 D-glucose , 11 . 6 Na-L-ascorbate , 3 . 1 Na-pyruvate , aerated with 95% O2/5% CO2 ) at 0°C ( Scheuss et al . , 2006 ) after perfusing the animal with the same solution . Slices of the separated hemispheres were incubated at 35°C for one hour and then stored at room temperature in normal ACSF ( in mM , 127 NaCl , 2 . 5 KCl , 25 NaHCO3 , 1 . 25 NaH2PO4 , 2 CaCl2 , 1 MgCl2 , 25 D-glucose , aerated with 95% O2/5% CO2 ) ( Scheuss et al . , 2006 ) until the recordings . Cell activity was measured by cell attached recordings using patch pipettes filled with ( in mM ) 10 KCl , 140 K-gluconate , 10 HEPES , 2 MgCl2 , 2 CaCl2 , 0 . 05 Alexa 594 , pH 7 . 25 . The seal resistance was 20–40 MΩ . For obtaining ChR2 dose-response curves , light-evoked activity at blindly selected L5 pyramidal neurons was measured at increasing illumination power ( 1 , 2 , 4 , 6 , 9 mW at objective back aperture; see section Imaging Experiments ) and pulse widths ( 2 and 5 ms ) in the same ACSF solution used for mapping experiments , but additionally containing 10 μM CPP . To verify the specificity of photostimulation of L5 neurons , the activity of neurons in other cortical layers was measured in response to the same light stimulation at L5 and in the same ACSF solution as used for mapping experiments . Experiments were performed at room temperature in normal ACSF containing in addition ( in µM ) 10 NBQX , 50 picrotoxin and 10 D-serine , on a custom built two-photon laser-scanning microscope controlled with Labview ( National Instruments , Austin , TX , USA ) based custom imaging software . L5 pyramidal cells with undetectable or low levels of ChR2-YFP expression in primary visual cortex were visually identified ( 2-photon imaging at 920 nm excitation ) and patched in whole-cell voltage clamp mode ( access resistance 10–30 MΩ; internal solution , in mM , 125 Cs-methanesulfonate , 10 HEPES , 10 Na2phosphocreatine , 4 MgCl2 , 4 Na2-ATP , 0 . 4 Na-GTP , 3 Na-L-ascorbate , 5 QX-314 , 10 TEA-Cl , 1 Fluo-5F , 0 . 03 Alexa 594 , pH 7 . 3 ) . Cells were filled for ≥10 min at resting potential and then depolarized to 10 mV above the NMDAR reversal potential to remove the Mg2+ block from NMDARs ( Oertner et al . , 2002 ) . Since in adulthood the occurrence of ‘silent’ synapses is effectively negligible ( Rumpel et al . , 1998 ) NMDA receptor mediated calcium signals are expected to identify functional synapses containing both AMPA and NMDA receptors . The photostimulus ( 470 nm LED; Rapp OptoElectronic GmbH , Wedel , Germany ) consisted of 3 pulses of 2 ms at 30 Hz ( 4 mW at the objective back-aperture ) and was applied via the fluorescence excitation path of the microscope to the full field of view . It evoked on average 1 . 57 ± 0 . 54 ( n = 12 ) action potentials selectively in 58% of the ChR2 expressing L5 neurons ( Figure 1—figure supplements 1 , 2 ) . Dendritic branches were systematically scanned with overlapping rectangular tiles ( 5×19 . 8 μm2 ) at 810 nm excitation . Rotation of the tiles was adjusted to fit spines and parent dendritic segment optimally in the field of view . For calcium imaging , 50 frames ( pixel size 100 x 100 nm2 ) of single z-planes were acquired at 10 Hz in green ( Fluo-5F ) and red ( Alexa 594 ) channel in the following sequence: First frame with laser shutter closed for measuring electrical offsets , 5 frames baseline before photostimulation , 2 frames with PMTs protected by a shutter during photostimulation , and 42 frames after the photostimulation . At every dendritic tile location , calcium imaging was repeated at least 3 times per z-plane ( interstimulus interval , ≥10 s ) at multiple z-planes to cover every spine , and an image stack was taken at higher resolution ( voxel size , 50 × 50 × 1000 nm3 ) . Data were analyzed with custom software in Matlab ( Mathworks , Natick , MA , USA; Source code 1 ) . Spine heads and their bases on the parent dendrite were marked with rectangular ROIs , and the dendritic segments were traced in reconstructions based on averaged red channel frames from calcium imaging or the maximum z-projections of the image stacks at each tile position stitched together based on cross-correlations . Spine calcium signals ( △G/R ) were calculated as change in fluorescence of the calcium indicator Fluo-5F ( ΔG ) normalized to the mean intensity of the structural marker Alexa 594 ( R ) averaged over the brightest 70% of ROI pixels in the red channel . Dendrite calcium signals at the spine bases were calculated identically but without any pixel selection . ΔG/R signals were smoothed ( 3-point moving average ) and the peak amplitude in the 2nd to 4th time point after the stimulus as well as the standard deviations in the baseline interval and during the last five time points were determined . A spine was classified as receiving L5 input ( L5 positive , L5+ ) if its peak ΔG/R exceeded in one trial the dendritic peak ΔG/R plus three times the higher value of the two standard deviations . We never observed large dendritic ΔG/R signals at input clusters , which could lead to false negative spine categorization according to this classification , because spine/dendrite coupling is usually low ( Sabatini et al . , 2002 ) and simultaneous activation of many spines within a cluster was very rare . The results of the classification algorithm for spine calcium signals were verified manually . Acquisitions with high noise in the calcium signal due to low fluorescence intensity , e . g . from small or out of focus spines , and without an evident difference between the spine and the dendrite calcium signal were classified as false-positives . Acquisitions with noticeable difference in spine and dendrite calcium signal , but classified by the algorithm as no response either due to a fast decaying spine calcium signal , or because spine and the dendrite calcium signal were of similar magnitude although the spine signal peak occurred earlier , were labeled as false-negatives . Acquisitions where a subjective decision was not possible were classified as ambiguous and treated as no response . In the individual traces ( n = 15293 acquisitions ) , the total discrepancy between manual verification and the classification by the algorithm was 2 . 37% , which translated into a total discrepancy of 6 . 46% with respect to the input classification ( 0 . 55% false negatives , 4 . 2% false positives , 1 . 71% ambiguous; n = 2168 spines ) . CASIM might miss synapses with very low release probability but otherwise the false negative rate is mostly determined by the fraction of presynaptic neurons not expressing ChR2 as with other approaches relying on the expression of markers of any kind ( e . g . Druckmann et al . , 2014; Rah et al . , 2013 ) . However , since undersampling the presynaptic population is expected to be random , this would not introduce artifacts of non-random structure in the mapped synapse distribution . Positions of spines on the dendrite were extracted from 3D reconstructions of dendrite traces in time-lapse acquisitions . The location of a spine was defined to be the point on the dendrite trace with the shortest orthogonal distance to the center of the spine head . The combinatorial cluster analysis substantially extends the classical pairwise analysis of nearest neighbor distances ( e . g . Druckmann et al . , 2014; Rah et al . , 2013 ) as it allows to identify and characterize individual clusters . The advantage of the combinatorial approach over the estimation of likelihoods using e . g . random reshuffling ( e . g . ( Takahashi et al . , 2012; Yadav et al . , 2012 ) is that it provides exact values for the small likelihoods involved , for which reliable estimates would require prohibitively large numbers of rounds of reshuffling . The combinatorial cluster analysis proceeds in five steps: To estimate the fraction of missed L5 inputs resulting from the slicing procedure we calculated the radial axon density A ( r ) ( Figure 2—figure supplement 3B ) of L5 pyramidal cell axons in mouse visual cortex based on the Sholl analysis by Blackman et al . , 2014 ( their Figure 2C , 2nd panel from left ) as ( 5 ) A ( r ) =number of crossings at r4πr2δr The average axon density arising from presynaptic L5 neurons at a distance r from a dendritic location is given by the radial axon density of L5 pyramidal cells at the distance of r from the soma . For calculating the remaining fraction of presynaptic neurons contributing to the axon density at a dendritic location at a depth of D in a slice of thickness T , three cases have to be distinguished ( Figure 2—figure supplement 3A ) . The neuron density is assumed to be uniform and therefore needs not to be explicitly considered in the following equations: Case 1: Spherical shell around dendritic location in slice is complete , 0 < r < D ( Figure 2—figure supplement 3A , top panel ) The volume of a spherical shell of radius r and thickness δr is ( 6 ) V ( r ) =4πr2δr Case 2: Spherical shell around dendritic location in slice with one cap missing , D < r < T-D ( Figure 2—figure supplement 3A , middle panel ) The volume of a spherical shell of radius r and thickness δr with one cap of height h = r-D subtracted is ( 7 ) V ( r ) = ( 4πr2−2πrh ) δr=2πr ( r+D ) δr Case 3: Spherical shell around dendritic location in slice with two caps missing , T-D < r < rmax ( Figure 2—figure supplement 3A , bottom panel ) The maximal radius rmax denotes here the maximal radial extent of the average axonal density . The volume of a spherical shell of radius r and thickness δr with top cap of height h1 = r-D and bottom cap of height h2 = r- ( T-D ) subtracted is ( 8 ) V ( r ) = ( 4πr2−2πrh1−2πrh2 ) δr=2πrTδr Combining these equations yields the remaining fraction of axonal density within a slice of thickness T as a function of the depth D in the slice as ( 9 ) F ( D ) =∫0D4πr2A ( r ) dr + ∫DT−D2πr ( r+D ) A ( r ) dr+ ∫T−Drmax2πrTA ( r ) dr∫0rmax4πr2A ( r ) dr From this we estimate that the fraction of the remaining L5 pyramidal cell axon density that can be activated in a slice is between 50% and 70% depending on the depth in the slice ( Figure 2—figure supplement 3C ) . This suggests that between 30% and 50% of L5 inputs are missed due to the slicing process . Spine and dendrite calcium signal peaks were compared to each other by Wilcoxon rank sum test . Z-tests were applied when comparing the median inter-input distance from the experiments to Monte Carlo simulations . In plots asterisks indicate p-values as * , p<0 . 05; , p<0 . 01; and * , p<0 . 001 . The values given in the text are mean ± SEM ( standard error of the mean ) , if not indicated otherwise .	Neurons in the brain exchange information through points of contact called synapses . If electrical activity arriving at a number of synapses exceeds a certain threshold , it can trigger an electrical impulse , which is transmitted to the next neuron . Synapses generally connect with branch-like structures called dendrites on the receiving neuron . However , little is known about how synapses are arranged on dendrites . Gökçe et al . have now used a technique called optogenetics to work out the exact arrangement of a type of synapse on neurons in a part of the mouse brain that is devoted to vision . Optogenetics takes advantage of light-activated proteins that can trigger electrical activity . Gökçe et al . used mice that had been genetically engineered to produce these proteins in specific neurons , and then deliberately triggered electrical activity simply by shining light on these neurons . The experiments also used another technique called two-photon calcium imaging to monitor the activity of single synapses in response to the electrical activity triggered by optogenetics . Gökçe et al . found that these neurons have clusters of four to fourteen synapses within a space of 30 micrometers along a dendrite . Synapses in clusters that are active at the same time can interact and thereby generate electrical signals more effectively than synapses spread across the dendrites . Further experiments are now needed to map the synapses between other kinds of neurons , and to map synapses from two different inputs at the same time .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "short", "report", "neuroscience" ]	2016	Clusters of synaptic inputs on dendrites of layer 5 pyramidal cells in mouse visual cortex
Sleep disturbances negatively impact numerous functions and have been linked to aggression and violence . However , a clear effect of sleep deprivation on aggressive behaviors remains unclear . We find that acute sleep deprivation profoundly suppresses aggressive behaviors in the fruit fly , while other social behaviors are unaffected . This suppression is recovered following post-deprivation sleep rebound , and occurs regardless of the approach to achieve sleep loss . Genetic and pharmacologic approaches suggest octopamine signaling transmits changes in aggression upon sleep deprivation , and reduced aggression places sleep-deprived flies at a competitive disadvantage for obtaining a reproductive partner . These findings demonstrate an interaction between two phylogenetically conserved behaviors , and suggest that previous sleep experiences strongly modulate aggression with consequences for reproductive fitness . Insufficient sleep impairs a wide range of essential processes such as cognition , alertness , metabolic activity , and immune function ( Foster and Wulff , 2005 ) . In addition , sleep disruptions influence emotional processing and can modulate affective state ( Banks and Dinges , 2007; Minkel et al . , 2011 , 2012 ) . Work over many decades has suggested an interaction between sleep loss and changes in aggressive behaviors ( Kamphuis et al . , 2012 ) , but even basic questions such as directionality of effect remain unresolved . Chronic enforced wakefulness or selective REM sleep deprivation have been linked to increased aggression in rodents , although in both cases it is not possible to distinguish the potential effect of sleep loss from prolonged physical activity and/or increased stress ( Kamphuis et al . , 2012 ) . In contrast , work in humans indicates that while measures of irritability increase with insufficient sleep , aggression itself is unaffected or reduced ( Kamphuis et al . , 2012; Cote et al . , 2013 ) . While human aggression often carries a negative connotation , it can provide a competitive advantage and thereby promote survival . However , dysregulated aggression and violence are significant public health concerns ( Anderson , 2012 ) , as is chronic sleep insufficiency ( Czeisler , 2013 ) , emphasizing the need to understand how these two processes affect one another . Both sleep and aggression are conserved across phylogeny . Drosophila has been established as a powerful model system for deconstructing the cellular and molecular basis of aggression , yielding novel insights into the neural basis of fighting behaviors ( Chen et al . , 2002; Asahina et al . , 2014 ) . Flies also exhibit sleep-like states ( Hendricks et al . , 2000; Shaw et al . , 2000 ) and , in response to sleep deprivation , demonstrate deficits in behaviors like learning and memory ( Seugnet et al . , 2008 ) . Whether sleep and aggression interact in the fly is unknown . Monoamines such as octopamine ( Crocker et al . , 2010 ) , which is the insect analog of norepinephrine , and dopamine are potent regulators of arousal in Drosophila ( Andretic et al . , 2005; Kume et al . , 2005; Lebestky et al . , 2009 ) . Distinct subpopulations of neurons of each type appear to independently modulate sleep and aggression . For octopamine , ASM neurons in the medial protocerebrum govern wake-promoting effects ( Crocker et al . , 2010 ) , while VUM cells in the anterior or posterior brain control social behavioral choice ( Certel et al . , 2010 , 2007 ) or fighting behaviors ( Zhou et al . , 2008 ) , respectively . Regarding dopamine , neurons projecting to the dorsal fan-shaped body control sleep ( Ueno et al . , 2012; Wu et al . , 2012 ) , and a number of distinct clusters modulate aggression , including PPM3 and T1 neurons ( Alekseyenko et al . , 2013 ) . In sum , both of these signaling systems would be well positioned to integrate information regarding internal state and environmental demand to optimize behavioral output at a given time . Sleep serves numerous vital functions , but aggression is also a critical behavior for acquisition of food , reproduction , and predator defense ( Anderson , 2012 ) . Disrupting sleep processes might impair the function of neuronal controls underlying aggression . Using Drosophila , we find that acute sleep deprivation strongly suppresses aggressive behaviors , while other social behaviors are unaffected . Reduced aggression occurs with different forms of sleep deprivation and is reversible with sufficient recovery sleep . Pharmacologic experiments reveal that an octopamine agonist specifically restores aggression in sleep-deprived flies , and we use genetic approaches to suggest that sleep loss itself , rather than manipulation of aggression neurons , is required for changes to aggression . Finally , we demonstrate that sleep deprived flies are at a disadvantage for reproductive success when competing against flies whose sleep has been unperturbed . To test how sleep deprivation affects aggressive behaviors in Drosophila , we focused on 4–7 day old Canton-S ( CS ) males in social isolation since shortly after eclosion . Flies were acutely sleep deprived for 12 hr overnight using mechanical stimulation , and aggression assays were performed the following morning , either by pairing flies within the same condition or pairing one control fly with a sleep-deprived fly ( ‘between conditions’ , for which males of each condition were tracked with a small dot of paint on the thorax ) . We found that acute sleep deprivation resulted in a profound suppression of aggression in both cases ( Figure 1A; Figure 1—source data 1 ) . In assays between deprived and non-deprived flies , non-deprived flies showed frequent lunges and were nearly always dominant over deprived flies ( Figure 1B ) . In contrast , flies that were sleep deprived overnight rarely lunged , even when attacked by a non-deprived fly , suggesting suppression of reactive aggression by sleep deprivation . In assays between 2 sleep-deprived flies , the males engaged in aggressive behaviors significantly less often than pairs of control flies ( Figure 1A; Videos 1 , 2 ) . The reduced aggression within condition did not simply derive from lack of social encounters , as sleep-deprived and control pairs spent similar amounts of time interacting during the beginning of the assay ( Figure 1—figure supplement 1 ) ; sleep-deprived flies could also be observed throughout the assay in close proximity without engaging in fighting behaviors , which rarely occurred in control flies . When lunging did occur , the latency to first lunge in pairs of sleep-deprived flies following initial social encounter was markedly increased compared to control flies ( Figure 1C ) , and dominance was rarely established . Together , these results suggest deficits in both proactive and reactive aggression , and demonstrate that sleep deprivation negatively impacts fighting behaviors . 10 . 7554/eLife . 07643 . 003Figure 1 . Acute sleep deprivation suppresses aggressive behaviors . ( A ) Quantification of aggression ( lunge count ) in control or sleep deprived male CS flies . ‘Between conditions’ indicates fights between a control fly and a sleep-deprived fly; ‘within condition’ is a fight between 2 sleep-deprived or 2 control flies . ‘Daytime deprivation’ and ‘recovery’ refers to fights between conditions ( n = 16 , 16 , 9 , 10 , 10 , 8 , 14 , 9 from left to right ) . ( B ) Percentage of flies in each condition establishing dominance ( fights between conditions ) . ( C ) Latency to first lunge following first social encounter ( fights are within condition; n = 9 control , 7 deprived ) . ( D ) Lunging follow sleep deprivation during the final 1 , 3 or 6 hr ( s ) of the night ( fights within condition; n = 8 , 8 , 13 , 9 , 12 , 12 from left to right ) . ( E ) Lunging following 6 hr of mechanical stimulation during the final 3 hr of night and first 3 hr of day ( fights within condition; n = 11 for both ) . ( F ) Recovery of aggression following prior sleep deprivation for 12 hr ( fights within condition; n = 10 , 10 , 18 , 17 , 19 , 18 , 15 , 12 from left to right ) . Box plots in this figure and all others represent median value ( horizontal line inside box ) , interquartile range ( height of the box , 50% of the data within this range ) , and minimum and maximum value ( whiskers ) . Bar graphs in this figure and all others are presented as mean ± s . e . m . p < 0 . 05; 1 way ANOVA with Tukey's ( A , D ) or Sidak's ( F ) post-hoc test; unpaired two-tailed Student's t-test ( C , E ) . SD = sleep deprivation . DOI: http://dx . doi . org/10 . 7554/eLife . 07643 . 00310 . 7554/eLife . 07643 . 004Figure 1—source data 1 . Quantification of lunging following mechanical sleep deprivation . DOI: http://dx . doi . org/10 . 7554/eLife . 07643 . 00410 . 7554/eLife . 07643 . 005Figure 1—figure supplement 1 . Quantification of percentage of time pairs of flies spent interacting , with or without prior sleep deprivation ( n = 12 pairs for control and deprived ) . Unpaired two-tailed Student's t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 07643 . 00510 . 7554/eLife . 07643 . 006Figure 1—figure supplement 2 . Sleep trace of CS male , showing high sleep amounts during both day and night ( n = 22 flies ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07643 . 00610 . 7554/eLife . 07643 . 007Figure 1—figure supplement 3 . Quantification of aggression following 12 hr of high temperature ( 31°C ) -induced sleep deprivation ( fights within condition at 25°C; n = 21 control , 28 deprived ) . p < 0 . 05; Unpaired two-tailed Student's t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 07643 . 00710 . 7554/eLife . 07643 . 008Figure 1—figure supplement 4 . Courtship index and copulation frequency of control or mechanically sleep-deprived males ( n = 15 control and deprived ) . Unpaired two-tailed Student's t-test ( courtship index ) ; Fischer's exact test ( copulation frequency ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07643 . 00810 . 7554/eLife . 07643 . 009Video 1 . Aggression assay between two control flies ( on regular food with a drop of yeast paste in the center ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07643 . 00910 . 7554/eLife . 07643 . 010Video 2 . Aggression assay between two sleep-deprived flies . DOI: http://dx . doi . org/10 . 7554/eLife . 07643 . 010 We next tested whether the link between sleep and aggression is specific to sleep loss during the night . The CS males used in our experiments show high sleep amounts during both day ( 450 . 77 ± 77 . 09 min ) and night ( 528 . 91 ± 109 . 74 min; Figure 1—figure supplement 2 ) , allowing us to test day and night sleep requirements . Mechanical sleep deprivation for 11 hr during the day with fighting assays performed during the final hour of the light period also reduced aggression ( Figure 1A ) and shifted dominance towards non-deprived flies ( Figure 1B ) , indicating that timing of sleep loss to day or night is not critical . How much sleep loss is required to affect subsequent fighting behaviors ? We mechanically sleep deprived flies for the final 1 , 3 , or 6 hr ( s ) of the night , followed by aggression assays the next morning . Focusing on assays within condition , we found that 6 hr of sleep loss resulted in suppression of fighting ( Figure 1D ) similar to 12 hr of sleep loss . However , sleep deprivation for the final 1 or 3 hr ( s ) of the night caused no reduction in fighting behavior ( Figure 1D ) , consistent with the timeframe of sleep loss required for learning deficits in flies ( Seugnet et al . , 2008 ) . To control for the possibility that mechanical stimulation/fatigue and not sleep loss per se is responsible for reduced aggression , we mechanically stimulated flies for a 6 hr period spanning the final 3 hr of night and first 3 hr of day . During this time , flies exhibit more wake in comparison to the final 6 hr of the night , only losing ∼3 hr of sleep ( sleep lost = 186 . 72 ± 61 . 03 in night/day transition; sleep during final 6 hr of night = 293 . 41 ± 61 . 49 ) . Continuous stimulation for 6 hr during the night/day transition period did not affect subsequent fighting ( Figure 1E ) , indicating that sleep deprivation and not mechanical stimulation is responsible for suppression of aggression . To rule out the possibility that acute sleep deprivation caused permanent impairment of aggressive behaviors , we examined lunging after 12 hr of overnight mechanical sleep deprivation followed by a 24 hr recovery period . Fighting returned to baseline levels with 1 day recovery after sleep loss ( Figure 1A ) , and dominance was equally likely for a previously sleep-deprived or non-deprived fly when paired against one another ( Figure 1B ) . Thus , neither sleep deprivation , nor the mechanical stimulation used to deprive , result in injury to the fly or long-lasting impairment . Next , we further investigated the timeframe for recovery following 12 hr of mechanical sleep deprivation . Deprived or control flies were assessed within condition at 1 , 3 , 6 , or 9 hr after overnight sleep deprivation ( Figure 1F ) . Following 1 and 3 hr of recovery sleep , aggression was still markedly suppressed . By 6 hr we observed recovery of fighting behaviors , with lunge counts no longer significantly reduced compared to the 6 hr control group . 9 hr after the end of sleep deprivation , levels of aggression remained indistinguishable from controls . Consistent with previous work ( Hoyer et al . , 2008 ) , we detected no differences in baseline aggression in control groups throughout the day ( Figure 1F ) . Do other methods of sleep deprivation similarly influence aggression ? We noted sleep in CS males to be exquisitely sensitive to high temperatures , with near total sleep loss when exposed to 31°C overnight . Temperature-dependent sleep deprivation also resulted in reduced lunging with aggression assays performed the following morning at 25°C ( Figure 1—figure supplement 3 ) , indicating that suppressed aggression is not specific to mechanical sleep deprivation . To examine whether sleep deprivation in the fly indiscriminately impairs other complex motor programs , we tested if courtship between a socially naive male and virgin female is affected . Consistent with previous work ( Seugnet et al . , 2011; Kayser et al . , 2014 ) , 12 hr of mechanical sleep deprivation during the night had no effect on courtship behavior the following morning as measured by courtship index or copulation frequency over a 10 min period ( Figure 1—figure supplement 4 ) . Thus , a distinct complex behavior remains intact following acute sleep loss , suggesting that like learning and memory , aggression is a specific behavior that is impaired following sleep deprivation . Octopamine and dopamine have been implicated in controlling both sleep/arousal and aggressive behaviors . We examined whether either of these monoamines play a role in coupling sleep deprivation to changes in aggression . First , we asked how thermogenetic sleep deprivation via overnight activation of octopaminergic or dopaminergic neurons affects next day fighting . The Drosophila thermosensitive cation channel dTrpA1 ( Hamada et al . , 2008 ) was expressed using Tdc2-GAL4 ( octopamine neurons ) or TH-GAL4 ( dopamine neurons ) , and flies were exposed to 29°C for 12 hr overnight . Activation of neurons using either GAL4 line resulted in near total sleep deprivation ( Figure 2—figure supplement 1 ) ; importantly , overnight exposure to 29°C in flies lacking either the GAL4 or UAS did not impact sleep ( Figure 2—figure supplement 2 ) , in contrast to CS males exposed to 31°C . Aggression was assessed the following morning at 23°C , below the threshold for TrpA1 activation . Sleep deprivation by activation of Tdc2+ or TH+ neurons caused a significant suppression of aggression the following day ( Figure 2A; Figure 2—source data 1 ) , while genetic and temperature controls with normal prior sleep showed no such effect ( Figure 2A; Figure 2—figure supplement 2 ) . Our results suggest that perturbation of octopamine and/or dopamine signaling as a result of sleep deprivation could impair aggressive behaviors . Using a pharmacologic approach we tested whether octopamine or dopamine agonists rescue reduced aggression . Flies were fed the octopamine agonist chlordimeform ( CDM ) or the dopamine agonist L-DOPA leading up to aggression assays that were preceded by thermogenetic sleep deprivation . CDM , but not L-DOPA , rescued aggression following sleep deprivation via activation of Tdc2+ neurons ( Figure 2B ) . Surprisingly , CDM also rescued aggression following sleep deprivation via activation of TH+ neurons , while L-DOPA did not ( Figure 2C ) . This effect is not specific to thermogenetic sleep deprivation , as CDM alone rescued aggression after mechanical sleep deprivation as well ( Figure 2—figure supplement 3 ) . These results suggest that aggression-relevant octopamine function is compromised downstream of sleep deprivation signals , regardless of method of sleep loss . 10 . 7554/eLife . 07643 . 011Figure 2 . Octopamine agonist CDM rescues reduced aggression following sleep deprivation . Quantification of aggression at 23°C in Tdc2-GAL4>UAS-TrpA1 or TH-GAL4>UAS-TrpA1 males that were thermogenetically sleep-deprived the prior night at 29°C ( SD , red bars ) or remained at 21°C ( Ctrl , white/gray bars ) ( n = 10 , 16 , 10 , 12 from left to right ) . ( B , C ) Rescue of suppressed aggression in Tdc2-GAL4>UAS-TrpA1 ( B ) or TH-GAL4>UAS-TrpA1 ( C ) males fed either CDM or L-DOPA and thermogenetically sleep-deprived ( B , n = 12 , 10 , 10; C , n = 12 , 12 , 12 from left to right ) . ( D ) Rescue of suppressed aggression in group-housed CS males fed either CDM or L-DOPA compared to males reared in isolation ( Control ) ( n = 9 , 9 , 12 , 12 from left to right ) . ( E ) Locomotor activity over 1 hr following exposure to 29°C the prior night ( n = 16 , 16 , 32 , 32 from left to right ) . All fights are within condition . p < 0 . 05; 1 way ANOVA with Sidak's ( A ) or Tukey's ( B–E ) post-hoc test . DOI: http://dx . doi . org/10 . 7554/eLife . 07643 . 01110 . 7554/eLife . 07643 . 012Figure 2—source data 1 . Quantification of lunging following thermogenetic sleep deprivation and pharmacologic rescue . DOI: http://dx . doi . org/10 . 7554/eLife . 07643 . 01210 . 7554/eLife . 07643 . 013Figure 2—figure supplement 1 . Sleep traces in TH-GAL4>UAS-TrpA1 ( n = 24 red , 20 black ) or Tdc2-GAL4>UAS-TrpA1 ( n = 10 red , 10 black ) flies loaded at ZT6 with temperature shift ( red trace ) to 29°C at ZT12 , compared to controls ( black trace ) remaining at 21°C . DOI: http://dx . doi . org/10 . 7554/eLife . 07643 . 01310 . 7554/eLife . 07643 . 014Figure 2—figure supplement 2 . ( Left ) Sleep traces in Tdc2-GAL4>UAS-TrpA1 ( black ) and GAL4 ( red ) and UAS ( blue ) controls loaded at ZT6 with temperature shift ( pink box ) to 29°C at ZT12 . ( Right ) Quantification of aggression at 23°C in GAL4 and UAS controls exposed to 29°C during the prior night ( n = 13 , 14 , 8 , 8 , 9 from left to right ) . p < 0 . 05; 1 way ANOVA with Tukey's post-hoc test . DOI: http://dx . doi . org/10 . 7554/eLife . 07643 . 01410 . 7554/eLife . 07643 . 015Figure 2—figure supplement 3 . Quantification of aggression in CS males fed either CDM or L-DOPA and mechanically sleep-deprived ( n = 17 , 20 , 24 from left to right ) . p < 0 . 05; 1 way ANOVA with Tukey's post-hoc test . DOI: http://dx . doi . org/10 . 7554/eLife . 07643 . 01510 . 7554/eLife . 07643 . 016Figure 2—figure supplement 4 . Quantification of aggression following 1 days or 3 days of drug exposure in isolated males . 1 way ANOVA with Tukey's post-hoc test . DOI: http://dx . doi . org/10 . 7554/eLife . 07643 . 016 To rule out the possibility that CDM is simply a more potent inducer of fighting behaviors , we investigated the effect of both drugs on baseline aggression and found no increase in fighting following either 1 or 3 days of drug feeding ( Figure 2—figure supplement 4 ) . Previous work has shown that social experience through group housing of flies also dramatically reduces aggression ( Wang et al . , 2008; Zhou et al . , 2008 ) , and we examined if a shared mechanism suppresses fighting in both cases . In contrast to sleep deprivation , reduced aggression secondary to group housing was similarly rescued either by CDM or L-DOPA ( Figure 2D ) , indicating that the drugs can function comparably to rescue aggression depending on the method of suppression . Moreover , while sleep deprivation and social experience both suppress aggression , they appear to do so through distinct though potentially overlapping mechanisms . Does sleep deprivation itself suppress aggression , or does reduced fighting stem from decreased locomotor activity after sleep loss ? Following thermogenetic sleep deprivation we quantified motor activity . As expected , sleep-deprived flies were less active than non-deprived controls ( Figure 2E ) , presumably due to increased homeostatic sleep drive . While only CDM rescued aggressive behaviors following sleep deprivation ( Figure 2B ) , we found that both CDM and L-DOPA rescued locomotor activity back to control levels ( Figure 2E ) . Together , these experiments dissociate motor activity from aggression following sleep deprivation , and indicate that sleep loss itself impairs aggressive behaviors . A group of 2–4 Tdc2+ neurons in the posterior fly brain within the VUM cluster near the suboesophageal ( SOG ) ganglion have been implicated in the role of octopamine modulation of aggression ( Zhou et al . , 2008 ) . These neurons can be targeted genetically using Tdc2-GAL4 with the Cha-GAL80 suppressor ( Figure 3A; Figure 3—source data 1 ) ( Zhou et al . , 2008 ) . Suppressed aggression following overnight activation of all Tdc2+ neurons with TrpA1 might be independent of sleep deprivation , and derive from hyperstimulation and subsequent quiescence of the Tdc2+ Cha- aggression cells . To rule out this possibility , we activated these neurons for 12 hr overnight at 29°C , followed by next day aggression assays at 23°C . Activation of Tdc2+ Cha- neurons did not cause sleep deprivation ( Figure 3B ) , consistent with research localizing the wake-promoting octopamine neurons to dorsal brain regions ( Crocker et al . , 2010 ) . Importantly , sustained activation of Tdc2+ Cha- neurons also did not impair aggression the following day ( Figure 3C ) . Together these results show that reduced aggression following overnight activation of Tdc2+ neurons is not caused by hyperstimulation of Tdc2+ Cha- neurons , and further suggest that sleep deprivation per se is required for suppression of aggression . 10 . 7554/eLife . 07643 . 017Figure 3 . Sleep loss is required for suppressed aggression following octopaminergic activation . ( A ) Images of Tdc2+ Cha- neurons in brains from Tdc2-GAL4;Cha-GAL80>UAS-CD8::GFP flies immunostained for GFP ( green ) and nc82 ( magenta ) . Arrows indicate 2–4 VUM neuron cluster in posterior brain . Scale bar = 100 µm . ( B ) Sleep traces in Tdc2-GAL4;Cha-GAL80>UAS-TrpA1 ( red ) , Tdc2-GAL4>UAS-TrpA1 ( black ) , and UAS-TrpA1;Cha-GAL80 ( blue ) flies with temperature shift ( pink box ) to 29°C at ZT12 ( n = 12 flies for all conditions ) . ( C ) Quantification of aggression in Tdc2-GAL4>UAS-TrpA1 males exposed to elevated temperature ( and sleep deprived ) overnight compared to Tdc2-GAL4;Cha-GAL80>UAS-TrpA1 males , and temperature controls ( fights within condition; n = 14 , 15 , 16 , 15 flies , from left to right ) . p < 0 . 05; 1 way ANOVA with Sidak's post-hoc test . DOI: http://dx . doi . org/10 . 7554/eLife . 07643 . 01710 . 7554/eLife . 07643 . 018Figure 3—source data 1 . Quantification of lunging following thermogenetic activation of Tdc2+Cha- neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 07643 . 018 In addition to a role in promoting aggression , octopamine also modulates the choice between courtship and aggression in males ( Certel et al . , 2010 , 2007 ) . A distinct subpopulation of VUM octopamine neurons that co-express the male variant of the sex determination factor Fruitless ( Fru ( M ) ) in the anterior SOG have been implicated in this behavioral choice ( Certel et al . , 2010 , 2007 ) . We assessed whether sleep deprivation might influence social behavior decision-making in addition to aggression itself . Either 2 sleep-deprived or 2 control male CS flies ( 4–7 days old; reared in isolation ) were placed with a virgin CS female , and we determined percent time spent courting the female in comparison to the other male . Previous work has shown that disturbing the function of Tdc2+ Fru ( M ) + neurons results in abnormal elevation of male–male courtship , whereas near exclusive male-female courtship with intermale aggression is found under normal conditions ( Certel et al . , 2010 , 2007 ) . Following mechanical sleep deprivation overnight , we found that flies overwhelmingly courted females , with no difference in the minimal time engaged in male–male courtship compared to pairs of non-deprived control males ( Figure 4A , B; Figure 4—source data 1 ) . Thus social behavioral choice is normal between pairs of sleep-deprived flies . 10 . 7554/eLife . 07643 . 019Figure 4 . Suppressed aggression following sleep deprivation impairs reproductive fitness . Courtship index towards the female target ( A ) or male ( B ) in control ( white , n = 32 ) or sleep-deprived ( gray , n = 31 ) flies during a competitive courtship assay ( within condition ) . ( C ) Percentage of time spent in chains/clusters during the competitive courtship assay with control ( n = 29 ) or sleep-deprived ( n = 30 ) flies ( within condition ) . ( D ) Percentage of assays in which the control or sleep-deprived male first copulates with ( ‘wins’ ) the female target in a competitive courtship assay between conditions ( n = 116 assays , 5 independent experiments ) . ( E ) As in ( D ) but with control vs sleep-deprived ± drug rescue males ( control vs SD + no drug , n = 20 assays; control vs SD + L-DOPA , n = 28 assays; control vs SD + CDM n = 27 assays; 3 independent experiments ) . *p < 0 . 01 , p < 0 . 05; Unpaired two-tailed Student's t-test ( A–C ) or two-tailed Binomial test ( D , E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07643 . 01910 . 7554/eLife . 07643 . 020Figure 4—source data 1 . Courtship and competitive copulations measures following sleep deprivation . DOI: http://dx . doi . org/10 . 7554/eLife . 07643 . 02010 . 7554/eLife . 07643 . 021Figure 4—figure supplement 1 . Percentage of assays in which the control or drug condition male first copulates with the female target in a competitive courtship assay between conditions ( control vs L-DOPA , n = 30 assays; control vs CDM n = 30 assays; 3 independent experiments ) . p = 0 . 36 for control vs L-DOPA , p = 0 . 86 for control vs CDM; two-tailed Binomial test . DOI: http://dx . doi . org/10 . 7554/eLife . 07643 . 021 In the behavioral choice assays , we also noticed what appeared to be decreased competition for the female target between sleep-deprived flies in comparison to non-deprived control conditions . Non-deprived flies were frequently observed to interrupt courtship activities of the other male , and at times , engage in fighting behaviors away from the female altogether . These behaviors were less frequently observed between sleep-deprived males . To quantify the reduced male–male competition for a female mate in sleep-deprived flies , we determined the percent of time that the 2 males and 1 female were observed clustered together or in a ‘chain’ formation , with both males actively courting the female concurrently . During these periods , males were clearly observed to engage in low intensity fighting behaviors like fencing , which sometimes escalated to brief high intensity tussling or lunging . We found that in non-deprived flies , 26% of time in the assay was spent in chains or clusters; this was reduced to 17 . 2% in pairs of sleep-deprived males ( Figure 4C; Videos 3 , 4 ) , suggesting that competitive social interactions are reduced following sleep deprivation . Taken together , our results demonstrate that while social behavioral choice is unaffected , impairments in aggression influence how males competitively court a female . 10 . 7554/eLife . 07643 . 022Video 3 . Competitive courtship assay between 2 control males ( white or red painted dot ) and a virgin female . DOI: http://dx . doi . org/10 . 7554/eLife . 07643 . 02210 . 7554/eLife . 07643 . 023Video 4 . Competitive courtship assay between 2 sleep-deprived males ( white or red painted dot ) and a virgin female . DOI: http://dx . doi . org/10 . 7554/eLife . 07643 . 023 Success in aggressive interactions has been correlated with fitness for mating ( Dow and von Schilcher , 1975 ) ( though not in flies bred to be hyper-aggressive [Dierick and Greenspan , 2006; Penn et al . , 2010] ) . Does reduced aggression following sleep deprivation have a functional consequence on reproductive fitness in a competitive environment ? We paired control and mechanically sleep-deprived CS males together with a virgin female and assayed copulation success . Males of each condition were tracked with a dot of paint on the thorax . In these assays , we found that control flies were more likely to ‘win’ copulation with the female compared to sleep deprived flies ( ∼63% vs 37%; Figure 4D ) . Thermogenetic sleep deprivation via activation of Tdc2+ neurons overnight similarly impaired next-day performance in the competitive copulation assay against a genetically identical male ( Figure 4E ) . We then tested whether competitive copulation deficits can be reversed with rescue of aggression after sleep deprivation . Tdc2-GAL4>UAS-TrpA1 males were fed CDM or L-DOPA and sleep deprived overnight by exposure to 29°C . Competitive copulation assays were performed the following morning at 23°C against non-deprived Tdc2-GAL4>UAS-TrpA1 males that received no drug . Consistent with rescue of aggressive behaviors , we found that CDM restored copulation success of sleep-deprived males competing against controls , while L-DOPA had no effect ( Figure 4E ) . Neither drug had a significant effect in assays against controls ( no drug ) in the absence of sleep manipulations ( Figure 4—figure supplement 1 ) . These results suggest that suppression of aggression following sleep loss is deleterious when competing with a non-deprived male for a reproductive partner . Here we show that sleep deprivation reduces aggression in Drosophila while other complex social behaviors are unimpaired . The suppression of aggression is independent of the method used to induce sleep loss and reversible with sufficient recovery sleep . We provide evidence that octopamine plays a role in mediating reduced aggression levels following sleep loss , and we find using competitive courtship assays that sleep deprivation places flies at a disadvantage for reproductive success due to insufficient aggression towards competing males . These results demonstrate that aggression is a specific social behavior that becomes dysregulated by sleep deprivation . Subjective experience in humans indicates that irritability increases with insufficient sleep , which is supported by validated measures of emotional reactivity and mood ( Banks and Dinges , 2007 ) ; however , irritability and aggression are likely dissociable in higher organisms . While there are well-validated measures of aggression as a lifetime personality trait , tools to measure changes to aggression in humans over relatively short-time periods ( hours to days ) are less well-established , as is measuring irritability in animal models independent of overt behavioral expression . Most human studies of the interaction between sleep and aggression have been correlational and utilized self-reported measures ( Kamphuis et al . , 2012 ) . Nonetheless , more recent human work is informative regarding dissociation of irritability and/or impulsivity from aggression itself: research supports a reduced capacity to inhibit impulsivity to negative stimuli after sleep deprivation ( Anderson and Platten , 2011 ) , while a study designed to measure reactive aggression in sleep-deprived men reported diminished aggression in response to provocation ( Cote et al . , 2013 ) . The rodent literature supports the contrasting hypothesis that sleep insufficiency leads to increased aggression ( Kamphuis et al . , 2012 ) . These differences might result from confounds such as the impact of sustained activity or stress rather than sleep deprivation per se , or REM sleep loss vs total sleep loss . Regardless , while the repertoire of emotional experience and output in humans is far more complex than in a laboratory animal , the central effect of sleep deprivation on aggression appears to be conserved between flies and humans . Segregating stress effects from sleep loss itself when sleep-depriving animals—including humans—presents obstacles . Using multiple types of sleep deprivation stimuli combined with genetic manipulations , our data suggest that the impact of sleep deprivation on aggression is causally related to sleep loss . Nonetheless , experimental paradigms that prevent sleep in the face of heightened sleep need are inherently stressful to an animal . The specificity with which aggressive social interactions are impaired as opposed to all social behaviors serves as a strong indicator that the effects we observe are not simply part of a broader stress response; moreover , other stressors are known to negatively impact courtship behaviors ( Patton and Krebs , 2001; Christie et al . , 2013 ) , suggesting that courtship is not uniquely immune to stress-related impairment . Aggression in Drosophila has long been appreciated to play a role in mate selection ( Sturtevant , 1915; Dow and von Schilcher , 1975 ) . Male flies fight more in the presence of a female , and recent work has shown that prior exposure to a female suppresses this effect ( Yuan et al . , 2014 ) ; the aforementioned study focused on aggressive behaviors occurring after male flies copulate with females in the arena , suggesting that aggression may function both to obtain a reproductive partner and possibly guard that partner after mating ( Yuan et al . , 2014 ) . Our data indicate that even in the setting of sleep deprivation , male flies will successfully court a female partner in the absence of competition . However , when competing for a female mate against a non-deprived control male , sleep deprivation impairs sexual fitness , likely due to reduced aggression towards the other male: pharmacologic restoration of aggression with an octopamine agonist likewise rescues sexual fitness . Importantly , reduced performance in the competitive copulation assay occurs in males that were isolated since eclosion , eliminating the possibility that previously-established hierarchical cues are involved and emphasizing the crucial role of sleep for normal function . Sleep loss is required for subsequent suppressed aggression , while prolonged prior excitation of aggression-relevant neurons has no effect on fighting the following day . How is sleep need conveyed to aggression loci ? Our results implicate octopaminergic signaling in this process , and suggest that neurons involved in setting aggression levels , but not male choice between courtship and aggressive behaviors , are specifically impacted: while intermale aggression is reduced with sleep deprivation , intermale courtship is not , by default , increased . Norepinephrine , the mammalian analog of insect octopamine , has a critical role in aggression in vertebrates; noradrenergic neurons are also important regulators of sleep-wake transitions . Identification of specific cellular subpopulations whose output is altered by sleep deprivation in Drosophila will enable mapping the neural circuits that relay sleep information to aggression centers , as well as investigation of the molecular signals that control deprivation-dependent changes to octopamine aggression neurons . The conserved cellular substrates of these behaviors between flies and mammals suggest such findings will be of relevance to understanding the neurobiological basis of aggression and impairment after sleep loss . In sum , our results indicate that two innate behaviors—sleep and aggression—are coupled , and suggest that molecular signals generated by sleep deprivation might be potential targets for selective modification of aggressive behaviors . TH-GAL4 , Tdc2-GAL4 , Cha-GAL80 are from laboratory stocks , and were outcrossed 8× into a w1118;CS background ( gift from D Anderson ) . Wild-type CS flies are a gift from E . Kravitz . UAS-dTrpA1 ( in w1118;CS background ) are a gift from D Anderson . Flies were maintained in bottles on standard food at 25° on a 12 hr:12 hr LD cycle . For sleep experiments , flies were loaded into glass tubes containing 5% sucrose and 2% agar at ∼ ZT6-8 by gentle aspiration ( if being used for a behavioral assay the following day ) . Locomotor activity was monitored using the Drosophila Activity Monitoring ( DAM ) system ( Trikinetics , Waltham MA ) . Activity was measured in 1 min bins and sleep was identified as 5 min of consolidated inactivity ( Hendricks et al . , 2000; Shaw et al . , 2000 ) . Data was processed using PySolo ( Gilestro and Cirelli , 2009 ) . Mechanical sleep deprivation was accomplished using a Trikinetics vortexer mounting plate , with shaking of monitors for 2 s randomly within every 20 s window for 12 hr during the night . Temperature-dependent sleep deprivation in CS males was at 31°C during the night; thermogenetic sleep deprivation was at 29°C . For Figure 2E , locomotor activity was quantified as total number of beam breaks from ZT1-2 . Flies were moved into isolation tubes ( 5 ml tubes , Falcon 352002 ) shortly after eclosion unless otherwise specified . Following sleep assays in DAM tubes , flies recovered for 30 min in isolation on regular food in isolation tubes , and then were moved to aggression arenas by gentle aspiration . Aggression assays were performed and scored as previously described ( Chen et al . , 2002; Certel and Kravitz , 2012 ) with minor modifications . Assays consisted of fights between 2 males in 1 well of a 12 well plate with a food cup in the center . Yeast paste or a buried headless female were placed in the center of the cup , which was well lit . The sides of each arena were coated in Fluon ( Bioquip Products , Rancho Dominguez , CA ) and lid with Rain-X ( ITW Global Brands , Houston , TX ) . Fights were recorded from above using a video camera ( Sony HDR-CX210 ) and lunges scored manually , blind to condition . For fights between conditions ( and all competitive copulation experiments ) male flies were labeled with a small dot of acrylic paint at least 24 hr prior to the assay; the paint color for each condition was randomized between experiments . Lunge count for ‘between condition’ assays ( control vs SD ) was quantified separately for each fly of the designated condition; lunge count ‘within condition’ was quantified as the combined number of lunges from both flies in the assay . Dominance ( scored only in ‘between condition’ assays ) was determined as repeated lunges by one fly followed by retreat of the other to the edge of the cup or off the cup altogether ( Alekseyenko et al . , 2014 ) . All assays were performed for 30 min at 25°C and 40–60% humidity unless otherwise specified . For Figure 1A , social interaction was defined as time spent with the flies within 1 body length of one another . Interaction was quantified during the first 5 min of the assay or until first lunge , whichever occurred first . For pharmacologic rescue experiments , flies were fed chlordimeform ( CDM; 0 . 05 mg/ml; Sigma ) or L-DOPA ( 3 mg/ml; R&D Systems ) mixed into 5% sucrose and 2% agar beginning ∼4 hr prior to lights off , and continued to feed on the drug throughout the night ( during 12 hr of thermogenetic sleep deprivation or mechanical sleep deprivation during the final 6 hr of night ) . At ZT0 ( end of sleep deprivation ) , flies recovered for 1 hr on drug/control mixed into regular food at the same concentration prior to aggression assays at ZT1 . For group-housing experiments , males were raised in groups of 10 flies for 4–5 days , then isolated on drug/control for 20 hr prior to assay at ZT1 . Virgin male flies were collected shortly after eclosion and housed in isolation . Female CS virgins ( 3–7 days post eclosion ) were used in all courtship assays . Following sleep assays , flies recovered for 30 min in isolation on regular food , and then a male and female were gently aspirated into a well-lit porcelain mating chamber ( 25 mm diameter and 10 mm depth ) covered with a glass slide . Courtship index ( CI ) was determined as the percentage of total amount of time a male was engaged in courtship activity during a period of 10 min or until successful copulation . Copulation frequency was calculated as percentage of flies in each condition that successfully copulated during the 10 min assay . For ‘within condition’ competitive courtship assays ( Figure 4A–C ) , 2 males and 1 virgin female were loaded into 1 well of a 12 well plate containing 5 ml of food ( Certel et al . , 2010 , 2007 ) . Courtship and copulation measures were determined for the male first demonstrating sustained courtship ( >20 s ) towards the female . Assays were recorded and scored blind to experimental condition . For competitive courtship assays between conditions , one male fly of each condition ( control or deprived ) and 1 virgin female were aspirated into a well , and the male that first copulated with the female was determined the ‘winner’ . Experiments were scored blind to condition . Brains were dissected in PBS , fixed in 4%PFA for 30 min at room temperature , and cleaned of remaining tissue in 0 . 3% PBST . Following 3 × 10 min washes in PBST , brains were blocked in 5% normal donkey serum ( NDS ) and incubated with primary antibody in block at 4° overnight . Following 3 × 10 min washes in PBST , brains were incubated with secondary antibody in block for 2 hr at room temperature . Following 3 × 10 min washes in PBST , brains were mounted in vectashield . Primary antibodies included: Mouse anti-nc82 ( 1:1000 , Developmental Studies Hybridoma Bank ) , Rabbit anti-GFP ( 1:1000 , Molecular Probes ) . Secondary antibodies included: FITC donkey anti-rabbit ( 1:500 , Jackson ) , Cy5 donkey anti-mouse ( 1:500 , Jackson ) . Brains were visualized with a TCS SP5 confocal microscope and images processed in NIH ImageJ . All analysis was done in GraphPad ( Prism ) . Individual tests and significance are detailed in figure legends .	We know from personal experience that sleepless nights can change the way we behave , sometimes making us more irritable and less adept at social interactions . However , it can be difficult to establish cause and effect: does a lack of sleep lead to altered behavior , or vice versa ? The fruit fly , Drosophila melanogaster , is a popular model organism for studying questions like this because its neural circuitry is relatively well understood . To explore the effects of lack of sleep on social behaviors , and in particular on aggression , Kayser et al . disrupted the sleep of male fruit flies using various techniques , such as shaking them during the night , and then observed how they behaved . The experiments revealed that sleep-deprived flies were less aggressive than flies with undisturbed sleep . Furthermore , sleep-deprived male flies were less successful at mating with female flies when they were in direct competition with a rested male fly . Normal behavior was restored when the sleep-deprived flies were allowed to recover lost sleep for as little as six hours before the next aggression assay . To investigate how sleep loss leads to a decrease in aggressive behavior , Kayser et al . used different drugs to treat the sleep-deprived flies . A drug activating the equivalent of the noradrenergic system in flies helped them to recover normal fighting behaviors despite a lack of sleep . In mammals , noradrenaline is a chemical that affects heart rate , sleep-wake patterns , aggression and a number of other phenomena . Although aggressive behavior is often perceived as negative in humans , it can be important for survival . Human brains and behaviors are obviously more complex than those of Drosophila . However , learning more about the neuronal circuits that control sleep and social behavior in fruit flies may lead to an improved understanding of these phenomena in humans and , in the longer term , the development of drugs that can influence or modulate aggressive behaviors .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Sleep deprivation suppresses aggression in Drosophila
Neurons lose intrinsic axon regenerative ability with maturation , but the mechanism remains unclear . Using an in-vitro laser axotomy model , we show a progressive decline in the ability of cut CNS axons to form a new growth cone and then elongate . Failure of regeneration was associated with increased retraction after axotomy . Transportation into axons becomes selective with maturation; we hypothesized that selective exclusion of molecules needed for growth may contribute to regeneration decline . With neuronal maturity rab11 vesicles ( which carry many molecules involved in axon growth ) became selectively targeted to the somatodendritic compartment and excluded from axons by predominant retrograde transport However , on overexpression rab11 was mistrafficked into proximal axons , and these axons showed less retraction and enhanced regeneration after axotomy . These results suggest that the decline of intrinsic axon regenerative ability is associated with selective exclusion of key molecules , and that manipulation of transport can enhance regeneration . Axon regeneration fails in the adult mammalian CNS due to a combination of extrinsic inhibitory cues and an inadequate intrinsic regenerative response ( Fawcett et al . , 2012; Liu et al . , 2011 ) . Long-distance regeneration can only be achieved if axons have a high intrinsic growth ability ( Liu et al . , 2010; Cheah et al . , 2016 ) . Embryonic axons have this ability , and can elongate for long distances if immature neurons are transplanted into the adult CNS environment ( Lu et al . , 2012; Reier et al . , 1986 ) . However with maturation , adult CNS neurons lose much of this intrinsic regeneration ability , and axons such as those of the corticospinal tract show limited growth even in a permissive environment such as a peripheral nerve graft ( Bradke et al . , 2012; Geoffroy et al . , 2016 ) ( Richardson et al . , 1984 ) . Several changes occur during neuronal differentiation that might explain this maturational reduction in growth ability . Amongst these a key factor that changes radically with maturity is the establishment of selective transport to axons and dendrites ( Bentley and Banker , 2016; Britt et al . , 2016; Franssen et al . , 2015; Maeder et al . , 2014; Petersen et al . , 2014 ) . Polarised transport of proteins is required for the correct molecules to travel to pre-and postsynaptic sites and in order for axons and dendrites to possess different structures and functions ( Britt et al . , 2016; Maeder et al . , 2014 ) . A consequence of polarised transport is that several key growth-related molecules including integrins , trkB and IGF receptor become excluded from cortical axons as they mature ( Andrews et al . , 2016; Franssen et al . , 2015; Hollis et al . , 2009a; Hollis et al . , 2009b ) . This development of polarised transport therefore provides a possible mechanism for the maturational loss of intrinsic regeneration ability . Our previous work has focused on integrins as an example of a molecule necessary for efficient regeneration and growth ( Andrews et al . , 2009; Tan et al . , 2011; Cheah et al . , 2016 ) , and has shown that these receptors are progressively excluded from cortical axons to become exclusively somatodendritic both in vitro and in vivo ( Franssen et al . , 2015; Andrews et al . , 2016 ) . Conversely , sensory and retinal ganglion cell neurons , which can be more easily manipulated to make them regenerate , continue to transport integrins into their axons into adulthood ( Andrews et al . , 2016 ) . Integrins are transported into axons in recycling endosomes marked by the small GTPases rab11 and ARF6 ( Eva et al . , 2010; Eva et al . , 2012; Franssen et al . , 2015 ) . Rab11 is implicated in the trafficking of a number of growth associated molecules , including Trks A and B ( Ascaño et al . , 2009 ) . Integrins , Trks and rab11 are all excluded from cortical CNS axons in-vivo ( Andrews et al . , 2016; Hollis et al . , 2009a; Sheehan et al . , 1996 ) . Rab11 is involved in axon elongation in young CNS and adult PNS neurons regenerating in vitro , and is required for correct growth cone function as its targeted removal leads to growth cone collapse ( van Bergeijk et al . , 2015; Eva et al . , 2010 ) . These findings led us to ask whether selective exclusion from axons of the rab11 vesicles that transport these growth-related molecules contributes to their maturational loss of intrinsic regenerative ability , and whether replenishment and/or modulation of activation of rab11 can enable regeneration . Previous studies of selective transport have used an in vitro model in which cortical neurons mature and exclude integrins from their axons ( Franssen et al . , 2015; Petersen et al . , 2014 ) . We further developed this culture model for regeneration studies , and demonstrated that axons in these cultures lose the ability to regenerate with maturity , and that this loss is intrinsic to the axons rather than due to environment . We found that the characteristics of retraction bulb formation define the probability of regenerating a growth cone , but that subsequent axon elongation is controlled differently . The behaviour of rab11 was then investigated: we find that it becomes excluded from axons as neurons mature in-vitro , but that restoring the presence of rab11 in axons modifies retraction and enhances regeneration of mature axons . Embryonic CNS neurons can differentiate in culture and provide a model for maturation-related changes ( Barbati et al . , 2013 ) . We asked whether this type of culture could also be used to model the developmental loss of intrinsic regenerative capacity . Dissociated embryonic day 18 ( E18 ) rat cortical neurons were grown in the presence of astrocyte feeder cultures ( Kaech and Banker , 2006 ) . Neuronal maturity was tracked by examining the electrical properties of neurons , their pattern of gene expression through mRNA sequencing , and immunolabelling of cytoskeletal maturity markers . Electrophysiological maturation was examined by patch clamping neurons at 4 , 8 , 16 , and 24 days in vitro ( DIV ) ( Figure 1 , Figure 1—figure supplement 1 ) . During time in culture , resting membrane potential lowered to −55 mV ( Figure 1A ) and membrane capacitance increased ( Figure 1B ) . Only 29 . 4 ± 10 . 6% of neurons were able to fire action potentials in response to steps of depolarizing current at 4 DIV , but the percentage increased to 100% by 24 DIV ( Figure 1C , E ) . Concurrently the action potential spikes turned from single to multiple ( Figure 1E ) , increased in amplitude ( Figure 1—figure supplement 1A ) , decreased in duration ( Figure 1—figure supplement 1B ) and spike threshold increased ( Figure 1—figure supplement 1D ) while input resistance decreased ( Figure 1—figure supplement 1E ) . All of these are changes associated with the maturation of neurons in vivo . Interestingly , the main statistical difference for these parameters was observed between 8 DIV and 16 DIV . As for network development , compatible with a previous study ( Kay et al . , 2011 ) , co-localization of both excitatory and inhibitory synaptic markers was observable by 10DIV ( Figure 1—figure supplement 1F ) . However , spontaneous activity increased between 16 DIV and 24 DIV ( Figure 1D ) . RNA sequencing showed progressive changes in many genes towards expression patterns typical of mature neurons . From the mRNA expression data we picked two useful maturity markers which are cytoskeletal-related molecules and show robust immunostaining: doublecortin and low molecular weight neurofilament . Immunostaining showed a decrease in doublecortin and an increase in low molecular weight neurofilament with neuronal maturation similar to the changes in their mRNA levels ( Figure 1—figure supplement 2A , B , C ) . Ingenuity pathway analysis of genes with increasing expression showed many changes in molecules involved in synapse formation ( Figure 1—figure supplement 2D ) , whereas many of the genes with decreasing expression were involved in neuronal development ( Figure 1—figure supplement 2E ) . Full results are available on NCBI GEO DataSets ( RRID:SCR_005012 , accession no: GSE92856 ) . Neurons started to develop axon initial segments visualized by immunostaining for neurofascin and Ankyrin-G as early as 3 DIV , and the structure consolidated with increasing DIV ( Figure 1—figure supplement 3A ) . There was little neuronal loss in the cultures up to DIV24 , and neurons showed no cleaved caspase 3 expression after 3 weeks in culture ( Figure 1—figure supplement 3B , C ) . These results demonstrate that neurons in this culture model show maturational changes similar to those of neurons in vivo and remain viable . In vitro laser axotomy was performed at 4 , 16 and 24 DIV and the events that follow axotomy were recorded . To distinguish individual axons from the large number of surrounding processes , small numbers of neurons were transfected with GFP and before axotomy fluorescent live neurofascin staining of the axon initial segment ( AIS ) was used in all cases to identify axons from dendrites ( Figure 1—figure supplement 3A ) . Events after axotomy unfold over several hours for CNS axons ( compared with sensory axons which usually regenerate within one hour ) . Cultures were therefore imaged using time-lapse microscopy for 10 ~ 20 hr . Axotomy was followed by either neuronal death , branch loss or retraction bulb formation ( Figure 2A , B ) . Neurons were considered dead if axon disintegration or cell disruption was seen within 10 hr of axotomy , and were excluded from further analysis: this occurred in 15 ~ 25% of axotomies with no variation with DIV ( Figure 2B , C ) . Branch loss was the retraction of the cut axon branch to the nearest branch point with no formation of a retraction bulb: this occurred in 10 ~ 15% of axotomies ( Figure 2B , C ) which were also excluded from further analysis . Retraction bulb formation was the commonest result of axotomy . Starting immediately after axotomy , the GFP signal within the axon disappeared proximal to the cut for up to 1000 µm . Following this , the cytosol would reflux up to the newly sealed axon tip and start accumulating to form a retraction bulb which was usually motile; this usually took from 1 to 3 hr ( see below ) ( Figure 2A , C ) . The appearance of the retraction bulb under fluorescence imaging and phase contrast were identical . After successful formation of a retraction bulb three outcomes were seen ( Figure 3 ) ; regeneration of a new axon from the bulb ( usually following a different track ) ( Figure 3A , B ) , failure of regeneration ( although the retraction bulbs were usually motile ) , or ectopic growth ( formation of a new axon branch from the side of the cut axon , usually within 100 µm of the bulb ) ( Figure 3B ) . The probability of these events changed radically with neuronal maturity; in 3–5 day old neurons over 70% showed axon regeneration , but by 23–30 days less than 10% regenerated ( Figure 3C ) . Typical videos of successful regeneration in a DIV4 neuron and failure of regeneration in a DIV30 neuron are shown in ( Videos 1 and 2 ) . In the further analysis we have combined the results of regeneration and ectopic growth into an overall regeneration score . To describe regeneration more precisely , we also measured six regenerative factors; retraction distance , retraction bulb formation time , regeneration ratio , regeneration initiation time , regeneration length , and growth cone area and these were measured for proximal and distal axotomies . ( measures defined in Table 1 ) . The dynamics of retraction bulb formation varied considerably with maturity . The retraction distance increased with DIV ( Figure 4 , Figure 4—figure supplement 1A ) and the time taken to form a retraction bulb increased from 1 . 39 ± 0 . 21 hr in 4 DIV neurons to 2 . 68 ± 0 . 20 hr in 24 DIV neurons ( p<0 . 0001 , Figure 4—figure supplement 1B ) , correlating with retraction distance ( Figure 4—figure supplement 1C ) . The retraction distance had a log Gaussian distribution and we used a log10 conversion for statistical analysis , finding an increase and change in distribution between 4 DIV and 24 DIV ( log10 4 DIV: 1 . 65 ± 0 . 06 , 24 DIV: 2 . 16 ± 0 . 03 , p<0 . 0001 , Figure 4A . The same data are plotted on a non-log scale in Figure 4—figure supplement 1A–E ) . Interestingly a biphasic distribution with long and short retractors could be seen at 16 DIV , suggesting a transition period , and by 24DIV all axons were long retractors . We analysed these long ( ≧ 70 µm ) and short ( <70 µm ) retracting neurons separately in our further analysis of retraction distance and regeneration . For retraction distances , in 4 DIV neurons and short retracting 16 DIV neurons the position of axotomy was not a factor; short retracting axons were seen after both proximal and distal axotomy ( Figure 4B ) . However the in long retracting 16 DIV neurons and all 24 DIV neurons , distal axotomy led to a greater retraction distance on average than proximal axotomy ( >600 µm for 16 DIV and >400 µm for 24 DIV ) ( Figure 4B Figure 4—figure supplement 1E ) . We next focused on regeneration . The DIV of neurons had a powerful negative impact on regeneration , with only 8% of 24 DIV neurons that formed a retraction bulb regenerating compared to 63% in the 4 DIV group ( p<0 . 0001 , Figures 3C and 4C ) . Longer retracting neurons showed a general tendency for less regeneration ( Figure 4G ) , consistent with a previous in vivo result ( Canty et al . , 2013 ) . Therefore , we categorized axons according to short or long retraction , and by proximal or distal axotomy ( Figure 4C ) . Distal axotomy led to less regeneration in the long retraction group of 16 DIV neurons ( proximal 64% vs distal 11% , p=0 . 0002 ) and in 24 DIV neurons which all showed long retraction ( proximal 24% vs distal 2% , p=0 . 0116 ) , but distal axotomy did not affect regeneration success in the short retraction groups ( p=0 . 0325 ) ( Figure 4C ) . In summary , with increasing DIV retraction distance increased and so did regeneration failure , particularly in axons that showed long retraction and after distal axotomy . We examined other growth cone behaviour measurements related to regeneration , including regeneration initiation time which increased with DIV and growth cone area which declined ( Figure 4D , F ) . The length of the regenerated axon 2 hr after regeneration initiation also declined with DIV ( Figure 4E ) . These factors mainly changed between 4 and 16 DIV , and within the 16 DIV axons , retraction and axotomy distance did not influence these measures ( Figure 4—figure supplement 1F , G , H ) . Taken together , it can be concluded that ( 1 ) long retraction is prognostic for poorer regeneration initiation , and ( 2 ) the neuron’s ability to successfully initiate regeneration does not dictate how rapidly the axon will then elongate . Having established a model which shows regenerative decline with DIV , we asked whether this was due to a build-up of inhibition in the culture environment or to an intrinsic loss of regeneration ability in the axons . To address this , we plated newly harvested immature neurons transfected with GFP onto mature 21 DIV cultures without changing the medium , and allowed them to grow for 4 days , resulting in a 4 DIV neuron surrounded by a 25 DIV environment ( Figure 5 ) . Interestingly , 4 DIV neurons on 25 DIV cultures had longer and more complex axons than 4 DIV neurons on PDL ( Figure 5A–C ) . When axotomized , 4 DIV neurons on 25 DIV cultures behaved very similarly to 4 DIV neurons on PDL , with short retraction ( Figure 5D ) , and high regeneration success ( 54% for 4 DIV neurons on PDL , 46% for 4 DIV neurons on 25 DIV cultures , and 2% for 24 DIV neurons , p<0 . 0001 , Figure 5E ) . These results demonstrate that maturation of the culture environment does not explain the observed regenerative decline with DIV . Having achieved an in vitro model in which to study the loss of intrinsic axon regeneration with maturation , we explored potential mechanisms . Based on our previous work , our hypothesis was based on the selective axon transportation that develops with maturity , directing some molecules to dendrites others to axons . We asked whether this process leads to exclusion of molecules from the axon that are necessary for regeneration . We focused on rab11 because it is associated with the recycling endosomes responsible for bringing growth receptors and integrins into axons , and because these endosomes are reported to be restricted to a somatodendritic distribution in-vivo ( Sheehan et al . , 1996; Eva et al . , 2010; 2012 ) . The distribution of endogenous rab11 was studied by immunostaining , which revealed that at 4 DIV rab11 is present equally in both axons and dendrites , but becomes exclusively somatodendritic by 16 DIV ( axon stem/dendrite ratio: 4 DIV 2 . 05 ± 0 . 21 , 16 DIV 0 . 23 ± 0 . 03 , p=0 . 0001 , Figure 6A , B ) . The immunofluorescence intensity of rab11 in the cell bodies did not change between DIVs ( Figure 6C ) , and neither did the mRNA level ( Figure 1—figure supplement 2F ) , whereas the axon stem/cell body ratio decreased and dendrite/cell body ratio increased between 4 DIV and 16 DIV ( Figure 4B , D ) . Collectively these results show that rab11 becomes progressively excluded from axons but not dendrites as neurons mature . In order to understand the transport dynamics that lead to the selective distribution of rab11 , we examined rab11 trafficking in axons and dendrites at 16 DIV using transfection of GFP-tagged rab11a . Previous studies have demonstrated different trafficking mechanisms for the active and inactive forms of rab11 ( Welz et al . , 2014 ) , so tagged wild type ( WT ) , dominant negative ( DN ) , and constitutively active ( CA ) forms of rab11a were transfected . Overexpression of all forms caused a degree of mis-trafficking in neurons , leading to some rab11 being found in proximal axons and an increase in the proximal axon/dendrite ratio for total rab11 ( GFP 0 . 62 ± 0 . 05 , WT 0 . 63 ± 0 . 04 , DN 0 . 61 ± 0 . 04 , CA 0 . 51 ± 0 . 05 , Figure 7A , B ) . However , the quantity declined rapidly with distance from the cell body and the transfected rab11 failed to reach distal axons compared to GFP , confirming an active but slightly leaky exclusion mechanism for rab11 vesicles in axons ( Figure 7C , D ) . Previously we have shown that integrin-containing vesicles move mostly retrogradely in mature axons ( Franssen et al . , 2015 ) , so we examined rab11 vesicle movements using time lapse imaging , illustrated in kymographs . Similar to integrins , there was a predominance of retrograde movement in 16 DIV axons , and little anterograde vesicle movement . In dendrites , however movement was equal in both directions ( Figure 7E , F ) . Predominantly retrograde transport in axons was seen for all rab11 forms , most prominently with the DN ( Figure 7F ) . The average vesicle velocity for both anterograde and retrograde movement was also slower in axons for WT and CA compared to dendrites ( Figure 7—figure supplement 1A , B ) . Taken together , these results show that rab11 becomes selectively excluded from axons by 16 DIV , and that the overall direction of rab11a transport prioritizes exclusion from axons and transport into dendrites by 16 DIV . As shown above , newly plated neurons show robust rapid axon growth and a high proportion of them regenerate after axotomy . Since this matches the period when axonal rab11 is still abundant , we asked if axons could regain their regenerative state if rab11 was returned to the axon at 16 DIV . In principle this should enable transport of many growth-related molecules into the mature axon . Taking advantage of mis-trafficking after overexpression to re-introduce rab11 into axons , we axotomized axons of rab11 WT , DN and CA transfected neurons in the proximal zone . Staining for total rab11 in transfected neurons confirmed that transfection had increased total rab11 in the proximal axons , and the proximal axon/dendrite ratio of immunofluorescence intensity was increased ( Figure 7A , B ) . Axons were cut in the proximal region where rab11 was now present and their regeneration behaviour was measured . Laser axotomy led to an accumulation of fluorescent rab11 for all forms in the retraction bulb ( Figure 8A , Figure 8 , Video 3 ) , showing that transfected rab11 is present during the reorganization of the severed stump . Interestingly , in some cases this accumulation started even before ( ~30 min ) the reflux of the cytosolic GFP or bulb formation ( Figure 8A , Video 3 ) . We then analysed the regenerative measures as in the previous experiments . As with GFP-transfected neurons , retraction distance and time to form a retraction bulb was correlated ( Figure 8B ) , and the retraction distance of rab11a transfected neurons showed a biphasic distribution for all forms ( Figure 8C ) , so we again separated neurons into short and long retraction groups . The percentages of axons in the short and long retraction groups did not change between GFP and the other rab11 forms . But strikingly , for all rab11 forms , overexpression resulted in decreased retraction in the long retraction group ( Figure 8E ) . The presence of rab11 also influenced growth cone regeneration , and the long retraction group of rab11 neurons had an improved regeneration ratio , but only with the WT and DN form suggesting the importance of the inactivated state of rab11 ( GFP 11% , WT 38% , DN 38% , CA 13% , GFP vs WT p=0 . 0174 , GFP vs DN p=0 . 0342 , Figure 8F ) . No pro-regeneration effect was observed in the short retraction group ( Figure 8F ) . For other measures , regeneration initiation time was unchanged ( Figure 8G ) , but the length of the axons after 2 hr was increased with rab11 WT and CA ( Figure 8H ) . Growth cone area was also enlarged in regenerating neurons transfected with rab11 WT , and a trend was seen with the CA but did not reach significance ( Figure 8I ) . In summary , increasing rab11 in the axons of 16 DIV neurons led to enhancement of regeneration especially in the long retraction group . The activation state of rab11 was a significant factor , with the DN form promoting growth cone formation and the CA form stimulating subsequent elongation . Our hypothesis was that rab11 forced into axons by overexpression would carry with it molecules that promote axon regeneration . We therefore transfected neurons with rab11-GFP or control GFP at DIV10 , the time at which partitioning of transport between axons and dendrites becomes apparent . At DIV 16 the neurons were stained for GFP and for a growth-associated receptor normally excluded from mature neurons . For this growth-related molecule normally carried in rab11 vesicles we selected α5 integrin ( Caswell and Norman , 2006; Eva et al . , 2010 ) ; Gardiner et al . , 2007; Hülsbusch et al . , 2015 ) , which is present in neurons and can be detected with a reliable antibody . The level of α5 immunostaining was measured by drawing a linear AOI along the axons ( identified by staining the AIS with neurofascin ) and dendrites and subtracting the same AOI moved to the background . The results were plotted as axon/dendrite intensity ratio ( Figure 9B ) and absolute level ( Figure 9C ) . As above , overexpression of rab11-GFP led to its presence in the axons over approx . the proximal 200 μm ( Figure 9A ) . The distribution of α5 integrin changed similarly . In control GFP-transfected neurons α5 integrin was not detectable in axons ( Figure 9A–C ) , while in rab11-GFP-transfected neurons the staining intensity was similar between proximal axons and dendrites ( Figure 9A–C ) . This experiment shows that rab11 is able to carry growth-related receptors into axons . Next , we investigated whether the relationships between maturation , rab11 distribution , axonal expression and regeneration are conserved in human neurons , taking advantage of recent advances that have made it possible to generate human dopaminergic neurons in vitro . The human embryonic stem cell ( hESC ) line RC17 ( Roslin Cells ) was induced to dopaminergic neuronal differentiation using a protocol in which differentiated post-mitotic neurons are generated within day ( d ) 35 after neural induction ( Kirkeby et al . , 2012 ) , and the cells were then allowed to mature in vitro for up to d65 . As in the rodent neurons , laser-mediated axotomy revealed a maturation-related loss of capacity for axonal regeneration . Between d35-40 , successful axonal regeneration was seen in over 70% of neurons , but this dropped to 46 . 9% and 39 . 6% between d45-55 and d55-65 respectively ( Figure 10A ) . To determine whether this maturation-related loss of regenerative capacity was associated with a reduction in rab11 trafficking into the axon , we expressed rab11-GFP in the human neurons . At d41 , the level of rab11-GFP fluorescence was similar in axons and dendrites . At d51 and d65 rab11-GFP was just detectable in the axons but there was a significant drop relative to levels in the dendrites ( Figure 10B , C ) . We asked whether , as in the rodent neurons , rab11 overexpression could restore the loss of axonal regeneration potential . Transfection of rab11 did not affect the rate of axonal regeneration in d35-40 neurons , but a significant increase in the number of axons showing regeneration was seen in d45-55 axons ( Figure 10A ) with an associated increase in the extent of elongation at 2 hr post-axotomy ( Figure 7D ) These results confirm that , as in rodent neurons , there is an age-dependent loss of axonal regeneration in human CNS neurons that is associated with loss of axonal rab11 . In order to analyse why CNS neurons lose their ability to regenerate with maturity , and to develop regeneration-promoting treatments , a manipulable in vitro model is needed . We describe an in-vitro single neuron axotomy model , in which there is a progressive decline in the intrinsic regeneration ability of the axons of cortical neurons as they mature . The neurons in these cultures can be transfected and regeneration-promoting effects of pharmaceuticals and other interventions tested ( unpublished observations ) . We describe here an analysis which demonstrates that the development of polarised neuronal transport is an important factor in the loss of regeneration ability , through exclusion of rab11 recycling endosomes ( which contain key growth molecules ) from axons . The main influence on regeneration in this study was neuronal maturity , represented by days in vitro ( DIV ) . This is in line with many in vivo studies which show that the regenerative ability of CNS axons is lost with maturity . Neuronal maturity is also a key factor in vivo , where grafted embryonic neurons have shown extensive growth in the adult brain and spinal cord , yet mature neurons show no growth ( Jakeman and Reier , 1991; Kim et al . , 2006 ) ( Lu et al . , 2012 ) . In C . elegans axons also lose regenerative ability with age , although the mechanism may be rather different ( Tang and Chisholm , 2016; Byrne et al . , 2014 ) . Even when regeneration is stimulated by manipulation of signalling pathways using Phosphatase and Tensin Homolog Deleted from Chromosome 10 ( PTEN ) deletion , the knockdown is much more effective if it is performed during the growth phase of cortical neurons rather than adulthood ( Du et al . , 2015; Geoffroy et al . , 2016 ) . However some regenerative events are seen after CNS axotomy in adulthood . In the corticospinal tract and other pathways extensive lateral sprouting can occur ( Bareyre et al . , 2004 ) , particularly in primates ( Rosenzweig et al . , 2010 ) , and growth into embryonic grafts can occur ( Bernstein-Goral and Bregman , 1993; Kadoya et al . , 2016 ) . Detailed quantification of the events following axotomy revealed behaviours that change at different DIV . The retraction distance after axotomy increased with maturity; at 4 DIV retraction distances were short , but by 24 DIV they were much longer , with a mixture of short and long retractors at 16DIV . Retraction distance predicted regeneration with long retractors seldom regenerating , especially after distal axotomy . The correlation between long retraction and regeneration failure has also been shown in vivo after cutting intracortical axons by laser ( Canty et al . , 2013 ) . We suggest that the increased retraction distance with maturity may reflect changes in cytoskeletal dynamics together with decreased anterograde transport of materials required to maintain the axons . At DIV16 , where we see both long and short-retracting axons indicating a mixed degree of maturity in neurons as they are becoming polarized and selective transport is being established ( Franssen et al . , 2015 ) . By DIV 24 these changes are fully established and all axons show long retraction after axotomy and almost complete loss of regenerative ability . The speed of axon growth after initiation of regeneration and the size of regenerated growth cones declined earlier than these retraction changes , the changes being complete by 16 DIV . Growth cone size is affected by many environmental and intrinsic factors , does not correlate with axon growth speed ( Harris et al . , 1987; Hur et al . , 2012; Tosney and Landmesser , 1985 ) , and is probably not a useful general indicator of regenerative ability . We focused on the hypothesis that the developmental change in regenerative ability in axons is related to the progressive exclusion of growth-related molecules from axons ( Bentley and Banker , 2016; Britt et al . , 2016 ) . In particular , previous work has shown that integrins ( Franssen et al . , 2015 ) and Trks ( Hollis et al . , 2009a ) become excluded from axons after they mature , as are most postsynaptic proteins . Precise control of the location of intracellular molecules is necessary so that axons or dendrites can have different structures and functions . Many growth-related molecules are transported into axons via the recycling pathway in vesicles marked by the small GTPase rab11 ( Lasiecka and Winckler , 2011 ) ( Baetz and Goldenring , 2013; Welz et al . , 2014 ) . In this study , we show that rab11 is present in immature axons but becomes restricted to a somatodendritic distribution with maturity , correlating with the loss of ability to regenerate axons in both rodent cortical neurons and human dopaminergic neurons . As with many molecules the selective transport mechanism is somewhat leaky , which could indicate passive diffusion of vesicles and molecules into axons followed by selective removal by retrograde transport . Overexpression of rab11 leads to mis-trafficking and therefore leads to the presence of some rab11 in the proximal axons of mature neurons . This allowed us to ask whether the presence of rab11 affects regeneration . In our study , rab11 WT and CA enhanced regeneration length and growth cone size , and this is in line with previous studies where these molecules have been shown to be involved in axon and dendritic growth ( Park et al . , 2006 ) . Our results also demonstrated that along with WT the DN mutant enhances formation of a new growth cone to initiate regeneration whereas the CA does not . Other studies have also confirmed the protrusion-initiating properties of rab11 DN ( Shirane and Nakayama , 2006; Ramel et al . , 2013 ) , suggesting that the different forms of rab11 can have different effects on different processes including vesicle transport , trafficking and actin reorganization and can therefore affect different phases of regeneration . Rab11 overexpression also enhanced regeneration in human dopaminergic neurons . The regeneration-promoting effect of rab11 overexpression is assumed to be due to the transport into axons of the many growth-related molecules that are transported in this class of endosome ( Welz et al . , 2014 ) . As an example , we showed that α5 integrin is transported into the proximal axons of rab11 overexpressing neurons . Rab11 transport is controlled in several ways . A transport complex of rab11 , Arf6 and JIP3/4 can associate with kinesin or dynein depending on the activation state of Arf6 ( Montagnac et al . , 2009 ) and rab11 associates with kinesin through activated protrudin ( Matsuzaki et al . , 2011 ) . Enhancing the level of PIP3 through PTEN knockdown has been a successful method for inducing regeneration ( Liu et al . , 2010; Du et al . , 2015 ) . Our results fit with this story , because PIP3 has effects on transport through GAPs and GEFs and on membrane trafficking ( Macia et al . , 2008; Randazzo et al . , 2001 ) . It is probable that there is a cycle , where transport of receptors allows PI3K activation and PIP3 generation , which in turn enhances transport and trafficking ( Cheng et al . , 2011 ) . Manipulation of rab11 transport using these methods can enhance regeneration in mature cortical neurons ( unpublished observations ) , and will be the basis of future in vivo investigations . The changes in transport that we have shown affect regeneration , but CNS axons nevertheless show some regenerative activity after damage , particularly lateral sprouting and growth into embryonic tissues . It is unlikely that these events can happen without the presence of a receptor on the axons and a corresponding ligand in the environment . These receptors , currently unidentified , must continue to be transported into mature axons , presumably independently of rab11 . Overall this study supports the concept that CNS axons lose their intrinsic ability to regenerate with maturation . Mature axons lose regeneration-associated molecules due to the development of polarised transportation , which directs the many growth-related molecules in rab11 vesicles to dendrites and away from axons . Modifying selective axon transportation could therefore be a strategy for enhancing intrinsic regeneration ability in the adult CNS .	The nerves in the brain and spinal cord can be damaged by trauma , stroke and other conditions . Damage to these nerve fibres can destroy the connections they form with each other , which may lead to paralysis , loss of sensation and loss of body control . If we could stimulate the regeneration and reconnection of the damaged nerve fibres then neurological function could be restored . However , although embryonic nerve fibres can regenerate when they are transplanted into the adult central nervous system , this regenerative ability appears to be lost as the nerve fibres mature . To investigate when and why nerve fibres lose the ability to regenerate , Koseki et al . first developed a tissue culture assay in which individual nerve fibres were cut with a laser and imaged for several hours to track their regeneration ( or failure to regenerate ) . The results demonstrate that nerve fibres from the central nervous system progressively lose the ability to grow and regenerate as they mature . To investigate why mature nerve fibres cannot regenerate , Koseki et al . measured whether nerve fibres can transport some of the molecules needed for growth and regeneration to sites of damage . This showed that the compartments in which some key growth molecules are transported become excluded from mature nerve fibres . These compartments are marked by a protein called rab11 , and Koseki et al . found that forcing rab11 back into mature nerve fibres restored their ability to regenerate . There is still a lot of work needed before these findings can lead to a new regeneration treatment for patients , but it is a crucial step forwards . Furthermore , the assay developed by Koseki et al . could be used to develop and test such treatments .	[ "Abstract", "Introduction", "Results", "Discussion" ]	[ "neuroscience" ]	2017	Selective rab11 transport and the intrinsic regenerative ability of CNS axons
The nicotinic acetylcholine receptor ( nAChR ) is a major target of autoantibodies in myasthenia gravis ( MG ) , an autoimmune disease that causes neuromuscular transmission dysfunction . Despite decades of research , the molecular mechanisms underlying MG have not been fully elucidated . Here , we present the crystal structure of the nAChR α1 subunit bound by the Fab fragment of mAb35 , a reference monoclonal antibody that causes experimental MG and competes with ~65% of antibodies from MG patients . Our structures reveal for the first time the detailed molecular interactions between MG antibodies and a core region on nAChR α1 . These structures suggest a major nAChR-binding mechanism shared by a large number of MG antibodies and the possibility to treat MG by blocking this binding mechanism . Structure-based modeling also provides insights into antibody-mediated nAChR cross-linking known to cause receptor degradation . Our studies establish a structural basis for further mechanistic studies and therapeutic development of MG . The nicotinic acetylcholine receptor ( nAChR ) at the neuromuscular junction ( NMJ ) is a ligand-gated ion channel that mediates rapid signal communication between spinal motor neurons and the muscle cells . This receptor is also a major target of autoimmune antibodies in patients with myasthenia gravis ( MG ) , an autoimmune disease that afflicts more than 20 in 100 , 000 people ( Lindstrom , 2000; Vincent et al . , 2001 ) . MG is the first and , so far , only autoimmune disease with well-defined autoantigen target; the binding of nAChR by MG antibodies leads to complement-mediated lysis of the postsynaptic structure and internalization of the receptor , thereby disrupting neuromuscular transmission ( Engel and Arahata , 1987; Drachman et al . , 1978; Gomez et al . , 2010 ) . The majority of MG cases can be diagnosed by the detection of autoantibodies to human muscle nAChR , and current treatment options include the use of acetylcholine esterase inhibitors , non-specific immunosuppressive drugs , plasmapheresis and thymectomy . Most of these treatments are for symptomatic control except for thymectomy that may lead to disease remission . Other therapeutic approaches to treating MG , such as nAChR-specific immunosuppressive therapy ( Luo and Lindstrom , 2015 ) , need to be explored . The autoimmune nature of MG was first suggested by the discovery of experimental autoimmune myasthenia gravis ( EAMG ) induced in rabbits immunized with nAChR purified from Electrophorus electricus ( Patrick and Lindstrom , 1973 ) . Subsequent studies with passive transfer of MG patient serum or purified nAChR antibodies to induce EAMG further established nAChR antibodies as the major pathological agents of MG ( Toyka et al . , 1975; Lindstrom et al . , 1976 ) . In fact , more than 85% of MG patients carry nAChR antibodies ( Lindstrom , 2000; Vincent et al . , 2001; Meriggioli and Sanders , 2009 ) . However , the total amount of nAChR antibodies in the serum of MG patients does not seem to correlate with disease severity , suggesting that various nAChR antibodies that bind different regions on nAChR may contribute differently to this disease ( Somnier , 1993; Berrih-Aknin , 1995; Mossman et al . , 1988; Tzartos et al . , 1998 ) . Mammalian muscle nAChR has a pentameric structure composed of two α1 , one β1 , one δ , and one ε ( adult form ) or γ ( fetal form ) subunit ( s ) ( Unwin , 2005 ) . Extensive studies suggest that antibodies to α1 play a major role in MG pathology ( Sideris et al . , 2007; Tzartos et al . , 2008 , 1987; Kordas et al . , 2014 ) . Furthermore , more than half of all autoantibodies in MG and EAMG bind an overlapping region on the nAChR α1 subunit , known as the main immunogenic region ( MIR ) ( Tzartos et al . , 1998 ) . The MIR is defined by the ability of a single rat monoclonal antibody ( mAb ) , mAb35 , to inhibit the binding of about 65% autoantibodies from MG patients or rats with EAMG ( Tzartos and Lindstrom , 1980; Tzartos et al . , 1982 , 1983 ) . Subsequent studies have mapped MIR to a peptide region that spans residues 67–76 on nAChR α1 ( Barkas et al . , 1988; Tzartos et al . , 1988 ) . Monoclonal antibodies directed to the MIR can passively transfer EAMG and possess all the key pathological functions of serum autoantibodies from MG patients ( Tzartos et al . , 1987 ) . Moreover , a recent study showed that titer levels of MIR-specific antibody from MG patients , rather than the total amount of nAChR antibodies , correlate with disease severity ( Masuda et al . , 2012 ) . These observations suggest that antibodies binding to the MIR on nAChR α1 play a major role in the pathogenesis of MG ( Tzartos et al . , 1998 ) . The myasthenogenic role of nAChR was established more than four decades ago . Since then , extensive efforts have been put into characterizing the interactions between MG antibodies and nAChR using biochemical ( Barkas et al . , 1988; Tzartos et al . , 1988; Das and Lindstrom , 1989; Saedi et al . , 1990; Papadouli et al . , 1990 , 1993; Luo et al . , 2009; Morell et al . , 2014 ) , structural ( Dellisanti et al . , 2007a; Beroukhim and Unwin , 1995; Kontou et al . , 2000; Poulas et al . , 2001 ) , and modeling approaches ( Kleinjung et al . , 2000 ) . These studies aimed to understand the basic mechanisms of MG and also the structure/function of nAChR in order to develop effective diagnosis and treatment for MG . However , exactly how antibodies bind and functionally affect nAChR has not been fully elucidated since no high-resolution structure of the complex between MG antibodies and nAChR was available . Here we describe the first crystal structure of muscle nAChR α1 subunit bound by an EAMG antibody at 2 . 61 Å resolution and present detailed analyses of the molecular interactions in myasthenia gravis . These structural analyses , in the context of the large amount of biochemical and functional data from previous MG research , provide unprecedented insights into the molecular mechanisms of MG and a basis for developing more effective diagnosis and treatment for this debilitating disease . mAb35 was chosen for structural analysis because it shares many functional characteristics with serum antibodies from MG patients and has been used as a reference MG antibody in extensive biochemical and functional studies ( Tzartos et al . , 1998 , 1981 ) . Although mAb35 is derived from rat immunized with Electrophorus AChR , it competes with more than two thirds of serum antibodies from MG patients ( Tzartos et al . , 1982 ) . At the functional level , mAb35 binds complement causing focal lysis of the postsynaptic membrane , cross-links AChRs thereby increasing their internalization , and can passively transfer EAMG ( Tzartos et al . , 1987 ) . To facilitate crystallization , we used the Fab fragment of mAb35 ( Fab35 ) and also included α-bungarotoxin ( α-Btx ) to stabilize flexible regions of nAChR α1 ECD that may hinder crystallization . We used a mutant of nAChR α1 ECD that contains three stabilization mutations , referred to as α211 as described previously ( Dellisanti et al . , 2007a ) . Although mAb35 does not bind well to the denatured receptor , it has been shown to bind natively folded nAChR α1 with high affinity ( Kd =~2 nM ) ( Luo et al . , 2009 ) . Our native PAGE analysis showed that nAChR α1 ECD , α-Btx and Fab35 can form a well-defined ternary complex ( Figure 1a ) , and the ternary complex of Fab35/nAChR α1 ECD/α-Btx was stable during purification by gel filtration . We obtained ternary complex crystals with both human and mouse nAChR α1 ECD and solved the structures ( 2 . 61 Å and 2 . 70 Å resolution , respectively ) by molecular replacement ( Supplementary file 1 ) . A detailed comparison between the two complex structures is presented in Figure 1—figure supplements 1 and 2 . Here , we focus our structural description and analysis on the antibody/receptor interface , which appears identical between the human and mouse complexes . 10 . 7554/eLife . 23043 . 003Figure 1 . The ternary complex of nAChR α1 ECD bound by Fab35 and α-Btx . ( a ) Gel shift assay . Native PAGE showed the formation of the ternary complex of nAChR α1 ECD , α-Btx and Fab35 . Lane 1: nAChR α1 ECD alone ( labeled as α1 ECD ) , Lane 2: α-Btx alone , Lane 3: Fab35 alone , Lane 4: nAChR α1 ECD plus α-Btx , Lane 5: nAChR α1 ECD plus Fab35 , and Lane 6: nAChR α1 ECD plus α-Btx plus Fab35 . Note that α-Btx in Lane 2 and Fab35 in Lane 3 were not visible because both proteins migrated upward due to their net positive charges under the experimental condition . ( b ) Ribbon representation of nAChR α1 ECD ( α1: cyan ) bound by α-Btx ( green ) and Fab35 ( heavy chain , H: yellow and light chain , L: magenta ) . The variable domains ( VH and VL ) and the constant domains ( CH and CL ) of Fab35 are indicated accordingly . This color scheme is kept the same throughout illustration unless noted otherwise . ( c ) Surface representation of the ternary complex . ( d ) Zoomed-in view of the binding interface . The complementarity determining regions ( CDRs ) of the heavy chain ( CDR-H1 , CDR-H2 , and CDR-H3 ) are indicated as H1 , H2 , and H3 , respectively . Those of the light chain ( CDR-L1 , CDR-L2 and CDR-L3 ) are indicated as , L1 , L2 , and L3 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 23043 . 00310 . 7554/eLife . 23043 . 004Figure 1—figure supplement 1 . Key structural features of the human nAChR α1 ECD . Our study also generated the first atomic picture of the human nAChR α1 ECD , which appears very similar to the previously characterized mouse nAChR α1 ECD ( PDB ID , 2QC1 ) ( Lindstrom , 2000 ) . Many functionally important structural features observed in the mouse nAChR α1 ECD are also conserved in the human α1 ECD as shown here . ( a ) The hydrophilic residues ( Thr52 and Ser126 ) and a bound water molecule buried inside the beta sandwich core of the ECD . The 2Fo-Fc electron density map ( shown in blue ) is countered at one sigma level . The electron density of water is shown as the Fo-Fc omit map ( shown in green ) countered at three sigma level . ( b ) The N-linked glycan ( at the Asn141 ) bridging the Cys-loop and the loop C . The 2Fo-Fc electron density map ( shown in blue ) is contoured at one sigma level . DOI: http://dx . doi . org/10 . 7554/eLife . 23043 . 00410 . 7554/eLife . 23043 . 005Figure 1—figure supplement 2 . Structural differences between the human and mouse nAChR α1 ECDs . A significant structural difference between the human and mouse nAChR α1 ECDs is the binding interface of α-Btx . ( a ) In the mouse nAChR α1 ECD , Phe189 is inserted into a surface pocket of α-Btx . ( b ) In the human nAChR α1 ECD , Thr189 , is too small to fill in the α-Btx pocket . These observations are consistent with predictions from previous analyses of the nAChR α1/α-Btx complex ( Vincent et al . , 2001 ) . The 2Fo-Fc electron density map ( shown in blue ) was contoured at one sigma level . DOI: http://dx . doi . org/10 . 7554/eLife . 23043 . 00510 . 7554/eLife . 23043 . 006Figure 1—figure supplement 3 . Structural comparison of mouse nAChR α1 ECDs in the ternary complex of Fab35/nAChR α1 ECD/α-Btx and the binary complex of nAChR α1 ECD/α-Btx . ( a ) Superposition of the mouse nAChR α1 ECD from the Fab35/nAChR α1 ECD/α-Btx complex ( blue ) and the nAChR α1 ECD/α-Btx complex ( green ) ( PDB ID , 2QC1 ) ( Lindstrom , 2000 ) using the Cα backbone of the nAChR α1 ECD . ( b ) Detailed comparison of side chain orientation of residues involved in Fab35 binding . DOI: http://dx . doi . org/10 . 7554/eLife . 23043 . 006 The Fab35 binds to nAChR α1 in an upright orientation , away from the α-Btx ( Figure 1b and c ) . The Fab35 binding sites on nAChR α1 include the MIR and the N-terminal helix; the buried solvent accessible area of the complex is 899 Å2 . Fab35 has the canonical IgG antibody structure where the complementarity determining regions ( CDRs ) from the heavy chain , CDR-H2 and CDR-H3 , and the light chain , CDR-L3 , form the binding site of nAChR α1 ( Figure 1d ) . The interface of their interaction is characterized by mutual insertion of loops into pockets of binding partners . On the receptor side ( Figure 2a ) , the MIR loop inserts deeply into a surface pocket between the variable domains of the heavy and light chains ( VH and VL ) , whereas the N-terminal helix sits into a groove on the surface of the heavy chain . On the Fab35 side ( Figure 2b ) , the CDR-H3 inserts into a surface pocket formed by the N-terminal helix , the loop following the N-terminal helix , the MIR and the loop preceding the MIR . 10 . 7554/eLife . 23043 . 007Figure 2 . Mutual insertion of loops into pockets of binding partners . ( a ) The MIR loop of nAChR α1 inserts into a surface pocket between the variable domains of the heavy and light chains ( VH and VL ) of Fab35 ( orange ) while the N-terminal helix sits into a groove on the surface of the heavy chain . ( b ) The CDR-H3 ( H3 ) from the heavy chain of Fab35 inserts into a surface pocket between the MIR loop and the N-terminal helix on the nAChR α1 ECD . DOI: http://dx . doi . org/10 . 7554/eLife . 23043 . 007 Superposition of the structures of mouse nAChR α1 ECD in the ternary Fab35/nAChR α1 ECD/α-Btx complex with that in the binary nAChR α1 ECD/α-Btx complex ( PDB ID , 2QC1 ) ( Dellisanti et al . , 2007a ) shows that the nAChR α1 structure remains the same in the two complex states ( Figure 1—figure supplement 3a ) . Moreover , the orientation of the side chain of most nAChR α1 residues involved in Fab35 binding ( see below ) is similar between the two complexes ( Figure 1—figure supplement 3b ) . This structural comparison suggests that Fab35 recognizes and binds a well-defined and preformed conformation of the nAChR α1 ECD . Residues from Fab35 and nAChR α1 that are within 4 . 5 Å from each other at the binding interface were mapped as contacting residues . As shown in Supplementary file 2 , Fab35-binding residues on nAChR α1 are mostly located on the MIR loop ( highlighted in light green in the table ) and the N-terminal helix ( highlighted in yellow in the table ) . While the MIR loop extensively interacts with residues from both the heavy and light chains of Fab35 , the N-terminal helix interacts exclusively with residues from the heavy chain . The contacting analysis also revealed several residues on nAChR α1 that make numerous contacts to Fab35 . Four such ‘hotspots’ of binding were identified: Asn68 and Asp71 from the MIR loop and Arg6 and Lys10 from the N-terminal helix . As described below , each of these four ‘hotspots’ anchors an extensive network of interactions that display remarkable chemical complementarities ( Figure 3 ) . The interface interactions are well defined by electron densities ( Figure 3—figure supplement 1a–e ) . 10 . 7554/eLife . 23043 . 008Figure 3 . Detailed interactions at the interface between Fab35 and nAChR α1 ECD . ( a ) Binding interactions at the Asp71 site of α1 ( located at the MIR ) . ( b ) Binding interactions at the Asn68 site of α1 ( located at the MIR ) . ( c ) Binding interactions surrounding Arg6 and Lys10 of α1 ( located at the N-terminus of α1 ) . ( d ) Binding interactions mediated by His3 of α1 ( located at the N-terminus of α1 ) . ( e ) Binding interactions at the CDR-H3 loop of Fab35 . Interacting residues are represented by stick model and are colored according to their protein subunits . Water molecules are represented by red spheres . DOI: http://dx . doi . org/10 . 7554/eLife . 23043 . 00810 . 7554/eLife . 23043 . 009Figure 3—figure supplement 1 . Fo-Fc omit maps of the interface between Fab35 and nAChR α1 ECD . The Fo-Fc omit maps of representative side chains and main chains ( only interacting ones ) of the Fab35-alpha1 residues and waters at the binding interface . The Fo-Fc omit maps are shown in green and countered at two sigma level for ( a ) and ( e ) and three sigma level for ( b ) – ( d ) . The orientation of each figure panel is the same as Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 23043 . 009 Asp71 forms a salt bridge with Arg50 of VH and a hydrogen bond with Tyr95 of VL ( Figure 3a ) . Asp71 also forms hydrogen bonds with two interfacial water molecules , H2O 4 and H2O 5 . H2O 4 in turn forms hydrogen bonds to the main chain amide of Asn68 of the α1 and the main chain carbonyl group of Ala103 of VH . H2O 5 in turn forms hydrogen bonds to the main chain carbonyl group of Tyr91 of VL and the main chain amide of Asn105 of VH . The adjacent Tyr72 can be considered as part of the Asp71 ‘hotspot’: Tyr72 not only mediates the packing interactions between the MIR loop and the N-terminal helix but also makes extensive contacts to the antibody , including hydrogen bonding to Ala103 and van der Waals contacts to Trp52 , Val58 , and Asn105 of VH . At Asn68 , another hotspot site ( Figure 3b ) , the amide group of Asn68 side chain forms three hydrogen bonds with VL ( the carbonyl of Tyr91 main chain and the amide and carbonyl of Gly94 main chain ) . The side chain of Asn68 also makes van der Waals contact to Ile92 and Asn93 of VL . Both Asn68 and Asp71 together , extending from the tip of the MIR loop , insert deeply into the antigen-binding site of Fab35 and make extensive contacts with Fab35 residues . The N-terminal helix of nAChR α1 engages in extensive interactions with VH of Fab35 . These interactions are centered at Arg6 and Lys10 residues ( Figure 3c ) . The guanidinium head group of Arg6 forms bidendate hydrogen bonds with Asp54 and cation-π stacking with Trp52 of VH . The aliphatic side chain of Arg6 also makes van der Waals contacts to the aromatic ring of Trp52 . This interaction network is extended by the nearby nAChR α1 residues , Lys10 and His3 ( Figure 3c and d , respectively ) . Lys10 forms salt bridges with Asp53 of VH ( Figure 3c ) ; its side chain also makes van der Waals contacts to numerous VH residues ( not shown ) . His3 makes a water-mediated ( H2O 1 ) hydrogen bond with Trp52 and van der Waals contact to Val58 of VH ( Figure 3d ) . An interesting structural feature of the antibody/receptor interface is the insertion of the CDR-H3 loop into a surface pocket on nAChR α1 ( Figure 3e ) . The tip of the CDR-H3 loop , including Arg102 and Ala103 , makes extensive van der Waals contacts to the surrounding receptor residues . The guanidinium group of Arg102 is sandwiched by the carboxylic amide of Asn64 side chain and the carbonyl of Asp14 main chain of nAChR α1 in a parallel orientation that may favor π-stacking . Arg102 also forms water-mediated ( H2O 82 ) hydrogen bonds with the main chain carbonyl groups of Leu11 and Asn64 of nAChR α1 . Adjacent to the CDR-H3 interaction site , Tyr63 of nAChR α1 forms a hydrogen bond with Lys50 of the CDR-L2 , which is stabilized by a cation-π interaction with Tyr32 of the CDR-L1 . Lys50 of the CDR-L2 also engages in electrostatic interaction with Glu23 of nAChR α1 . These interactions expand the binding interface from the MIR and the N-terminal helix to the loop region between the N-terminal helix and the β-strand β1 ( residues 15–23 ) . Whether different MG mAbs bind nAChR through conserved or divergent mechanisms is an important question relevant to understanding the disease mechanism and developing therapeutics . To address this question , we compared the structure of Fab35 with that of two other MG mAbs ( Fab198: PDB ID , 1FN4 and Fab192: PDB ID , 1C5D ) ( Kontou et al . , 2000; Poulas et al . , 2001 ) . Interestingly , superposition of the structure of Fab198 onto that of Fab35 in the ternary complex shows that these two Fabs share not only a conserved immunoglobulin fold but also a similar antigen-binding site ( Figure 4a ) . As such , the MIR loop fits well into the pocket surrounded by the CDR-H2 , CDR-H3 and CDR-L3 loops of Fab198 , as predicated by previous modeling studies ( Kleinjung et al . , 2000 ) . The CDR-H2 loop of Fab198 is also in position to interact with the N-terminal helix adjacent to the MIR ( Figure 4b ) . Even more remarkably , many key α1-binding residues in Fab35 are also conserved in Fab198 and they appear to make similar contacts to nAChR α1 in the modeled Fab198/nAChR α1 binding interface ( Figure 4a and c ) . These residues include Trp47 ( CDR-H2 ) , Arg50 ( CDR-H2 ) , and Tyr95 ( CDR-L3 ) at the center of the MIR-binding pocket , and Trp52 and Asp54 ( both CDR-H2 ) which interact with the N-terminal helix . However , in contrast to the above structural similarities , the CDR-H3 loops between Fab198 and Fab35 differ significantly in length and sequence ( Figure 4b and c ) . As a result , the CDR-H3 loop of Fab198 is too short to interact with the surface pocket of nAChR α1 , which is , in the case of Fab35 , occupied by the corresponding CDR-H3 loop ( Figure 4—figure supplement 1a ) . These structural analyses suggest that mAb35 and mAb198 share a high similarity in binding mechanism to the core MIR/N-terminal helix region , but differ in the periphery of the binding interface . 10 . 7554/eLife . 23043 . 010Figure 4 . Structural comparisons among MG mAbs . ( a ) Superposition of Fab198 ( Poulas et al . , 2001 ) ( heavy chain: purple and light chain: dark green ) onto Fab35 in the Fab35/nAChR α1/α-Btx ternary complex using the Cα backbone . ( b ) Detailed comparison of the binding interface . The residues are colored according to their protein subunits . Note that key α1-binding residues in Fab35 , including Trp47 , Arg50 , Trp52 and Asp54 of VH and Tyr95 of VL are conserved in Fab198 , and seem to be able to make similar contacts to nAChR α1 in the modeled interface . The CDR-H3 loop of Fab198 ( purple ) is substantially shorter than that of Fab35 ( yellow ) , as indicated by arrows . ( c ) Structure-based sequence alignment of the nAChR α1-binding loops ( CDR-H2 , CDR-H3 and CDR-L3 ) between Fab35 , Fab198 and Fab192 ( Kontou et al . , 2000 ) . Residues shaded in light green are involved in nAChR α1 binding in Fab35 , some of these ( in bold font and colored in red ) are conserved in Fab198 or Fab192 . Note that Fab35 and Fab198 share a high similarity in their nAChR α1-binding CDR-H2 and CDR-L3 loops , but differ significantly in CDR-H3 . On the other hand , Fab192 differs significantly from Fab35 and Fab198 , especially in the CDR-H2 and CDR-H3 loops ( See also Figure 4—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23043 . 01010 . 7554/eLife . 23043 . 011Figure 4—figure supplement 1 . Structural comparison between Fab35 and Fab198/Fab192 . ( a ) A surface model showing that the CDR-H3 loop of Fab198 ( PDB ID , 1FN4 ) ( purple ) ( Engel and Arahata , 1987 ) is too short to interact with the surface pocket of nAChR α1 , which is , on the other hand , occupied by the corresponding CDR-H3 loop of Fab35 ( yellow ) . ( b ) The structure of Fab192 ( PDB ID , 1C5D ) ( heavy chain , red; light chain , blue ) ( Drachman et al . , 1978 ) was superimposed onto that of the Fab35 in the Fab35/nAChR α1/α-Btx ternary complex using the Cα backbone . It shows substantial differences between Fab192 and Fab35 . DOI: http://dx . doi . org/10 . 7554/eLife . 23043 . 011 On the other hand , superposition of the structure of Fab192 onto that of Fab35 in the ternary complex reveals substantial differences between them ( Figure 4—figure supplement 1b ) . Although the constant domains ( CH and CL ) of these two Fabs align very well structurally , the variable domains ( VH and VL ) show a significant rotational twist , such that the MIR loop does not fit into the antigen-binding site of Fab192 ( Figure 4—figure supplement 1b ) . Moreover , the key α1-binding residues of Fab35 , such as Arg50 and Trp52 of CDR-H2 , are not conserved in Fab192 ( Figure 4c ) . This structural comparison suggests that Fab192 differs significantly from Fab35 in terms of their binding mechanisms to nAChR α1 , confirming and extending the differences that were previously recognized between mAb35 and mAb192 ( Luo et al . , 2009 ) . Another important question of clinical and mechanistic relevance is the binding specificity of MG mAbs to different nAChR subunits and to nAChR α1 from different species . To address the first part of this question , Fab35-contacting residues were mapped onto aligned sequences of human nAChR subunits ( Figure 5a ) . A subset of nAChR family members , including α2 , α3 , α5 and β3 , has either identical or homologous residues at the key Fab35-binding positions . These analyses suggest that Fab35 and other MIR-directed mAbs may be able to bind these nAChR subunits . On the other hand , the other nAChR members have divergent sequences at the Fab35-binding sites . For example , the key Fab35-binding residues in the MIR of nAChR α1 , Asn68 and Asp71 , are replaced by Asp and Gln in nAChR α9 , respectively . On the N-terminal helix of nAChR α1 , the key Fab35-binding residues , Arg6 and Lys10 , are also replaced by Lys and Asp in nAChR α9 , respectively . Superposition of the recently solved structure of the nAChR α9 ECD ( PDB ID , 4UY2 ) ( Zouridakis et al . , 2014 ) with that of the α1 in the Fab35-bound complex showed that the sequence divergence could disrupt both the shape and chemical complementarity of the binding interface ( Figure 5b ) . Consistent with the above structural analyses , we have shown by native PAGE that Fab35 indeed binds specifically to nAChR α1 but not α9 ( Figure 5c ) . 10 . 7554/eLife . 23043 . 012Figure 5 . Specificity of antibody-receptor binding . ( a ) Multiple sequence alignment of the N-terminal α helix ( left ) and the MIR ( right ) of human nAChR family members . The sequence of human nAChR α1 ( hα1 ) in the crystal structure is underlined . Abbreviation follows as t ( torpedo ) and h ( human ) . The Fab35-contacting profile for human nAChR α1 , indicating how many Fab35 residues are directly contacting with each particular residue of human nAChR α1 , is shown above the sequence . The aligned sequences are colored based on the contacting profile , with red color indicating highly contacting residues ( ‘hotspots’ ) . ( b ) Superposition of nAChR α9 ( orange ) ( Zouridakis et al . , 2014 ) onto the nAChR α1 in the Fab35/nAChR α1/α-Btx ternary complex showing the disrupted binding interface . ( c ) Native PAGE showing the binding specificity of Fab35 . Lane 1: α211 ( labeled as α1 ) , Lane 2: α9 , Lane 3: α-Btx , Lane 4: α211 plus α-Btx , Lane 5: α9 plus α-Btx , Lane 6: Fab35 , Lane 7: α211 plus Fab35 , and Lane 8: α9 plus Fab35 . Note that α-Btx in Lane 3 and Fab35 in Lane 6 were not visible because both proteins are positively charged and migrated upward under the native gel electrophoresis condition . Lanes with α-Btx were included as positive controls ( Lanes 3–5 ) . Lanes 4 and 5 show that both nAChR α1 and α9 bind α-Btx . Note that the α9/α-Btx complex has a smaller shift than the α1/α-Btx complex . Lanes 7 and 8 show that Fab35 binds α1 but not α9 . ( d ) Multiple sequence alignment of the N-terminal α helix ( left ) and the MIR ( right ) of nAChR α1 from several species along with the Fab35-contacting profile as in ( a ) . Abbreviation follows as b ( bovine ) , r ( rat ) , m ( mouse ) , c ( chicken ) , t ( torpedo ) and x ( Xenopus ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23043 . 012 For the second part of the question , a similar structure-based sequence comparison shows that Fab35-binding residues are conserved in nAChR α1 across a wide range of species ( Figure 5d ) , which is consistent with the high cross-reactivity of autoantibodies in MG and EAMG between different species ( Tzartos et al . , 1998 , 1982; Luo et al . , 2009; Gomez et al . , 1981 ) . However , in Xenopus nAChR α1-b , Asn68 and Asp71 are substituted by Asp and Lys , respectively . Based on our ternary structure , such N68D substitution would disrupt the hydrogen bonds between the side chain amide group of Asn68 and both the main chain carbonyl of Tyr91 and Gly94 in the CDR-L3 ( Figure 3b ) . In addition , the D71K substitution would introduce positive charge repulsion with Arg50 in the CDR-H2 and numerous steric clashes ( Figure 3a ) . Consistent with these structural observations , previous studies have shown that Xenopus nAChR α1 indeed does not bind Fab35 ( Saedi et al . , 1990 ) . The nAChR receptors are pentamers of identical or homologous subunits . To see how neighboring subunits to the α1 may affect the antibody-receptor interactions , we analyzed the binding of Fab35 to a nAChR pentamer by structure-based modeling using our previously solved structure of the α7/AChBP chimera ( PDB ID , 3SQ9 , 3 . 1 Å resolution ) ( Li et al . , 2011 ) as the nAChR ECD pentamer ( Figure 6a and b ) . This model suggests that Fab35 , by binding to the extruding tip of an α1 subunit , makes no direct contact to the neighboring subunits . Moreover , Fab35 binds nAChR α1 at a site that is far away from the ligand-binding site , consistent with the observation that MIR-directed antibodies generally do not affect the channel function ( Tzartos et al . , 1981; Gomez et al . , 1981; Tamamizu et al . , 1996 ) . 10 . 7554/eLife . 23043 . 013Figure 6 . Modeling the binding of Fab35 to a nAChR pentamer . ( a ) Superposition of the Fab35/nAChR α1 ECD/α-Btx ternary complex on one subunit of the α7/AChBP chimera pentamer ( blue ) ( PDB ID , 3SQ9 ) ( Li et al . , 2011 ) using the Cα backbone of ECDs as the reference . ( b ) Zoomed-in view of the contact between Fab35 and α7/AChBP Chimera . Fab35 makes no direct contact to the neighboring subunits in the pentamer . ( c ) Binding of a complete antibody ( Fab+Fc ) to a nAChR pentamer . The ternary Fab35/nAChR α1 ECD/α-Btx complex structure was used to guide the docking of an intact IgG1 antibody ( PDB ID , 1IGY ) ( in surface model ) ( Harris et al . , 1998 ) onto the human α4β2 nicotinic receptor ( PDB ID , 5KXI , 3 . 94 Å resolution ) ( Morales-Perez et al . , 2016 ) . The two Fab domains ( Fab-I and Fab-II ) and the Fc region of IgG1 are indicated as shown . Each heavy chain is colored as blue and cyan . Each light chain is colored as yellow and orange . ECD of nAChR α1 is shown in magenta . DOI: http://dx . doi . org/10 . 7554/eLife . 23043 . 01310 . 7554/eLife . 23043 . 014Figure 6—figure supplement 1 . Modeling the binding of a complete MG mAb to full-length nAChR ( s ) . ( a ) Modeling of the binding of a complete antibody ( Fab+Fc ) to a nAChR pentamer using the Torpedo nAChR as a template . The ternary Fab35/nAChR α1 ECD/α-Btx complex structure was used to guide the docking of an intact IgG1 antibody ( PDB ID , 1IGY ) ( in surface model ) ( Gomez et al . , 2010 ) onto the full-length Torpedo nAChR pentamer ( PDB ID , 2BG9 ) ( in ribbon model ) ( Luo and Lindstrom , 2015 ) . The two Fab domains ( Fab-I and Fab-II ) and the Fc region of IgG1 are indicated as shown . The extracellular domain of Torpedo nAChR α1 subunit is colored in magenta . Each heavy chain is colored as blue and cyan . Each light chain is colored as yellow and orange . ( b ) Cross-linking of two nAChR pentamers by a single MG antibody . This model shows that this cross-linking event may cause the curving of the cell membrane . ( c ) A hypothetical model wherein the cross-linking of nAChRs by MG antibodies leads to inward membrane curvature and internalization of nAChRs . DOI: http://dx . doi . org/10 . 7554/eLife . 23043 . 014 We also modeled the binding of a complete mAb35 ( Fab+Fc ) to the full-length nAChR pentamer using the structure of an intact IgG1 ( PDB ID , 1IGY ) ( Harris et al . , 1998 ) , which is the same IgG1 subtype as mAb35 , and the Torpedo AChR pentamer ( PDB ID , 2BG9 , 4 Å resolution ) ( Unwin , 2005 ) or the human α4β2 nicotinic receptor ( PDB ID , 5KXI , 3 . 94 Å resolution ) ( Morales-Perez et al . , 2016 ) as templates . The model built on the Torpedo AChR ( Figure 6—figure supplement 1a ) is very similar to that built on the α4β2 nAChR ( Figure 6c ) . The primary differece between the models is that the N-terminal helix of the α1 subunit in the Torpedo AChR appears to adopt a different orientation from the conserved conformation adopted by the corresponding helix in a number of nAChR structures ( Unwin , 2005; Dellisanti et al . , 2007a; Morales-Perez et al . , 2016 ) . The source of this structural difference is currently unknown but our analysis shows that it has little effect on the overall structure of the modeled full antibody-receptor complex . The modeled complex structure shows that the antibody is projected away from the central pore of the receptor , consistent with previous EM analyses ( Beroukhim and Unwin , 1995 ) . After putting all of the molecular components on proper scale and orientation , our model of the full-length antibody/receptor complex suggests that , as a result of steric and geometric constraints , the two Fabs from a single mAb35 antibody are unable to bind to the two α1 subunits within the same nAChR pentamer ( Figure 6c ) . Consequently , MG antibodies will bind two α1 subunits from different pentamers , thereby cross-linking the nAChRs . Our modeling analyses are consistent with previous sucrose density gradient studies showing that mAb35 and similar mAbs cannot bind the two α1 subunits within the same pentamer but can cross-link adjacent nAChR pentamers ( Conti-Tronconi et al . , 1981 ) . Considering that each muscle nAChR pentamer contains two α1 subunits and the relatively high density of nAChRs at the postsynaptic membrane , such inter-pentamer cross-linking mediated by MG antibodies could lead to a super high-order of antibody/receptor complex at the neuromuscular junction . Previous studies have shown that the binding of nAChR by the divalent MG antibodies rather than the monovalent Fab fragment leads to accelerated degradation of the receptor proteins ( Drachman et al . , 1978 ) . This observation suggests that cross-linking of the nAChRs is a critical step in receptor degradation . Our analyses suggest that receptor cross-linking is an intrinsic property of mAb35 and mAb35-like MG antibodies . This cross-linkig could lead to the formation of large antibody-receptor complexes that disrupt the structure and function of the neuromuscular junction and induce the degradation of the nicotinic receptor proteins . Our crystal structure of Fab35 bound to the nAChR α1 ECD provides the first atomic view of the detailed interactions between an EAMG antibody and the nAChR . Our structure reveals that the MIR loop inserts deeply into the antigen-binding pocket of Fab35 and that the adjacent N-terminal helix makes extensive contacts with the CDR-H2 loop of the heavy chain . The binding interface structure and detailed interactions observed in the crystal can be cross-validated with existing biochemical data . Earlier studies mapped the core region of MIR to residues 67–76 ( Barkas et al . , 1988; Tzartos et al . , 1988; Das and Lindstrom , 1989 ) . More recent studies using natively folded nAChR α1/α7 chimera proteins ( Luo et al . , 2009 ) or GFP-fused protein fragments ( Morell et al . , 2014 ) showed that the N-terminal helix ( residues 1–14 ) is also important for high-affinity MG antibody binding . These studies further indicated that other regions of nAChR , including the loop following helix 1 ( residues 15–32 ) ( Luo et al . , 2009 ) and the β5-β6-loop packing against the MIR ( residues 110–115 ) ( Morell et al . , 2014 ) , also contribute to the binding of some MG antibodies . Our structures reveal that the antibody-receptor binding interface indeed centers on the MIR and the N-terminal helix and also includes peripheral regions such as residues 15–23 . Although the β5-β6-loop ( residues 110–115 ) does not contact Fab35 in our structure , it may contribute to antibody binding indirectly by maintaining the natively folded structure of MIR ( Morell et al . , 2014 ) . Specific residues on MIR ( Papadouli et al . , 1990 , 1993; Bellone et al . , 1989 ) and the N-terminal helix ( Morell et al . , 2014 ) have also been analyzed using peptide/protein fragments containing mutations at specific positions for their roles on MG antibody binding . These studies showed that Asn68 and Asp71 of MIR are essential for MG antibody binding while the surrounding residues including Pro69 and Tyr72 showed partial effect when mutated . The essential role of Asn68 and Asp71 was further confirmed by site-directed mutagenesis of N68D and D71K in the intact receptor ( Saedi et al . , 1990 ) . On the N-terminal helix of Torpedo nAChR α1 , two exposed residues , Arg6 and Asn10 , which correspond to Arg6 and Lys10 in human nAChR α1 , respectively , are found to be critical to MG antibody binding by mutational analyses ( Morell et al . , 2014 ) . Most of the nAChR residues found to be essential for antibody binding by mutagenesis studies , including Asn68 and Asp71 from the MIR and Arg6 and Lys10 from the N-terminal helix , indeed correspond to interaction ‘hotspots’ at the Fab35/nAChR α1 interface . These ‘hotspot’ residues anchor multiple interaction networks at the interface and make extensive contacts to the antibody . Although biochemical mapping of antibody-binding residues on nAChR α1 were performed with different antibodies ( e . g . mAb210 and mAb132A ) ( Barkas et al . , 1988; Tzartos et al . , 1988; Das and Lindstrom , 1989; Saedi et al . , 1990; Papadouli et al . , 1990 , 1993; Luo et al . , 2009; Morell et al . , 2014 ) , it is remarkable that these biochemical data agree so well with our crystal structure , suggesting that many MIR-directed antibodies may share high similarities in their binding sites on the nAChR . It has been suggested that EAMG and MG antibodies may bind epitopes different from MIR and that these MG antibodies may be competed off by mAb35 through steric effect rather than direct epitope competition ( Luo et al . , 2009 ) . Our structures ( Figure 1 and Figure 6c ) indeed support this possibility , which is represented by mAb192 ( also see below ) . However , our structural analyses also reveal that many MIR residues at the center of the antibody-receptor are important for the high affinity binding of a variety of MG antibodies ( e . g . , mAb35 , mAb210 , and mAb132A ) . This is a rather surprising finding given the potential heterogeneity of nAChR antibodies mentioned above . An important implication of this finding is that molecular mimicries of the MIR and its immediate surrounding regions could be developed to bind a significant fraction of MIR-directed MG autoantibodies . Such molecules could be useful leads for developing diagnostics and therapeutics for MG ( Tzartos et al . , 1998; Masuda et al . , 2012; Sophianos and Tzartos , 1989 ) . MIR-directed antibodies display a wide range of binding properties . Some such antiboies ( e . g . mAb35 ) exclusively bind natively folded receptor and others ( e . g . mAb210 ) are also capable of binding denatured receptor or isolated MIR peptides ( Luo et al . , 2009; Morell et al . , 2014 ) . The binding mechanisms between different MG antibodies may have subtle or significant differences . Our comparative structural analyses indicate that Fab35 and Fab198 share highly similar binding mechanisms to nAChR α1 , especially in the MIR/N-terminal helix core region . On the other hand , Fab192 seems to have very different nAChR-binding mechanisms from Fab35 and Fab198 even though Fab192 can be competed off by mAb35 through steric effect . Most MIR-directed mAbs , such as mAb35 , bind preferably to folded receptors . Our structure shows that the complete epitope consists of two separated peptide regions ( the MIR loop and the N-terminal helix ) that are required to fold together properly for optimal binding . These structural observations are consistent with previous studies of chimeras and mutants showing that the native conformation of the MIR that permits binding of mAbs 35 and 198 depends on the interaction of the N-terminal helix with the MIR loop ( Luo et al . , 2009; Dellisanti et al . , 2007a ) . Moreover , for mAb35 , a significant amount of binding energy may derive from the insertion of its CDR-H3 loop to the surface pocket on nAChR α1 , whose structure can only form in the natively folded receptor . This unique feature of mAb35 is consistent with the observation that mAb35 is particularly conformation sensitive in binding to nAChR α1 ( Luo et al . , 2009 ) . Our modeling analyses indicate that neighboring subunits in the nAChR pentamer do not make direct contact to Fab35 , suggesting that the binding interactions observed in our crystal structures represent most , if not all , of those in the native complexes between mAb35 and the full-length nAChR pentamer . However , it is known that antibodies in MG serum and mAbs bind mature and pentameric nAChR more tightly than the unassembled nAChR α1 monomer ( Luo et al . , 2009; Merlie and Lindstrom , 1983; Conroy et al . , 1990 ) , and there is evidence that MIR plays an important role in initiating conformational maturation of subunits prior to their assembly into pentameric AChR receptors ( Luo et al . , 2009 ) . Structural differences between monomeric α1 and α1 in a muscle nAChR heteropentamer could account for the different binding affinities . Although the N-terminal helix of the α1 subunit in the full length Torpedo nAChR has a significantly different orientation from that in the monomeric mouse/human α1 crystal structures ( Unwin , 2005; Dellisanti et al . , 2007a ) , it is not clear if this structural difference , which may be due to the limited resolution ( 4 Å ) of the full length Torpedo nAChR , is realistic and hence responsible for the different binding affinities . However , we cannot rule out the possibility that subtle structural differences between the monomeric and pentameric forms of mammalian nAChR α1 could affect the binding affinity , especially for MG autoantibodies , like mAb35 , that are particularly conformation sensitive . Furthermore , the monomeric α1 may be more dynamic than its counterpart in the fully assembled pentamer , and the conformational flexibility could reduce the binding affinity through entropic effects . Our studies reveal two structural insights into the antibody-nAChR interaction that may have important mechanistic and clinical implications . The first is the conservation of the key Fab35-binding residues in a number of nAChR family members , including α2 , α3 , α5 and β3 . This observation suggests potential cross-reactivity of α1-derived MG antibodies to these nAChR family members . As discussed above , because many MIR-directed mAbs share the same binding residues on nAChR α1 , this cross-reactivity may not be limited to mAb35 , but can also occur among other MG autoantibodies . Consistent with this notion , it has been shown that some MG mAbs bind neuronal nAChR subunits . mAb35 was shown to bind chicken nAChR α3 and also suggested to bind human α2 , α3 , α5 and β3 ( Conroy and Berg , 1995 ) . Another MIR-directed antibody , mAb210 , has been used to bind human α5 and β3 ( Kuryatov et al . , 2008; Wang et al . , 1996 ) . The wide expression and diverse physiological functions of nAChR members within and outside the neuronal system are being increasingly recognized , raising an intriguing question whether the cross-reactivity of nAChR with autoantibodies has broader pathological effects than currently recognized . The second is the structural basis of antibody-mediated receptor cross-linking . Our crystal structures reveal a well-defined orientation of the bound antibody with respect to the receptor due to the relatively rigid binding interface . Based on this structural feature , we modeled the complex of full-length nAChR pentamer bound by the intact MG antibody , which suggests that MG antibodies are unlikely to bind the two α1 subunits within the same muscle nAChR pentamer , but rather two α1 subunits from different pentamers , thereby cross-linking the nAChR receptors ( Tzartos et al . , 1981; Conti-Tronconi et al . , 1981 ) . These modeling analyses provide a structural support for previous functional observations that MG antibody-mediated nAChR cross-linking accelerates the degradation of the receptor proteins ( Drachman et al . , 1978 ) . We further noticed that MG antibodies are unlikely to bind two nAChR pentamers oriented vertically on a flat membrane surface even when considering the hinge flexibility of the antibody ( Figure 6—figure supplement 1b ) . Assuming that the nAChR pentamers in the membrane cannot be tilted freely , a potential effect of antibody-mediated receptor cross-linking is membrane curvature . This raises an intriguing question if such distortion of the membrane structure could play a role in the internalization and degradation of nAChRs at the neuromuscular junction ( Figure 6—figure supplement 1c ) . While the detailed mechanisms by which antibody-mediated receptor cross-linking induces the receptor degradation remain to be elucidated , molecular mimicries of the MIR should prevent such cross-linking and degradation of nAChR by competitive binding to MG antibodies ( Sophianos and Tzartos , 1989 ) . Our studies suggest that it is possible to develop drug molecules to inhibit the binding of a large fraction of MG antibodies to nAChR and related pathological immune reactions , and the crystal structures presented here provide a basis for developing such drug molecules . The mouse α211 construct was provided by Dr . Zuo-Zhong Wang , Zilkha Neurogenetic Institute , Department of Cell and Neurobiology , Keck School of Medicine , University of Southern California . The detailed construction information is as previously described ( Yao et al . , 2002 ) . Briefly , a Flag-tag and a His-tag were added at the N-terminus and the C-terminus , respectively , for higher expression and purification purposes . The construct was truncated at the 211th amino acid from the N-terminal of nAChR α1 subunit without a signal sequence , and three point mutations ( V8E , W149R , and V155A ) were introduced for improved solubility and stability ( Dellisanti et al . , 2007a; Chen , 2010 ) . The human α211 construct was designed based on the mouse α211 construct , and the synthesized cDNA was ordered from GenScript . The gene codons were optimized for yeast , and cloned into a pPICZαA vector using the EcoRI and XbaI sites . Both mouse and human α211 constructs were linearized by digesting with SacI restriction enzyme ( New England Biolabs ) and transformed into KM71H of P . pastoris ( Invitrogen ) by electroporation . The transformants were plated on YPDS plates , which contained 100 µg mL−1 Zeocin . Plates were incubated at 30°C for 3–5 days until colonies formed , and several colonies were restreaked on fresh YPDS plates . Pre-inoculation was made by seeding a single colony in 30 mL BMGY medium . The culture was incubated at 30°C with shaking overnight . About 5–7 mL of this culture was used to inoculate 500 mL of BMGY in a 2 L baffled flask ( total 2 L of culture ) . The inoculated culture was incubated at 30°C with shaking to OD600 value of 6 . Cells were harvested by centrifugation at 3000×g for 15 min at room temperature . The supernatant was discarded , and cell pellets were resuspended in 400 mL of BMMY medium for induction . The resuspended culture was divided between two 2 L baffled flasks ( 200 mL each ) and incubated at 20°C with shaking for 72 hr . 100% methanol was added every 24 hr to a final concentration of 0 . 5% ( v/v ) to induce protein expression . After 72 hr of induction , cells were harvested by centrifuging at 6000×g for 20 min at room temperature . Protein purification proceeded with the supernatant as the protein was secreted . Ni-NTA agarose beads ( QIAGEN ) were incubated with the supernatant at 4°C overnight with end-over-end rotation . The protein was eluted with elution buffer ( 50 mM NaH2PO4 , pH 7 . 8 , 0 . 5 M KCl , 10% ( v/v ) glycerol , and 500 mM imidazole ) after washing with washing buffer containing 20 mM imidazole and 0 . 1% Triton X-100 to remove loosely bound proteins . The eluted protein was concentrated and ran over a size exclusion column ( Superdex 75 10/300 GL , GE Healthcare ) with 20 mM HEPES , pH 7 . 5 and 150 mM NaCl buffer for further purification . After each peak fraction was analyzed by OD280 measurement and SDS-PAGE , fractions containing α211 were pooled and concentrated for further experiments . Hybridoma cells of mAb35 were purchased from American Type Culture Collection ( ATCC ) . The cells were maintained in DMEM medium containing 1 . 97 g L−1 NaHCO3 and 10% fetal bovine serum ( FBS ) . The cells were cultured in a 37°C incubator with 5% CO2 and subcultured every 2 to 3 days with cell density between 1 × 105 and 1 × 106 cells mL−1 . For protein production , the cell culture was incubated at 37°C for several days until the medium color changed to yellow . After 7–10 days of incubation at 37°C , the cell culture was harvested by centrifuging at 6000×g for 15 min . Affinity purification was performed using Protein G Sepharose 4 Fast Flow ( GE Healthcare ) . The supernatant and beads were incubated at room temperature for 2 hr with rotation . The beads were washed with washing buffer ( 20 mM sodium phosphate , pH 7 . 0 ) , and the protein was eluted with elution buffer ( 0 . 1 M glycine-HCl , pH 2 . 7 ) . Due to the low pH of the elution buffer , a neutralizing buffer ( 1 M Tris-HCl , pH 9 . 0 ) was added to the collection tubes ( 60 to 200 µL mL−1 elute ) prior to collection . After checking the presence of the protein with SDS-PAGE gels , the protein elution was concentrated and ran over a size exclusion column ( Superdex 200 10/300 GL , GE Healthcare ) . The fractions of protein peak were pooled and concentrated for further study . Purified mAb35 was buffer-exchanged into digestion buffer ( 20 mM sodium phosphate , pH 7 . 0 , 10 mM EDTA and 20 mM cysteine-HCl; adjust pH to 7 . 0 right before use ) using Zeba Spin columns ( Thermo Scientific ) , and the resulting sample was incubated with immobilized papain beads ( Thermo Scientific ) . The sample was rotated at 30°C overnight , and the flow through was collected . The protein was identified by SDS-PAGE , and then the sample was concentrated down to ~500 µL while exchanging buffer to Mono Q Buffer A ( 20 mM sodium phosphate , pH 7 . 0 ) . An anion exchange column ( Mono Q HR 5/5 , GE Healthcare ) was used to separate the Fab portion from the rest using a salt gradient of 0–100% Buffer B ( 20 mM sodium phosphate , pH 7 . 0 and 1 M NaCl ) . Fab fractions ( Fab35 ) were collected and concentrated for gel shift assay and α211 complex purification . α211 , α-bungarotoxin ( α-Btx ) and Fab35 were mixed in an equimolar ratio , and the mixture was incubated on ice for 1 hr . For the binding specificity experiment , α9-3Mut , which contains three point mutations ( V189W , S191F and G193S ) in the loop C that were designed to enhance the binding of α-Btx , was prepared in the same manner as α211 ( Kaori Noridomi , Ph . D . dissertation , University of Southern California , 2015 ) . A 10% native PAGE gel was run at 4°C for 3 . 5 hr at 100–120 V , ~15 mA , with 1× TBE buffer . Bands were detected with Coomassie Blue staining . α211 , α-Btx and Fab35 were mixed at a 1:1 . 5:1 . 5 molar ratio , and the mixture was incubated on ice for 1 hr . Ni-NTA purification was performed to remove excess α-Btx and Fab35 . The elution was run over a size exclusion column ( Superdex 200 10/300 GL , GE Healthcare ) with 20 mM HEPES , pH 7 . 5 and 150 mM NaCl buffer . Two peaks were obtained , and fractions of each peak were pooled separately and concentrated for crystallization . The purified and concentrated ternary complex of α211 ( both mouse and human ) /Fab35/α-Btx was diluted to 2 . 5 mg mL−1 . Crystals were screened using Crystal Screen , Crystal Screen 2 and Index kits ( Hampton Research ) by a hanging drop method at room temperature . Each reservoir contained 0 . 5 mL of screening solution , and each drop contained 0 . 5 μL of protein and 0 . 5 μL of reservoir solution . Three conditions gave initial hits , and one condition was selected to optimize a crystallizing condition . The optimized condition was 0 . 1 M sodium cacodylate trihydrate , pH 6 . 5 , 0 . 1–0 . 15 M calcium acetate hydrate and 18–20% ( w/v ) PEG 8K . Rod-shaped crystals were grown as bundle with the size of 10 μm × 20–100 μm × 200 μm . Crystals were harvested using harvest solution ( 0 . 2 M calcium acetate hydrate , 0 . 1 M sodium cacodylate trihydrate , pH 6 . 5 and 30% ( w/v ) PEG 8K ) and cryo-solution ( harvest solution +20% glycerol ) as follows . Crystals were transferred to different concentrations of harvest/cryo mixture solution to protect crystals from osmotic shock ( in order of 100% harvest solution , 3:1 harvest/cryo solution , 1:1 harvest/cryo solution , 1:3 harvest/cryo solution and 100% cryo solution ) . Each incubation time was approximately 5–10 min . The cryoprotected crystals were fished using 100-300 μm Hampton CryoLoop ( Hampton Research ) and flash-cooled in liquid nitrogen . Diffraction data were collected at the Advanced Photon Source ( APS ) beamline 23-ID-B at Argonne National Laboratory using a 10 µm × 10 µm beam ( λ = 1 . 0332 Å , 12 . 000 keV ) with the attenuation factor of 5 . 0 and a MARmosaic 300 CCD detector . The detector distance was 300 . 0 mm . The oscillation range and the exposure time per frame were 0 . 5° and 2 . 0 s , respectively . Data were processed and scaled using HKL2000 package ( Otwinowski and Minor , 1997 ) . Both human and mouse complex crystals belong to the space group C2 , with unit cell dimensions of a = 160 . 0 Å , b = 42 . 1 Å , c = 136 . 5 Å , β = 117 . 1°; and a = 159 . 9 Å , b = 42 . 0 Å , c = 137 . 6 Å , β = 116 . 5° , respectively . The crystal structures were solved by molecular replacement using PHASER MR ( McCoy et al . , 2007 ) in CCP4 ( Collaborative Computational Project Number 4 , 1994 ) and the coordinates of α211/α-Btx complex ( PDB ID , 2QC1 ) and Fab198 ( PDB ID , 1FN4 , modified to poly-Ala ) ( Dellisanti et al . , 2007a; Poulas et al . , 2001 ) . Refmac5 was used for the final refinement of the structures ( Murshudov et al . , 2011 , 1997 ) . The amino acid sequence of mAb35 was obtained from an antibody sequencing company , MCLAB Molecular Laboratories ( San Francisco , CA ) , and the sequence of Fab35 was added using Coot ( Emsley et al . , 2010 ) . Additional model building in Fab35 was carried out with O ( Jones et al . , 1991 ) . Crystallographic analysis and refinement statistics are summarized in Supplementary file 1 . Coordinates and structure factors have been deposited in the Protein Data Bank under accession codes PDB: 5HBT ( Fab35/human nAChR α1 ECD/α-Btx ) and 5HBV ( Fab35/mouse nAChR α1 ECD/α-Btx ) .	Myasthenia gravis is a disease that causes chronic weakness in muscles . It affects more than 20 in every 100 , 000 people and diagnosis is becoming more common due to increased awareness of the disease . However , most current treatments only temporarily relieve symptoms so there is a need to develop more effective therapies . The disease occurs when the immune system produces molecules called antibodies that bind to and destroy a receptor protein called nAChR . This receptor is normally found at the junctions between nerve cells and muscle cells , and its destruction disrupts communication between the nervous system and the muscle . However , it is not known exactly how these antibodies bind to nAChR , partly due to the lack of a detailed three-dimensional structure of the antibodies and nAChR together . The human nAChR protein is made up of several subunits , including one called alpha1 that is the primary target of Myasthenia gravis antibodies . Noridomi et al . used a technique known as X-ray crystallography to generate a highly detailed three-dimensional model of the structure of the alpha1 subunit with an antibody from rats that acts as in a similar way to human Myasthenia gravis antibodies . The structure reveals the points of contact between the antibodies and a core region of the nAChR alpha1 subunit and suggests that many different Myasthenia gravis antibodies may bind to nAChR in the same way . These findings may aid the development of drugs that bind to and disable Myasthenia gravis antibodies to relieve the symptoms of the disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "neuroscience" ]	2017	Structural insights into the molecular mechanisms of myasthenia gravis and their therapeutic implications
Avian influenza outbreaks have been occurring on smallholder poultry farms in Asia for two decades . Farmer responses to these outbreaks can slow down or accelerate virus transmission . We used a longitudinal survey of 53 small-scale chicken farms in southern Vietnam to investigate the impact of outbreaks with disease-induced mortality on harvest rate , vaccination , and disinfection behaviors . We found that in small broiler flocks ( ≤16 birds/flock ) the estimated probability of harvest was 56% higher when an outbreak occurred , and 214% higher if an outbreak with sudden deaths occurred in the same month . Vaccination and disinfection were strongly and positively correlated with the number of birds . Small-scale farmers – the overwhelming majority of poultry producers in low-income countries – tend to rely on rapid sale of birds to mitigate losses from diseases . As depopulated birds are sent to markets or trading networks , this reactive behavior has the potential to enhance onward transmission . Livestock production systems have been a major driver of novel pathogen emergence events over the past two decades ( Gao et al . , 2013; Guan et al . , 2002; Rohr et al . , 2019 ) . The conditions enabling the emergence and spread of a new disease in the human population partly depend on human behavioral changes , like hygiene improvements or social distancing , in the face of epidemiological risks ( Funk et al . , 2010 ) . The same observation applies to disease emergence and spread in livestock populations as farmers adapt their farm management to maximize animal production and welfare while limiting cost in a constantly changing ecological and economic environment ( Chilonda and Van Huylenbroeck , 2001 ) . Poultry farming generates substantial risk for emergence of novel infectious diseases . It is now the most important source of animal protein for the human population and the industry is changing rapidly ( FAOSTAT , 2019 ) . The link between poultry sector expansion and pathogen emergence is exemplified by the worldwide spread of the highly pathogenic form of avian influenza ( AI ) due to the H5N1 subtype of influenza A , after its initial emergence in China in 1996 ( Guan et al . , 2002; Guan and Smith , 2013 ) . Highly Pathogenic Avian Influenza ( HPAI ) causes severe symptoms in the most vulnerable bird species ( including chicken , turkey , and quail ) , with mortality rates as high as 100% reported in broiler flocks ( OIE , 2018 ) . Some subtypes of AI viruses have caused infection in humans , including H5N1 , H5N6 , H7N9 and H9N2 , with potentially severe illness and , in the cases of H7N9 and H5N1 , a high case-fatality rate ( Chen et al . , 2013; Claas et al . , 1998; Peiris et al . , 1999; Yang et al . , 2015 ) . So far , reports of human-to-human transmission of these subtypes of influenza have been either absent or anecdotal , but the risk that they make the leap to a human pandemic is a persistent if unquantifiable threat to public health ( Imai et al . , 2012 ) . While HPAI does not persist in poultry populations in most affected countries , it has become endemic in parts of Asia and Africa and is periodically re-introduced into other areas like Europe and North America ( Lai et al . , 2016; Li et al . , 2014 ) . In affected countries , major factors influencing HPAI epidemiology appear to be farm disinfection , poultry vaccination , and marketing of potentially infected birds through trade networks , all of which depend on farmers’ management decisions ( Biswas et al . , 2009; Desvaux et al . , 2011; Fasina et al . , 2011; Henning et al . , 2009; Kung et al . , 2007 ) . It is still unclear how and to what extent changes in outbreak risk or mortality risk affect the behavior of poultry farmers . An anthropological study in Cambodia showed that high levels of farmer risk awareness associated with HPAI did not translate into major changes in their farming practices ( Hickler , 2007 ) . Qualitative investigations conducted in Vietnam , Bangladesh , China , and Indonesia reported that farmers sometimes urgently sell or cull diseased poultry flocks as a way to mitigate economic losses , but evidence of this behavior’s onward epidemiological impact was not available ( Biswas et al . , 2009; Delabouglise et al . , 2016; Padmawati and Nichter , 2008; Sultana et al . , 2012; Zhang and Pan , 2008 ) . Additionally , it is unknown whether poultry farmers increase application of disinfection practices or vaccination rates against avian influenza in response to disease outbreaks occurring in their flocks . Changes in farm management caused by variations in epidemiological risk have not been quantified for any livestock system that we are aware of , primarily because of the lack of combined epidemiological and behavioural data in longitudinal studies of livestock disease ( Hidano et al . , 2018 ) . Ifft et al . , 2011 compared the evolution of chicken farm sizes and disease prevention in administrative areas with different levels of HPAI prevalence in Vietnam , and Hidano and Gates , 2019 modelled the effect of cattle mortality and production performance on the frequency of sales and culling in New Zealand dairy farms . One limitation of these two studies is that the dynamics were observed over year-long time steps , which does not allow for a precise estimation of the timing of farmer response after the occurrence of disease outbreaks and the potential feedback effect of this response onto the resulting outbreaks or epidemics . Vietnam has suffered human mortality and economic losses due to HPAI . The disease has been endemic in the country since its initial emergence in 2003–2004 ( Delabouglise et al . , 2017 ) . Small-scale poultry farming is practiced by more than seven million Vietnamese households , mostly on a scale of fewer than 100 birds per farm ( General Statistics Office of Vietnam , 2016 ) . In addition to HPAI , other infectious diseases severely affect this economic sector , including Newcastle disease , fowl cholera , and Gumboro , which are all endemic despite the availability of vaccines for their control ( OIE , 2019 ) . We present a longitudinal study of small-scale poultry farms where we aimed to characterize the effect of disease outbreaks on livestock harvest rate ( i . e . rate of removal by sale or slaughter ) and on two prevention practices , vaccination and farm disinfection . This longitudinal farm survey was conducted on small-scale poultry farms in the Mekong river delta region of southern Vietnam ( Delabouglise et al . , 2019 ) . Fifty three farms were monitored from June 2015 to January 2017 . Monthly questionnaires were used to collect farm-level information on poultry demographics ( number , introduction , death and departure of birds ) , mortality ( cause of death , observed clinical symptoms ) and management by farmers . The main poultry species kept on these farms was chicken , with ducks and Muscovy ducks as the other two primary relevant species held . Farmers kept an average number of 79 chickens , 53 ducks and 7 Muscovy ducks per farm over the 20 month study period . Each farm’s poultry were classified into ‘flocks’ , defined as groups of birds of the same age , species , and production type . Figure 1 illustrates the farms’ structure and dynamics . Broiler chicken flocks were kept for 15 . 5 weeks on average after which most chickens were harvested and a minority was consumed or kept on the farm for breeding and egg production ( Delabouglise et al . , 2019 ) . We fit mixed-effects general additive models ( MGAM ) with three different dependent variables: a ‘harvest model’ of the probability of harvesting ( i . e . selling or slaughtering ) chicken broiler flocks at a particular production stage ( data points are flock-months ) , an ‘AI vaccination model’ of the probability of performing AI vaccination on chicken broiler flocks which had never received AI vaccination ( data points are flock-months ) , and a ‘disinfection model’ of the probability of disinfecting farm facilities ( data points are farm-months ) . Disease outbreaks were included in each model as independent categorical variables . Disease outbreaks refer to the occurrence of poultry mortality attributable to an infectious disease in the corresponding farm at different time intervals before the corresponding month . Specifically , outbreaks were defined by the death of at least two birds of the same species with similar clinical symptoms in the corresponding farm in the same month , one month prior , and two months prior . For the harvest model , only outbreaks in chickens were considered . For the AI vaccination model , outbreaks in chickens and outbreaks in any other species were included as two separate covariates . For the disinfection model , outbreaks in any of the species present in the farm were considered . In chickens , outbreaks with ‘sudden deaths’ ( i . e . the death of chickens less than one day after the onset of clinical symptoms ) are considered as being indicative of HPAI infection ( Mariner et al . , 2014 ) . Therefore , we created two sub-categorical variables for outbreaks in chickens , with sudden deaths ( OS , ‘outbreaks sudden’ ) and with no sudden deaths ( ONS , ‘outbreaks not sudden’ ) . The three dependent variables are likely influenced by several other farm- , flock- , and time-related factors , justifying the inclusion of control covariates which are reported in Table 1 and described in detail in the ‘Materials and methods’ . A total of 1656 broiler chicken flock-months were available for analysis . They belonged to 391 chicken flocks present on 48 farms . In 18 . 8% of flock-months non-sudden outbreaks ( ONS ) were observed in chickens on the same farm , 1 . 6% of flock-months saw sudden outbreaks ( OS ) in chickens on the same farm , and 7 . 2% of flock-months saw disease outbreaks in poultry of other species on the same farm ( Table 1 ) . The percentages are very similar for outbreaks occurring one month prior and two months prior since they are averaged over similar sets of months , with differences mostly related to outbreak frequency in the two first months and two last months of the study period . Additional descriptive statistics on control covariates are described in Table 1 . Out of 1656 broiler chicken flock months , 1503 flock-months were selected for the harvest analysis after excluding data points with new-born chicks and flock-months in which all the chickens had died ( see Materials and methods ) . No harvest occurred in 995 flock-months , complete harvest occurred in 258 flock-months , and partial harvest occurred in 250 flock-months . The probability of harvest during a month , with partial harvests weighted appropriately , was 23 . 9% . Excluding flock-months of already vaccinated chickens ( and some with missing data ) , 1318 flock-months were selected for the AI vaccination analysis ( see Materials and methods ) . AI vaccination was performed in 7 . 5% ( 99/1318 ) of flock-months . The 99 vaccinated flocks were from 29 different farms ( out of 48 farms keeping broiler chickens ) . For the disinfection model , 858 farm-months belonging to 52 farms were included ( see Materials and methods ) . During 552 farm-months the farm was fully disinfected , during 259 farm-months the farm was not disinfected at all , and during 47 farm-months disinfection was performed for some ( but not all ) of the flocks present in the farm . The probability of disinfection during a month , with partial disinfections weighted appropriately , was 67 . 4% . The best fit statistical models and their parameter values are summarized in Table 2 . Fitted spline functions cannot be elegantly summarized by their coefficients and are displayed graphically in Figures 2 and 3 . The harvest model showed support for associations between flock- and farm-level covariates , particularly the difference between flock age and age at maturity and the probability of harvesting broiler chickens . The model explained 34 . 2% of the observed deviance . There was no statistical support for a temporal auto-correlation of the probability of harvest of broiler chicken flocks on a given farm ( Table 2 ) . As the interaction term between flock size ( n ) and outbreak occurrence was significant ( p<0 . 01 ) but difficult to interpret ( displayed in Supplementary file 1 ) , we separated the flocks into large and small . A threshold value of 16 birds per flock gave the lowest Akaike Information Criterion ( AIC ) ( when using a categorical variable indicating small flock or large flock ) , and flocks of 16 birds or fewer ( 52% of all flocks ) were designated as small while flocks of 17 or more ( 48% of all flocks ) were designated as large . As expected , the probability of harvest was found to be strongly dependent on the difference ( δt ) between the flock age and the anticipated age at maturity , with older flocks being more likely to be sold . The probability of harvest was close to zero when δt < −15 weeks , i . e . flocks that are more than 15 weeks away from maturity . The probability of harvest increased steeply from δt = −10 to δt = 0 . For δt > 0 ( flocks past their age at maturity ) , the probability of harvest was consistently high but lower than 100% and did not depend on age . Larger flocks had a steeper increase in harvest probability as a function of δt; once past the age at maturity ( δt > 0 ) , the estimated probability of harvest for large flocks was higher ( interquartile range: 41–61% ) than for small flocks ( interquartile range: 30–41% ) ( Figure 2 ) . Disease outbreaks substantially affected the likelihood of harvest of broiler chickens . The probability of harvest of small flocks was significantly higher on farms that had experienced a non-sudden outbreak ( ONS ) in chickens in the same month ( odds ratio ( OR ) = 2 . 06; 95% confidence interval ( CI ) : 1 . 23–3 . 45 ) or the previous month ( OR = 2 . 06; 95% CI: 1 . 17–3 . 62 ) and was lower on farms that had experienced an ONS in chickens two months prior ( OR = 0 . 41; 95% CI: 0 . 19–0 . 92 ) . The probability of harvest of small flocks was much higher on farms that had experienced a sudden outbreak ( OS ) in the same month ( OR = 9 . 34: 95% CI: 2 . 13–40 . 94 ) . We used the fitted model to predict the mean harvest proportion in the study population with and without outbreak . Estimated mean harvest proportions of small flocks were 17% ( no outbreak ) , 28% ( ONS ) , and 56% ( OS ) when considering outbreaks occurring in the same month; this corresponded to harvest increases of 56% and 214% for ONS and OS outbreaks , respectively . Estimated mean harvest proportion was 18% ( no outbreak ) and 28% ( ONS ) when considering outbreaks one month prior; this corresponded to a 56% increase in harvest in case of ONS one month prior . Mean harvest proportions were 20% ( no outbreak ) and 11% ( ONS ) when considering outbreaks two months prior , indicating a 47% decrease in harvest in case of ONS two months prior . For large flocks , ONS in chickens ( in any month current or previous ) did not have any effect on the harvest of broiler chickens ( the removal of ONS variables decreased the model AIC ) . The occurrence of OS in chickens one month prior may be positively associated with early harvest with an estimated 76% increase in harvest proportion ( OR = 3 . 89; 95% CI: 0 . 82–18 . 46; p=0 . 09 ) . However , we do not have sufficient statistical power to support this association . In the last six months of data collection , farmers were asked to indicate the destination of harvested birds . Based on these partial observations , flocks harvested during or one month after outbreaks in chickens ( OS or ONS ) were more likely to be sold to traders and less likely to be slaughtered at home ( Table 3 ) . The likelihood of harvest was also positively correlated with the number of other broiler chickens present on the farm ( Supplementary file 1 , p < 0 . 01 ) . It was not found to be affected by the concomitant introduction of other flocks , vaccination status , or calendar time ( T ) . The farm random effect was significant for large flocks ( σ = 0 . 74; 95% CI: 0 . 47–1 . 17 ) and not significant for small flocks . The number of outbreaks with sudden deaths is relatively small ( 11 small flock-months and 14 large flock-months occurred on farms experiencing an OS in the same month ) and OS are potentially subject to misclassification , depending on how regularly farmers check on their chickens . Therefore , in order to ensure the robustness of our result , we conducted a separate analysis with merged OS and ONS categories . The results are displayed in Supplementary file 2 and Figure 2—figure supplement 1 . The probability of harvest of small flocks was significantly higher on farms that had experienced an outbreak in chickens in the same month ( Odds ratio ( OR ) = 2 . 34; 95% CI: 1 . 43–3 . 81 ) or the previous month ( OR = 1 . 96; 95% CI: 1 . 14–3 . 37 ) and was lower in farms that had experienced an outbreak in chickens two months prior ( OR = 0 . 45; 95% CI: 0 . 22–0 . 92 ) . For large flocks , there was no statistical support for outbreaks in chickens having an effect on the harvest of broiler chickens . The AI vaccination model showed support for an effect of flock size on vaccination , while explaining 71 . 9% of the observations’ deviance . The likelihood of broiler chicken vaccination against AI strongly increased with flock size; probability of vaccination was almost zero for flocks of 16 birds or fewer and nearly 100% for flocks of more than 200 birds ( Figure 3A ) . Vaccination was preferentially performed at 4 . 3 weeks of age ( Figure 3B ) . Flocks kept indoors or in enclosures had a substantially higher chance of being vaccinated than flocks scavenging outdoors ( OR = 24 . 6; CI: 6 . 32–95 . 6 ) . Harvested flocks were less likely to receive an AI vaccination ( OR = 0 . 01; CI: 0–0 . 37 ) . The likelihood of AI vaccination was dependent on calendar time: it increased over the September-January period and decreased during the rest of the year ( Figure 3C ) . There was no statistical support for a temporal auto-correlation of the probability of vaccination of broiler chicken flocks against AI on a given farm . The farm random effect was significant ( σ = 2 . 86; CI: 1 . 88–4 . 35 ) . We failed to obtain convergence when fitting the specific effects of OS and ONS in chickens , so we used an aggregate variable ‘outbreak in chickens’ instead ( Table 2 ) . Broiler chicken flocks were more likely to be vaccinated if an outbreak had occurred in the same month in other species ( OR = 4 . 62; CI: 1 . 08–19 . 72; p=0 . 04 ) and less likely to be vaccinated if an outbreak had occurred two months prior in chickens ( OR = 0 . 27; CI: 0 . 08–0 . 89; p=0 . 03 ) . These two effects were weakly significant and should be interpreted with caution ( Table 2 ) . The coefficients for interaction terms between outbreak occurrence and flock size were not significantly different from zero . The number of broiler Muscovy ducks present in the farm had a negative effect ( p=0 . 03 ) and the number of layer ducks and layer Muscovy ducks had a positive effect ( both p=0 . 03 ) on the probability of AI vaccination ( Table 2 ) . The disinfection model showed evidence that larger farms were more likely to report routine disinfection of their premises; the model explained 61 . 9% of the observations’ deviance . Probability of disinfection on farms was auto-correlated in time ( likelihood ratio test for 1 month AR-model on residuals; p<0 . 0001 ) ; this was not observed for the harvest or vaccination models ( both p>0 . 3 ) . Consequently , the disinfection model was improved by fitting an AR-1 autoregressive model using the ‘gamm’ routine of the ‘mgcv’ R package . The estimated AR-1 autoregressive coefficient was high ( ρ = 0 . 71 ) . The likelihood of disinfection of farm facilities increased with the number of layer-breeder hens ( OR = 1 . 3; CI: 1 . 12–1 . 51; p=0 . 001 ) , layer-breeder ducks ( OR = 1 . 25; CI: 1 . 02–1 . 53; p=0 . 03 ) , and to a lesser extent broiler chickens ( OR = 1 . 07; CI: 1 . 01–1 . 13; p=0 . 02 ) present on the farm ( Table 2 ) . Farm disinfection appeared to have a seasonal component . It was least likely in October-November and most likely in the January-April period ( Figure 3D ) . It was not found to be affected by the occurrence of outbreaks ( no decrease in AIC when including outbreak occurrence ) . Regions like the Mekong river delta combine high human population density , wildlife biodiversity , and agricultural development . As such , they are considered hotspots for the emergence and spread of novel pathogens ( Allen et al . , 2017 ) . The high density of livestock farmed in semi-commercial operations with limited disease prevention practices further increases the risk of spread of emerging pathogens in livestock and their transmission to humans ( Henning et al . , 2009 ) . In-depth studies of poultry farmers’ behavioral responses to disease occurrence in animals are needed to understand how emerging pathogens – especially avian influenza viruses – may spread and establish in livestock populations and how optimal management policies should be designed . To the best of our knowledge , this study is the first to provide a detailed and quantified account of the dynamics of livestock management in small-scale farms and its evolution in response to changing epidemiological risks shortly after disease outbreaks occur . While our analysis was performed on a geographically restricted area , the decision-making context of the studied sample of farmers is likely to be applicable to a wide range of poultry producers in low- and middle-income countries . Small-scale poultry farming , combining low investments in infrastructure , no vertical integration , and subject to limited state control on poultry production and trade , is common in most regions affected by avian influenza , in Southeast Asia , Egypt , and West Africa ( Burgos et al . , 2008a; Hosny , 2006; Obi et al . , 2008; Sudarman et al . , 2010 ) . Additional longitudinal surveys using a similar design should be carried out in other countries and contexts to assess the presence or absence of the behavioral dynamics observed here . In our longitudinal study , owners of small chicken broiler flocks resorted to early harvesting of poultry , also referred to as depopulation , as a way to mitigate losses from infectious disease outbreaks . The revenue earned from the depopulation of flocks might be low , either because birds are still immature or because traders use disease symptoms as an argument to decrease the sale price . Nevertheless , depopulation allows the farmer to avoid a large revenue loss resulting from disease-induced mortality or the costs of management of sick or dead birds . More importantly , farmers avoid the cost of feeding chickens at high risk of dying and prevent the potential infection of subsequently introduced birds . Our results also suggest that the depopulation period , which lasts approximately two months , is followed by a ‘repopulation’ period during which farmers lower their harvest rate , possibly to increase their pool of breeding animals in order to repopulate their farm . The epidemiological effect of chicken depopulation is likely twofold: on the one hand it may slow the transmission of the disease on the farm , since the number of susceptible and infected animals is temporarily decreased ( Boni et al . , 2013 ) ; on the other hand , since most poultry harvested during or just after outbreaks were sold to itinerant traders or in markets , depopulation increases the risk of dissemination of the pathogens through trade circuits ( Delabouglise and Boni , 2020 ) . There is epidemiological evidence that poultry farms can be contaminated with HPAI through contact with traders who purchase infectious birds and that infectious birds can contaminate other birds at traders’ storage places and in live bird markets ( Biswas et al . , 2009; Fournié et al . , 2016; Kung et al . , 2007 ) . Overall , chicken depopulation may reduce local transmission at the expense of long-distance dissemination of the pathogen . The rapid sale of sick birds also exposes consumers and actors of the transformation and distribution chain ( traders , slaughterers , retailers ) to an increased risk of infection with zoonotic diseases transmitted by poultry , like avian influenza ( Fournié et al . , 2017 ) . Large flocks appear to be less readily harvested upon observation of disease mortality . Farmers may depopulate large flocks only upon observation of sudden deaths , but the number of observations in our study is too small to demonstrate statistical significance of this effect . The likely reason for this difference is that the sale and replacement of larger flocks incurs a higher transaction cost . While small flocks are easily collected and replaced by traders and chick suppliers in regular contact with farmers , the rapid sale of larger flocks probably requires the intervention of large-scale traders or several small-scale traders with whom farmers have no direct connection , and who may offer a lower price per bird . When farm production increases , farmers tend to rely on pre-established agreements with traders , middlemen , or hatcheries on the sale dates in order to reduce these transaction costs , giving them little possibility to harvest birds at an earlier time ( Catelo and Costales , 2008 ) . The timing of harvest of broiler chickens is also affected by farm-related factors , as shown by the significance of the farm random effect in large flocks . Indeed , farmers have different economic strategies , some aiming at optimizing farm productivity and harvesting broilers as soon as they reach maturity , and others using their poultry flocks as a form of savings and selling their poultry whenever they need income or when prices are high ( ACI , 2006 ) . For the latter category , the sale of chickens presumably depends on variables which were not captured in this study , like changes in market prices , economic shocks affecting the household , a human disease affecting a member of the household , or celebrations . Those variables should be captured in future surveys in order to improve the predictive power of harvest models . Another limit of the model is the use of a proxy of the chicken weight combining age , age at maturity , and flock size , rather than the actual weight , which is difficult to monitor in a longitudinal study of this size . While government-supported vaccination programs have been proposed as a suitable tool to control AI in small scale farms with little infrastructure ( FAO , 2011 ) , in this survey AI vaccination was almost exclusively performed in large flocks kept indoors or in an enclosure . Vaccination against AI is believed to be inexpensive for farmers as vaccines are supplied for free by the sub-department of animal health of Ca Mau province and performed by local animal health workers . However , vaccination may still involve some fixed transaction cost as farmers have to declare their flocks to the governmental veterinary services beforehand . Also it is possible that small flocks , being less likely to be sold to distant larger cities ( Tung and Costales , 2007 ) , are less likely to have their vaccination status controlled , making their vaccination less worthwhile from the farmers’ perspective . Crucially , it is these smaller flocks that are more likely to be sold into trading network during outbreaks . Finally , farmers' willingness to expand their production , invest in farm infrastructure , and implement AI prevention are likely correlated . Farms with a large breeding-laying activity tend to invest more in preventive actions ( disinfection and vaccination ) compared to farms specialized in broiler production . This may reflect a higher individual market value of layer-breeder hens compared to broiler chicks , making their protection more worthwhile . While vaccination against AI and disinfection appear to depend on individual farmer attitude , as shown by the significance of the farm random effects , they still vary over time when viewed across all farms ( Figure 1 ) . Contrary to harvesting behavior , these preventive actions have a seasonal component ( Figure 3C and D ) indicating a willingness to maximize the number of vaccinated broiler chickens and the protection against other diseases during the January-March period . The January-March period is the period of lunar new year celebrations in Viet Nam , commonly associated with higher poultry market prices and an increased risk of disease transmission , as has been observed for avian influenza ( Delabouglise et al . , 2017; Durand et al . , 2015 ) . In response , farmers tend to invest more in disease prevention practices at this time and veterinary services provide more vaccines and disinfectant for free . Farm disinfection has a significant temporal autocorrelation component and is unaffected by disease outbreaks , indicating that farmers are slower at adapting this practice to changing conditions . Some events may affect the frequency of vaccination and disinfection on a long time frame . For example , the peak in AI vaccination observed at the end of 2015 can be interpreted as a part of a long-term response to the high HPAI incidence reported in early 2014 ( Delabouglise et al . , 2017 ) . The time period of the present study is too short to provide a statistical support for these long term dynamics . The data from this study were recorded at farm level on monthly basis , which limits the risk of recall bias . It was an easy task for farmers participating in the survey to report the number of deaths and associated clinical symptoms . We cannot , however , totally exclude the risk of misclassification of disease outbreaks , especially the misclassification of outbreaks in chickens as ‘sudden’ , as it is influenced by the frequency of inspection of chickens flocks by farmers and other members of the households . The main result of the study is that , as poultry flock size decreases , farmers increasingly rely on depopulation rather than preventive strategies to limit economic losses due to infectious diseases . In the current context , depopulation mainly results in the rapid transfer of potentially infected chickens to trade systems , increasing the risk of pathogen dissemination . In response , governments may use awareness campaigns directed at actors of poultry production systems to communicate information on the public health risks associated with the trade of infected birds . However , if the economic incentives for depopulating are high enough , communication campaigns may fail to produce noticeable results . Small-scale farmers could play an active role in the control of emerging infectious diseases if they were given the opportunity to depopulate their farm upon disease detection without disseminating pathogens in trade circuits , as theoretical models predict that depopulation can maintain a disease-free status in farming areas ( Delabouglise and Boni , 2020 ) . Policymakers may be able to encourage the establishment of formal trade agreements enabling and encouraging ‘virtuous’ management of disease outbreaks in poultry . For example , in some areas of Vietnam , poultry originating from farms experiencing disease outbreaks are partly used as feed for domestic reptiles ( farmed pythons and crocodiles ) or destroyed with the support of larger farms ( Delabouglise et al . , 2016 ) . The last 23 years of emerging pathogen outbreaks and zoonotic transmissions failed to prepare us for the epidemiological catastrophe that we are witnessing in 2020 . Multiple subtypes of avian influenza viruses have crossed over into human populations since 1997 ( Gao et al . , 2013; Lai et al . , 2016 ) , all resulting from poultry farming activities . Small-scale poultry farming is likely to be maintained in low- and middle-income countries as it provides low-cost protein , supplemental income to rural households , and is supported by consumer preference of local indigenous breeds of poultry ( Burgos et al . , 2008a; Epprecht , 2005; Sudarman et al . , 2010 ) . If we ignore the active role that poultry farmers play in the control and dissemination of avian influenza , we may miss another opportunity to curtail an emerging disease outbreak at a stage when it is still controllable . An observational longitudinal study was conducted in Ca Mau province in southern Vietnam ( Delabouglise et al . , 2019; Thanh et al . , 2017 ) with the collaboration of the Ca Mau sub-Department of Livestock Production and Animal Health ( CM-LPAH ) . Fifty poultry farms from two rural communes were initially enrolled and three additional farms were subsequently added to the sample in order to replace three farmers who stopped their poultry farming activity . The two communes were chosen by CM-LPAH based on ( 1 ) their high levels of poultry ownership , ( 2 ) their history of HPAI outbreaks , and ( 3 ) likelihood of participation in the study ( Thanh et al . , 2017 ) . Study duration was 20 months , from June 2015 to January 2017 . Monthly Vietnamese-language questionnaires were used to collect information on ( 1 ) number of birds of each species and production type , ( 2 ) expected age of removal from the farm , ( 3 ) number of birds introduced , removed , and deceased in the last month , ( 4 ) clinical symptoms associated with death , ( 5 ) vaccines administered , ( 6 ) type of poultry housing used , and ( 7 ) disinfection activity . Each farm’s poultry were classified into ‘flocks’ , defined as groups of birds of the same age , species , and production type ( Delabouglise et al . , 2019 ) . Because individual poultry cannot be given participant ID numbers in a long-term follow-up study like this , a custom python script was developed to transform cross-sectional monthly data into a longitudinal data set on poultry flocks ( Nguyen-Van-Yen , 2017 ) . Recruitment was designed to have a mix of small ( 20–100 birds ) and large ( >100 birds ) farms and a mix of farms that were ‘primarily chicken’ and ‘primarily duck’ . As multiple poultry species were present on most farms , the chicken and duck farm descriptors were interpreted subjectively . The enrollment aim was to include 80% small farms among chicken farms and 50% small farms among ducks farms; there was approximately equal representation of chicken and ducks farms , but many could have been appropriately classified as having both chickens and ducks . As the residents in the two communes were already familiar with CM-LPAH through routine outreach and inspections , all invitees agreed to study participation . The farm sizes and poultry compositions were representative of small-scale poultry ownership in the Mekong delta regions , but other potential selection biases in the recruitment process could not be ascertained . No sample size calculation was performed for the behavioral analysis presented here , as we had no baseline estimates of sale patterns or disease prevention activities . The duration and size of the study was planned to be able to observe about 1000 poultry flocks ( all species and production types included ) . For the ‘harvest model’ and ‘AI vaccination model’ , we focused our analysis on broiler chicken flocks , since chicken was the predominant species in the study population , the overwhelming majority of chicken flocks were broilers , and their age-specific harvest was easier to predict than the harvest of layer-breeder hens . Additionally , only six layer-breeder chicken flocks were vaccinated against AI during the study period . Observations made in the two first months of the study were discarded since , during these two months , it was unknown whether farms had previously experienced outbreaks . In the ‘disinfection’ model , observations were farm-months . A total of 876 farm-months were available for inclusion in the model . We removed farm-month with missing data on disinfection performed by farmers ( 18 farm-months ) so 858 farm-months were used to fit the disinfection model . In the ‘harvest’ and ‘AI vaccination’ models , observations were chicken broiler flock-months . We selected all chicken flock-months more than 10 days old at the time of data collection and classified by farmers as ‘broilers’ . A total of 1656 flock-months were available for inclusion in the model . In the “harvest model we removed flock-months which were less than 20 days old at the time of data collection . This 20 day threshold was chosen because some newborn flocks below this age were partly sold , not for meat consumption but for management on other farms . Also , we removed flock-months where no chickens were available for harvest because they had all died in the course of the month ( 25 flock-months ) . In total , 153 flock-months were removed and 1503 flock-months were used to fit the harvest model . In the ‘AI vaccination’ model , we removed flock-months of flocks which had already been vaccinated against avian influenza in a previous month , since vaccination is usually performed only once ( among the 338 vaccinated flocks , only eight were vaccinated a second time ) . We also removed flock-months whose housing conditions were not reported ( four flock-months ) . In total , 338 flock-months were removed and 1318 flock-months were used to fit the AI vaccination model . A disease outbreak was defined as the death of at least two birds of the same species – on the same farm , in the same month , with similar clinical symptoms – as this may indicate the presence of an infectious pathogen on the farm . Our definition of outbreaks with sudden deaths encompassed all instances of outbreaks where chicken deaths were noticed without observation of any symptoms beforehand . Since farmers , or their family , check on their poultry at least once per day , it was assumed that these ‘sudden deaths’ corresponded to a time period of less than one day between onset of symptoms and death . For both the harvest and AI vaccination models , we assumed the effect of outbreaks on the dependent variable may be affected by the size of the considered flock ( n ) . Consequently , we included this interaction term in the analysis . The three dependent variables are likely affected by several farm- , flock- , and time-related factors , justifying the inclusion of several control covariates in the multivariable models , summarized in Table 1 . For the harvest model , the main control variable is , logically , ( 1 ) the body weight of chickens , as broiler chickens are conventionally harvested after a fattening period upon reaching a given weight . Since the chicken weight was not collected during the survey , we used the difference between the current flock age t and the anticipated age at maturity t* indicated by farmers in the questionnaire . Hereafter we use δt = t – t* for this difference . The shape of the function linking δt and harvest may depart from linearity and is affected by the chicken breed , which determines the growth performance . Since information on chicken breed was not collected we used the age at maturity t* and the logarithm of flock size ( log ( n ) ) as proxy indicators of the growing performance of the breed and built a proxy body weight variable as a multivariate spline function of δt , t* and n ( Burgos et al . , 2008b ) . 20% of flock-months had missing value for t* . Since there was little within-farm variation in t* ( 2 months of difference at most between two flocks of the same farm ) , missing values were replaced by the median t* in the other flocks of the corresponding farm . ( 2 ) The calendar time T was included as an additional smoothing spline term , since harvest may also be influenced by market prices which vary from one month to the next . Control variables included as standard linear terms were ( 1 ) the number of chickens kept for laying eggs or breeding - famers with a large breeder-layer activity may want to keep some broilers chickens in the farm for replacing the breeding-laying stock , making them less likely to harvest broilers; ( 2 ) the number of broiler chickens simultaneously present in the same farm in other flocks; ( 3 ) the number of chicken flocks introduced in the same month; ( 4 ) the number of chicken flocks introduced in the previous month – farmers with a high number of broilers chickens or many recently introduced broiler flocks may want to sell their current flocks faster in order to limit feeding expenses and workload; ( 5 ) the vaccination status of the flock against AI; ( 6 ) the vaccination status of the flock against Newcastle Disease ( ND ) – farmers may keep their vaccinated flocks for a longer period as they are at lower risk of being affected by an infectious disease . We assumed the effect of outbreaks on the dependent variable may be affected by the size of the considered flock ( n ) . Consequently , we included an interaction term between outbreaks and log ( n ) in the analysis . For the AI primo-vaccination model , control variables included as smoothing splines were ( 1 ) the flock age t - vaccination may be preferentially done early in the flock life , ( 2 ) the flock size n , and ( 3 ) the calendar time T - vaccination activities may be intensified at particular times of the year . Control variables included as standard linear terms were ( 1 ) the type of housing ( free-range or confinement in pens or indoor ) which affects the convenience of vaccination; ( 2 ) the proportion of the flock harvested in the same month - farmers might be less willing to vaccinate flocks being harvested; and the size of populations of ( 3 ) broiler chickens , ( 4 ) layer-breeder chickens , ( 5 ) broiler ducks , ( 6 ) layer-breeder ducks , ( 7 ) broiler Muscovy ducks and ( 8 ) layer-breeder Muscovy ducks kept in other flocks - farmers’ perceived risk of AI and attitude towards vaccination may be influenced by the size of the poultry population at risk for AI and production type . We assumed the effect of outbreaks on the dependent variable may be affected by the size of the considered flock ( n ) . Consequently , we included an interaction term between outbreaks and log ( n ) in the analysis . For the disinfection model , control variables included as smoothing splines were ( 1 ) the calendar time T - disinfection activities may be intensified at particular times of the year . Control variables included as standard linear terms were the size of populations of ( 1 ) broiler chickens , ( 2 ) layer-breeder chickens , ( 3 ) broiler ducks , ( 4 ) layer-breeder ducks , ( 5 ) broiler Muscovy ducks and ( 6 ) layer-breeder Muscovy ducks - the farmers’ attitude towards prevention may be influenced by the size of the poultry population at risk of disease . We assumed that the events of interest , namely harvest , AI vaccination , and disinfection were drawn from a binomial distribution and used a logistic function to link their probability to a function of the independent covariates . Flocks were either fully vaccinated for AI or not at all , so the AI vaccination variable for flock-months took only the value 0 or one and was , therefore , treated as binary . Partial flock harvest ( the harvest of only a fraction of the chickens in a given flock ) and partial farm disinfection ( the disinfection of facilities for only a fraction of the poultry flocks present in the farm ) occurred in a minority of observations . Therefore , the number of chickens harvested per flock-month and the number of poultry flocks disinfected per farm-month were treated as binomial random variables with a number of trials equal to the flock size ( for harvest ) and the number of flocks per farm ( for disinfection ) . To ensure that the model was not conditioned on the size of flocks and number of flocks per farm , prior weights equal to the inverse of the flock size and the number of flocks in the farm ( i . e . the number of trials ) were used in the binomial harvest model and disinfection model , respectively . The extent of over- or under-dispersion in the data was investigated by fitting a quasi-binomial model in parallel ( Papke and Wooldridge , 1996 ) . The resulting dispersion parameters were 0 . 76 ( harvest model ) and 0 . 77 ( disinfection model ) , indicating moderate underdispersion , and that the estimates of our analyses are conservative . Some of the included effects are non-linear in nature , and we needed to account for the intra-farm autocorrelation of the dependent variables . We therefore used a mixed-effects general additive model ( MGAM ) implemented in R with the ‘mgcv’ package ( Wood et al . , 2016 ) . This enabled us to model the combined effect of δt , t* , and flock size ( n ) on harvest time; the effect of t and n on AI vaccination; and the effect of calendar time ( T ) on all the dependent variables , as penalized thin plate regression splines ( Wood , 2017 ) . We specifically chose these variables because they are presumably the most important factors influencing the dependent variables and their effect could possibly be highly non-linear . All other covariates were included as parametric regression terms . We also modelled the individual effects of farms on the dependent variables as random effects . The complete models linking the logit Yij of probability of realization of an event and the set of explanatory variables , for a flock-month i ( harvest , vaccination for AI ) or a farm-month i ( disinfection ) in a farm j , are described by the following set of equations: Harvest model ( flock-month level ) : ( 1 ) Yij=α+∑m=02βONS−mXijONS−m+∑m=02βOS−mXijOS−m+fδt ( δtij , tij∗ , log ( nij ) ) +fT ( Tij ) +∑k=16βkXijk+ϕj+εij AI vaccination model ( flock-month level ) : ( 2 ) Yij=α+∑m=02βONS−mXijONS−m+∑m=02βOS−mXijOS−m+∑m=02βOD−mXijOD−m+ft ( log ( tij ) ) +fn ( log ( nij ) ) +fT ( Tij ) +∑k=18βkXijk+ϕj+εij Disinfection model ( farm-month level ) : ( 3 ) Yij=α+∑m=02βO−mXijO−m+fT ( Tij ) +∑k=16βkXijk+ϕj+εij The model parameters are α the model intercept; β the parametric coefficients; f a thin-plate spline function; Xk the general notation for variables with linear effects; XO-m , XOS-m , XONS-m and XOD-m , categorical variables denoting presence or absence of an outbreak in the same farm m months prior in any species ( O ) , in chickens with sudden deaths ( OS ) , in chickens with no sudden deaths ( ONS ) , and in different species ( OD ) respectively; n the flock size; t the current age of the flock; t* the age at maturity of the flock anticipated by the farmer; δt the difference between current age and age at maturity; T the calendar time; φ the farm random effect; ε the residual error term . Some variables with a highly skewed distribution ( Table 1 ) were transformed . Current age ( t ) and flock size ( n ) being strictly positive , they were log-transformed . Farm populations of broiler and layer-breeders of different species being null or positive , they were square-root transformed . Covariates included in the multivariate spline function for body weight ( δt , t* , log ( n ) ) were centered and standardized . Interaction terms between outbreak categorical variables and flock size log ( nij ) were added in the Harvest and AI vaccination models . Excessive multi-collinearity between covariates was assessed by estimating their variance inflated factor using the ‘usdm’ R package ( Naimi et al . , 2014 ) . We fitted the complete models using the whole set of covariates using restricted maximum likelihood estimation . We then used a backward-forward stepwise selection , based on AIC comparison , to eliminate the variables with non-significant effects ( Hosmer and Lemeshow , 2000 ) . Arguably , one farmer is likely to maintain the same farm management from one month to the next despite changes in influential covariates . Therefore , for each model , we tested the presence of farm-level temporal autocorrelation by fitting two linear regression models on the deviance residuals , with a fixed constant effect and with and without intra-farm AR-1 time autocorrelation structure and comparing the two model fits with a log-likelihood ratio test . For the ‘disinfection’ model , the fit was significantly improved by including the autocorrelation term while for the two other models it was not . Therefore , we implemented the same model fitting protocol for the ‘disinfection’ model with an additional intra-farm AR-1 time autocorrelation term on the dependent variable . We used the ‘gamm’ routine of the ‘mgcv’ package for this purpose ( Wood , 2017 ) . Since ‘gamm’ models for binomial data are fitted with the penalized quasi-likelihood approach , the AIC metric is not suitable to compare such models . Instead , we implemented a stepwise removal of covariates whose t-test returned the highest probability of type one error ( p-value ) until all remaining covariates had a p-value lower than 20% . All analyses and graphical representations were performed with R version 3 . 6 . 1 ( R Development Core Team , 2014 ) . The collaboration between the investigators ( authors ) and the Ca Mau sub-Department of Livestock Production and Animal Health ( CM-LPAH ) was approved by the Hospital for Tropical Diseases in Ho Chi Minh City , Vietnam . The CM-LPAH , which at the province-level is the equivalent of an ethical committee for studies on livestock farming , specifically approved this study .	The past few decades have seen the circulation of avian influenza viruses increase in domesticated poultry , regularly creating outbreaks associated with heavy economic loss . In addition , these viruses can sometimes ‘jump’ into humans , potentially allowing new diseases – including pandemics – to emerge . The Mekong river delta , in southern Vietnam , is one of the regions with the highest circulation of avian influenza . There , a large number of farmers practice poultry farming on a small scale , with limited investments in disease prevention such as vaccination or disinfection . Yet , it was unclear how the emergence of an outbreak could change the behavior of farmers . To learn more , Delabouglise et al . monitored 53 poultry farms , with fewer than 1000 chickens per farm , monthly for over a year and a half . In particular , they tracked when outbreaks occurred on each farm , and how farmers reacted . Overall , poultry farms with more than 17 chickens were more likely to vaccinate their animals and use disinfection practices than smaller farms . However , disease outbreaks did not affect vaccination or disinfection practices . When an outbreak occurred , farmers with fewer than 17 chickens tended to sell their animals earlier . For instance , they were 214% more likely to send their animals to market if an outbreak with sudden deaths occurred that month . Even if they do not make as much money selling immature individuals , this strategy may allow them to mitigate economical loss: they can sell animals that may die soon , saving on feeding costs and potentially avoiding further contamination . However , as animals were often sold alive in markets or to itinerant sellers , this practice increases the risk of spreading diseases further along the trade circuits . These data could be most useful to regional animal health authorities , which have detailed knowledge of local farming systems and personal connections in the communities where they work . This can allow them to effect change . They could work with small poultry farmers to encourage them to adopt efficient disease management strategies . Ultimately , this could help control the spread of avian influenza viruses , and potentially help to avoid future pandemics .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "epidemiology", "and", "global", "health", "microbiology", "and", "infectious", "disease" ]	2020	Poultry farmer response to disease outbreaks in smallholder farming systems in southern Vietnam
Mammalian HMG-CoA reductase ( HMGCR ) , the rate-limiting enzyme of the cholesterol biosynthetic pathway and the therapeutic target of statins , is post-transcriptionally regulated by sterol-accelerated degradation . Under cholesterol-replete conditions , HMGCR is ubiquitinated and degraded , but the identity of the E3 ubiquitin ligase ( s ) responsible for mammalian HMGCR turnover remains controversial . Using systematic , unbiased CRISPR/Cas9 genome-wide screens with a sterol-sensitive endogenous HMGCR reporter , we comprehensively map the E3 ligase landscape required for sterol-accelerated HMGCR degradation . We find that RNF145 and gp78 independently co-ordinate HMGCR ubiquitination and degradation . RNF145 , a sterol-responsive ER-resident E3 ligase , is unstable but accumulates following sterol depletion . Sterol addition triggers RNF145 recruitment to HMGCR via Insigs , promoting HMGCR ubiquitination and proteasome-mediated degradation . In the absence of both RNF145 and gp78 , Hrd1 , a third UBE2G2-dependent E3 ligase , partially regulates HMGCR activity . Our findings reveal a critical role for the sterol-responsive RNF145 in HMGCR regulation and elucidate the complexity of sterol-accelerated HMGCR degradation . Editorial note: This article has been through an editorial process in which the authors decide how to respond to the issues raised during peer review . The Reviewing Editor's assessment is that all the issues have been addressed ( see decision letter ) . Cholesterol plays a critical role in cellular homeostasis . As an abundant lipid in the eukaryotic plasma membrane , it modulates vital processes including membrane fluidity and permeability ( Hannich et al . , 2011; Haines , 2001 ) and serves as a precursor for important metabolites including steroid hormones and bile acids ( Payne and Hales , 2004; Chiang , 2013 ) . The cholesterol biosynthetic pathway in mammalian cells also provides intermediates for essential non-steroid isoprenoids and therefore requires strict regulation ( Goldstein and Brown , 1990 ) . The endoplasmic-reticulum ( ER ) resident , polytopic membrane glycoprotein 3-hydroxy-3-methylglutaryl coenzyme A reductase ( HMGCR ) is central to this pathway , catalysing the formation of mevalonate , a crucial isoprenoid precursor . As the rate-limiting enzyme in mevalonate metabolism , HMGCR levels need to be tightly regulated , as dictated by intermediates and products of the mevalonate pathway ( Johnson and DeBose-Boyd , 2018 ) . The statin family of drugs , which acts as competitive inhibitors of HMGCR , represents the single most successful approach to reducing plasma cholesterol levels and therefore preventing atherosclerosis-related diseases ( Heart Protection Study Collaborative Group , 2002 ) . Understanding how HMGCR is regulated is therefore of fundamental biological and clinical importance . Cholesterol , together with its biosynthetic intermediates and isoprenoid derivatives , regulates HMGCR expression at both the transcriptional and posttranscriptional level ( Johnson and DeBose-Boyd , 2018 ) . Low cholesterol induces transcriptional activation of HMGCR through the sterol response element binding proteins ( SREBPs ) which bind SREs in the promoter region ( Osborne , 1991 ) . In a cholesterol-rich environment , SREBPs are inactive and held in the ER in complex with their cognate chaperone SREBP cleavage-activating protein ( SCAP ) in association with the ER-resident Insulin-induced genes 1/2 ( Insig-1/2 ) anchor proteins ( Dong and Tang , 2010; Yabe et al . , 2002 ) . A decrease in membrane cholesterol triggers dissociation of the SCAP-SREBP complex from Insigs and translocation to the Golgi apparatus , where the SREBP transcription factor is proteolytically activated by Site-1 and Site-2 proteases , released into the cytosol and trafficked to the nucleus ( reviewed in Horton et al . , 2002 ) . Low sterol levels therefore dramatically increase both HMGCR mRNA and extend HMGCR protein half-life , ensuring the resultant elevated enzyme levels stimulate the supply of mevalonate to re-balance cholesterol homeostasis ( Goldstein and Brown , 1990; Brown et al . , 1973 ) . Once cholesterol levels are restored , excess HMGCR is rapidly degraded by the ubiquitin proteasome system ( UPS ) in a process termed sterol-accelerated degradation ( Hampton et al . , 1996; Ravid et al . , 2000; Sever et al . , 2003a ) . This joint transcriptional and translational regulation of HMGCR is controlled by a host of ER-resident polytopic membrane proteins and represents a finely balanced homeostatic mechanism to rapidly regulate this critical enzyme in response to alterations in intracellular cholesterol . While the ubiquitin-mediated , post-translational regulation of HMGCR is well-established , the identity of the critical mammalian ER-associated degradation ( ERAD ) E3 ubiquitin ligase ( s ) responsible for sterol-accelerated HMGCR ERAD remains controversial . In yeast , S . cerevisiae encodes three ERAD E3 ubiquitin ligases , of which Hrd1p ( HMG-CoA degradation 1 ) , is named for its ability to degrade yeast HMGCR ( Hmg2p ) in response to non-sterol isoprenoids ( Hampton et al . , 1996; Bays et al . , 2001 ) . The marked expansion and diversification of E3 ligases in mammals makes the situation more complex , as in human cells there are 37 putative E3 ligases involved in ERAD , few of which are well-characterised ( Kaneko et al . , 2016 ) . Hrd1 and gp78 represent the two mammalian orthologues of yeast Hrd1p . Hrd1 was not found to regulate HMGCR ( Song et al . , 2005; Nadav et al . , 2003 ) . However , gp78 was reported to be responsible for the sterol-induced degradation of HMGCR as ( i ) gp78 associates with Insig-1 in a sterol-independent manner , ( ii ) Insig-1 mediates a sterol-dependent interaction between HMGCR and gp78 , ( iii ) overexpression of the transmembrane domains of gp78 exerted a dominant-negative effect and inhibited HMGCR degradation , and ( iv ) , siRNA-mediated depletion of gp78 resulted in decreased sterol-induced ubiquitination and degradation of HMGCR ( Song et al . , 2005 ) . The same laboratory subsequently suggested that the sterol-induced degradation of HMGCR was mediated by two ERAD E3 ubiquitin ligases , with TRC8 involved in addition to gp78 ( Jo et al . , 2011 ) . However , these findings remain controversial as , despite confirming a role for gp78 in the regulation of Insig-1 ( Lee et al . , 2006; Tsai et al . , 2012 ) , an independent study found no evidence for either gp78 or TRC8 in the sterol-induced degradation of HMGCR ( Tsai et al . , 2012 ) . Therefore , the E3 ligase ( s ) responsible for the sterol-accelerated degradation of HMGCR remain disputed . The introduction of systematic forward genetic screening approaches to mammalian systems ( Carette et al . , 2009; Wang et al . , 2014 ) has made the unbiased identification of E3 ubiquitin ligases more tractable , as demonstrated for the viral ( van den Boomen and Lehner , 2015; van de Weijer et al . , 2014; Stagg et al . , 2009 ) and endogenous regulation of MHC-I ( Burr et al . , 2011; Cano et al . , 2012 ) . To identify the E3 ligases governing HMGCR ERAD , we applied a genome-wide forward genetic screen to a dynamic , cholesterol-sensitive reporter cell line , engineered to express a fluorescent protein fused to endogenous HMGCR . This approach identified cellular genes required for sterol-induced HMGCR degradation , including UBE2G2 and the RNF145 ERAD E3 ubiquitin ligase . The subtle phenotype observed upon RNF145 depletion alone suggested redundant ligase usage . A subsequent , targeted ubiquitome CRISPR/Cas9 screen in RNF145-knockout cells showed RNF145 to be functionally redundant with gp78 , the E3 ligase originally implicated in HMGCR degradation . We confirmed that loss of gp78 alone showed no phenotype , while loss of both E3 ligases significantly inhibited the sterol-induced ubiquitination and degradation of HMGCR . Complete stabilisation required additional depletion of a third ligase - Hrd1 . We find that endogenous RNF145 is an auto-regulated , sterol-responsive E3 ligase which is recruited to Insig proteins under sterol-replete conditions , thus promoting the regulated ubiquitination and sterol-accelerated degradation of HMGCR . Our data resolve the controversy of the E3 ligases responsible for the post-translational regulation of HMGCR and emphasise the complexity of the mammalian ubiquitin system in fine-tuning sterol-induced HMGCR turnover and cholesterol homeostasis . To identify genes involved in the post-translational regulation of HMGCR , we engineered a cell line in which Clover , a bright fluorescent protein ( Lam et al . , 2012 ) , was fused to the C-terminus of endogenous HMGCR , generating an HMGCR-Clover fusion protein ( Figure 1A ) . The resulting HMGCR-Clover Hela single-cell clone expresses a dynamic , cholesterol-sensitive fluorescent reporter that is highly responsive to fluctuations in intracellular cholesterol . Basal HMGCR-Clover levels in sterol-replete tissue culture media were undetectable by flow cytometry ( Figure 1B ) and phenocopy endogenous WT HMGCR expression ( Figure 1C , compare lanes 1 and 4 ) . Following overnight sterol depletion , a ~ 25 fold increase in HMGCR-Clover expression was detected ( shaded grey to blue histogram in Figure 1D , Figure 1C ( lanes 2 and 5 ) ) , representing a combination of increased SREBP-induced transcription and decreased sterol-induced HMGCR degradation . Reintroduction of sterols induced the rapid degradation of HMGCR-Clover ( ~80% decrease within 2 hr ) , confirming the sterol-dependent regulation of the reporter ( blue to red histogram in Figure 1D ) . Residual , untagged HMGCR detected by immunoblot analysis in the reporter cells under sterol-depleted conditions suggested that at least one HMGCR allele remained untagged ( Figure 1C , compare lanes 2 and 5 ) , which was confirmed by PCR-amplification and sequencing of the genomic locus ( Figure 1—figure supplement 1A–C ) . The unmodified allele allowed us to monitor both tagged and untagged forms of HMGCR . Inhibiting the enzymatic activity of HMGCR with mevastatin also stabilised HMGCR-Clover expression , as did inhibition of the proteasome ( bortezomib ) or p97/VCP ( NMS-873 ) ( Figure 1E ) , confirming the rapid , steady-state degradation of the HMGCR reporter . Furthermore , we showed that CRISPR/Cas9-mediated ablation of both Insig-1 and −2 together induced a dramatic increase in HMGCR-Clover expression , equivalent to levels seen following sterol depletion ( Figure 1F ) . Under these conditions , the SREBP-SCAP complex is not retained in the ER , leading to constitutive SREBP-mediated transcription of HMGCR-Clover , irrespective of the sterol environment . CRISPR-mediated gene disruption of either Insig-1 or −2 alone caused only a small , steady-state rescue of HMGCR-Clover ( Figure 1F ) , which was more pronounced with the loss of Insig-1 than Insig-2 . While Insig-1-deficient cells were unable to completely degrade HMGCR upon sterol addition , only a minor defect in HMGCR degradation was seen in the absence of Insig-2 ( Figure 1F ) , suggesting that Insig-1 is dominant over Insig-2 under these conditions . Finally , we confirmed that HMGCR-Clover was appropriately localised to the ER by confocal microscopy ( Figure 1G ) . Thus , HMGCR-Clover is a dynamic , cholesterol-sensitive reporter , which rapidly responds to changes in intracellular cholesterol and is regulated in a proteasome-dependent manner . To identify genes required for the sterol-induced degradation of HMGCR , we performed a genome-wide CRISPR/Cas9 knockout screen in HMGCR-Clover cells . We took advantage of the rapid decrease in HMGCR-Clover expression following sterol addition to cells starved overnight ( 16 hr ) of sterols ( Figure 1D ) , and enriched for rare genetic mutants with reduced ability to degrade HMGCR-Clover in response to sterols . To this end , HMGCR-Clover cells were transduced with a genome-wide CRISPR/Cas9 knockout library comprising 10 sgRNAs per gene ( Morgens et al . , 2017 ) . Mutagenised cells were first depleted of sterols overnight; sterols were then reintroduced for 5 hr , at which point rare mutant cells with reduced ability to degrade HMGCR-Clover upon sterol repletion were enriched by fluorescence-activated cell sorting ( FACS ) ( Figure 2A , gating shown in Figure 2—figure supplement 1A ) . This process was repeated eight days later to further purify the selected cells . The enriched population contained only a small percentage of cells ( 1 . 96% after sort #1 , 24 . 49% after sort #2 ) with increased steady-state HMGCR-Clover expression ( green filled histogram in Figure 2B ) . However , the majority of sterol-starved cells from this selected population showed impaired degradation of HMGCR-Clover after addition of sterols ( compare red versus orange filled histogram ( Figure 2B , compare lanes 6 and 9 in Figure 2—figure supplement 1B ) ) . The broad distribution of this histogram ( Figure 2B red histogram ) suggested that the enriched cell population contains a variety of mutants which differ in their ability to degrade HMGCR-Clover . The sgRNAs in the selected cells , and an unselected control library , were sequenced on the Illumina HiSeq platform ( Figure 2A ( viii ) ) . Using the RSA algorithm , we identified a set of 11 genes , which showed significant enrichment ( -logP >5 ) in the selected cells ( Figure 2C ) . Many of these are known to be required for the sterol-induced degradation of HMGCR ( König et al . , 2007 ) . The screen identified the E2 ubiquitin conjugating enzyme UBE2G2 and its accessory factor AUP1 , which recruits UBE2G2 to lipid droplets and membrane E3 ubiquitin ligases ( Klemm et al . , 2011; Jo et al . , 2013; Spandl et al . , 2011; Christianson et al . , 2011 ) , as well as both Insig-1 and −2 ( Yabe et al . , 2002; Yang et al . , 2002; Sever et al . , 2003a ) . The role of the remaining hits is summarized ( Table 1 ) and validation of selected hits is shown ( Insig-1/2 , Figure 1F; UBE2G2 , EHD1 , GALNT11 , LDLR and TECR , Figure 2 – figure supplement 1C/D ) . Strikingly , the only ER-resident E3 ubiquitin ligase to emerge from the screen is the poorly characterised RNF145 . RNF145 shares 27% amino acid identity with TRC8 , which is one of the E3 ligases ( together with gp78 ) previously suggested to ubiquitinate HMGCR ( Jo et al . , 2011 ) . Interestingly , RNF145 also harbours a YLYF motif at its N-terminus , which is similar to the YIYF motif present in the sterol-sensing domain ( SSD ) of SCAP and HMGCR required for their binding to the Insig proteins ( Yang et al . , 2002; Sever et al . , 2003a; Jiang et al . , 2018; Cook et al . , 2017; Zhang et al . , 2017 ) . The presence of the YLYF motif suggested that RNF145 might itself interact with the Insig proteins and therefore represented a promising candidate from our genetic screen . To see if we could validate the role of RNF145 in HMGCR degradation , we designed four independent sgRNAs , either targeting RNF145 individually or as a pool . Under cholesterol-replete conditions , no accumulation of the HMGCR-reporter was observed in RNF145-depleted cells ( top and middle rows , Figure 2D ) , but a small and highly reproducible decrease in HMGCR-Clover degradation was seen following re-introduction of sterols ( red histograms , bottom row in Figure 2D ) , emphasising the utility of the endogenous fluorescent reporter in identifying subtle phenotypes . Since the identity of the E3 ubiquitin ligases regulating HMGCR turnover remains controversial , the modest effect of RNF145 loss on HMGCR-Clover sterol-induced degradation suggested the involvement of additional ligase ( s ) . Our screen therefore identified both known and novel components implicated in sterol-dependent HMGCR ERAD . If a second E3 ligase is partially redundant with RNF145 , its effect should be unmasked in RNF145-deficient cells . We therefore generated a focussed subgenomic sgRNA library targeting 1119 genes of the ubiquitin-proteasome system as described in ‘Materials and methods’ , including 830 predicted E3 ubiquitin ligases , and used this library to screen for genes required for the degradation of HMGCR in RNF145-deficient HMGCR-Clover cells ( Figure 3—figure supplement 4B , lane two for knockout validation ) . Due to the reduced complexity of this focussed library , only a single FACS enrichment step was used ( Figure 3A , red histogram ) . Strikingly , this screen identified gp78 ( gene name: AMFR ) ( Figure 3B , Table 2 ) , the E3 ubiquitin ligase previously implicated in HMGCR degradation ( Jo et al . , 2011; Song et al . , 2005; Fang et al . , 2001 ) . Taking a combined knockout strategy we asked whether gp78 and RNF145 are together responsible for HMGCR degradation . As predicted by the genetic approach ( Figure 3C ( ii ) ) , there was no difference in sterol-induced HMGCR-Clover degradation between control and gp78-depleted HMGCR-Clover cells . Gp78 was not , therefore , a false-negative from our initial , genome-wide CRISPR/Cas9 screen ( Figure 2C ) . Individual knockout of RNF145 again showed that sterol-induced HMGCR-Clover degradation was mildly impaired in RNF145-depleted cells ( Figure 3C ( iii ) ) . However , sgRNA-mediated targeting of gp78 together with RNF145 ( Figure 3C ( iv ) , see Figure 3—figure supplement 4A and B lane three for knockout validation ) , resulted in a significant increase in both steady-state HMGCR-Clover ( Figure 3C ( iv ) grey to green filled histograms ) and an inability to degrade HMGCR-Clover upon addition of sterols to sterol-starved cells ( Figure 3C ( iv ) blue to red histogram ) , a phenotype comparable to UBE2G2 deletion ( Figure 3C ( v ) ) . Our results therefore suggest a partial functional redundancy between gp78 and RNF145 and imply that both ligases can independently regulate the sterol-induced degradation of HMGCR . To confirm that the phenotypes observed in RNF145- and gp78-deficient HMGCR-Clover cells were representative of endogenous , wild type HMGCR regulation , we deleted RNF145 and/or gp78 from WT HeLa cells and monitored endogenous HMGCR by immunoblot analysis . The sterol-induced degradation of HMGCR was assessed in four RNF145 knockout clones , derived from two different sgRNAs ( validation in Figure 3—figure supplement 1A , B ) . No difference in the sterol-induced degradation of HMGCR was seen in these RNF145 knockout clones ( Figure 3—figure supplement 2 , compare lanes 6 and 7–10 ) . The subtle effect on HMGCR-Clover expression revealed by flow cytometry ( Figures 2D and 3C ) may not be detected by the less sensitive immunoblot analysis . Similarly , loss of gp78 alone ( Figure 3—figure supplement 3A for sgRNA validation ) did not affect HMGCR degradation ( Figure 3D , compare lanes 6 and 7–10 ) , but loss of gp78 together with RNF145 resulted in a significant rescue of steady state HMGCR ( Figure 3E , Figure 3—figure supplement 3C ) . Following sterol addition , gp78/RNF145 double-knockout clones showed a marked ( although still incomplete ) reduction in sterol-induced HMGCR degradation ( Figure 3F , compare lanes 7 + 8 with 9 –12 ) . These data validate the phenotypes exhibited by the HMGCR-Clover reporter cell line and confirm a role for both gp78 and RNF145 in the sterol-induced degradation of endogenous HMGCR . To determine whether RNF145 E3 ubiquitin ligase activity is required for HMGCR degradation , we complemented a population of gp78/RNF145 double-knockout HMGCR-Clover cells ( Figure 3—figure supplement 4B , lane three for knockout validation ) with either epitope-tagged wild type RNF145 , or a catalytically-inactive RNF145 RING domain mutant ( C552A , H554A ) ( Figure 4A ) . The pronounced block in the sterol-induced degradation of HMGCR-Clover was at least partially rescued by expression of wild type , but not the RNF145 RING domain mutant ( Figure 4B , compare blue to red histogram ) . The E3 ligase activity of RNF145 is therefore critical for HMGCR ERAD . Endogenous RNF145 has a short half-life ( ~2 hr ) and displayed rapid , proteasome-mediated degradation ( Figure 5A ( i ) ) , an observation confirmed in multiple cell lines ( Figure 5—figure supplement 1A ) . This rapid turnover of endogenous RNF145 contrasts sharply with the stability of endogenous gp78 , which shows little degradation over the 10 hr chase period ( Figure 5A ( i ) ) . Although RNF145 and gp78 both target HMGCR for degradation , the two ligases did not appear to be co-regulated as RNF145 stability was unaffected by gp78 and vice-versa ( Figure 5A ( i , ii ) , Figure 5—figure supplement 1B ) . However , endogenous RNF145 was stabilised by deletion of its cognate E2 enzyme UBE2G2 ( Figure 5B ) , and , furthermore , the catalytically-inactive RING domain mutant expressed in RNF145-deficient cells ( ΔRNF145 #4 + RNF145-V5 ( mut ) ) exhibited greater abundance at steady-state compared with its wild type counterpart ( Figure 3—figure supplement 4C ) . Together these data show that RNF145 is intrinsically unstable and rapidly turned over in an auto-regulatory manner . Since RNF145 is rapidly turned over , we aimed to determine whether RNF145 gene transcription was sterol-responsive . Sterol depletion induced RNF145 ( ~2 . 99 ± 0 . 65 fold increase , p=0 . 0009 ) mRNA expression as well as HMGCR ( ~12 . 26 ± 3 . 16 fold increase , p=0 . 0004 ) mRNA expression ( Figure 5C ) . This accumulation of endogenous RNF145 was suppressed following the addition of methyl beta-cyclodextrin ( MBCD ) -complexed cholesterol ( Chol/MBCD ) to the starvation media ( Figure 5D ) , whereas gp78 abundance remained unaltered ( Figure 5—figure supplement 1D ) . RNF145 is therefore a unique , sterol-regulated E3 ubiquitin ligase whose expression is dependent on the cellular sterol status . The Insig proteins provide an ER-resident platform for sterol-dependent interactions between HMGCR and its regulatory components ( Dong et al . , 2012 ) . Since RNF145 is sterol-regulated and degrades HMGCR we initially wanted to know if RNF145 interacts with HMGCR . We found that in sterol-replete but not sterol-deplete conditions , endogenous HMGCR co-immunoprecipitates both epitope-tagged RNF145 ( Figure 6A , Figure 3—figure supplement 4C lane three for relative RNF145-V5 levels upon reconstitution ) , as well as endogenous RNF145 ( Figure 6B ) . Initial attempts to ascertain whether this interaction between RNF145 and HMGCR was direct , or mediated via the Insig proteins were challenging due to the low expression levels of endogenous RNF145 . We circumvented this problem by performing the co-immunoprecipitation in UBE2G2 knockout cells , which express increased levels of endogenous RNF145 ( Figure 5B ) . Under these conditions , RNF145 showed a clear , sterol-dependent interaction with Insig-1 , correlating with RNF145’s association with HMGCR ( Figure 6C ) . Importantly , endogenous RNF145 is not , therefore , continually bound to Insig-1 , but , like HMGCR , associates with Insig-1 in a sterol-dependent manner . Binding of RNF145 to Insig-1 was HMGCR-independent ( Figure 6D ) and , in the absence of Insigs , RNF145 was unable to bind HMGCR ( Figure 6E , see Figure 6—figure supplement 1A–C for generation of Insig-1+2 knockout cells ) . Insigs are therefore indispensable for the interaction between RNF145 and HMGCR . These findings emphasize the central role of Insig proteins as scaffolds in the sterol-induced engagement of HMGCR by RNF145 . Despite our two genetic screens identifying a requirement for RNF145 and gp78 in HMGCR degradation ( Figures 2C and 3B ) , the combined loss of these two ligases failed to completely inhibit sterol-induced HMGCR degradation ( Figure 3C ( iv ) ; Figure 7A ( ii ) ) . Furthermore , ablation of UBE2G2 in RNF145/gp78 double-knockout cells exacerbated the sterol-dependent degradation defect ( Figure 7A ( iv ) ) , predicting the role for an additional E3 ubiquitin ligase ( s ) utilising UBE2G2 in HMGCR degradation . We therefore assessed whether ablation of either of the two remaining ER-resident E3 ligases known to use UBE2G2 , TRC8 ( van de Weijer et al . , 2017 ) and Hrd1 ( Kikkert et al . , 2004 ) , exacerbated the HMGCR-degradation defect in RNF145/gp78 double-knockout cells ( Figure 7B , Figure 7—figure supplements 1B and 2B for knockdown validation ) . While the loss of TRC8 had no effect on HMGCR-Clover expression , the loss of Hrd1 in RNF145/gp78 double-knockout cells increased steady-state HMGCR-Clover expression and caused a complete block in the sterol-accelerated degradation of HMGCR-Clover ( Figure 7B ( ii ) , Figure 7—figure supplement 1A for validation with individual independent sgRNAs ) . This additive effect of Hrd1 depletion on the sterol-induced turnover of endogenous HMGCR was independently confirmed by immunoblot analysis ( Figure 7C , compare lanes 2 , 4 and 6 ) and was observed as early as 60 min after sterol addition ( Figure 7—figure supplement 1D , compare lanes 7 and 9 ) . Importantly , depletion of Hrd1 , alone or in combination with depletion of either gp78 or RNF145 , did not affect HMGCR-Clover degradation ( Figure 7—figure supplement 1C ) . Moreover , TRC8 depletion affected neither steady-state HMGCR-Clover expression , nor sterol-induced HMGCR-Clover degradation ( Figure 7B ( iii ) ) . Indeed , despite a functional TRC8 depletion ( Figure 7—figure supplement 2B for validation of TRC8 depletion ) ( Stagg et al . , 2009 ) , we could detect no role for TRC8 , depleted either alone or in combination with RNF145 , in the sterol-induced degradation of HMGCR ( Figure 7—figure supplement 2A ) . In summary , gp78 with RNF145 are the only combination of ligases whose loss inhibited HMGCR degradation . Hrd1 depletion also delays sterol-induced HMGCR degradation , but only in the absence of RNF145 and gp78 . As a complete block of sterol-accelerated HMGCR degradation required the depletion of all three UBE2G2-dependent E3 ubiquitin ligases , we determined how the sequential depletion of these ligases affected the ubiquitination status of HMGCR . The combined loss of RNF145 with gp78 showed a dramatic reduction in HMGCR ubiquitination , but a complete loss of ubiquitination required the depletion of all three ligases ( Figure 7D ) . As predicted , depletion of UBE2G2 also caused a marked decrease in HMGCR ubiquitination . Taken together , these results demonstrate the remarkable plasticity of the HMGCR-degradation machinery . The generation of a dynamic , cholesterol-sensitive endogenous HMGCR reporter cell line allowed an unbiased genetic approach to identify the cellular machinery required for sterol-accelerated HMGCR degradation . This reporter cell line has the advantage of being able to identify both complete and partial phenotypes and helps explain why the identity of the E3 ubiquitin ligases responsible for the sterol-accelerated degradation of HMGCR has remained controversial . We find that three E3 ubiquitin ligases - RNF145 , gp78 and Hrd1 - are together responsible for HMGCR degradation ( Figure 8 ) . The activity of the two primary ligases , RNF145 and gp78 is partially redundant as the loss of gp78 alone did not affect HMGCR degradation , while loss of RNF145 showed only a small reduction on HMGCR degradation . In the absence of both RNF145 and gp78 , a third ligase , Hrd1 , can compensate and partially regulate HGMCR degradation , but this effect of Hrd1 is only revealed in the absence of both RNF145 and gp78 , and in no other identified combination . Initial reports of a role for gp78 in HMGCR degradation , either alone ( Song et al . , 2005 ) or in combination with TRC8 ( Jo et al . , 2011 ) , were not reproduced in an independent study ( Tsai et al . , 2012 ) and so this important issue has remained unresolved . Our initial genome-wide screen successfully identified a single E3 ligase ( RNF145 ) as well as many of the components known to be required for sterol-accelerated HMGCR degradation ( e . g . Insig-1/2 , UBE2G2 , AUP1 , FAF2; Figure 2C ) ( Sever et al . , 2003b; Miao et al . , 2010; Jo et al . , 2013 ) , thus validating the suitability of this genetic approach . For a small number of validated hits from our screen ( Figure 2C , Figure 2—figure supplement 1D ) , the effects on sterol-accelerated HMGCR degradation were unanticipated and likely reflect wide-ranging alterations to the protein and lipid environment . Trans-2 , 3-enoyl CoA reductase ( TECR ) catalyses the final steps in the synthesis of very long-chain fatty acids ( VLCFAs ) ( Moon and Horton , 2003 ) as well as the saturation step in sphingolipid degradation ( Wakashima et al . , 2014 ) . Polypeptide N-acetylgalctosaminyltransferase 11 ( GALNT11 ) initiates protein O-linked glycosylation , suggesting that a protein involved in HMGCR regulation requires O-linked glycosylation ( Schwientek et al . , 2002 ) . Interestingly , our screen revealed that loss of the LDLR impaired HMGCR-Clover degradation . This finding is unexpected as the cholesterol added to the cells to induce HMGCR degradation was not in the form of LDL . EH domain-containing protein 1 ( EHD1 ) is required for the internalisation and recycling of several plasma membrane receptors , including the LDLR ( Naslavsky et al . , 2007; Naslavsky and Caplan , 2011 ) and loss of EDH1 impairs LDLR trafficking with decreased intracellular cholesterol levels ( Naslavsky et al . , 2007 ) . The other significant hits in the screen ( PPAP2C , FER ) have not been validated . The screen also identified the E2 ubiquitin conjugating enzyme UBE2G2 and the E3 ubiquitin ligase RNF145 . Depletion of UBE2G2 prevented HMGCR degradation , implying that all ligases involved in HMGCR degradation utilise this E2 enzyme . In contrast , and despite being a high confidence hit in our screen , depletion of RNF145 caused a highly reproducible but small inhibition of sterol-accelerated HMGCR degradation , confirming the sensitivity of the screen to detect partial phenotypes and predicting the requirement for at least one additional UBE2G2-dependent ligase . A subsequent , targeted ubiquitome library screen in an RNF145-knockout reporter cell line confirmed a role for gp78 in HMGCR degradation . Gp78 has previously been shown to use UBE2G2 as its cognate E2 enzyme in the degradation of ERAD substrates ( Chen et al . , 2006 ) . During preparation of this manuscript , the combined involvement of RNF145 and gp78 in Insig-mediated HMGCR degradation in hamster ( CHO ) cells was also reported ( Jiang et al . , 2018 ) , confirming the role for these ligases in other species . The availability of an RNF145-specific polyclonal antibody provides further insight into the expression and activity of endogenous RNF145 , without the concerns of overexpression artefacts . RNF145 is an ER-resident E3 ubiquitin ligase with several unique features that make it well-suited for HMGCR regulation . A challenge facing all proteins responsible for cholesterol regulation is that the target they monitor , cholesterol , resides entirely within membranes . Like HMGCR and SCAP , RNF145 contains a putative sterol-sensing domain in its transmembrane region ( Cook et al . , 2017 ) , suggesting that sterols may facilitate RNF145’s association with Insigs . In contrast to Jiang et al . , 2018 , who reported a constitutive , sterol-independent association between ectopically expressed RNF145 and Insig-1 or −2 , we find that endogenous RNF145 interacts with endogenous Insig-1 in a sterol-dependent manner ( Figure 6C ) , as reported for the interaction of SCAP and HMGCR with the Insig proteins ( Lee et al . , 2007 ) . The binding of RNF145 to Insig-1 is HMGCR-independent ( Figure 6D ) . Furthermore , in the absence of Insigs , the RNF145-HMGCR association is lost ( Figure 6E ) , implying that the interaction between these two proteins is absolutely Insig-dependent . Therefore , sterols trigger the recruitment of RNF145 to HMGCR via Insigs , leading to HMGCR ubiquitination and degradation . This ability of RNF145 to rapidly bind Insigs following sterol availability supports a key role for this ligase in HMGCR regulation . A striking feature of RNF145 is its short half-life and rapid proteasome-mediated degradation , which contrasts with the long-lived gp78 ( Figure 5A/B , Figure 5—figure supplement 1B ) . RNF145 is an intrinsically unstable ligase whose half-life is regulated through autoubiquitination and was not prolonged on binding to Insig proteins ( data not shown ) . Its stability and turnover is RING- and UBE2G2-dependent , but independent of either the gp78 ( Figure 6A- C ) or Hrd1 E3 ligase ( Figure 5—figure supplement 1C ) . As cells become sterol-depleted , the transcriptional increase in RNF145 ( Figure 5C/D ) likely anticipates the need to rapidly eliminate HMGCR , once normal cellular sterol levels are restored . While this sterol-dependent transcriptional increase in RNF145 expression may at first seem counterintuitive , under sterol-deplete conditions the RNF145 ligase is not engaged with Insigs or its HMGCR substrate . The build-up of RNF145 predicts the restoration of sterol concentrations allowing RNF145 to immediately engage with , and degrade its HMGCR substrate . Thus the build-up of RNF145 anticipates its critical role in the restoration of cellular cholesterol homeostasis . RNF145 transcription was reported to be regulated by the sterol-responsive Liver X Receptor ( LXR ) family of transcription factors ( Cook et al . , 2017; Zhang et al . , 2017 ) , which transcriptionally activate cholesterol efflux pumps ( ABCA1 , ABCG1 ) ( Costet , 2000; Edwards et al . , 2002 ) and the IDOL E3 ubiquitin ligase , which targets the LDLR for degradation ( Zelcer et al . , 2009 ) . Pharmacological treatment of HeLa cells with the LXR inducer ( GW3965 ) increased protein levels of ABCA1 , but RNF145 transcript levels were not significantly increased ( Figure 5 – figure supplement 2A/B ) . In HeLa cells , therefore , the increased expression of RNF145 following cholesterol starvation is not primarily driven by the LXR pathway . While it is not unusual for more than one ligase to be required for substrate ERAD degradation ( Christianson and Ye , 2014; Morito et al . , 2008; Stefanovic-Barrett et al . , 2018 ) , the redundancy in HMGCR turnover is intriguing . This may simply reflect the central role of HMGCR in the mevalonate pathway and the importance of a fail-safe mechanism of HMGCR regulation to both maintain substrates for non-sterol isoprenoid synthesis and prevent cholesterol overproduction . Alternative explanations can also be considered , particularly as the properties of RNF145 and gp78 are so different . Under sterol-deplete conditions gp78 also regulates the degradation of Insig-1 , but following addition of sterols , the association of Insigs with SCAP displaces Insigs binding to gp78 ( Yang et al . , 2002; Lee et al . , 2006 ) . Different Insig-associated complexes are therefore likely to co-exist within the ER membrane , under both sterol-replete and -deplete conditions , and will reflect the sterol microenvironment of the ER ( Goldstein et al . , 2006 ) . Under these circumstances it might be advantageous to have more than one ligase regulating HMGCR . Alternatively , gp78 may provide basal control of the reductase , which can then be ‘fine-tuned’ by the sterol-responsive RNF145 , reflecting the sterol concentration of the local ER environment . Further understanding of the stoichiometry and nature of the different Insig complexes within the ER membrane will be important . While all cells need to regulate their intracellular cholesterol , the contribution of each ligase to sterol regulation may also depend on their differential tissue expression . In this regard , liver-specific ablation of gp78 in mice has been reported to lead to increased steady-state levels of hepatocyte HMGCR ( Liu et al . , 2012 ) , whereas gp78 knockout MEFs show no apparent impairment in HMGCR degradation ( Tsai et al . , 2012 ) . Further delineation of the contribution of each ligase to HMGCR degradation in different tissues and cell types will be important . A role for the Hrd1 E3 ligase in HMGCR regulation was unanticipated , and both orthologues ( gp78 and Hrd1 ) of yeast Hrd1p , which regulates yeast HMGCR ( Hmg2p ) , are therefore involved in mammalian HMGCR turnover . The best recognised function of Hrd1 is the ubiquitination of misfolded or unassembled ER-lumenal and membrane proteins targeted for ERAD ( Sato et al . , 2009; Tyler et al . , 2012; Christianson et al . , 2008 ) . Our finding that Hrd1 is only involved in HMGCR regulation when the other two ligases are absent , suggests that under sterol-rich conditions , and in the absence of RNF145 or gp78 , conformational changes in the sterol-sensing domains of HMGCR may lead to a less ordered state and be recognised and targeted by the Hrd1 quality control pathway . Ligand-induced selective and reversible local misfolding in Hmg2p , dubbed ‘mallostery’ , is a suggested mode of recognition by Hrd1p ( Wangeline et al . , 2017; Wangeline and Hampton , 2018 ) . The mechanism underlying recognition of HMGCR by Hrd1 is unclear , and whether the Hrd1 complex directly recognises sterol-induced structural changes as seen with Hmg2 degradation in yeast is unknown . Hrd1 might utilise the Insig proteins as scaffolds for HMGCR binding . This is partially borne out in the complete rescue of HMGCR in Insig-1 and −2 depleted cells ( Figure 1F ) . These mechanisms are not mutually exclusive and suggest that the contributions by different ligases may represent a regulated misfolding event as part of a ligand-mediated control of HMGCR stability . Further investigation is needed to clearly determine their contribution to HMGCR regulation . In summary , our unbiased approach to identify proteins involved in sterol-regulated HMGCR degradation resolves the ambiguity of the responsible E3 ubiquitin ligases , and uncovers additional control points in modulating the activity of this important enzyme in health and disease . Single guide RNAs ( sgRNAs ) were cloned into pSpCas9 ( BB ) −2A-Puro V1 ( Addgene #48139 , deposited by Dr . Feng Zhang ) , pSpCas9 ( BB ) −2A-Puro V2 ( Addgene #62988 , deposited by Dr . Feng Zhang ) as previously described ( Ran et al . , 2013 ) . The genome-wide sgRNA library ( Morgens et al . , 2017 ) was a kind gift from the Bassik lab ( Stanford University ) . To generate the ubiquitome sgRNA library , sgRNAs ( sgRNA sequences in Supplementary file 1 ) were cloned into pKLV-U6gRNA ( BbsI ) -PGKpuro2ABFP ( Addgene # 50946 ) as reported previously ( Doench et al . , 2016 ) . To generate RNF145 expression plasmids , the RNF145 CDS , PCR amplified from an RNF145 IMAGE clone ( Source Bioscience , Nottingham , UK ) , was cloned into pHRSIN-PSFFV-GFP-PPGK-HygromycinR ( BamHI , NotI ) ( Demaison et al . , 2002 ) , replacing GFP with the transgene . To create RNF145-V5 , the RNF145 CDS was Gibson-cloned into pHRSIN-PSFFV-PPGK-HygromycinR containing a downstream in-frame V5-tag . RNF145-V5 RING domain mutations ( C552A , H554A ) were introduced by PCR amplification of RNF145-V5 fragments with primers encoding C552A and H554A mutations and RNF145 ( C552A , H554A ) -V5 was introduced into pHRSIN-PSFFV-PPGK-HygromycinR by Gibson assembly . FLAG-NLS-Cas9 was cloned from the lentiCRISPR v2 ( Sanjana et al . , 2014 ) ( Addgene #49535 , deposited by Feng Zhang ) into pHRSIN . pSFFV MCS ( + ) pSV40 Blast ( BamHI , NotI ) . The following compounds were used in this study: Dulbecco’s Modified Eagle’s Medium high glucose ( DMEM; Sigma-Aldrich , 6429–500 ml ) , foetal calf serum ( FCS; Seralab ( catalogue no: EU-000 , SLI batch: E8060012 , Supplier batch: A5020012 ) and Life Technologies ( catalogue no: 10270 , lot: 42G4179K ) ) , lipoprotein-deficient serum ( LPDS; biosera , FB-1001L/100 ) , mevastatin ( Sigma-Aldrich , M2537-5MG ) , mevalonolactone ( Sigma-Aldrich , M4467-1G ) , cholesterol ( Sigma-Aldrich , C3045-5G ) , 25-hydroxycholesterol ( Sigma-Aldrich , H1015-10MG ) , methyl-β-cyclodextrin ( MBCD; Sigma-Aldrich , 332615–1G ) , GW3965 HCl ( Sigma-Aldrich , G6295 ) , bortezomib/PS-341 ( BostonBiochem , I-200 ) , ( S ) -MG132 ( Cayman Chemicals , 10012628 ) , NMS-873 ( Selleckchem , s728501 ) , digitonin ( Merck , 300410–5 GM ) ( 1% digitonin for immunoprecipitation experiments was generated by using the soluble supernatant of a 2% digitonin solution which had been left rotating with a small amount of CL4B beads overnight ) , cycloheximide ( Sigma-Aldrich , C-7698 ) , IgG SepharoseTM 6 Fast Flow ( GE Healthcare , 17-0969-01 ) , ProLong Gold Antifade Mountant with DAPI ( Thermo Fisher ) , bovine serum albumin ( BSA; Sigma-Aldrich , A4503-10G ) , Protein A-SepharoseR ( Sigma-Aldrich , P3391-1 . 5G ) , iodoacetamide ( IAA; Sigma-Aldrich , I1149-5G ) , cOmplete protease inhibitor ( EDTA-free; Roche , 27368400 ) , phenylmethylsulfonyl fluoride ( PMSF; Roche , 20039220 ) , V5 peptide ( Sigma-Aldrich , V7754-4MG ) , N-ethylmaleimide ( NEM; Sigma-Aldrich , E3876-5G ) , puromycin ( Cayman Chemicals , 13884 ) , hygromycin B ( Invitrogen , 10687010 ) , Penicillin-Streptomycin ( 10 , 000 U/mL; Thermo Fisher , 15140122 ) . Antibodies specific for the following targets were used for immunoblotting analysis: Insig-1 ( rabbit; Abcam , ab70784 ) , Hrd1 ( rabbit; Abgent , AP2184a ) , TRC8 ( rabbit; Santa Cruz , sc-68373 ) , tubulin ( mouse; Sigma , T9026 ) , VCP ( mouse; abcam , ab11433 ) , β-actin ( mouse; Sigma-Aldrich , A5316 ) , calnexin ( mouse; AF8 , kind gift from M Brenner , Harvard Medical School ) , calreticulin ( rabbit; Pierce , PA3-900 ) , HMGCR ( mouse; Santa Cruz , sc-271595 ) , HMGCR ( rabbit; Abcam , ab174830 ) , gp78 ( rabbit; ProteinTech , 16675–1-AP ) , Insig-1 ( rabbit; Abcam , ab70784 ) , RNF145 ( rabbit; ProteinTech , 24524-I-AP ) , V5 ( mouse; Abcam , ab27671 ) , VU-1 ubiquitin ( mouse; Life Sensors , VU101 ) , UBE2G2 ( mouse; Santa Cruz , sc-100613 ) , GFP ( rabbit; Thermo Fisher Scientific , A11122 ) , KDEL ( mouse; Enzo , 10C3 ) , HRP-conjugated anti-mouse and anti-rabbit ( goat; Jackson ImmunoResearch ) , TrueBlot Anti-Rabbit-HRP ( Rockland , 18-8816-31 ) , TrueBlot Anti-Mouse-HRP ULTRA ( Rockland , 18-8817-30 ) . Alexa Fluor 488 ( goat anti-rabbit; Thermo Fisher ) , Alexa Fluor 568 ( goat anti-mouse; Thermo Fisher ) were used as secondary antibodies for immunofluorescence microscopy . Anti-MHC-I ( W6/32; mouse ) and Alexa Fluor 647 ( rabbit anti-mouse; Thermo Fisher ) were used for cytofluorometric analysis . HeLa , HEK-293T , and HepG2 cells were maintained in DMEM +10% FCS+penicillin/streptomycin ( 5% CO2 , 37°C ) . HeLa cells were obtained from ECACC . HEK-293T and HepG2 cells were obtained from ATCC . HeLa and HepG2 cells were authenticated by STR profiling ( Eurofins Genomics ) . All cell lines tested mycoplasma negative ( Lonza MycoAlert ) . Transfection of HeLa cells was performed using the TransIT-HeLa MONSTER kit ( Mirus ) according to the manufacturer’s instructions . In sort , cells were seeded at low confluency in 12-well tissue culture plates and the next day transfection mix ( 1 µg DNA , 3 µl TransIT-HeLa reagent +2 µl MONSTER reagent in OptiMEM ( Gibco ) ) was added . Alternatively , reverse transfection was performed by seeding 3 . 5105 cells per well of a 12-well plate to the transfection mix on the day of transfection . For co-transfection of multiple sgRNA plasmids , equal amounts of each plasmid were added up to 1 µg . CRISPR/Cas9-mediated genomic editing was performed according to Ran et al . ( Ran et al . , 2013 ) . For generation of knockout cell lines , cells were transfected with pSpCas9 ( BB ) −2A-Puro ( PX459 ) V1 . 0 or V2 . 0 ( Addgene #48139 , and #62988 respectively; deposited by Dr . Feng Zhang ) containing a sgRNA specific for the targeted gene of interest . Guide RNA sequences are listed in Supplementary file 2 . Cells were cultured for an additional 24 hr before selection with puromycin ( 2 µg/ml ) at low confluency for 72 hr . The resulting mixed knockout populations were used to generate single-cell clones by limiting dilution or fluorescence-assisted single-cell sorting . A detailed list of single cell knockout clones used in this study can be found in Supplementary file 3 . Gene disruption was validated by immunoblotting , immunoprecipitation and/or targeted genomic sequencing . An HMGCR-Clover knock-in donor template was created by Gibson assembly of ~1 kb flanking homology arms , PCR-amplified from HeLa genomic DNA , and the NsiI and PciI digested backbone from pMAX-GFP ( Amaxa ) cloned into the loxP-Ub-Puro cassette from pDonor loxP Ub-Puro ( a kind gift from Ron Kopito , Stanford University ) . Each arm was amplified using nested PCR . The 5' arm was amplified using 5’-GATGCAGCACAGAATGTTGGTAG-3’ and 5’-CAATGCCCATGTTCCAGTTCAG-3’ , followed by 5’-CAATGCCCATGTTCCAGTTCAG-3’ and 5’-CAGCTGCACCATGCCATCTATAG-3’ . The 3' arm was amplified using the following primer pairs: 5’-CCAAGGAGCTTGCACCAAGAAG-3’ and 5’-CTAAGGTCCCAGTCTTGCTTG-3’ . The product served as template for a subsequent PCR step using the primers 5’-CCAAGGAGCTTGCACCAAGAAG-3’ and 5’-GTCACCCTCATCTAAGCAAC-3’ . Overhangs required for Gibson assembly were introduced by PCR . HeLa cells were co-transfected with Cas9 , sgRNA targeting immediately downstream of the HMGCR stop codon and donor template . Three different donor templates were simultaneously transfected , each differing in the drug resistance marker ( puromycin , hygromycin and blasticidin ) . The transfected cells were treated with the three antibiotics five days post transfection until only drug-resistant cells remained . The resulting population was transfected with Cre-recombinase in pHRSIN MCS ( + ) IRES mCherry pGK Hygro . mCherry positive cells were single-cell cloned by FACS . To confirm the knock-in of a myc- and Clover-tag downstream of the HMGCR coding sequence , genomic DNA was isolated from HeLa HMGCR-Clover cells using the Quick-gDNA MicroPrep kit ( Zymo Research ) . The genomic sequence encoding the myc- and Clover-tags and flanking 5’ and 3’ homology regions were amplified using the following primer combination: 5’- ACTATTCATCTACTGTAGTTCCAAGTTAAAATTCTACACTC-3’ , 5’- GCATGTAAAGCACTAAACTGTGTTCAGATCTGAGGAGTC-3’ . PCR products were separated by agarose gel electrophoresis , gel excised and analysed by Sanger sequencing . HEK-293T cells were transfected with a lentiviral expression vector , the packaging vectors pCMVΔR8 . 91 and pMD . G at a ratio of 1:0 . 7:0 . 3 using TransIT-293 ( Mirus ) as recommended by the manufacturer . For production of CRISPR library virus , HEK-293T cells were transfected as above in 15 cm tissue culture plates . 48 hr post transfection , virus-containing media was collected , filtered ( 0 . 45 µm pore size ) and directly added to target cells or frozen ( −80°C ) for long-term storage . Typically , cells were transduced in 6-well tissue culture plates at an M . O . I . <1 and selected with puromycin ( 2 µg/ml ) or hygromycin B ( 200 µg/ml ) . To generate HeLa HMGCR-Clover stably expressing Cas9 , HeLa HMGCR-Clover cells were transduced with pHRSIN-PSFFV-Cas9-PPGK-HygromycinR and stable integrants selected with hygromycin B . Cas9 activity was confirmed by transduction with pKLV encoding a β−2-microglobulin ( B2M ) -targeting sgRNA followed by puromycin selection . MHC-I surface expression was assessed by flow cytometry in puromycin-resistant cells five days post transduction . Typically , ~90% reduction of cell surface MHC-I expression was observed . To identify CRISPR-induced frame-shift mutations , genomic DNA was extracted from wild type HeLa cells and RNF145 CRISPR clones using the Quick-gDNA MicroPrep kit ( Zymo Research ) followed by nested PCR of the genomic region 5’ and 3’ of the predicted sgRNA binding site . One in each primer pair for the second PCR was 5’ modified with 6-FAMTM ( fluorescein , Sigma-Aldrich ) . Primer sequences were as follows: For sgRNA #8 PCR1_Forward: CAGAATGCTCACTAGAAGATTAG , PCR1_Reverse: GTAGTATACGTTCTCACATAG , PCR2_Forward: GTGATGTAGACACTCACCTAC and PCR2_Reverse: GTGACAACCTATTAGATTCGTG . PCR products were detected using an ABI 3730xl DNA Analyser . Cells were collected by trypsinisation and analysed using a FACS Calibur ( BD ) or an LSR Fortessa ( BD ) . Flow cytometry data was analysed using the FlowJo software package . Cells resuspended in sorting buffer ( PBS + 10 mM HEPES +2% FCS ) were filtered through a 50 µm filter , and sorted on an Influx machine ( BD ) , or , for the ubiquitome CRISPR/Cas9 screen , on a FACS Melody ( BD ) . Sorted cells were collected in DMEM +50% FCS and subsequently cultured in DMEM +10% FCS+penicillin/streptomycin . For MHC-I flow cytometric analysis , cells resuspended in cold PBS were incubated with W6/32 ( 20 min , 4°C ) , washed twice and then incubated with Alexa-647-labelled anti-mouse antibody ( 15 min , 4°C ) . Cells were washed twice and resuspended in PBS . For genome-wide and ubiquitome-library CRISPR/Cas9 knockout screens , 108 and 1 . 2107 HeLa HMGCR-Clover ( Cas9 ) or ΔRNF145 #6 ( Cas9 ) , respectively , were transduced at M . O . I . ~ 0 . 3 by spinfection ( 750xg , 60 min , 37°C ) . Transduction efficiency was determined via flow-cytometry-based measurement of mCherry ( genome-wide screen ) or BFP ( ubiquitome screen ) expression 48–72 hr post infection . Transduced cells were enriched by puromycin selection ( 2 µg/ml ( genome-wide screen ) , 1 µg/ml ( ubiquitome-library screen ) ) . On day 8 ( genome-wide screen ) or day 7 ( ubiquitome-library screen ) post transduction , cells were rinsed extensively with PBS and cultured overnight in starvation medium ( DMEM +10% LPDS+10 µM mevastatin +penicillin/streptomycin ) before sterol addition ( 2 µg/ml 25-hydroxycholesterol and 20 µg/ml cholesterol for 5 hr ) . An initial FACS selection ( ‘sort #1’ ) on cells expressing high levels ( ~0 . 3–0 . 6% of overall population ) of HMGCR-Clover ( HMGCR-Cloverhigh ) was performed . 2105 ( genome-wide screen ) and ~105 ( ubiquitome-library screen ) sorted cells were pelleted and DNA was extracted using the Quick-gDNA MicroPrep kit ( Zymo Research ) . To gauge sgRNA enrichment , DNA was extracted from 3106 ( genome-wide library screen ) or 6106 ( ubiquitome library screen ) cells pre-sort using the Gentra Puregene Core kit A ( Qiagen ) . Cells in the genome-wide screen were subjected to a second round of sterol deprivation and sort ( see above ) after expansion of initially 2 . 5105 sorted cells for 8 days . Sorted cells were cultured until 5106 cells could be harvested for genomic DNA extraction using the Gentra Puregene Core kit A ( Qiagen ) . Individual integrated sgRNA sequences were amplified by two sequential rounds of PCR , the latter introducing adaptors for Illumina sequencing ( Supplementary file 4 ) . Sequencing was carried out using the Illumina HighSeq ( genome-wide screen ) and MiniSeq ( ubiquitome-library screen ) platforms . Illumina HiSeq data was analysed as described previously ( Timms et al . , 2016 ) . Guide RNA counts were analysed with the RSA algorithm under default settings ( König et al . , 2007 ) . Of note , a gene’s calculated high significance value and therefore high enrichment in the selected population does not necessarily reflect its importance relative to genes with lower significance values/enrichment , since gene disruption can be incomplete or lethal phenotypes might evade enrichment . Whole-cell RNA was isolated with the RNeasy Plus Mini Kit ( Qiagen , Venlo , Netherlands ) and reverse transcribed using Oligo ( dT ) 15 primer ( Promega , C110A ) and SuperScriptTM III reverse transcriptase ( Invitrogen ) . Transcript levels were determined in triplicate using SYBR Green PCR Master Mix ( Applied Biosystems ) in a real time PCR thermocycler ( 7500 Real Time PCR System , Applied Biosystems ) . Primers used for target amplification can be found in Supplementary file 5 . RNA quantification was performed using the ΔΔCT method . GAPDH transcript levels were used for normalization . Raw data can be found in Figure 5—source data 1 and Figure 5—figure supplement 2—source data 1 . Typically , HeLa cells at ~50% confluency were washed five times with PBS and cultured for 16–20 hr in starvation medium ( DMEM +10% LPDS+10 µM mevastatin +50 µM mevalonate +penicillin/streptomycin ) before addition of 25-hydroxycholesterol ( 2 µg/ml ) and cholesterol ( 20 µg/ml ) to analyse sterol-accelerated protein degradation . Complexation of cholesterol ( 2 . 5 mM ) with MBCD ( 25 mM ) was performed according to Christian et al . ( Christian et al . , 1997 ) . An emulsion of cholesterol powder ( final: 2 . 5 mM ) and an MBCD solution ( 25 mM ) was produced by vortexing and tip sonication ( 1 min in 10 s intervals ) , and continuously mixed for 16 hr at 37°C . The solution was sterile filtered ( 0 . 45 μm PVDF pore size ) and stored at −20°C . Sterols were prepared by resuspension in ethanol or complexation with MBCD ( see above ) . Mevalonate was prepared by adding 385 µl 2 . 04 M KOH to 100 mg mevalonolactone ( Sigma ) . The solution was heated ( 1 hr , 37°C ) and adjusted to a 50 mM stock solution . Cells were collected mechanically in cold PBS or by trypsinisation , centrifuged ( 1000xg , 4 min , 4°C ) , and cell pellets resuspended in lysis buffer ( 1% ( w/v ) digitonin , 1x cOmplete protease inhibitor , 0 . 5 mM PMSF , 10 mM IAA , 2 mM NEM , 10 mM TRIS , 150 mM NaCl , pH 7 . 4 ) . After 40 min incubation on ice , lysates were centrifuged ( 17 . 000xg , 15 min , 4°C ) , the post-nuclear fraction isolated and protein concentration determined by Bradford assay . Samples were adjusted with lysis buffer and 6 x Laemmli buffer +100 mM dithiothreitol ( DTT ) and heated at 50°C ( 15 min ) . Samples were separated by SDS-PAGE and transferred to PVDF membranes ( Merck ) for immunodetection . Membranes were blocked in 5% milk +PBST ( PBS + 0 . 2% ( v/v ) Tween-20 ) ( 1 hr ) and incubated with primary antibody in PBST +2% ( w/v ) BSA at 4°C overnight . For detection from whole-cell lysate , membranes were incubated in peroxidase ( HRP ) -conjugated secondary antibodies . For detection of immunoprecipitated proteins , TrueBlot HRP-conjugated secondary antibodies ( Rockland ) were used . Immunoprecipitated RNF145 was detected using Protein A-conjugated HRP . Cells were seeded to 15 cm tissue culture plates ( 4106 cells per plate ) . The following day , cells were washed five times with PBS and cultured in starvation medium ( DMEM +10% LPDS+10 µM mevastatin +50 µM mevalonate +penicillin/streptomycin ) for 20 hr . To prevent HMGCR membrane extraction and degradation , starved cells were treated with NMS-873 ( 50 µM ) 0 . 5 hr prior to sterol addition ( 2 µg/ml 25-hydroxycholesterol and 20 µg/ml cholesterol for 1 hr ) and collection in cold PBS . Cells were lysed in IP buffer 1 ( 1% ( w/v ) digitonin , 10 µM ZnCl2 , 1x cOmplete protease inhibitor , 0 . 5 mM PMSF , 10 mM IAA , 2 mM NEM , 10 mM TRIS , 150 mM NaCl , ph 7 . 4 ) , post-nuclear fractions isolated by centrifugation ( 17 . 000xg , 4°C , 15 min ) adjusted to 0 . 5% ( w/v ) digitonin and pre-cleared with IgG SepharoseTM 6 Fast Flow ( 1 hr ) . Endogenous RNF145 and V5-tagged RNF145 were immunoprecipitated at 4°C overnight from 3 to 6 mg whole-cell lysate using Protein A-Sepharose and anti-RNF145 or V5 antibody , respectively . Beads were collected by centrifugation ( 1500xg , 4 min , 4°C ) , washed for 5 min with IP buffer 2 ( 0 . 5% ( w/v ) digitonin , 10 µM ZnCl2 , 10 mM Tris , 150 mM NaCl , pH 7 . 4 ) and 4 × 5 min with IP buffer 3 ( 0 . 1% ( w/v ) digitonin , 10 µM ZnCl2 , 10 mM TRIS , 150 mM NaCl , pH 7 . 4 ) . Proteins whose interaction with RNF145 was labile in the presence of 1% ( v/v ) Triton X-100 were recovered by eluting twice for 30 min with 20 µl TX100 elution buffer ( 1% ( v/v ) Triton X-100 +2 x cOmplete protease inhibitor in 10 mM TRIS , 150 mM NaCl pH 7 . 4 ) at 37°C under constant agitation . Immunoprecipitated RNF145 was subsequently eluted in 30 µl 2x Laemmli buffer +3% ( w/v ) DTT at 50°C ( 15 min ) . RNF145-V5 and associated complexes were recovered by two sequential elutions with V5 elution buffer ( 1 mg/ml V5 peptide +2 x cOmplete protease inhibitor in 10 mM TRIS , 150 mM NaCl pH 7 . 4 ) for 30 min at 37°C under continuous agitation . Eluted samples were adjusted with Laemmli buffer and denatured at 50°C ( 15 min ) . Cells were sterol-depleted ( 20 hr ) , treated with 20 µM MG132 and left for 30 min before addition of sterols ( 2 µg/ml 25-hydroxycholesterol and 20 µg/ml cholesterol for 1 hr ) or EtOH ( vehicle control ) . Immunoprecipitation of ubiquitinated HMGCR was performed as described above from 1 mg whole-cell lysate and using rabbit α-HMGCR ( Abcam , ab174830 ) . Proteins were eluted in 30 µl 2x Laemmli buffer +100 mM DTT at 50°C ( 15 min ) . For immunoblotting of ubiquitin with mouse VU-1 α-ubiquitin ( Life Sensors , VU101 ) , the PVDF membrane was incubated with 0 . 5% ( v/v ) glutaraldehyde/PBS pH 7 . 0 ( 20 min ) and washed 3x with PBS prior to blocking in 5% ( w/v ) milk +PBS + 0 . 1% ( v/v ) Tween-20 . Cells were grown on coverslips , fixed in 4% PFA ( 15 min ) , permeabilised in 0 . 2% ( v/v ) Triton X-100 ( 5 min ) and blocked with 3% ( w/v ) BSA/PBS ( 30 min ) . Cells were stained with primary antibody diluted in 3% ( w/v ) BSA/PBS ( 1 hr ) , washed with 0 . 1% ( w/v ) BSA/PBS , followed by staining with secondary antibody in 3% BSA/PBS ( 1 hr ) , an additional washing step ( 0 . 1% ( w/v ) BSA/PBS ) and embedded using ProLong Gold Antifade Mountant with DAPI ( Thermo Fisher ) . Images were acquired using an LSM880 confocal microscope ( Zeiss ) at 64x magnification . Statistical significance was calculated using the unpaired Student’s t-test . Sequencing data from CRISPR/Cas9 knockout screens presented in this study have been deposited at the Sequence Read Archive ( SRA ) ( genome-wide screen: SRP151225; ubiquitome screen: SRP151107 ) .	Cholesterol is a fatty molecule that is essential for our health; for example , it is a component of the outer membrane that surrounds every cell in our body . Yet , it also has a reputation for clogging arteries and causing heart attacks and strokes . Our organism can adjust the amount of cholesterol it creates through an enzyme called HMGCR , which is found in all cells . Switching off HMGCR , for instance by taking drugs called statins , reduces the amount of cholesterol made by cells . To regulate the activity of HMGCR , the body uses proteins known as E3 ubiquitin ligases , which can label the enzyme for destruction . However , the identity of the ligases that target HMGCR is a matter of intense debate . Here , Menzies , Volkmar et al . addressed this issue by using an approach called a genome-wide CRISPR forward genetic screen . First , HMGCR was marked inside the cells with a fluorescent tag to watch how its levels change in response to different amounts of cholesterol . Then , each gene in the cell was deleted , and the effects recorded . This allowed Menzies , Volkmar et al . to find the genes responsible for the rapid destruction of HMGCR . The experiments revealed that the E3 ubiquitin ligases RNF145 and gp78 are independently responsible for the degradation of the majority of HMGCR , with a third ligase , Hrd1 , getting involved if the first two are absent . In particular , RNF145 builds up when a cell is starved of cholesterol , but it immediately marks HMGCR for destruction once cholesterol becomes more abundant . This ligase can therefore both sense and respond to the amount of cholesterol in a cell , making it a perfect candidate for regulating HMGCR based on what the body needs . Identifying the proteins that adjust the levels of HMGCR sheds light on how a cell controls the amount of cholesterol it creates . This knowledge could be relevant in the fight against the health problems associated with this molecule .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "research", "communication", "cell", "biology" ]	2018	The sterol-responsive RNF145 E3 ubiquitin ligase mediates the degradation of HMG-CoA reductase together with gp78 and Hrd1
Genetic variations in the myeloid immune receptor TREM2 are linked to several neurodegenerative diseases . To determine how TREM2 variants contribute to these diseases , we performed structural and functional studies of wild-type and variant proteins . Our 3 . 1 Å TREM2 crystal structure revealed that mutations found in Nasu-Hakola disease are buried whereas Alzheimer’s disease risk variants are found on the surface , suggesting that these mutations have distinct effects on TREM2 function . Biophysical and cellular methods indicate that Nasu-Hakola mutations impact protein stability and decrease folded TREM2 surface expression , whereas Alzheimer’s risk variants impact binding to a TREM2 ligand . Additionally , the Alzheimer’s risk variants appear to epitope map a functional surface on TREM2 that is unique within the larger TREM family . These findings provide a guide to structural and functional differences among genetic variants of TREM2 , indicating that therapies targeting the TREM2 pathway should be tailored to these genetic and functional differences with patient-specific medicine approaches for neurodegenerative disorders . The precise molecular determinants and mechanisms underlying neurodegenerative diseases remain uncertain , but recent large-scale genetic sequencing projects have uncovered new candidates associated with these diseases . For example , whole genome and whole exome sequencing has identified point mutations in the gene encoding the protein TREM2 ( triggering receptor expressed on myeloid cells 2 ) that correlate with a significantly increased risk of developing Alzheimer’s disease ( AD ) ( Guerreiro et al . , 2013b; Jonsson et al . , 2013 ) . In particular , the TREM2 R47H variant is associated with a risk that is similar to that associated with APOE4 , previously the only well-established risk factor for late-onset AD . This mutation correlates with increased cerebrospinal fluid ( CSF ) tau levels , a well-established risk factor for AD ( Cruchaga et al . , 2013 ) , and subsequent studies have identified this mutation in patients who have frontal temporal dementia ( FTD ) , Parkinson’s disease ( PD ) ( Rayaprolu et al . , 2013 ) , and sporadic amyotrophic lateral sclerosis ( ALS ) ( Cady et al . , 2014 ) . A recent study of more than 1 , 600 late-onset AD ( LOAD ) and cognitively normal brains revealed that microglia-specific networks , including those containing TREM2 , are those most significantly dysregulated in LOAD ( Zhang et al . , 2013 ) . Thus , even in the absence of rare variants in TREM2 that increase AD risk , TREM2-containing pathways play a significant role in disease . Recent studies in AD mouse models that are deficient in TREM2 confirm that loss of TREM2 function contributes to classic AD pathology and demonstrates a crucial role for TREM2 in central nervous system ( CNS ) biology ( Jay et al . , 2015; Wang et al . , 2015; Ulrich et al . , 2014 ) . TREM2-deficient AD mice display fewer activated microglia , and the absence of TREM2 prevents microglia proliferation and promotes microglia apoptosis , which was correlated with increased accumulation of Aβ plaques ( Wang et al . , 2015; Jay et al . , 2015 ) . Microglia in TREM2-deficient AD mice displayed less activation and did not engulf Aβ plaques . This impacts the density of Aβ plaques and promotes diffuse Aβ structures , which in turn are more neurotoxic , and contributes to the accumulation of classic AD pathology ( Wang et al . , 2016; Yuan et al . , 2016 ) . These findings highlight a crucial role for TREM2 in maintaining CNS homeostasis . Therefore , understanding how these risk variants affect TREM2 function and contribute to the pathogenesis of neurodegenerative diseases is vital to the development of therapies targeting these devastating conditions . TREM2 is an innate immune receptor expressed on dendritic cells ( DCs ) , resident macrophages such as osteoclasts and microglia , infiltrating ( Jay et al . , 2015 ) and inflammatory ( Wu et al . , 2015 ) macrophages , and CSF monocytes ( Colonna and Wang , 2016 ) . It is a type one receptor protein consisting of an extracellular V-type Ig domain , a short stalk , a transmembrane domain that associates with the adaptor protein DAP12 for signaling , and a cytoplasmic tail ( Figure 1a ) ( Colonna , 2003 ) . TREM2 has historically been shown to play an anti-inflammatory role in vitro by antagonizing the production of inflammatory cytokines from bone-marrow-derived macrophages ( BMDMs ) and dendritic cells ( BMDDCs ) in response to FcR ( Hamerman et al . , 2006 ) and Tlr signaling ( Turnbull et al . , 2006; Ito and Hamerman , 2012 ) . Likewise , TREM2 participates in phagocytosis of apoptotic cells in cultured microglia and reduces the production of inflammatory cytokines ( Takahashi et al . , 2005 ) . However , TREM2-expressing macrophages can also promote inflammatory disease in the brain ( Jay et al . , 2015 ) and lung ( Wu et al . , 2015 ) . The identity of a physiologic TREM2 ligand ( TREM2-L ) remains uncertain , although several classes of molecules have been proposed , including bacterial carbohydrates ( Daws et al . , 2003; Quan et al . , 2008 ) , sulfoglycolipids ( Phongsisay et al . , 2015 ) , nucleic acids ( Kawabori et al . , 2015 ) , phospholipids ( Cannon et al . , 2012; Wang et al . , 2015 ) and proteins ( Stefano et al . , 2009; Takegahara et al . , 2006; Yoon et al . , 2012; Atagi et al . , 2015; Bailey et al . , 2015 ) . Additionally , previous studies have identified cells that express a TREM2-L , including astrocytes ( Daws et al . , 2003 ) , DCs ( Ito and Hamerman , 2012 ) , BMDMs ( Hamerman et al . , 2006 ) , neurons and apoptotic cells ( Hsieh et al . , 2009 ) . This growing body of literature underscores the case for immune deregulation , specifically involving TREM2-associated pathways in neurodegenerative and inflammatory diseases ( Golde et al . , 2013 ) . 10 . 7554/eLife . 20391 . 003Figure 1 . Crystal structure of the human TREM2 ectodomain . ( a ) Schematic of TREM2 cell-surface association with adapter protein DAP12 , which contains an Immuno Tyrosine Activation Motif ( ITAM ) . Engagement of TREM2-L by the ectodomain of TREM2 induces signaling . Domain boundaries are indicated . ( b ) TREM2 ectodomain in two orientations with disease-linked residues shown as sticks . The positions of AD risk variants are shown in magenta , whereas Nasu-Hakola disease ( NHD ) mutations are shown in cyan . The N-acetylglucosamine ( NAG ) is shown as green sticks . ( c ) Table of TREM2 disease-linked mutations , associated disease , and calculated solvent accessible surface exposure for the side-chain ( calculated using Naccess ) , along with statistical correlations to AD ( OR = odds ratio; MAF = mean allele frequency ) ( from Jin et al . , 2014 ) . Table is highlighted with same color scheme as Figure 1b . Validated AD risk variants ( R47H and R62H ) are not marked . Potential AD risk variants are denoted with an asterisk . ( d ) Side-by-side stereo view of difference electron density ( 2mFo-DFc contoured at 2σ ) for the N79-NAG . ( e–g ) Difference electron density ( 2mFo-DFc contoured at 2σ ) for the surface-exposed AD-associated mutation positions ( e ) R47 , ( f ) R62 and ( g ) T96 . DOI: http://dx . doi . org/10 . 7554/eLife . 20391 . 00310 . 7554/eLife . 20391 . 004Figure 1—figure supplement 1 . Analysis of TREM2 glycosylation , comparison of TREM2 monomers in the crystal structure , and SA-omit maps of AD-linked residues . ( a ) Immunoblot ( anti-6His ) analysis of secreted WT hTREM2 ectodomains . From left to right: natively glycosylated and secreted hTREM2 ectodomain , hTREM2 ectodomain secreted from kifunensine-treated cells , EndoHf-digested ectodomains from kifunensine-treated cells ( which have a single N-acetylglucosamine residue at the two N-linked sites ) , and PNGaseF-digested TREM2 ectodomain from kifunensine-treated cells ( from which the entire glycan has been removed ) . ( b ) The hTREM2 asymmetric unit ( ASU ) contains two monomers of the TREM2 ectodomain . Note that the NAG ( shown in sticks ) participates in the interface , suggesting that the full-length glycan would disrupt this interaction observed in the crystal . ( c ) Chain A and B superimposed ( Cα RMSD < 1 Å ) . Chain A is magenta and Chain B is cyan . The only notable difference between the two monomers is the loop in chain B , which makes lattice contacts with symmetry-related B-H103 and is slightly shifted compared to chain A , which does not make lattice contacts . ( d–g ) Simulated annealing composite omit maps ( 2mFo-DFc ) contoured at 2σ around residues ( d ) R47 , ( e ) R62 , ( f ) T96 , and ( g ) N79-NAG . DOI: http://dx . doi . org/10 . 7554/eLife . 20391 . 00410 . 7554/eLife . 20391 . 005Figure 1—figure supplement 2 . Packing neighbors for the TREM2 AD risk variant R47 and the TREM2 NHD mutants Y38 , T66 , and V126 . ( a ) Packing environment for surface residue R47 ( magenta carbon sticks and surface ) . The gray carbon surface shows all residues that contact R47 . Residues whose side chains make Van-der Waals contacts with the R47 side chain are shown as sticks and are labeled . ( b ) Packing environment for the buried residue Y38 ( cyan carbon sticks and surface ) . The gray carbon surface shows all residues that contact Y38 . Residues whose side chains make Van der Waals contacts with the Y38 side chain are shown as sticks and are labeled . ( c ) Packing environment for the buried residue T66 ( cyan carbon sticks and surface ) . The gray carbon surface shows all residues that contact T66 . Residues whose side chains make Van-der Waals contacts with the T66 side chain are shown as sticks and are labeled . ( d ) Packing environment for buried residue V126 ( cyan carbon sticks and surface ) . The gray carbon surface shows all residues that contact V126 . Residues whose side chains make Van der Waals contacts with the V126 side chain are shown as sticks and are labeled . DOI: http://dx . doi . org/10 . 7554/eLife . 20391 . 005 Intriguingly , genetic variations in TREM2 are associated with two distinct groups of neurodegenerative diseases . Homozygous mutations including early-stop codons ( Paloneva et al . , 2003; Soragna et al . , 2003 ) , splice site mutations ( Numasawa et al . , 2011; Chouery et al . , 2008 ) , the coding stalk mutations D134G and K186N ( Paloneva et al . , 2002 ) , and the coding ectodomain mutations Y38C , T66M , and V126G ( Guerreiro et al . , 2013a , 2013c; Le Ber et al . , 2014 ) cause either NHD , characterized by early-onset dementia , demyelination , and bone cyst lipoma ( Paloneva et al . , 2002; Colonna , 2003 ) , or a frontotemporal dementia variant with severe loss of brain matter but lacking the bone manifestations . By contrast , TREM2 mutations associated with AD contribute to disease risk as heterozygous variants . In addition to R47H , the coding mutation R62H is associated with increased risk of AD in independent studies ( Jin et al . , 2014; Ridge et al . , 2016 ) . These two variants have the strongest risk link to AD . N68K and D87N have also been identified in AD patients , but because these mutations are very rare , their risk remains uncertain ( Guerreiro et al . , 2013b; Jonsson et al . , 2013 ) . In addition , the mutation T96K has been linked to a decreased risk of AD , but this mutation is also too rare to allow verification of this observation ( Jin et al . , 2014 ) . This curious segregation of disease phenotypes by distinct mutations occurring within the same protein suggest that these mutations should have divergent effects on the structure and function of TREM2 . In order to understand the molecular basis of how different mutations within TREM2 can lead to distinct neurodegenerative diseases , we performed structural , biophysical , and functional studies of wild-type ( WT ) and mutant TREM2 proteins . To facilitate these studies , we developed a novel mammalian expression system to produce the natively folded and glycosylated TREM2 ectodomains in milligram quantities ( Kober et al . , 2014 , 2015 ) . We determined the crystal structure of the TREM2 ectodomain at 3 . 1 Å resolution and found that the disease-linked mutations exhibit distinct structural patterns , suggesting that they would impact TREM2 function through alternate mechanisms . To investigate this hypothesis , we carried out extensive biophysical , cellular , and functional assays . We found that NHD-causing mutations impact TREM2 protein folding and stability , whereas AD risk variants decrease binding to the cellular TREM2-ligand ( TREM2-L ) . Furthermore , these AD-linked mutations appear to epitope map a disease-relevant functional surface on TREM2 that facilitates binding to cell-surface TREM2-L . These findings demonstrate two distinct loss-of-function mechanisms for TREM2 , illuminate a disease-relevant functional surface on TREM2 , and pave the way for the development of patient-specific molecular therapies for the treatment of distinct neurodegenerative diseases . We previously reported the crystallization of the WT human TREM2 ectodomain ( amino acids 19–134 ) purified from a mammalian cell expression system ( Kober et al . , 2014 ) . These crystals diffracted to 3 . 1 Å ( Table 1 and Table 2 ) . The TREM2 ectodomain is a V-type Ig domain containing two disulfide bonds , the canonical Ig disulfide between residues C36 and C110 ( βB–βF ) and an additional linkage between C51 and C60 ( βC-βC’ ) ( Figure 1b ) . A single N-acetylglucosamine ( NAG ) glycan that remains after Endo Hf treatment ( Figure 1—figure supplement 1a ) is clearly visible in the electron density at residue N79 ( Figure 1d and Figure 1—figure supplement 1g ) . The nine β-strands characteristic of proteins within this Ig domain class are present , and two short α-helix passages are observed spanning β-sheets B–C and E–F . The asymmetric unit ( ASU ) contains two monomers of hTREM2 in a parallel arrangement , but the limited buried surface area ( ~400 Å2 ) and chemical nature of the interface suggest that this dimer would not exist in solution ( Figure 1—figure supplement 1b ) . Furthermore , these two monomers are quite similar ( Cα RMSD of 0 . 65 Å , Figure 1—figure supplement 1c ) , therefore , our analysis of hTREM2 described henceforth will refer to chain A . 10 . 7554/eLife . 20391 . 006Table 1 . Determination of resolution cut-off by evaluation of meaningful data . To evaluate the functional resolution of data past 3 . 3 Å , data were scaled to 3 . 0 Å resolution . Data were successively truncated at 0 . 1 Å intervals from 3 . 4 Å to 3 . 0 Å and the initial 3 . 3 Å solution was used to initiate molecular replacement , rigid body refinement and XYZ refinement in PHENIX without manual refinements . After refinement at the higher resolution , the resulting model was used to calculate R and Rfree values at the immediate lower resolution . If this resulted in a better model , as judged by Rfree , the higher resolution data are useful for improving the model . Data at resolutions that improved the model are highlighted in green while data at resolutions that worsened the model are highlighted in red . DOI: http://dx . doi . org/10 . 7554/eLife . 20391 . 006Refine at: 3 . 33 . 23 . 13 . 0Calculate at: 3 . 326 . 26/31 . 6625 . 30/31 . 473 . 226 . 75/31 . 8826 . 79/31 . 803 . 127 . 71/32 . 4727 . 64/32 . 873 . 028 . 30/32 . 5210 . 7554/eLife . 20391 . 007Table 2 . Data collection and refinement statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 20391 . 007Human TREM2 ectodomainData collection Space groupP 64 2 2Cell dimensions a , b , c ( Å ) 125 . 76 , 125 . 76 , 183 . 70 α , β , γ ( ° ) 90 , 90 , 120Resolution ( Å ) 50 . 00–3 . 10 ( 3 . 21–3 . 10 ) Rsym0 . 11 ( 1 . 00 ) Mean I / σI 21 . 0 ( 1 . 76 ) Completeness ( % ) 99 . 88 ( 99 . 75 ) Redundancy12 . 9 ( 13 . 4 ) Refinement Resolution ( Å ) 50 . 00–3 . 10 No . reflections16 , 285Rwork / Rfree0 . 2605/0 . 2736 No . atoms1 , 784 Protein1 , 756 Carbohydrate28B-factors93 . 48 Protein92 . 39 Carbohydrate162 . 1R . m . s . deviations Bond lengths ( Å ) 0 . 006 Bond angles ( ° ) 1 . 18Values in parentheses are for highest-resolution shell . We analyzed the structure to map the location of disease-linked point mutations and noticed an intriguing pattern . The side chains of mutations causing NHD ( Y38 , T66 and V126 ) are all buried within the core of the Ig fold , whereas the side chains of verified AD risk variants ( R47H and R62H ) and possible AD risk variants ( N68K , D87N and T96K ) all lie on the protein surface ( Figure 1b ) . Quantification using solvent-accessible surface calculations verified this qualitative observation ( Figure 1c ) . Using structural analysis , we can hypothesize how the buried NHD mutations could negatively impact protein folding . Y38 is adjacent to C36 , which forms an intramolecular disulfide with C110 , so the Y38C mutation most likely disrupts correct disulfide formation ( Figure 1—figure supplement 2b ) . T66 immediately follows the C’ β-sheet and is tightly packed inside the core of the protein , with the side chain hydroxyl engaging the backbone amide of K48 ( Figure 1—figure supplement 2c ) . The T66M mutation would sterically disrupt this packing and probably destabilize the protein . V126 is located on β-sheet G; it is entirely buried and contributes to a hydrophobic core of the buried residues F24 , D104 , A105 , and Y108 . Removal of the side chain by the V126G mutation would disrupt this packing and probably destabilize this hydrophobic core ( Figure 1—figure supplement 2d ) . In addition , sequence analysis reveals that Y38 and V126 are conserved within the TREM family , implying that they are probably required to preserve the common fold within this family of receptors ( Figure 5c ) . The T66 residue is less conserved in the TREM family , but is highly conserved in TREM2 proteins across species . In stark contrast to the NHD mutants , all of the residues implicated in the development of AD are surface exposed ( Figure 1b , c , e–g ) and engage in very few structure-stabilizing contacts . R47 , the residue with the strongest link to AD as the R47H variant , is at the end of the loop preceding β-sheet C . It is well-ordered in the structure with the side chain lying parallel along the surface of the protein , and the side-chain amines engage the carbonyl oxygen of T66 ( Figure 1e , Figure 1—figure supplement 2d , and Figure 1—figure supplement 2a ) . R62 , another strong AD risk factor as the R62H variant , comes after the loop marked by the C51-C60 disulfide bond ( Figure 1f and Figure 1—figure supplement 2e ) . Its side chain faces outward and makes no polar contacts . N68 is also surface exposed and its side chain makes no obvious polar contacts within the molecule . D87 is at the end of β-strand D; its side chain extends from the protein surface and the side-chain carboxyl engages the backbone amide of G90 . This interaction probably stabilizes the βD-E loop . T96 is on the end of β-strand E and its side chain points away from the surface and engages the side chain of Q33 ( Figure 1g and Figure 1—figure supplement 2f ) . On the basis of this structural analysis of neurodegenerative disease mutations in TREM2 , we hypothesized that NHD mutants ( which are buried ) would affect protein folding and stability and thus decrease their surface expression , whereas AD risk variants ( which lie on the protein surface ) would not affect surface expression and instead probably impact ligand binding . To begin to test tour hypothesis , we expressed each TREM2 ectodomain variant and analyzed their secretion and solution properties using size exclusion chromatography . WT hTREM2 ectodomain consistently eluted as a monomer ( Figure 2a and b ) , in accordance with our crystallographic analysis . The verified ( R47H and R62H ) and possible ( N68K , D87N and T96K ) AD-risk variants all elute at the same volume as WT TREM2 ( Figure 2—figure supplement 1g–l ) , suggesting that these mutations do not drastically alter the folding or oligomerization of TREM2 in solution . Furthermore , all of the TREM2 AD risk variant ectodomains migrated as monomers in the absence of reducing agent ( Figure 2b ) . By contrast , although the buried NHD mutants Y38C , T66M , and V126G were secreted , they elute much earlier , consistent with aggregation and misfolding of these proteins ( Figure 2—figure supplement 1m–o ) . Consistent with this observation , SDS-PAGE analysis of TREM2 NHD mutants in the absence of reducing agent showed that these ectodomains were largely produced as covalent dimers or trimers , indicating that misfolding of these mutants promoted the formation of aberrant intermolecular disulfide bonds ( Figure 2c ) . 10 . 7554/eLife . 20391 . 008Figure 2 . Chromatographic and surface expression analysis of WT and variant TREM2 . The schematic depicts the domain of TREM2 used in the experiments shown in ( a–c ) . ( a ) Gel filtration chromatography profile ( Superdex 200 10/300 GL ) of purified WT human TREM2 ectodomain showing a single monomeric peak . ( b ) SDS-PAGE analysis of purified TREM2 AD risk variants prepared with ( + ) and without ( - ) reducing agent ( β-mercaptoethanol ) . ( c ) Western blot analysis of TREM2 NHD mutants secreted from transfected 293F cells prepared with ( + ) and without ( - ) reducing agent ( β-mercaptoethanol ) . The NHD mutants Y38C , T66M , and V126G all migrate as higher MW oligomers under non-reduced conditions , indicating that they are misfolded and linked by aberrant intermolecular disulfide bonds . Purified WT TREM2 is shown for comparison . Representative of more than three independent expressions . The second schematic depicts FLAG TREM2 used in the experiments shown in ( d–h ) . ( d–e ) Surface expression of WT and variant TREM2 in 293F cells assayed by flow cytometry . Full-length FLAG-WT or mutant TREM2 were co-transfected with mDAP12 into 293F cells . Surface expression was measured by either ( d ) a polyclonal serum or ( e ) an anti-FLAG antibody . Normalized TREM2 indicates the normalized fraction of cells staining positive over DAP12-only background . Data are from four ( d ) or five ( e ) independent experiments . Bars are color-coded in the same way as the residues in Figure 1b ( cyan = NHD mutant; magenta = AD risk variant ) . Error bars are SEM . Significance was determined by ANOVA with Bonferroni post-test correction . ( p<0 . 05 , p<0 . 01 , and *p<0 . 001 ) . ( f ) Western blot analysis of whole-cell lysates of 293F cells used in ( d ) and ( e ) showing expression levels of WT , NHD mutant ( Y38C , T66M and V126G ) , and AD risk variant ( R47H , R62H , N68K , D87N and T96K ) TREM2 . Note that the NHD mutants are more highly expressed than the AD risk variants . ( g ) Confocal microscopy of 293F cells co-transfected with DAP12 and FLAG-TREM2 . Cells were fixed and either permeabilized ( left ) or non-permeabilized ( right ) and stained with anti-FLAG antibody ( red ) . ( h ) FLAG-TREM2 full-length constructs were co-transfected with DAP12 into 293T cells and expression analyzed by anti-FLAG immunoblot . Samples were prepared by suspending cells in reducing SDS loading buffer and boiling for no more than 5 min . We observe SDS-resistant aggregate bands for the NHD variants , which were largely absent in WT and AD risk variants . The T96K variant shows some light aggregation , consistent with the slight shift ( ~5°C ) in denaturation temperature for that variant ( Figure 3 ) . Schematic depicts FLAG TREM2 signaling assay employed in the experiments shown in ( i–j ) . ( i ) TREM2 signaling analyzed by phosphor-ERK1/2 and ERK1/2 immunoblot in RAW264 . 7 macrophage cells transfected with WT or variant FLAG-TREM2 . ( RAW264 . 7 cells express endogenous DAP12 . ) 24 hr post-transfection , cells were stimulated with anti-FLAG antibody ( 1:100 ) for the indicated length of time . ERK/pERK content was assessed by immunoblot . In these cells , we observe pERK at 10 min post-stimulation . However , only WT sustained signaling 30 min post-stimulation . ( j ) TREM2 signaling analyzed in Cos-7 cells as in ( c ) . Cos-7 cells were co-transfected with DAP12 and TREM2 . In these cells , there is a non-specific antibody response at 10 min . However , as in the RAW264 . 7 cells , we observe sustained WT signaling at the later time point which is diminished in R47H and Y38C variants . DOI: http://dx . doi . org/10 . 7554/eLife . 20391 . 00810 . 7554/eLife . 20391 . 009Figure 2—figure supplement 1 . Analysis of WT and variant TREM2 ectodomains by size exclusion chromatography . ( a-f ) Preparative SEC chromatograms of WT and validated ( R47H and R62H ) and possible ( N68K , D87N and T96K ) AD risk variant hTREM2 ectodomains harvested from 293F suspension cells . Fractions were pooled as shown in ( a ) for SDS-PAGE analysis ( r ) . Fraction volumes corresponding to 1 ) aggregate; , 2 ) oligomer , and 3 ) monomer TREM2 are indicated . Asterisks ( ) in ( a ) indicate elution volumes of MW standards . Fraction 3 contains purified hTREM2 ectodomains that elute as a monomer . This fraction was concentrated and re-run . ( g–l ) Analytical SEC chromatograms of purified WT and validated and possible AD risk variant hTREM2 ectodomains showing that they elute as monomers without detectable aggregation . ( m–o ) Preparative SEC chromatograms of hTREM2 NHD mutant ectodomains harvested from 293F suspension cells . These mutants do not yield folded TREM2 protein and largely elute as aggregates and oligomers . ( p ) SDS-PAGE analysis of WT and validated and possible AD risk variant TREM2 ectodomain preparative traces . Fractions 1 , 2 , and 3 are indicated as denoted in ( a ) . TREM2 NHD mutant ectodomains were analyzed by anti-6His western blots because they have minimal yields of purified protein that are not readily detectable by SDS-PAGE staining . The AD risk variants elute similarly to WT , while NHD mutants elute as aggregates/oligomers . DOI: http://dx . doi . org/10 . 7554/eLife . 20391 . 00910 . 7554/eLife . 20391 . 010Figure 2—figure supplement 2 . hTREM2 surface expression probed using a commercial antibody . ( a ) hTREM2 surface expression was probed by flow cytometry using a commercial anti-TREM2 antibody ( R&D ) . The results are comparable to results obtained using our anti-TREM2 polysera ( see Figure 2d ) . In this experiment , anti-FLAG staining was observed for all variants as in Figure 2e ( not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20391 . 010 Given that our structural analysis suggested that the TREM2 NHD mutants should be misfolded , we were surprised to observe that these ectodomains bypassed cellular quality control and were secreted . We therefore devised an assay using flow cytometry to investigate whether these point mutants in full-length TREM2 impair cell surface expression . DAP12 was co-expressed with FLAG-tagged WT or mutant TREM2 full-length proteins and surface expression was measured either by an anti-FLAG antibody or by anti-TREM2 sera raised against our secreted , folded protein . Whole-cell expression levels of the TREM2 NHD mutants were slightly higher than those of WT or AD risk variant TREM2 ( Figure 2f ) . However , as predicted by our structural analysis , anti-TREM2 staining was decreased for the NHD mutants Y38C , T66M , and V126G , but not for the AD risk variants R47H , R62H , N68K , D87N , and T96K ( Figure 2d ) . Similar observations were made using a commercial anti-TREM2 antibody ( R&D , Figure 2—figure supplement 2 ) . By contrast , the anti-FLAG antibody showed surface expression for all TREM2 mutants ( Figure 2e ) . We next examined cells expressing TREM2 WT , NHD mutant Y38C , and AD risk variant R47H by immuofluorescence confocal microscopy in the presence or absence of permeabilization . We found that the surface expression pattern in non-permeabilized cells was similar for all cell types , whereas in permeabilized cells , the NHD mutant Y38C displayed diffuse intracellular accumulation , in comparison to WT and R47H which displayed a distinct punctate staining pattern ( Figure 2g ) . Taken together , these results suggest that NHD mutant proteins do retain some level of surface expression; but the protein is misfolded or aggregated and thus is not recognized by conformation-specific TREM2 antibodies . Consistent with these observations , we found significant amounts of SDS-resistant aggregates by anti-FLAG immunoblotting when expressing the full-length TREM2 NHD mutants in 293T cells ( Figure 2h ) . Finally , we evaluated how these different classes of mutations impact TREM2 signaling by crosslinking the different versions of FLAG-TREM2 using anti-FLAG antibody . Consistent with our observation of FLAG-TREM2 on the cell surface for all the variant proteins , we found that while both the Y38C and R47H variants have some signaling potential , as assessed by phosphorylated-ERK blotting in RAW264 . 7 and Cos-7 cells , neither of these variants sustains signaling as efficiently as WT TREM2 ( Figure 2i , j ) . The loss of sustained signaling by the disease variants may suggest that these mutant proteins have aberrant folding kinetics ( and thus are replenished more slowly at the surface than WT proteins ) or cannot bind co-factors that may be required to sustain signaling . Altogether , these experiments demonstrate that the buried residue changes linked to NHD cause misfolding and aggregation , with variable impact on surface expression , as would be suggested by our structural analysis . We next sought to evaluate whether the AD-linked surface mutations affect protein structure or stability using sensitive solution techniques . First , we used circular dichroism ( CD ) spectroscopy to analyze whether AD-linked variants induced large conformational changes in the TREM2 ectodomain . For these experiments , we chose to analyze the R47H and R62H mutations , as they represent the most significant TREM2 risk factors identified to date , as well as the T96K mutation , which is a sporadically occurring mutation that is not significantly associated with AD . In order to evaluate whether the point mutations induced conformational changes , we collected CD spectra on the proteins , analyzed for large changes , and compared the ratio in minima at 214 nm ( due to β sheets ) to minima at 233 nm ( due to tryptophans ) , which are characteristic features of the CD spectra for Ig folds ( Sikkink and Ramirez-Alvarado , 2008 ) . We found that R62H and T96K displayed CD spectra similar to WT proteins while , surprisingly , R47H showed a subtle , yet statistically significant difference in the 214 nm/233 nm ratio ( Figure 3a and b ) . Thus the R47H mutation in TREM2 seems to induce a small , but measurable , conformational change , whereas the R62H and T96K mutations do not induce any structural changes detectable by CD . 10 . 7554/eLife . 20391 . 011Figure 3 . Structure and stability analysis by CD and DSF of human WT and AD-linked surface variant TREM2 ectodomains . ( a ) CD spectra of human WT TREM2 ( blue ) and R47H ( red ) , R62H ( brown ) , and T96K ( green ) variants . ( b ) Ratio of CD minima at 214 nm and 233 nm reveals that R47H has an altered CD spectrum . Minima ratios were measured on five ( WT and R47H ) or three ( R62H and T96K ) independent protein preparations , respectively . ( c ) Thermal melt temperatures measured by DSF . ( d ) WT and variant TREM2 thermal denaturation measured by CD at 225 nm while increasing the temperature from 20°C to 90°C with or without 1 mM DTT . DOI: http://dx . doi . org/10 . 7554/eLife . 20391 . 01110 . 7554/eLife . 20391 . 012Figure 3—figure supplement 1 . Structure and stability analysis by CD and DSF of human WT and AD-linked surface variant TREM2 ectodomains . ( a ) WT TREM2 melt curves measured in 5°C intervals . TREM2 shows a denaturation transition at 225 nm ( black line ) . ( b ) Table of denaturation temperatures measured by CD or DSF and chemical denaturation concentrations of guanidine hydrochloride ( GuHCl ) measured by CD . CD denaturation experiments were performed on at least three independent protein preparations . DSF experiments were done in triplicate on at least two independent protein preparations . All error bars are SEM . Significance was determined by ANOVA with Bonferroni post-test correction . ( **p<0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20391 . 012 We next investigated the impact of these mutations on thermal stability using differential scanning fluorimetry ( DSF ) ( Niesen et al . , 2007 ) . Each surface variant produced slightly lower denaturation temperatures , with R47H and R62H marginally reduced when compared to WT proteins and T96K having the lowest denaturation temperatures ( Figure 3c ) . As a control , the same experiments were carried out in the presence of 1 mM DTT , which induced the expected shift to lower denaturation temperatures by disrupting the disulfide bonds . We also used thermal denaturation CD to assess the thermal stability of these proteins . By obtaining full CD scans at 5°C increments , we observed that TREM2 has a denaturation transition at 225 nm , consistent with other Ig domains ( Sikkink and Ramirez-Alvarado , 2008 ) , and this wavelength was monitored in subsequent denaturation experiments ( Figure 3—figure supplement 1a ) . WT TREM2 and all the TREM2 variants tested showed reversible thermal denaturation as they reacquired their native spectra rapidly after cooling from 90°C to 20°C , but their ability to refold was ablated in the presence of DTT ( not shown ) . In contrast to DSF measurements , thermal denaturation by CD did not show a difference between WT , R47H , and R62H proteins; however , the T96K variant has a pronounced shift towards lower denaturation temperature ( Figure 3d ) . In the presence of DTT , all proteins had lower denaturation temperatures ( Figure 3d ) . We further assayed stability using chemical denaturation by titration of guanidine-HCl ( GuHCl ) in CD experiments . The R47H and R62H proteins melt at slightly lower GuHCl concentrations than the WT protein , while the T96K variant unfolds at lower GuHCl concentrations than R47H and R62H , consistent with the thermal melt experiments ( Figure 3—figure supplement 1b ) . Thus , as predicted by our structural analysis , the AD-risk surface variants do not induce large conformational changes in TREM2 , nor do they dramatically impact protein stability , but they can induce subtle changes in both parameters . Our structural analysis suggested that AD risk variants might impact function by altering ligand binding due to the surface presentation of the mutated residues . Recently , TREM2 was shown to signal following stimulation of reporter cells by plated phospholipids , and the R47H mutation resulted in loss of signaling ( Wang et al . , 2015 ) . To test the direct binding of TREM2 to phospholipids in a cell-free system , we used purified protein in solid-state ELISA and liposome sedimentation assays . Consistent with the results of previous work ( Cannon et al . , 2012 ) , we detected TREM2 binding to phospholipids by both ELISA and liposome sedimentation . However , we did not observe differences in phospholipid discrimination or in direct lipid binding between WT and AD variant TREM2 ectodomains ( Figure 4—figure supplement 1a–c ) . We therefore investigated binding to a cell surface ligand ( TREM2-L ) . Although no endogenous protein ligand for TREM2 has yet been identified , several cell types have been reported to express a ligand on the basis of staining with a TREM2-Fc fusion construct ( Ito and Hamerman , 2012; Hamerman et al . , 2006; Daws et al . , 2003; Hsieh et al . , 2009; Stefano et al . , 2009 ) . In order to evaluate the effect of AD-linked mutations on TREM2 binding to TREM2-L , we constructed a novel TREM2 protein cell-binding reagent containing a site-specific BirA-biotinylation sequence on the C-terminus of TREM2 , which could then be complexed with PE-labeled streptavidin to create a tetrameric cell-staining reagent ( Figure 4a ) . In initial experiments , we found that both human and mouse TREM2/SA-PE tetramers were able to bind Neuro2A ( N2A ) and THP-1 cells ( Figure 4b–d , f , g and Figure 4—figure supplement 2a , c ) . To validate the specificity of our reagent , we cultured THP-1 cells with various immune stimuli and found that overnight treatment with PMA/ionomycin dramatically decreased binding of TREM2/SA-PE tetramers to THP-1 cells ( Figure 4b and c ) , while the myeloid cell marker CD45 remained unchanged ( not shown ) . Other stimuli tested ( overnight exposure to IL-13 , Poly I:C , LPS , or M-CSF ) did not alter staining ( Figure 4—figure supplement 2a ) . PMA/ionomycin treatment is the first identified stimulus that ablates TREM2 staining , and this confirms that TREM2 recognizes a specific cell-surface ligand . We next sought to interrogate the endogenous target recognized by TREM2 on the cell surface , and found that TREM2 binding is sensitive to proteinase treatment of cells prior to incubation with our staining reagent ( Figure 4d and Figure 4—figure supplement 2b ) . Treatment with proteinase K abolished binding while the more specific proteases chymotrypsin and elastase had an intermediate effect . Proteinase K treatment did not affect cell viability , which was >99% by 7-aminoactinomycin D ( 7-AAD ) staining ( Figure 4—figure supplement 2f ) . This demonstrates that TREM2-L contains a protein component . In addition , given the high pI of TREM2 , and the frequent reports of anionic potential ligands for TREM2 , we reasoned that proteoglycans containing highly sulfated glycosaminoglycans ( GAGs ) may facilitate cell-surface binding . We tested binding to wild-type CHO-K1 cells , which express heparan and chondroitin sulfates , and to CHO-745 cells , which are deficient in GAG maturation and surface expression ( Esko et al . , 1985 ) . In this experiment , GAG-dependent cell binding of the TREM2 tetramer was observed ( Figure 4e ) . Next , we used heparinases ( which also cleave heparan sulfates ) and chondroitinase ABC to ask whether TREM2 selectively binds either type of GAG . In CHO ( Figure 4e ) , THP-1 ( Figure 4f ) and N2A ( Figure 4g ) cells , treatment with heparinases had a pronounced effect on cell surface binding while chondroitinase only slightly diminished binding , together implicating GAGs , specifically heparin sulfate , as a major component of the cell surface TREM2-L . 10 . 7554/eLife . 20391 . 013Figure 4 . AD risk variants in TREM2 alter binding to cell surface TREM2-L . ( a ) Schematic outlining the flow cytometry experiments in ( b–g ) . TREM2 ectodomains were specifically biotinylated at the C-terminus and probed for binding to TREM2-L on the surface cells , which was reduced by pre-treating cells with proteinases or heparinases . ( b–c ) Flow cytometry analysis of either ( b ) mTREM2/SA-PE tetramer or ( c ) hTREM2/SA-PE tetramer staining of THP-1 cells ± PMA/ionomycin treatment . ( d ) Flow cytometry analysis of hTREM2/SA-PE tetramer staining of THP-1 cells pretreated by various proteases . ( e ) mTREM2 staining of CHO-745 ( GAGless ) and CHO-K1 cells . CHO-K1 cells were treated with a cocktail of heparinases , chondroitinase ABC , or both . ( f–g ) mTREM2 staining of ( f ) THP-1 and ( g ) N2A cells treated with heparinases , chondroitinases , or both . ( h ) Schematic outline of control cell-binding experiments with the TREM-like proteins TREML2 and TREML4 . On the left is a schematic of receptor versions of TREM2 , TREML2 , and TREML4 , along with calculated pI values for the respective ectodomains . ( i ) Tetramer staining of TREM family ectodomains to N2A cells . ( j ) Schematic outlining monomeric TREM2-6His cell-staining experiments . Cells were stained with monomeric TREM2-6H and detected using PE-labeled anti-6H antibody . ( k ) Representative plot of THP-1 staining by hTREM2-6H analyzed by flow cytometry . ( l ) MFI of anti-6His staining of SEC-purified hTREM2 WT and variant ectodomains pre-incubated with THP-1 cells at the indicated concentrations . All data are representative of at least two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 20391 . 01310 . 7554/eLife . 20391 . 014Figure 4—figure supplement 1 . TREM2 lipid binding as assessed by phospholipid ELISA and liposome sedimentation . ( a ) Mouse TREM2 phospholipid binding measured by streptavidin-HRP detection of biotinylated mTREM2 ectodomain binding to plated phospholipids . Mean and SEM of three independent experiments done in triplicate . Each value was calculated by normalizing the lipid signal to the MeOH-only background for each protein . ( b ) Human TREM2 ectodomain phospholipid-binding ELISA . Mean and SEM are one of two representative experiments done in triplicate . ( c ) Immunoblot analysis of liposome pellets shows equivalent binding by disease-variants and preferential binding of anionic phospholipids . Representative of three independent experiments . ( d ) Alignment of the ectodomains of hTREM2 , T-cell immunoglobulin and mucin domain-containing protein 4 ( TIM4 , UniProt Q96H15 ) , TIM1 ( UniProt Q96D42 ) and CD300a ( UniProt Q9UGN4 ) . Conserved and invariant residues labeled as in Figure 5 . The black box denotes the ‘WFND’ motif required for phospholipid binding by TIM and CD300 ectodomains . ( e ) Human TREM2 and TIM4-phosphatydilserine crystal structures ( pdb: 3BIB ) superimposed , with the key lipid-binding motif residues shown as black sticks . Note that the conserved TREM2 D119 is facing away from the putative lipid-binding pocket , whereas the TIM4 D100 faces inward towards the pocket to coordinate a Ca2+ ion . ( f ) mTREM2-lipid binding was assessed by ELISA in the presence of 5 mM Mg2+ , Ca2+ , or EDTA . Representative of two independent experiments with both mouse and human TREM2 . DOI: http://dx . doi . org/10 . 7554/eLife . 20391 . 01410 . 7554/eLife . 20391 . 015Figure 4—figure supplement 2 . TREM2 binding to mammalian cells . ( a ) Flow cytometry ( FC ) analysis of mouse TREM2 ectodomain/SA-PE tetramers binding to THP-1 cells following stimulation with different compounds . Only PMA/ionomycin reduces binding . ( b ) FC analysis of mouse TREM2/SA-PE tetramers binding to 293F cells , which is ablated by the pre-treatment of cells with proteinase K and reduced by chymotrypsin and elastase . ( c ) Flow cytometry analysis of WT , R47H , R62H , and T96K mTREM2/SA-PE tetramers binding Neuro2A cells ( N2A ) . ( d ) Flow cytometry analysis of WT , R47H , and T96K mTREM2/SA-PE binding THP-1 cells . ( e ) Competition binding to THP-1 cells . WT hTREM2/SA-PE tetramers were applied to THP-1 cells pre-incubated with SEC-purified hTREM2 ectodomains at decreasing concentrations . The highest concentration is 1 . 5 mg/mL , with two-fold dilution points . All staining results are representative of at least two independent experiments . ( f ) Cell viability analysis of proteinase-K-treated THP-1 cells . Forward and side scatter analysis of THP-1 cells selected for staining analysis . No dead cells were detected in this singlet population ( 7-AAD staining ) . Analysis of all events showed >95% viability . DOI: http://dx . doi . org/10 . 7554/eLife . 20391 . 015 Next , we tested the impact of TREM2 AD risk variants on TREM2-L binding in three different formats . For these experiments , we chose to compare WT TREM2 , the validated TREM2 AD risk variants ( R47H and R62H ) , and the possibly protective AD risk variant ( T96K ) . We utilized anti-6H detection of purified monomeric hTREM2 ectodomains bound to THP-1 cells ( Figure 4j , k , and l ) , tetramer staining of WT and variant mTREM2 binding to N2A and THP-1 cells ( Figure 4—figure supplement 2c , d ) , and competition assays in which SEC-purified monomers competed with SA-PE-labeled WT tetramers ( Figure 4—figure supplement 2e ) . Consistent with our hypothesis , the AD-linked variant R47H displayed markedly diminished binding to N2A and THP-1 cells . In stark contrast , the potentially AD-protective T96K variant significantly increased binding . The R62H variant reduced binding as measured in the anti-6H format , but the reduction in binding was less dramatic when measured by competition or tetramer staining , suggesting an intermediate decrease in affinity . Thus , as suggested by our structural analysis , the AD risk variant R47H TREM2 negatively impacts binding to cell surface TREM2-L , whereas the variant T96K results in increased cellular binding . These data , together with our direct phospholipid binding experiments , suggest that TREM2 AD-risk variants retain phospholipid binding , and instead impact binding to cell surface GAGs . As the R47H risk variant is able to bind phospholipids directly , but does not signal in cellular assays of TREM2 signaling triggered by plated phospholipids ( Wang et al . , 2015 ) , we suggest that TREM2 interaction with GAGs in cis is likely required to orient or cluster TREM2 to mediate signaling upon stimulation by phospholipids . Alternatively , phospholipids may function to orient TREM2 on the cell surface for proper presentation to GAGs and/or additional as of yet undefined protein surface receptors . Our cell surface TREM2-L binding experiments categorized the naturally occurring TREM2 mutations as loss-of-binding ( R47H and R62H ) and enhancement-of-binding ( T96K ) compared to WT , and showed that these mutations influence TREM2 interactions with cell surface GAGs . We returned to analysis of our crystal structure in order to interpret these observations . Upon mapping the electrostatic surface of hTREM2 , we noticed a large basic surface that was not present on the surface of other TREM family receptors for which coordinates are available ( mTREM1 , PDB 1 U9K , [Kelker et al . , 2004a]; hTREM1 , PDB 1SMO , [Kelker et al . , 2004b]; and TLT-1 , PDB 2FRG [Gattis et al . , 2006] ) ( Figure 5a , b and d–f ) . Furthermore , sequence analysis shows that the residues constituting this basic patch are highly conserved within TREM2 , but not within the rest of the TREM family , suggesting that this interface has evolved specifically toward a role in TREM2 function ( Figure 5c ) . To demonstrate this , we prepared staining tetramers for other TREM-like family members with similar overall high pI values calculated for the ectodomain ( Figure 4h ) . Consistent with our hypothesis , TREM2 ectodomain tetramers bound to N2A cells , while tetramers of ectodomains from TREM-like 2 ( TREML2 ) and TREM-like 4 ( TREML4 ) did not ( Figure 4i ) . This cluster of residues contains most of the AD-linked surface mutations ( including R47H and R62H ) . Additionally , the T96K mutation is located adjacent to this basic patch and would therefore extend it , providing a structural explanation for the gain of function observed with this variant ( Figure 5b ) . To test whether additional basic residues located within this surface influence TREM2-L binding , we selected two other basic residues that are conserved only in TREM2 ( R76 and R77 ) , and assayed the ability of these variants to bind TREM2-L using our flow cytometry assay . We found that both a R76D and a R77D mutation decreased binding to TREM2-L on THP-1 cells in a manner similar to the validated AD risk variants ( R47H and R62H ) ( Figure 5g ) . Together , the data indicate that this surface represents a conserved functional interface on TREM2 that participates in cell-surface ligand binding and that this surface is unique to TREM2 within the TREM family . 10 . 7554/eLife . 20391 . 016Figure 5 . AD risk variants reveal a functional surface on TREM2 . ( a ) Electrostatic surface of hTREM2 and ( b ) hTREM2 rotated 90° . The extended basic patch on hTREM2 is outlined with a dashed line . ( c ) Structure-based sequence alignment of human ( h ) and mouse ( m ) TREM family members . Secondary structure assigned using the DSSP server . Invariant residues are shown in magenta and conserved residues are shown in yellow . Disulfide bonds are numbered in green . NHD residues are highlighted with black boxes and AD residues in blue boxes . Basic patch residues are labeled with blue asterisks . ( d ) mTREM-1 ( 1 U9K ) , ( e ) hTREM-1 ( 1SMO ) , and ( f ) hTLT-1 ( 2FRG ) aligned with hTREM2 as in ( a ) . Electrostatic potential was plotted on the solvent-accessible surface using the AMBER force field and the PDB2PQR server within the APBS Pymol plugin . Scale is −6 . 0 kT/e ( blue ) to +6 kT/e ( red ) . ( g ) Flow cytometry analysis of WT , R76D , and R77D hTREM2/SA-PE binding to TREM2-L on THP-1 cells . Representative of two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 20391 . 016 Here we elucidate how distinct point mutations in TREM2 give rise to different neurodegenerative diseases via two separate loss-of-function mechanisms . Our studies indicate , as presaged by our crystal structure , that NHD-linked mutant residues are buried and impact protein folding and stability , while AD-linked variant residues are on the protein surface and do not diminish the stability or surface expression of the molecule but instead impact ligand engagement ( Figure 6a ) . Several variants have been reported for TREM2 , but some are very rare , so it is not clear whether they associate with AD risk . The structure-based studies presented here provide a framework with which one could make predictions as to the variant's impact on function . Our observation of detectable misfolded cell-surface TREM2 NHD mutant proteins was surprising , given that one would expect that the cellular quality-control machinery should degrade misfolded proteins and prevent their surface expression . In addition , previous reports noted decreased surface expression for these mutants ( Park et al . , 2015; Kleinberger et al . , 2014 ) . Here , we used an approach that probes surface expression using both non-conformational ( anti-FLAG ) and conformational ( anti-hTREM2 ) antibodies . We found that both the NHD-associated and AD-linked mutant proteins reach the cell surface , however the NHD-associated TREM2 mutants are probably not functional as they are either misfolded or aggregated , and thus are not recognized by the conformational-dependent antibodies . These results are supported by our analysis of the solution behavior of theTREM2 ectodomains of these mutant proteins . An alternative explanation could be that the FLAG tag or overexpression drives the surface expression of these mutated proteins; but we do not think that is the case since we found that the TREM2 ectodomains carrying these mutations ( which contain a different tag that is on the opposite terminus ) were also secreted . These findings are of potential therapeutic value , as the discovery of molecular chaperones that rescue the folding of and restore function to these mutants could be pursued for patient-specific NHD treatments . 10 . 7554/eLife . 20391 . 017Figure 6 . Models developed from the current data . ( a ) Current model for the role of TREM2 ectodomain point mutations in neurodegenerative diseases . WT TREM2 is surface expressed in complex with DAP12 and engages TREM2-L normally . AD risk variants do not impair surface expression , but do impair ligand binding . They occur heterozygously , so this would lead to overall reduced TREM2 function in individuals carrying these risk variants . NHD mutations cause misfolding , which leads to aggregation and impaired surface expression . These mutations occur homozygously , leading to no TREM2 function in individuals carrying these mutations . ( b ) Current model for the engagement of possible TREM2 ligands investigated here . Membrane-bound or soluble TREM2 ( sTREM2 ) produced by proteolytic cleavage likely engages GAGs via the basic patch and engages phospholipids ( PL ) via the hydrophobic patch . In addition , there is a protein component of TREM2-L . DOI: http://dx . doi . org/10 . 7554/eLife . 20391 . 01710 . 7554/eLife . 20391 . 018Figure 6—figure supplement 1 . hTREM2 has an extended hydrophobic surface . ( a ) hTREM2 oriented as in Figure 5a and ( b ) hTREM2 rotated 90° towards the viewer to show the hydrophobic ‘top’ of the protein . Protein surface shown with residues colored by hydrophobicity scoring according to the color_h Pymol script , where red is hydrophobic is and white is hydrophilic . DOI: http://dx . doi . org/10 . 7554/eLife . 20391 . 018 In contrast to the buried NHD mutant proteins , the AD risk variants lie on the surface of the protein and do not severely impact the protein folding or stability of TREM2 . However , it should be noted that the R47H mutation does induce a subtle change in secondary structure detectable by CD . The AD mutations probably alter TREM2 function by impairing ligand binding . Recent studies in AD mouse models that are deficient in TREM2 confirm that loss of TREM2 function contributes to classic AD pathology and suggest a crucial role for TREM2 in microglia function ( Jay et al . , 2015; Wang et al . , 2015; Ulrich et al . , 2014 ) . TREM2-deficient AD mice display fewer activated microglia , and the absence of TREM2 prevents microglia proliferation and promotes microglia apoptosis , which is correlated with increased accumulation of Aβ plaques ( Wang et al . , 2015; Jay et al . , 2015 ) . Microglia in TREM2-deficient AD mice display less activation and do not engulf Aβ plaques , which reduces the density of Aβ plaques and promotes diffuse Aβ structures , which , in turn , are more neurotoxic and contribute to the accumulation of classic AD pathology ( Wang et al . , 2016; Yuan et al . , 2016 ) . These models suggest a crucial role for TREM2 in microglia biology and in the turnover of Aβ . On the basis of the conserved nature of AD-linked basic surface residues and our analysis of cell-surface ligand binding for multiple cell lines , we favor the concept that AD risk variants lead to impaired binding of TREM2-L , which in turn perturbs microglia function . With this in mind , we investigated how AD risk variants in TREM2 affect binding to previously identified and newly hypothesized ligands based on the analysis of ourcrystal structure . The many diverse ligands reported for TREM2 may in fact suggest that TREM2 is capable of recognizing multiple ligands and that under different conditions , their respective affinities might tune signaling accordingly . For example , low-affinity ligands may induce a tonic inhibitory signal while ligation of a high-affinity ligand could result in an activating response . Indeed , anionic bacterial carbohydrates ( Daws et al . , 2003 ) , phospholipids ( Wang et al . , 2015 ) , myelin lipids ( Poliani et al . , 2015 ) , and even purified DNA ( Kawabori et al . , 2015 ) all activate TREM2 reporter cell lines . This is not surprising in light of our crystal structure , which reveals an extended basic patch on the surface of TREM2 that would be capable of forming electrostatic interactions with these anionic ligands . In contrast to highly selective phospholipid binding by other Ig domains ( Kobayashi et al . , 2007; Miyanishi et al . , 2007; Santiago et al . , 2007; Simhadri et al . , 2012 ) , purified TREM2 shows only broad discrimination for anionic ( PA , PG , PI , PS ) over more neutral ( PC , PE ) lipids ( Figure 4—figure supplement 1a , b and c ) . Structurally , Ig domains with selective lipid binding feature a canonical FG loop motif that contains large hydrophobic residues , absent in TREM2 , which mediate membrane insertion and acyl chain binding ( Tietjen et al . , 2014 ) ( Figure 4—figure supplement 1d , e ) . Moreover , while TREM2 does share a conserved aspartic acid residue from this motif , which chelates a required Ca2+ ion for the TIM and CD300a proteins , TREM2-lipid binding is not sensitive to EDTA , unlike TIM and CD300 ( Figure 4—figure supplement 1e and f ) . Thus , while we do observe direct binding of TREM2 to phospholipids , it does not appear that TREM2 binds phospholipids with high selectivity or using the same binding mode as other phospholipid-binding Ig domains . By contrast , it should be noted that there is a long and narrow hydrophobic patch on the end of TREM2 most distal from the plasma membrane ( Figure 6—figure supplement 1 ) that is the most likely lipid-binding surface . Interestingly this patch is adjacent to the basic patch , which could mediate interaction with anionic phospholipid head groups ( Figure 6b ) . These hypotheses will need to be tested in future studies using sensitive binding assays ( such as surface plasmon resonance ) and crystallographic studies of TREM2-phospholipid recognition . In contrast to recent studies employing reporter cells expressing TREM2 R47H ( Wang et al . , 2015 ) , we did not observe any large differences in binding to phospholipids for any of the validated TREM2 AD risk variants . In light of our experimental observations , we believe this is best explained by the hypothesis that another interaction ( such as TREM2 binding to GAGs , which we show here is impacted by these AD risk variants ) is required to mediate phospholipid signaling; thus in the context of cellular experiments , TREM2 R47H shows diminished signaling , while in direct lipid-binding assays it performs in the same way as WT TREM2 . This is potentially supported by a recent report that TREM2 binds to isolated PS , but does not bind apoptotic Jurkat T cells , which present PS on their cell surface but may also be deficient in GAGs ( Bailey et al . , 2015 ) . The results presented here suggest that the point mutations in TREM2 participate in neurodegenerative disease pathogenesis through distinct molecular mechanisms . We found that the validated AD risk variants in TREM2 impact binding to a cell-surface ligand ( TREM2-L ) . This ligand does appear to contain a protein component , as pre-treatment of cells with proteases decreases or prevents the binding of TREM2 to these cells . This interaction also appears to be highly GAG-mediated , as TREM2 cell-staining tetramers do not bind CHO 745 cells , which do not express GAGs . This interaction seems most dependent on heparan sulfate , as treatment with heparin sulfatases remove much of the observed binding . As the AD risk variants most notably impact binding to TREM2-L , we reason that this interaction is involved in the pathogenesis of AD . In support of this , we find that the mutations R47H and perhaps R62H deleteriously impact TREM2-L recognition . By contrast , T96K appears to be a potential gain-of-function mutation in our binding experiments . Interestingly , this mutation appears at a much higher frequency than R47H and R62H , and in some studies it has been reported to be enriched in European-American AD cases compared to controls; overall it is associated with a decreased odds ratio for the development of late-onset AD ( Figure 1c ) , again suggesting the TREM2-L binding function is coupled to AD risk . Altogether , the structural and functional analysis of these variants illuminates a surface epitope that is specific to TREM2 , that is not conserved among the other members of the TREM family , and that mediates binding to a cell-surface ligand . Additional point mutations localized to that surface also resulted in loss of binding , supporting this functional role . Our results suggest that this ligand may have a specific protein component in addition to GAGs . On that note , while this manuscript was in preparation , it was reported that TREM2 binds lipid-loaded ApoE and that this interaction facilitates uptake of Aβ loaded into lipoparticles by microglia ( Yeh et al . , 2016 ) . This report also suggests that TREM2 AD risk variants decrease binding to ApoE , but binding affinities were not quantitated . However , we do not suspect that ApoE is the protein component of cell surface TREM2-L because it is a soluble rather than a surface-bound protein . Our study demonstrates that TREM2 can specifically bind GAGs , utilizing the basic patch; other TREM-like proteins with the same overall calculated pI , but lacking the basic patch residues , do not bind GAGs . We also demonstrate that TREM2 AD risk variants affect binding to GAGs . Future studies will need to focus on how TREM2 binding to GAGs and ApoE are coordinated and how they impact TREM2 function in microglia . Our results illustrate two distinct paths to loss-of-function in TREM2 , so how might these different mechanisms result in either AD or NHD ? In attempting to explain this , it is important to take into account genetics-based dosage affects ( Figure 6a ) . In NHD , the mutations are homozygous and the TREM2 is either deleted or misfolded , leading to a complete loss of function , and more serious and early-onset neurodegenerative disease . In AD , the mutations are heterozygous and impair function ( but do not completely abolish it ) , leading to less severe and late onset neurodegenerative disease . Under this hypothesis , it should be possible that a heterozygous NHD mutant would give increased risk for AD . Consistent with this , there have been reports of rare AD cases containing heterozygous NHD variants , including Q33X , Y38C , and T66M ( Guerreiro et al . , 2013b ) . In addition , TREM2 may function not only as a receptor but also as a signaling molecule as it can be cleaved . We have recently shown that the soluble TREM2 may be a survival signal in bone-marrow macrophages ( Wu et al . , 2015 ) . Interestingly , we have observed that the absence of TREM2 in an AD mouse model prevents microglia proliferation and promotes microglia apoptosis , suggesting that soluble TREM2 could play a similar role in the CNS ( Wang et al . , 2015 ) . Accordingly , soluble TREM2 can be detected in the CNS , and we have shown that some AD risk variants , but not the NHD mutant proteins are readily detected in the CSF of patients ( Piccio et al . , 2016 ) . Because the AD risk variant proteins are expressed and stable enough to be soluble factors , they may retain some signaling ability and prevent more serious diseases . These hypotheses will require animal knock-in studies for full evaluation . Therefore , a comprehensive analysis of all TREM2 variants will be essential to understanding TREM2 function in both neurodegenerative diseases and other inflammatory diseases , and to design targeted therapies accordingly . Full-length wild-type ( WT ) human TREM2 as well as R47H , N68K , D87N , and T96K mutants in pMX-3p plasmid and mouse Trem2 cDNA were used as PCR templates . Mammalian cell expression constructs of human TREM2 wild-type , R47H , D87N , T96K and mouse TREM2 ectodomains were produced as previously described ( Kober et al . , 2014 , 2015 ) by subcloning from these templates into the pHLsec vector , which contains an optimized signal sequence and a C-terminal 6-histidine tag for purification . Similarly , primers were designed to amplify the full-length human TREM2 gene with an N-terminal FLAG peptide . Additional mutants ( human Y38C , R62H , T66M , R76D , R77D; mouse R47H , R62H , and T96K ) were generated in the pHsec constructs using either the QuikChange Lightning Site-Direct Mutagenesis Kit ( Agilent . Santa Clara , CA ) or the Q5 Mutagenesis Kit ( NEB ) ( primer sequences in Table 3 ) . WT and mutant human and mouse TREM2 constructs containing site-specific biotinylation sites were generated by EcoRI-KpnI restriction digest of the pHLsec inserts followed by ligation into the pHLAvitag3 vector , which encodes a C-terminal BirA biotin ligase biotinylation sequence followed by a 6-histidine tag . All constructs were verified by sequencing . 10 . 7554/eLife . 20391 . 019Table 3 . Primers used in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 20391 . 019hTREM2ectodomainGAAACCGGTCACAACACCACAGTGTTCCAGGGCCGGGGTACCCAGGGGGTCTGCCAGCACCTCCAChTREM2Y38CCTGCAGGTGTCTTGCCCCTGTGACTCCATGAAGCACTAGTGCTTCATGGAGTCACAGGGGCAAGACACTTGCAGhTREM2T66MTGCCAGCGTGTGGTCAGCATGCACAACTTGTGGCTGCGCAGCCACAAGTTGTGCATGCTGACCACACGCTGGCAhTREM2V126GTCAGGAAGGTCCTGGGGGAGGTGCTGGCAGATCTGCCAGCACCTCCCCCAGGACCTTCCTGAhTREM2R62H+CCCAGTCCAGCATGTGGTCAGCACCCTTCTCTCCCAGCTGGChTREM2R76D+GTCCTTCCTGGACAGGTGGAATGGGAGCAGCCACAAGTTGTGChTREM2R77D+CTTCCTGAGGGACTGGAATGGGAGCACAGGACAGCAGCCACAAGTTGFLAG-hTREM2GAAACCGGTGATTATAAAGATGATGATGATAAACACAACACCACAGTGTTCCAGGGCCGGGGTACCTCACGTGTCTCTCAGCCCTGGCAGmTREM2 ectodomainGAAACCGGTCTCAACACCACGGTGCTCGGGGTACCTTGGTCATCTAGAGGGTmTREM2R62H+CCCATGCCAGCATGTGGTGAGCACCCTCCTCACCCAGCTGCmTREM2R47HAAGCACTGGGGGAGACACAAGGCCTGGTGTCGGCCGACACCAGGCCTTGTGTCTCCCCCAGTGCTTmTREM2T96KCTTGCTGGAACCGTCACCATCAAGCTGAAGAACCTCCAAGCCGGTACCGGCTTGGAGGTTCTTCAGCTTGATGGTGACGGTTCCAGCAAGmDAP12CCGGAATTCGCCACCATGGGGGCTCTGGAGCCCTCCTGGCGGGGTACCTCATCTGTAATATTGCCTCTGTGTAll primers 5’−3’ with the coding direction primer listed first*Designed for Quikchange Lightning Mutagenesis+Designed for NEB Q5 mutagenesis The WT human TREM2 ectodomain was expressed in Freestyle 293F cells in the presence of 1 µg mL−1 kifunensine , deglycosylated with EndoHf , and purified for crystallization in a single-step using Ni-NTA resin as previously described ( Kober et al . , 2014 ) . Protein was concentrated to 10 mg ml−1 in buffer containing 20 mM HEPES pH 7 . 4 and 150 mM NaCl . Crystals were grown by hanging drop vapor diffusion by mixing 1:1 with well solution containing 100 mM HEPES 7 . 0 , 2 . 1 M NaCl , 0 . 2 M MgCl2 and 0 . 2 M NDSB-201 . Crystals were cryoprotected in mother liquor containing 20% ethylene glycol and flash frozen under a nitrogen stream at −160°C . Data were collected at the Advanced Photon Source , beamline 19-ID ( Argonne National Lab , Chicago ) . A molecular replacement solution was found with PHASER using mouse TREM-1 ectodomain ( 1 U9K ) ( Kelker et al . , 2004a ) as the probe , locating two molecules in the asymmetric unit ( ASU ) . Data were initially scaled and processed at 3 . 3 Å using HKL2000 ( Otwinowski and Minor , 1997 ) . The initial model was substantially improved by iterative rounds of manual rebuilding in COOT ( Emsley et al . , 2010 ) and refinement using Phenix ( Adams et al . , 2010 ) . Next , resolution was extended using the method described by Karplus and Diederichs ( Karplus and Diederichs , 2012 ) ( Table 1 ) . The 3 . 3 Å model was used as a molecular replacement probe on data scaled to 3 . 4 Å and 3 . 3 Å , and subsequently subjected to automated refinement in Phenix without manual intervention . The 3 . 3 Å solution was then used to calculate Rfree and Rwork at 3 . 4 Å . Rfree was lower at 3 . 4 Å using the model refined at 3 . 3 Å compared to the model refined at 3 . 4 Å , providing evidence that 3 . 3 Å data improved the model . Similarly , data were again extended to 3 . 2 Å and finally 3 . 1 Å . Data past 3 . 1 Å did not improve the model as determined by an increased Rfree calculated at 3 . 1 Å using the model refined on 3 . 0 Å data . Final model refinement occurred by iterative building and refinement in Phenix . NCS and secondary structure restraints were used during refinement , and TLS refinement of B-factors was applied in later rounds . The structure is complete with only one N-terminal residue and four C-terminal residues not visible in the electron density ( Figure 1b ) . Ramachandran statistics are 97 . 7% favored , 0% outliers , 2 . 3% allowed . Molprobity score was 1 . 66 ( 100th percentile ) and clashscore was 9 . 62 ( 96th percentile ) for the final model . Calculations on the final model were performed using Naccess ( Hubbard and Thornton , 1993 ) to measure the solvent accessibility of side chains and HBPLUS ( McDonald and Thornton , 1994 ) to identify hydrogen bonds . LigPlot+ ( Laskowski and Swindells , 2011 ) was used to analyze side-chain contacts . For structure-based alignment , amino acid sequences were aligned using Clustal Omega ( Sievers et al . , 2011 ) and residue conservation scored by ESPript ( Robert and Gouet , 2014 ) . All crystallographic and analysis software used were compiled and distributed by the SBGrid resource ( Morin et al . , 2013 ) and diffraction images were archived with the SB Data Grid ( Meyer et al . , 2016 ) . For protein expression , Freestyle 293F cells were cultured at 8% CO2 in serum-free 293Freestyle media supplemented with Glutamax and penicillin/streptomycin ( Pen/Strep , Gibco by ThermoFisher , Waltham , MA ) . Human monocyte THP-1 cells were cultured at 5% CO2 in RPMI media supplemented with 10% Fetal Bovine Serum ( FBS ) , 10 mM HEPES , 50 µM β-mercaptoethanol , L-glutamine and penicillin/streptomycin ( Pen/Strep ) . Mouse neuroblast N2A cells were cultured at 5% CO2 in MEM supplemented with 10% FBS , L-glutamine and Pen/Strep . CHO cells were cultured at 5% CO2 in Ham’s F12 supplemented with 10% FBS , L-glutamine and Pen/Strep . Protein expression for biophysical and functional studies was performed as described previously ( Kober et al . , 2014 ) . In brief , plasmid DNA was complexed at a ratio of 1:2 ( µg/µg ) with PEI-TMC25 to transfect Freestyle 293F suspension cells cultured in serum-free 293 Freestyle Media . One µg of plasmid DNA was used per 1 × 106 cells . Transfected cells were allowed to express protein for 72–96 hr . Supernatants were collected and protein was purified using Ni-NTA chromatography . The eluted protein was then further purified by gel filtration chromatography using an analytical s200 size-exclusion column ( GE ) run in a buffer containing 20 mM Tris pH 8 . 5 and 150 mM NaCl . Full-length FLAG-tagged hTREM2 WT or mutant genes in pHL vectors were co-transfected with mDAP12 in pcDNA3 . 1 vector into 293F cells using 293fectin . After 24 hr , cells were harvested and washed into FACs buffer ( 1% BSA in PBS ) . FLAG epitope expression was detected using a FITC-conjugated M2 anti-FLAG antibody ( 1:50 , Sigma-Aldrich . St . Louis , MO ) , and folded TREM2 surface expression was detected by staining with anti-TREM2 primary polysera ( 1:1000 , hamster ) followed by PE-anti-hamster secondary ( 1:200 eBioscience ) . Background was defined by cells transfected with mouse DAP12 only . Thermal stability was assessed by differential scanning fluorimetry ( DSF ) on protein purified by SEC . The Protein Thermal Shift kit ( Applied Biosystems ) was used according to the manufacturer’s instructions . Briefly , protein was concentrated to 0 . 5 mg mL−1 after buffer exchange into PBS . 5 µl of reaction buffer and 2 . 5 µl 8x fluorescent dye were added to 12 . 5 µL protein on ice . Melt-curve experiments were performed using Fast7500 qPCR machine ( ABI ) starting at 25°C and with continuous 1% ramp to 95°C ( roughly 1°C min−1 ) . The data were analyzed using Protein Thermal Shift software . Circular dichroism spectroscopy ( CD ) measurements were performed using a JASCO J-815 spectropolarimeter equipped with a Peltier temperature controller . Thermal denaturation experiments were carried out in 10 mM phosphate ( pH 7 . 0 ) and 150 mM NaF . For native and GuHCl-denaturation scans , a 1 mm path length cuvette was used and the protein concentration was 30 µM . For denaturation experiments , a 1 cm path length cuvette was used and the protein concentration was 3 µM . For reducing conditions , the buffer contained 1 mM DTT . Ellipticity was measured at 225 nm in 1°C steps from 20°C to 90°C at a rate of 1°C min−1 and melt-curve data were smoothed using JASCO software . For chemical stability experiments , purified human ectodomains were incubated with GuHCl ( Fluka 50933 ) at room temperature for at least 1 hr before measuring CD . Repeat measurements at later time points confirmed that equilibrium had been achieved . Lipid ELISA experiments were carried out essentially as described by Kobayashi et al . , ( 2007 ) . In brief , phospholipids were dissolved in methanol or methanol:chloroform as needed and diluted to 5 µg mL−1 in methanol . 100 µL was added to ELISA plates and allowed to air-dry . Wells were blocked with 3% BSA in PBS . Biotinylated WT or mutant TREM2 in 3% BSA-PBS were incubated overnight at 4°C . Plates were washed three times in PBS + 0 . 05% Tween 20 . Biotinylated TREM2 was detected by streptavidin-HRP ( R and D ) in 1% BSA-PBS for 2 hr at RT before final wash and developing with TMB Microwell Peroxidase Substrate ( KPL . Gaithersburg , MD ) . Absorbance was measured at 450 nm on a Gemini Plus plate reader ( Molecular Devices . Sunnyvale , CA ) . HEK293T cells were co-transfected with full-length FLAG-TREM2 and DAP12 for 24 hr before replating on glass slides . Cells were washed 2x with PBS , then fixed with 4% paraformaldehyde ( PFA ) in PBS for 5 min . Cells were then washed and blocked with animal-free blocker ( Vector Laboratories . Burlingame , CA ) for 1 hr at RT . For permeabilized cells , 0 . 1% triton x-100 was added to the blocking buffer and subsequent steps . Anti-FLAG antibody ( M2 , Sigma ) was added at 1:1000 overnight at 4°C . Cells were then washed twice , and anti-mouse secondary Alexa Fluor 555 conjugate ( Life Technologies ) was added at 1:200 for 1 hr . Cells were washed a final time and then mounted in VECTASHIELD H-1200 Mounting Medium with DAPI ( Vector Laboratories ) . Confocal microscopy was carried out using a Zeiss LSM 510 META Confocal Laser Scanning Microscope ( Carl Zeiss Microscopy , Thornwood , NY ) at 400x magnification . The images were acquired with LSM 4 . 2 software . Liposomes consisted of a base of 35:35:10 PtdCholine , PtdEthanolamine , and cholesterol with an additional 20% wt/wt of candidate phospholipids . Phospholipids were purchased from Sigma or Avanti dissolved in chloroform . Lipids were mixed in solvent-washed glass vials and solvent was evaporated under nitrogen stream . Lipids were resuspended in PBS , warmed at 37°C for 15 min followed by three freeze-thaw cycles on liquid N2 and 37°C H2O . For the sedimentation assay , 10 µg TREM2 proteins quantified by BCA were mixed with 100 µg liposomes and incubated for 1 hr at room temperature . Liposomes were sedimented by centrifugation at 16 , 800 x g at 4°C . The supernatant was removed and the pellet resuspended in SDS buffer for 6His immunoblot analysis . All statistics were calculated using GraphPad Prism5 .	Alzheimer’s disease is a neurodegenerative disease and the most common cause of dementia – characterized by memory loss and difficulties with thinking , problem solving and language – in the elderly . Individuals with rare mutations in the gene that encodes a protein called TREM2 have a substantial risk of developing Alzheimer’s disease in their mid-60s . A different set of mutations in the gene for TREM2 can cause a more severe degenerative brain disease known as Nasu-Hakola disease in much younger people . Proteins are made up of chains of building blocks called amino acids that need to fold into specific three-dimensional shapes to allow the protein to work properly . TREM2 is a signaling protein that is found on the surface of immune cells in the brain . Mutations causing Alzheimer’s and Nasu-Hakola disease result in the production of mutant TREM2 proteins that differ from the normal protein by only a single amino acid . It is not clear how different mutations affecting the same protein can give rise to two distinct neurodegenerative diseases . To address this question , Kober et al . used a range of techniques to study normal and mutant TREM2 proteins . First , a technique called X-ray crystallography – which makes it possible to construct three-dimensional models of proteins – revealed that the mutations responsible for Nasu-Hakola disease are buried deep within the core of the folded TREM2 protein . On the other hand , mutations associated with Alzheimer’s disease lie on the surface of the protein . Further experiments examined how these mutations alter the properties of TREM2 , revealing that mutations linked to Nasu-Hakola disease affect the ability of TREM2 to fold correctly and how stable its final shape is . This results in fewer TREM2 proteins being present on the surface of immune cells . In contrast , mutations associated with Alzheimer’s disease make it harder for TREM2 to bind to molecules known as glycosaminoglycans . The Alzheimer’s mutations affect a specific part of TREM2 that is not found in other closely related proteins . The findings of Kober et al . suggest that TREM2 binding to glycosaminoglycans is likely to be important in preventing Alzheimer’s disease . The next step following on from this work is to find out exactly how these interactions affect immune cells , which may aid the development of new therapies for this disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "neuroscience" ]	2016	Neurodegenerative disease mutations in TREM2 reveal a functional surface and distinct loss-of-function mechanisms
Trypanosomes are important disease agents of humans , livestock and cold-blooded species , including fish . The cellular morphology of trypanosomes is central to their motility , adaptation to the host’s environments and pathogenesis . However , visualizing the behaviour of trypanosomes resident in a live vertebrate host has remained unexplored . In this study , we describe an infection model of zebrafish ( Danio rerio ) with Trypanosoma carassii . By combining high spatio-temporal resolution microscopy with the transparency of live zebrafish , we describe in detail the swimming behaviour of trypanosomes in blood and tissues of a vertebrate host . Besides the conventional tumbling and directional swimming , T . carassii can change direction through a ‘whip-like’ motion or by swimming backward . Further , the posterior end can act as an anchoring site in vivo . To our knowledge , this is the first report of a vertebrate infection model that allows detailed imaging of trypanosome swimming behaviour in vivo in a natural host environment . Trypanosoma is a genus of flagellated protozoa , and most species proliferate in the blood and tissue fluids of their host . The best-known and most commonly studied trypanosomes are those that cause human diseases: Human African Trypanosomiasis caused by Trypanosoma brucei and American Trypanosomiasis caused by T . cruzi . However , trypanosomes infect animals from all vertebrate classes , including warm-blooded mammals and birds as well as cold-blooded amphibians , reptiles and fish ( Simpson et al . , 2006 ) . African trypanosomes , such as Trypanosoma brucei , are exclusively extracellular and are continuously exposed to the innate , humoral and cellular immune systems . T . brucei has become a textbook example of antigenic variation based on low frequency switches of the expressed variant surface glycoprotein ( VSG ) that dominates the cell surface and allows the population to escape the host adaptive immune system . In addition , individual cell survival is favoured by a process that cleans the cell surface of low levels of bound immunoglobulins through endocytosis and degradation ( Engstler et al . , 2007; Forlenza et al . , 2009 ) . Trypanosome motility is an integral part of this process as the hydrodynamic drag on antibody-VSG complexes , caused by the forward swimming motion of the cell , results in accumulation of the complexes at the posterior pole , close to the flagellar pocket , where endocytosis occurs ( Engstler et al . , 2007 ) . In addition , motility is essential for successful cell division , immune evasion and development in the host ( Broadhead et al . , 2006; Griffiths et al . , 2007; Ralston et al . , 2006; Shimogawa et al . , 2018 ) . In vitro studies using African trypanosomes have focused on the characterization of qualitative and quantitative parameters of trypanosome morphology and motility and emphasized the ability of trypanosome species to adapt to the various environments of their mammalian hosts ( Alizadehrad et al . , 2015; Bargul et al . , 2016; Heddergott et al . , 2012; Krüger and Engstler , 2015 ) . Thus far , detailed analysis of swimming behaviour , morphology and trypanosome-host cell interaction has been restricted to ex vivo , using conditions that best mimic the host environment , for example in blood taken from naïve or infected animals ( Bargul et al . , 2016; Beattie and Gull , 1997; Engstler et al . , 2007; Heddergott et al . , 2012; Hemphill and Ross , 1995; Shimogawa et al . , 2018; Skalický et al . , 2017; Sunter and Gull , 2016; Wakid and Bates , 2004 ) . It has not been feasible to recreate complex in vivo microenvironments , which , in the case of the vertebrate host , includes the streaming of the blood , vessels with heterogeneous size and endothelium composition , as well as changes occurring during the course of an infection , for example from the onset of anaemia . It still remains a challenge to mimic in vitro the dynamic conditions of the crowded and fast-moving bloodstream in which trypanosomes live . Given the importance of motility for trypanosome survival , analysis of trypanosome swimming behaviour in vivo , in a vertebrate host environment , is important to fully understand trypanosome biology and pathogenesis . Quantitative and qualitative approaches in vivo are hindered by the lack of transparency of mammalian vertebrate hosts . Nevertheless , using transgenic trypanosomes expressing a luciferase reporter protein it has been possible to monitor overall trypanosome distribution and to quantify total trypanosome load in mice ( Burrell-Saward et al . , 2015; Capewell et al . , 2016; Goyard et al . , 2014; McLatchie et al . , 2013 ) . Fluorescent trypanosomes have been used in vivo to visualize their presence in mouse skin , a possible reservoir of trypanosomes in the late phase of infection ( Capewell et al . , 2016 ) . Although the sensitivity of both systems allows for indirect visualization of trypanosome location or distribution , low spatio-temporal resolution and the lack of transparency of deep tissues limits the possibility for qualitative and quantitative analysis of trypanosome swimming behaviour in vivo . Finally , using a combination of multicolour light sheet fluorescence microscopy and high-speed fluorescence microscopy it has been possible to analyse the infection process and swimming behaviour of T . brucei ex vivo , in dissected tissues of the partially transparent tsetse fly vector ( Gibson and Peacock , 2019; Schuster et al . , 2017; Wang and Belosevic , 1994 ) . Nevertheless , to date , no method is available to study with sufficient resolution trypanosome swimming behaviour in vivo in a vertebrate host environment . In the current study , we have used Trypanosoma carassii infection of zebrafish , Danio rerio , to visualize trypanosome movement in the bloodstream and tissues of a vertebrate host . T . carassii infects a broad range of cyprinid fish ( Overath et al . , 1998 ) , is transmitted by blood-sucking leeches ( i . e . , Hemiclepsis marginata ) ( Lom and Dyková , 1992 ) , and lives extracellularly in the blood and tissue fluids of the fish ( Haag et al . , 1998; Lom and Dyková , 1992 ) . T . carassii can establish a long-term infection characterized by polyclonal B cell activation and recurrent waves of parasitaemia , without expression of a uniform VSG-like surface coat ( Agüero et al . , 2002; Joerink et al . , 2007; Overath et al . , 2001; Overath et al . , 1999 ) . Phylogenetically , T . carassii belongs to the aquatic clade of Trypanosoma , a sister group that diverged from the Trypanosomatid lineage prior to the divergence of the Stercorarian and Salivarian trypanosomes that infect mammals ( Simpson et al . , 2006; Stevens , 2008 ) . Morphologically , T . carassii isolated from fish has a single flagellum that emerges from the flagellar pocket at the cell posterior , is attached to the cell body and extends free at the anterior end , defining the anterior-posterior axis . Zebrafish have been used as a model for developmental biology , as well as biomedical and neurobiology research ( Asnani and Peterson , 2014; Blackburn and Langenau , 2014; Cronan and Tobin , 2014; Goessling and North , 2014; Miyares et al . , 2014; Nguyen-Chi et al . , 2014; Phillips and Westerfield , 2014; Renshaw and Trede , 2012; Schartl , 2014; Torraca et al . , 2014; Veinotte et al . , 2014; Zon and Cagan , 2014 ) , and are a fresh water cyprinid fish closely related to many of the natural hosts of T . carassii . The great advantage of zebrafish is the transparency of the larvae , and of some juvenile and adult stages . Furthermore , mutant and transgenic lines including those marking blood vessels and relevant immune cell lineages are available ( Benard et al . , 2015; Bertrand et al . , 2010; Ellett et al . , 2011; Langenau et al . , 2004; Lawson and Weinstein , 2002; Page et al . , 2013; Petrie-Hanson et al . , 2009; Renshaw et al . , 2006 ) . The transparency of the zebrafish allows high-resolution , real-time imaging of trypanosome movement and host-parasite interaction in vivo , under conditions that exactly represent the natural environment during infection . In this study , by combining T . carassii infection of transparent zebrafish with high-speed microscopy , we provide the first description of trypanosome swimming behaviour in vivo in a vertebrate host . Our observations reveal that trypanosomes can rapidly adapt their swimming behaviour to the heterogeneous host environments . It was not possible to assign one preferred swimming behaviour to trypanosomes in either the bloodstream , tissues or other body fluids . Conditions such as the presence or absence of the blood flow or of red blood cells , the speed of the flow , the size of the blood vessel , the type of endothelium or epithelium lining the vessels or the tissues , as well as the compactness of the tissues , all influenced the swimming behaviour . Furthermore , we show that trypanosomes can change direction through backwards swimming and through a ‘whip-like’ motion . Finally , we were able to capture a novel mechanism through which trypanosomes attach to host cells or tissues . These observations greatly expand our knowledge on trypanosome swimming behaviour and show that trypanosomes can rapidly adapt to match the host environment . Zebrafish were kept and handled according to the Zebrafish Book ( zfin . org ) and animal welfare regulations of The Netherlands . Adult zebrafish were reared at the aquatic research facility of Wageningen University and Research ( Carus ) . Zebrafish embryos and larvae were raised in egg water ( 0 . 6 g/L sea salt , Sera Marin , Heinsberg , Germany ) at 27°C with a 12:12 light-dark cycle . From 5 days post fertilization ( dpf ) until 14 dpf , larvae were fed once a day with Tetrahymena . Larvae older than 10 dpf were also fed daily with dry feed ZM-100 ( zmsystem , UK ) . The zebrafish lines used in this study included: wild type AB , optically transparent casper lines ( White et al . , 2008 ) , a transgenic Tg ( fli:egfp ) y1 line marking the vasculature ( Lawson and Weinstein , 2002 ) or crosses thereof . All animals were handled in accordance with good animal practice as defined by the European Union guidelines for handling of laboratory animals ( http://ec . europa . eu/environment/chemicals/lab_animals/home_en . htm ) . All animal work at Wageningen University was approved by the local experimental animal committee ( DEC number 2014095 ) . Trypanosoma carassii ( strain TsCc-NEM ) was previously cloned and characterized ( Overath et al . , 1998 ) and maintained in our laboratory by syringe passage through common carp ( Cyprinus carpio , R3xR8 strain; Irnazarow , 1995 ) . To this end , adult common carp were infected by intraperitoneal injection of 1 × 104 T . carassii; approximately 3 weeks post-infection and before the parasitaemia reached 1 × 106/mL , carp were euthanized in 0 . 6 g/L tricaine methane sulfonate ( TMS , Crescent Research Chemicals ) and bled via the caudal vein using a final concentration of 10–20 Units heparin/mL of blood . Trypanosomes in blood were imaged immediately or blood was kept at 4°C overnight in siliconized tubes . The following morning , trypanosomes were enriched at the interface between the red blood cells and plasma , and this buffy coat was recovered and centrifuged at 600xg for 8 min at room temperature . Trypanosomes were resuspended in RPMI without L-glutamine and phenol red ( Lonza , Verviers , Belgium ) . To separate trypanosomes from red blood cells , the suspension was loaded on top of a 100% Ficoll-Paque layer ( GE Healthcare , Uppsala , Sweden ) and centrifuged at room temperature for 20 min , at 800xg . Cells at the interphase were transferred to a new 50 mL tube and washed with RPMI . Trypanosomes were then resuspended at a density of 5 × 105–1×106/mL and cultured in 75 or 165 cm2 flasks in complete medium: 22 . 5% MEM with L-glutamine and phenol red ( Gibco ) , 22 . 5% Leibovitz’s L-15 medium without L-glutamine , with phenol red ( Lonza , Verviers , Belgium ) , 45% Hanks’ Balanced Salt solution ( HBSS ) with phenol red ( Lonza , Verviers , Belgium ) , 10% sterile water , completed with 10% pooled carp serum , 2% 200 mM L-glutamine ( Fisher Scientific ) and 1% penicillin-streptomycin solution ( 10 . 000:10 . 000 , Fisher Scientific ) . Cultures were incubated at 27°C without CO2 . Trypanosomes were kept at a density below 5 × 106/mL and subcultured one to three times a week . Using this medium , T . carassii was kept in culture without losing infectivity for up to 2 months . For zebrafish infection , trypanosomes were cultured for 1 week and never longer than 3 weeks . Cultured trypanosomes were centrifuged at 800xg for 5 min and resuspended in 2% polyvinylpyrrolidone ( PVP , Sigma-Aldrich ) prior to injection . PVP was used to increase the viscosity of the medium to ensure a homogenous trypanosome solution throughout the injection period . Trypanosome number in 1–2 nL drop size varied depending on the intended dose and was monitored at the beginning and after 50 injections using a Bürker counting chamber . Prior to injection , 5 dpf zebrafish larvae were anaesthetized with 0 . 017% ethyl 3-aminobenzoate methanesulfonate ( MS-222 , Tricaine , Sigma-Aldrich ) in egg water , and injected intravenously . Experimental groups received either T . carassii resuspended in PVP solution or PVP solution alone as a negative control . Injected larvae were directly transferred into pre-warmed egg water and kept in the incubator at 27°C . Viability was monitored daily . During the course of the optimization of the infection model , we noticed that some fish displayed lethargic behaviour and no escape reflex to a pipette , such fish usually had a high parasitaemia leading to death . These clinical signs were therefore used to monitor morbidity and the progression of infection . When necessary , fish were removed from the experiment and euthanized with an overdose of anaesthetic ( 0 . 4% MS-222 ) . At various time points after infection , three to six zebrafish larvae were sacrificed by an overdose of anaesthetic , pooled and transferred to RNA later ( Ambion ) . Total RNA isolation was performed with the Qiagen RNeasy Micro Kit ( QIAgen , Venlo , The Netherlands ) according to the manufacturer’s protocol . Next , 250–500 ng of total RNA was used as a template for cDNA synthesis using SuperScript III Reverse Transcriptase and random hexamers ( Invitrogen , Carlsbad , CA , USA ) , following the manufacturer’s instructions with an additional DNase step using DNase I Amplification Grade ( Invitrogen , Carlsbad , CA , USA ) . cDNA was then diluted 25 times to serve as template for real-time quantitative PCR ( RT-qPCR ) using Rotor-Gene 6000 ( Corbett Research , QIAgen ) , as previously described ( Forlenza et al . , 2012; Forlenza et al . , 2008 ) . Primers for zebrafish elongation factor-1α ( ef1a Fw: 3’-CTGGAGGCCAGCTCAAACAT-5’ and RV: 3’-ATCAAGAAGAGTAGTAGTACCG-5’; ZDB-GENE-990415–52 ) and T . carassii heat-shock protein-70 ( FW: 3’-CAGCCGGTGGAGCGCGT-5’ 3’-AGTTCCTTGCCGCCGAAGA-5’; GeneBank-FJ970030 . 1 ) were obtained from Eurogentec ( Liège , Belgium ) . Gene expression was normalized to the ef1a housekeeping gene and expressed relative to the time point PVP control . For imaging of T . carassii swimming behaviour in vitro , a high-speed camera was mounted on an automated DM6b upright digital microscope ( Leica Microsystems ) , controlled by Leica LASX software ( version 3 . 4 . 2 . ) and equipped with 100x oil ( NA 1 . 32 ) , 40x ( NA 0 . 85 , DIC ) and 20x ( NA 0 . 8 , DIC ) short distance objectives ( Leica Microsystems ) . For high-speed light microscopy a ( 12 bits ) Photron APX-RS High Speed Camera ( Photron , resolution ( 128 × 16 ) to ( 1024 × 1024 ) pixels ) , with Leica HC 1x Microscope C-mount Adapter was used , controlled by Photron FASTCAM Viewer ( PFV ) software ( version 3 . 5 . 1 ) . Images were acquired at a resolution of 900 × 900 or 768 × 880 pixels depending on the C-mount adapter . Trypanosomes were transferred to non-coated microscopic slides ( Superforst , Thermo Scientific ) , covered with a 24 × 50 mm coverslip and imaged immediately and for no longer than 10 min . Prior to imaging of T . carassii swimming behaviour in vivo , the high-speed camera was mounted on a DMi8 inverted digital microscope ( Leica Microsystems ) , controlled by Leica LASX software ( version 3 . 4 . 2 . ) and equipped with 40x ( NA 0 . 6 ) and 20x ( NA 0 . 4 ) long distance objectives ( Leica Microsystems ) . For high-speed light microscopy a ( 8 bits ) EoSens MC1362 High Speed Camera ( Mikrotron GmbH , resolution 1280 × 1024 pixels ) , with Leica HC 1x Microscope C-mount Camera Adapter was used , controlled by XCAP-Std software ( version 3 . 8 , EPIX inc ) . Images were acquired at a resolution of 900 × 900 or 640 × 640 pixels . Zebrafish larvae were anaesthetized with 0 . 017% MS-222 and embedded in UltraPure LMP Agarose ( Invitrogen ) on a microscope slide ( 1 . 4–1 . 6 mm ) with a well depth of 0 . 5–0 . 8 mm ( Electron Microscopy Sciences ) . Upon solidification of the agarose , the specimen was covered with three to four drops of egg water containing 0 . 017% MS-222 , by a 24 × 50 mm coverslip and imaged immediately . For all high-speed videos , image series were acquired at 240–500 frames per second ( fps ) and analysed using PFV software ( version 3 . 2 . 8 . 2 ) or MiDAS Player v5 . 0 . 0 . 3 ( Xcite , USA ) ; 240–250 fps were found optimal for imaging of trypanosome swimming behaviour in vitro , whereas 480–500 fps were used for in vivo imaging of infected zebrafish . Quantification of trypanosome length , swimming speed and directionality was performed with ImageJ-Fijii ( version 1 . 51 n ) using the MTrack plug-in . For livestream light microscopy ( acquisitions at 20 fps ) a DFC3000G camera ( Leica Microsystems ) was mounted on the DMi8 inverted digital microscope and controlled by the Leica LASX software . Images were acquired at a resolution of 720 × 576 or 1296 × 966 pixels . For fluorescence microscopy of Tg ( fli1:egfp ) y1 casper lines , marking the zebrafish vasculature in green , the Zeiss LSM-510 confocal microscope , with a 20x long-distance objective was used with the following settings: laser excitation = 488 nm with 73% transmission; HFT filter = 488 nm; BP filter = 505–550; detection gain = 800; amplifier offset = −0 . 01; amplifier gain = 1 . 1; bright field channel was opened with Detection Gain = 130; frame size ( pixels ) = 2048×2048; pinhole = 300 ( optical slice <28 . 3 μm , pinhole ø = 6 . 26 airy units ) . Videos were produced using CyberLink PowerDirector 16 . Previous in vitro studies reported on the heterogeneity in swimming behaviour of African trypanosomes , and on how this was dependent on the viscosity of the culture medium or host blood ( Bargul et al . , 2016; Engstler et al . , 2007; Heddergott et al . , 2012 ) . To investigate the swimming behaviour of T . carassii in fish blood or culture medium , we used high spatio-temporal resolution microscopy . For the initial description of trypanosome swimming behaviour in vitro , we adopted the classification and quantification method described previously ( Bargul et al . , 2016 ) : persistent swimmers , trypanosomes exhibiting a directional movement covering several hundreds of micrometres; tumblers , trypanosomes exhibiting a non-directional movement and travelling no further than their body length; and intermediate swimmers , trypanosomes alternating periods of directional and non-directional movement . Analysis of the swimming behaviour of trypanosomes in blood of infected carp revealed that up to 96 . 4% of T . carassii could be classified as tumblers ( Figure 1A–B , and Video 1 ( 00:00 - 45:19 s ) ) . The remaining trypanosomes ( 3 . 5% ) behaved as intermediate swimmers ( Video 1 , 00:00 - 45:19 s ) , and only 0 . 1% could be classified as persistent swimmers ( Figure 1A–B and Video 1 , 45:19 - 58:03 s ) . Persistent swimmers showed an average speed of 32 μm/s and could cover a ‘straight-line’ distance of up to 418 μm in 20 s , whereas intermediate swimmers showed a lower average speed of 14 μm/s ( Figure 1B ) . Although tumblers did not move any distance greater than their body length ( ~23 ± 2 . 4 μm , Figure 1C ) , they were still very mobile . Taking advantage of the spatio-temporal resolution of high-speed videography , the speed of displacement of the posterior end of tumblers was 14 μm/s , similar to that of intermediate swimmers ( Figure 1B ) . Comparison of the movement of freshly isolated trypanosomes kept in either carp serum or culture medium revealed comparable swimming behaviour ( Video 1 , 00:58 s - 1:15 min ) . Culture in medium or serum for a period of up to 2 months did not alter morphology , the proportions of tumblers , intermediate or persistent swimmers ( data not shown ) , suggesting that the swimming behaviour is an intrinsic property of trypanosomes in blood . In freshly drawn blood from infected carp , trypanosomes were observed anchored to red blood cells or to leukocytes . Attachment always occurred through their posterior end leaving the flagellum free to move ( Figure 2 and Video 2 , 00 - 28 s ) . Similarly , also when cultured , trypanosomes could attach to the flask or glass surface through their posterior end ( Video 2 , 28 - 55 s ) . Whether the site of attachment coincided with the cell membrane , flagellum base , or neck of the flagellar pocket was not readily clarified with the current image resolution . Finally , trypanosomes were also observed to alternate between forward and backward swimming ( Video 2 , 00:55 s - 1:15 min ) . To observe the swimming behaviour of T . carassii in a host , we developed an infection model in transparent zebrafish larvae . First , the susceptibility and kinetics of parasitaemia were determined . Infection of 5 day-post-fertilization ( dpf ) zebrafish resulted in an acute infection associated with low survival ( Figure 3A ) and high parasitaemia ( Figure 3B ) , independent of the infection dose . This confirms that zebrafish , similarly to other cyprinid fish , are susceptible to T . carassii infection . Having established that T . carassii can infect zebrafish , we took advantage of the transparency of zebrafish larvae to characterize trypanosome swimming behaviour in vivo . Zebrafish larvae were infected at 5 dpf with 200 T . carassii per fish and imaged using either live stream imaging ( 20 fps ) or high spatio-temporal resolution microscopy ( 500 fps ) at various time points after infection and in differently sized blood vessels . When parasitaemia was low , typically early during infection , trypanosomes were most readily detected in small to medium-sized blood vessels with reduced blood flow and a lower density of red blood cells , such as the tail tip loop or intersegmental capillaries ( ISCs ) . In the tail tip loop , trypanosomes were dragged passively by the bloodstream along with red blood cells ( Figure 4A , Video 3 , 0 - 42 s ) and were seen to either curl or stretch the cell body as well as occasionally propel their flagellum in the same or opposite direction to the blood flow ( Video 3 , 00:42 s - 01:57 min ) , but were never seen swimming faster than the flow . ISCs are narrow , with a diameter equivalent to a single red blood cell . In ISCs , trypanosomes were elongated with their flagellum in the opposite direction to the blood flow ( Figure 4B , Video 3 , 01:57 - 02:49 min ) ; the diameter of the vessel , the speed of the flow and the presence of colliding red blood cells within the ISC , force the trypanosomes to passively move forward in the direction of the flow . In larger diameter vessels with a strong blood flow and higher density of red blood cells , such as the cardinal caudal vein ( Video 3 , 02:49 - 04:11 min ) or artery ( Video 3 , 04:11 - 04:37 min ) , detection and description of swimming behaviour was greatly aided by the use of high spatio-temporal resolution microscopy . In these vessels as well , trypanosomes are dragged passively by the bloodstream , curling among the densely packed red blood cells . In real-time speed , only the occasional trypanosome was seen to slow down by rolling or bouncing against the vessel in the peripheral cell-free layer ( Figure 4C and Video 3 , 03:53 - 04:11 min ) . In general , the typical tumbling movement described in vitro was not observed in the fish , except for those locations where the blood flow was highly reduced or absent as in sharp turns of the tail tip ( Video 3 , 04:37 - 05:09 min ) , and in locations where leukocytes adhering to the endothelium would create a local disturbance of the flow rate allowing trypanosomes to tumble ( Video 3 , 05:09 - 05:24 min ) . Physiological changes in the blood flow or red blood cells density associated with the infection , allowed visualization of additional swimming behaviours in blood vessels . For example , in cases where the blood flow was temporarily interrupted , trypanosomes were able to swim directionally , repeatedly invert direction , or tumble ( Figure 5A , and Video 4 , 0 - 01:23 min ) . In cases where the blood flow continued but red blood cells were occluded , trypanosomes were able to persistently swim against the blood flow ( Figure 5B , and Video 4 , 01:23 - 02:13 min ) . We did not attempt to distinguish , in vivo , between intermediate and persistent swimmers because physical factors such as the presence of red blood cells , the length of a capillary or the stability of the blood flow could all interfere with the directionality of their movement . Instead , all trypanosomes observed to swim directionally in the blood , independent of the distance covered before changing direction , were considered swimmers . Altogether , we observed that T . carassii can adopt different swimming behaviours all greatly influenced by the blood flow , size of the blood vessel and presence of red blood cells . Most frequently , in blood vessels with an intact blood flow and high density of red blood cells , trypanosomes are dragged passively by the flow along with red blood cells , in a curling motion . Occasionally , when the blood flow or the number of red blood cells is reduced , trypanosomes can swim directionally and persistently ( swimmers ) in blood vessels . This indicates that , at least in vivo , it is not possible to assign a specific swimming behaviour to trypanosomes in blood vessels; on the contrary , trypanosomes rapidly adapt their swimming behaviour to changes in microenvironmental conditions . In addition to being dragged passively within blood vessels , trypanosomes could often be seen attached to the endothelium on the dorsal luminal side of the cardinal caudal vein ( referred to as caudal vein , Figure 6A ) . Remarkably , despite the strong blood flow , attachment ( anchoring ) could last for several seconds ( Video 5 , 0 - 01:11 min ) and involved a small area of the posterior end of the trypanosome , leaving the cell body and the flagellum free to move ( Video 5 , 01:11 - 01:59 min ) . Although anchoring could be observed already at 1 dpi , it was more easily detected at later stages of the infection . Anchoring was not the only mode of attachment , trypanosomes were also seen crawling along the vessel wall involving the entire cell body ( Figure 6B , and Video 5 , 01:59 - 02:38 min ) . Within blood vessels , anchoring occurred exclusively at the dorsal side of the caudal vein , whereas crawling could occur anywhere in the vein . No attachment or crawling was observed in arteries or capillaries , independently of the speed of the blood flow . Altogether , our observations show that T . carassii attaches to host cells through their posterior end , both in vitro ( Video 2 ) and in vivo , leaving the flagellum free to move , and suggest that the posterior end acts as an anchoring site . Whether the exact anchoring site corresponds to the flagellum base or to the neck of the flagellar pocket and whether it may possibly favour extravasation , could not be confirmed under the current conditions and will be the focus of further investigation . A characteristic of T . carassii infections is extravasation from blood vessels into surrounding tissues and tissue fluids ( Haag et al . , 1998; Lom and Dyková , 1992 ) . In zebrafish , this was observed at 1 dpi and allowed us to investigate the swimming behaviour in tissue fluids other than blood , including those of the peritoneal and heart cavities . In these locations , we could thus evaluate the swimming behaviour of trypanosomes in the absence of red blood cells and of blood flow . In these environments , the swimming behaviour was similar to that observed in vitro ( Video 1 and Figure 1 ) , where the majority of the trypanosomes were tumblers ( Figure 7 and Video 6 , 0 – 42 s ) . In a field of view of more than 100 trypanosomes in the peritoneal cavity , only three persistent swimmers could be identified ( Figure 7B and Video 6 , 00:42 s - 01:35 min ) . Furthermore , trypanosomes were seen anchored by their posterior end to cells of the peritoneal membrane in a manner similar to that observed in the caudal vein ( Video 6 , 34 - 42 sec ) . Next , we analysed the swimming behaviour of T . carassii in tissues: the tail tip , fins , muscle and interstitial space lining the blood vessels . Here , we observed no apparent consistency in swimming behaviour , and trypanosomes alternated between directional and non-directional swimming depending on the compactness of the tissue . For example , in the compact tissue of the fins , most trypanosomes were directional swimmers ( Figure 8A–B ) , although their path can often be interrupted or hindered by the compactness of the tissue . Swimmers moved at an average speed of 47 . 5 μm/s , covering up to 187 μm , before disappearing from view or colliding with an obstacle that resulted in a change in direction ( Video 7 , 0 – 47 s ) . Similar to the observations made in vitro ( Video 2 ) , trypanosomes could invert their swimming direction by swimming backwards ( Figure 8A–B , Video 7 , 00:47 s - 01:07 min ) . In vivo , backward swimming was only observed in the fins . In the interstitial space lining the cardinal blood vessels ( artery and vein ) , trypanosomes swim directionally or tumble ( Video 8 , 0 - 23 s ) , but can also use the space between cells to pin themselves and effectively invert swimming direction in a ‘whip-like’ motion , a movement distinct from the more random tumbling movement . ( Video 8 , 00:23 s - 01:17 min ) . Such ‘whip-like’ movement was also observed for trypanosomes swimming in ISC in which the blood flow is absent , and in more compact tissues such as the fin ( Figure 8C , Video 8 , 01:17 - 02:18 min ) . The ‘whip-like’ motion combines the swing of the flagellum along one plane , similar to the movement of tumblers on a glass surface ( Video 1 ) , accompanied by a 180°C rotation of the cell body along a third axis . This is indeed possible only in vivo where the cylindrical form of a capillary or interstitial space within a tissue allow the very flexible trypanosome cell body to move in three dimensions . Furthermore , in compact tissues that do not present a ready passage for trypanosomes , the persistent swimming translates into a drilling ( auger ) movement , which in some cases can lead to an enlargement of the space between somatic cells ( Figure 8D , Video 8 , 02:18 - 03:52 min ) . Altogether , in tissues and tissue fluids T . carassii can adopt all swimming movements and can adhere through the posterior end to endothelial cells of the peritoneal cavity . Besides the previously described tumbling and directional ( forward ) swimming , trypanosomes were also able to invert direction through a ‘whip-like’ motion or by backward swimming . Physiological changes associated with the progression of the infection can affect the conditions within blood vessels or host tissues , and thus influence trypanosome behaviour . In addition to extravasation ( Videos 6–8 ) , which occurred as early as 1 dpi , we observed onset of anaemia and vasodilation of blood vessels . Non-infected larvae have a strong and steady blood flow , with all blood vessels packed with red blood cells ( Figure 9A ) . In contrast , infected larvae progressively showed a decrease in the ratio between red blood cells and trypanosomes from 1 to 8 dpi ( Figure 9B–C ) , until in some cases red blood cells disappeared completely ( Figure 9D and Video 9 ) . Anaemia , therefore , is a hallmark of late stages of T . carassii infection in zebrafish larvae . Highly infected fish that are anaemic also showed vasodilation , a clinical sign typical of advanced stages of infection ( >3 dpi ) with T . carassii ( Figure 10A–C ) , most obviously observed in the caudal vein . The degree of vasodilation differed between individuals , and in extreme cases , the diameter of the caudal vein could be up to three times larger than that of control fish ( Figure 10D ) . Vasodilation also occurs in the caudal artery but to a lesser extent ( not shown ) . Interestingly , while the dilated blood vessels of larvae are packed with trypanosomes and have limited circulation , the number of extravasated trypanosomes is very low ( Video 9 ) . In this study we describe a trypanosome infection model in zebrafish . By combining the transparency of zebrafish larvae with high spatio-temporal resolution microscopy , we were able to describe in detail the in vivo swimming behaviour of T . carassii in blood , tissues and tissue fluids of a vertebrate host . In addition to non-directional tumbling and directional forward swimming , we also describe how in vivo trypanosomes can reverse direction through a ‘whip-like’ motion or by swimming backwards . Finally , we report a novel observation that the posterior end of T . carassii , possibly coinciding with the flagellum base or flagellar pocket’s neck , can act as an anchoring site . To our knowledge , this is the first report of the swimming behaviour of trypanosomes in vivo in a vertebrate host environment . Knowledge of trypanosome-host cell interaction , movement , and tropism in vertebrate hosts is important to understand trypanosome biology and pathology . To date , detailed analysis of trypanosome swimming behaviour and interaction with vertebrate host cells has only been possible using isolated blood or conditions that best mimic those of the host blood or tissues . Although the current in vitro approaches have brought a wealth of information on the quantitative aspects of trypanosome motility and potential immune evasion strategies , they could not fully reproduce the streaming nature of the blood , the heterogeneity of the bloodstream between and within vessels , the different sizes of blood vessels between which trypanosomes regularly alternate , the different types of endothelium lining arteries and veins and , finally , the changes that occur in the blood caused by the infection itself . In this study , we first investigated the motion of T . carassii in infected carp blood . Based on previously proposed descriptions of the swimming behaviour of salivarian trypanosomes in mouse blood ( Bargul et al . , 2016; Shimogawa et al . , 2018 ) , clearly more than 90% of T . carassii could be classified as tumblers , and this was independent of whether they were in whole blood , diluted in serum or culture medium . Our observations are in agreement with a previous study reporting that directional persistent swimming was not a prominent feature of T . brucei in whole blood from an immunocompetent infected mouse ( Shimogawa et al . , 2018 ) , but are in contrast to the study by Bargul and colleagues in which up to 30% persistent swimmers and 45% intermediate swimmers could be observed in blood films from immunosuppressed infected mice ( Bargul et al . , 2016 ) . Differences in T . brucei strains , host immune status as well as the use of whole blood or blood films may account for the observed discrepancies . Nevertheless , both studies could not reproduce the streaming of the blood and , thus , the observations were made under static conditions . Our initial in vitro observations led to the suggestion that the tumbling behaviour might be an intrinsic property of T . carassii in the bloodstream . However , our subsequent in vivo observations showed that this applies only to trypanosomes swimming in fluids with highly reduced or absent flow , for example the peritoneal fluid or blood in vessels in which the flow was slow or absent ( Figure 7A and Video 9 ) . In contrast , analysis of the swimming behaviour of T . carassii in the zebrafish bloodstream revealed that it is not possible to assign a single or predominant swimming behaviour . In vessels with a normal blood flow and in the presence of red blood cells , trypanosomes were dragged passively by the flow along with red blood cells . In the centre of large vessels , such as the cardinal artery or cardinal vein , the velocity of the flow does not allow either tumbling or directional swimming against or faster than the flow . Similarly , in narrow iISCs , trypanosomes are forced to move forward because of streaming of the blood or the presence of colliding red blood cells . Under these conditions , trypanosomes were never seen swimming faster than the passive movement of red blood cells as previously suggested ( Heddergott et al . , 2012; Langousis and Hill , 2014 ) . Furthermore , because trypanosomes are carried along with red blood cells by the flow , it is difficult to envisage how the red blood cells could represent anything more than very occasional obstacles or surfaces for mechanical interactions that would favour forward swimming ( Bargul et al . , 2016; Heddergott et al . , 2012 ) . At the periphery of the cardinal vein , there is a cell-free layer where the number of red blood cells and the speed of the blood flow is reduced ( Bagchi , 2007 ) . Here , trypanosomes could clearly slow themselves down by crawling , rolling or by temporarily anchoring themselves to the dorsal endothelium ( further discussed below ) . However , in arteries or small capillaries , adherence to the epithelium was never observed , even when the flow was reduced or completely absent . This suggests that not only the velocity of the flow but also the type of endothelium influences trypanosome swimming behaviour . Physiological changes occur as the infection progresses that modify conditions within blood vessels . The flow is strongly reduced or interrupted and the density of red blood cells is reduced by obstruction of blood vessels or anaemia . In these conditions , directional swimmers were observed that swam faster than the flow and others that swam against the flow , or that repeatedly changed direction within blood vessels ( Video 4 ) . Taken together , our observations show that it is not possible to generalize the swimming behaviour of trypanosomes in blood as they can rapidly adopt different swimming behaviours: tumbling , directional swimming or anchoring . These behaviours are all largely influenced by the size and type of vessel , the speed of the flow , the presence , or not , of red blood cells as well as the highly dynamic microenvironment within a vessel , for example the centre compared to the periphery , and the type of endothelium . Outside of the blood , it has been suggested that the directionality of swimming is influenced by the density or compactness of the tissue ( Bargul et al . , 2016; Sun et al . , 2018; Wheeler et al . , 2013 ) . Our observation of swimming behaviour in various tissues revealed that it is the size of the interstitial space through which the trypanosome has to swim that largely determines whether it will swim directionally , tumble or both . For example , within the compact tissue of the fins , where the space below epithelial cells is very narrow ( mesenchyme ) and the basement membrane is not deformable ( Mateus et al . , 2012 ) , the vast majority of trypanosomes swim directionally . This observation is in agreement with the swimming behaviour described for procyclic forms of T . brucei swimming through microfluidic devices smaller than their maximum cell diameter , mimicking potential size-limiting environments within host tissues ( Sun et al . , 2018 ) . In zebrafish fins , the space can be so restrictive that to invert direction the trypanosome is obliged to swim backwards ( Video 7 ) . Backward swimming was previously observed in vitro or ex vivo for T . brucei motility mutants or for wild type trypanosomes swimming in high-density medium or in whole blood ( Bargul et al . , 2016; Baron et al . , 2007; Branche et al . , 2006; Engstler et al . , 2007; Heddergott et al . , 2012; Shimogawa et al . , 2018 ) , and ex vivo for procyclic and mesocyclic stages of T . brucei parasites swimming in confined spaces within the midgut of the tsetse fly ( Schuster et al . , 2017 ) . The observation of T . carassii swimming in zebrafish tissues is the first report of backward swimming in a vertebrate host environment . In the less compact areas of the fins or in the tissue lining the blood vessels , in addition to directional swimmers ( forward or backwards ) , tumblers and trypanosomes that repeatedly inverted direction through a whip-like motion were observed ( Video 8 ) . When directional swimmers reached dead ends , the persistent forward swimming translated into either a drilling movement similar to the movement described for T . brucei in dead-end spaces within the midgut of the tsetse fly ( Schuster et al . , 2017 ) , or backwards swimming as observed in the fins . Altogether , these observations indicate that trypanosomes can adopt several swimming behaviours and that these are largely influenced by the compactness and confinement offered by the tissue . Perhaps one of the most interesting observations is the discovery that trypanosomes can anchor themselves to the vein endothelium by their posterior end , leaving the flagellum and the entire cell body free to move . Anchoring was observed as soon as 30 min to 1 h after T . carassii injection into zebrafish larvae , and by 1 dpi extravasation was observed . Under the current conditions , it was not possible to ascertain whether this adhesion mechanism favours or is even required for extravasation . So far , we have been unable to capture the exact moment of extravasation . The anchoring site on the posterior of the trypanosome leaves the cell body and flagellum free to move ( Video 2 and Video 5 ) , so it seems likely that the adhesion on the trypanosome occurs via the flagellum base or the neck of the flagellar pocket itself . The presence of specific adhesion molecules would at least partly explain how T . carassii can very rapidly anchor themselves , upon sudden collision with the endothelium , and remain in position for more than 15 s despite the very rapid blood flow and collisions with blood cells . Anchoring was only observed in the cardinal vein but trypanosomes were seen crawling on vein endothelium ( Video 5 ) through dynamic interactions that involved the cell membrane , not just the flagellum membrane , suggesting that the molecules required for whole-cell adherence might be different from those required for anchoring through the posterior end . This is reminiscent of leukocyte rolling , which also occurs only in veins and not in arteries . Altogether , the possibility to observe T . carassii behaviour both in vitro and in vivo , in the presence or absence of a strong hydrodynamic flow , and at various locations within the vertebrate host , demonstrated how the environmental conditions , especially the presence or absence of a flow , strongly influence the ability of the trypanosome to attach and the duration of the attachment . T . carassii anchors to zebrafish cells in a manner clearly distinct from that described for other trypanosomes . The major difference being the lack of extensive interaction between the trypanosome’s flagellum membrane and the host cell or artificial surface . Stable interaction involving large portions of the flagellum membrane has been described among others for haptomonads stages ( surface-attached ) of Paratrypanosoma confusum , or Leishmania promastigotes ( Figure 11A ) ( Skalický et al . , 2017; Wakid and Bates , 2004 ) . In these liberform parasites ( flagellum not laterally attached to the cell ) , adhesion occurs through an attachment pad forming from the bulge at the base of the flagellum . At least in vitro , the formation of the pad takes approximately 1 h , causing extensive remodelling of the flagellum itself , and effectively anchors the parasites to the surface , and in the case of P . confusum , also favours their division ( Skalický et al . , 2017 ) . Similarly , T . brucei epimastigotes divide while attached to the brush border of the salivary gland epithelium through extensive outgrowths of the non-cell-attached anterior part of the flagellum membrane ( Figure 11B ) ( Beattie and Gull , 1997; Langousis and Hill , 2014; Schuster et al . , 2017 ) . Such an adhesion mechanism was only recently captured ex vivo through high-speed videography of dissected tsetse fly salivary glands; however , the exact moment of attachment and the time required to establish the stable interaction in vivo were not reported ( Schuster et al . , 2017 ) . T . congolense was reported to adhere in vitro to bovine aorta endothelial ( BAE ) cell monolayers via extensive membrane protrusions ( filopodia ) of the membrane-attached flagellum ( Figure 11C ) ( Beattie and Gull , 1997; Hemphill and Ross , 1995 ) , an interaction that was shown to involve sialic acid residues on BAE cells ( Hemphill et al . , 1994 ) . Although adhesion was observed already at 1 h , the filopodia increased in size over a period of 24–48 h . Whether such a type of interaction occurs with similar kinetics also in vivo , in vessels with an intact blood flow , is yet to be confirmed . All the above attachment mechanisms are clearly geared towards creating a very stable interaction with a surface to either establish a permanent infection in the salivary glands of the insect host , or to possibly adhere to the vertebrate host endothelium . They all involve extensive modifications of the flagellum membrane that occur over time to increase the contact area between the ( para ) trypanosomes and the surface . Most of these interactions , however , were observed in vitro for ( para ) trypanosomes cultivated on glass or plastic surfaces or endothelial cell monolayers and studied by means of scanning or transmission electron microscopy , as well as in vitro binding assays ( Beattie and Gull , 1997; Hemphill and Ross , 1995; Hemphill et al . , 1994; Skalický et al . , 2017; Vickerman , 1969; Wakid and Bates , 2004 ) . Therefore , to what extent the kinetics of interaction and the size of the contact area described also apply to the more dynamic in vivo environment is yet to be ascertained . The interactions between T . carassii and zebrafish cells were limited to the tip of the posterior end of the trypanosome body ( Figure 11D ) , leaving the entire flagellum and trypanosome cell body free to rapidly move ( Video 2 ) . Despite the small surface involved , anchoring occurred very rapidly , suggesting a very strong , yet dynamic , type of interaction , the duration of which was influenced by the presence or absence of a strong hydrodynamic flow as well as colliding red blood cells . Given the in vivo dynamic conditions within a blood vessel , it seems unlikely that T . carassii would establish stable interactions that involve large portions of the trypanosome cell surface , as for example described in vitro for T . congolense . We were unable to determine whether the anchoring area of T . carassii corresponds to the flagellum base , flagellar pocket neck or to the cell body membrane . Nevertheless , because T . carassii adherence to red blood cells and to glass surfaces was also observed in vitro , it will be possible to investigate the adhesion mechanism at high resolution in the future . In conclusion , we describe here for the first time the swimming behaviour of a trypanosome in vivo in the natural environment of a vertebrate host . We report the complex and heterogeneous environment in which the trypanosomes reside and how this highly influences the swimming behaviour . We describe how it is not possible to assign specific behaviours to trypanosomes swimming in any of the host compartments , as the trypanosomes were extremely effective in rapidly adapting their motion to the highly dynamic host environment . We describe backward swimming and whip-like movements that allowed trypanosomes to invert direction as well as identifiedying the posterior end as a novel anchoring site that allows the cell to adhere to host cells in a manner different from those described to date for other trypanosomes . Altogether , establishment of the T . carassii zebrafish infection model in combination with the genetic tractability of the zebrafish and of trypanosomes , represent a unique possibility to address questions related to: 1 ) trypanosome swimming behaviour in vivo in the natural environment of a vertebrate host , 2 ) host-pathogen interaction , 3 ) trypanosome biology , 4 ) the effect of specific immune factors on the progression of the infection , 5 ) the effect of drugs on both the trypanosome and the host , and 6 ) immune evasion strategies of trypanosomes that do not present antigenic variation . For all these reasons , the T . carassii-zebrafish model holds the promise to become a valuable complementary model to those currently available , to study the complex biology of trypanosomes and their interaction with the vertebrate host .	Trypanosomes are one-celled parasites that cause the disease trypanosomiasis , which is also known as sleeping sickness . Trypanosomiasis is transmitted to humans and animals by a type of fly , known as tse-tse , which is commonly found in sub-Saharan Africa . A bite from the tse-tse fly transfers the trypanosome cells into the host’s bloodstream , where they spread from the blood to the internal organs and brain . This leads to a long-term illness , which can sometimes result in a coma and eventually death . Once in the blood trypanosomes move around using a structure similar to an underwater propeller called the flagellum . How the trypanosomes move and behave in the blood determines how the infection will progress . Until now it has only been possible to observe trypanosomes in plastic dishes or in blood drawn from infected patients . However , neither of these settings mimic the conditions of the bloodstream , and it is currently impossible to look inside human hosts to watch how trypanosomes move . To overcome this hurdle , Doro et al . infected zebrafish with Trypanosoma carassii , a close relative of the sub-Saharan trypanosomes that specifically infects fish . Zebrafish are transparent when young , making it possible to observe the parasite in the blood and tissues of live fish using a microscope . Doro et al . noticed that Trypanosoma carassii cells adapt to different environments in the host by using different swimming techniques . For example , in small capillaries trypanosomes were dragged along with the blood flow , whilst in larger vessels , when blood flow was slow or there were fewer red blood cells , trypanosomes actively swam against the current . The parasites were also able to change direction by using their flagella in a ‘whip-like’ motion . Lastly , it was discovered that Trypanosoma carassii could rapidly attach to blood vessel walls using one end of its cell body , even when blood flow was strong . This behaviour may help the parasites escape from the bloodstream into the surrounding tissues , making the infection more dangerous . Studying how trypanosomes infect zebrafish at this high level of detail provides new insights into how these parasites move and behave inside a host . An important question that remains to be answered , is how exactly the trypanosomes leave the bloodstream . A better understanding of the whole infection process may hint at new ways of fighting these deadly infections in future .	[ "Abstract", "Introduction", "Materials", "and", "methods", "Results", "Discussion" ]	[ "microbiology", "and", "infectious", "disease" ]	2019	Visualizing trypanosomes in a vertebrate host reveals novel swimming behaviours, adaptations and attachment mechanisms
Dopamine neurons in the ventral tegmental area use glutamate as a cotransmitter . To elucidate the behavioral role of the cotransmission , we targeted the glutamate-recycling enzyme glutaminase ( gene Gls1 ) . In mice with a dopamine transporter ( Slc6a3 ) -driven conditional heterozygous ( cHET ) reduction of Gls1 in their dopamine neurons , dopamine neuron survival and transmission were unaffected , while glutamate cotransmission at phasic firing frequencies was reduced , enabling a selective focus on the cotransmission . The mice showed normal emotional and motor behaviors , and an unaffected response to acute amphetamine . Strikingly , amphetamine sensitization was reduced and latent inhibition potentiated . These behavioral effects , also seen in global GLS1 HETs with a schizophrenia resilience phenotype , were not seen in mice with an Emx1-driven forebrain reduction affecting most brain glutamatergic neurons . Thus , a reduction in dopamine neuron glutamate cotransmission appears to mediate significant components of the GLS1 HET schizophrenia resilience phenotype , and glutamate cotransmission appears to be important in attribution of motivational salience . Dopamine ( DA ) neurons regulate several aspects of motivated behaviors ( Bromberg-Martin et al . , 2010; Salamone and Correa , 2012; Schultz , 2013 ) , and are involved in the pathophysiology of neuropsychiatric disorders ranging from drug dependence to schizophrenia ( Robinson and Berridge , 2008; Winton-Brown et al . , 2014 ) . Like most CNS neurons , DA neurons release multiple neurotransmitters ( Trudeau et al . , 2014 ) . They release DA with both slower modulatory actions ( Tritsch and Sabatini , 2012 ) , as well as faster signaling actions ( Ford et al . , 2009; Chuhma et al . , 2014 ) . They variously release glutamate ( GLU ) ( Hnasko and Edwards , 2012 ) and GABA ( Tritsch et al . , 2016 ) as cotransmitters , conferring both greater dynamic signaling range and heterogeneity in their synaptic actions , as well as differential susceptibility to endogenous and exogenous modulation ( Chuhma et al . , 2017 ) . Discerning the behavioral role of DA neuron GLU cotransmission has been challenging ( Morales and Margolis , 2017 ) . DA neuron GLU cotransmission has a crucial neurodevelopmental role . The abrogation of GLU cotransmission via a DA transporter ( DAT ) -driven conditional knockout ( cKO ) of vesicular GLU transporter 2 ( VGLUT2 , encoded by Slc17a6 ) ( Hnasko et al . , 2010; Stuber et al . , 2010 ) impairs survival and axonal arborization of DA neurons in vitro , and compromises the development of the mesostriatal DA system in vivo leading to a 20% decrease in the number of DA neurons ( Fortin et al . , 2012 ) . GLU cotransmission also plays an important role in modulating DA release by enhancing packing of DA into vesicles ( Hnasko et al . , 2010 ) via vesicular synergy ( El Mestikawy et al . , 2011 ) . Functionally , DAT VGLUT2 cKO show about a 25% reduction in electrically-evoked DA release and about a 35% reduction in DA content in the nucleus accumbens ( NAc ) ( Hnasko et al . , 2010; Fortin et al . , 2012 ) . Behaviorally , DAT VGLUT2 cKO show modest deficits in emotional and motor behaviors ( Birgner et al . , 2010; Fortin et al . , 2012 ) , normal reinforcement learning drive by DA neuron activation but decreased response vigor ( Wang et al . , 2017 ) , a blunted response to psychostimulants ( Birgner et al . , 2010; Hnasko et al . , 2010 ) , and a paradoxical increase in sucrose and cocaine seeking ( Alsiö et al . , 2011 ) . Whether the behavioral phenotypes of DAT VGLUT2 cKO mice are due to the impact of the VGLUT2 deficit on DA neuron development , DA transmission , or GLU synaptic actions is not clear . Phasic activity of DA neurons projecting to the NAc encodes the incentive salience of reward-predicting cues and invigorates cue-induced motivated behaviors ( Bromberg-Martin et al . , 2010; Flagel et al . , 2011 ) . At the synaptic level in the striatum , DA neurons make the strongest GLU connections in the NAc shell to cholinergic interneurons ( ChIs ) ( Chuhma et al . , 2014; Mingote et al . , 2015 ) . When DA neurons are driven at burst firing frequencies — mimicking their in vivo phasic firing — their GLU postsynaptic actions drive synchronized burst-pause sequences in ChIs ( Chuhma et al . , 2014 ) that are likely to be important in salience encoding . Dysregulated DA neuron firing is thought to disrupt salience processing leading to the development of psychotic symptoms ( Kapur , 2003; Winton-Brown et al . , 2014 ) . The hyperdopaminergic state associated with positive symptoms of schizophrenia is modeled in rodents by amphetamine sensitization ( Peleg-Raibstein et al . , 2008 ) , which enhances the motivational salience of drug-associated stimuli ( Robinson et al . , 2016 ) . Interestingly , amphetamine sensitization as well as gestational MAM treatment , a validated rodent model of schizophrenia , selectively enhance activity of VTA neurons projecting to NAc shell ( Lodge and Grace , 2012 ) , a key brain region associated with motivational salience ( Ikemoto , 2007 ) , where DA neurons make the strongest GLU connections ( Mingote et al . , 2015 ) . Dysregulation in salience processing is also thought to underlie the disruption of latent inhibition ( LI ) seen in schizophrenia ( Weiner , 2003 ) . Disruption of LI is replicated in rodents by amphetamine-induced increases in DA neuron activity ( Young et al . , 2005 ) , in particular increases in DA neuron phasic firing ( Covey et al . , 2016 ) . Although DA neuron GLU signals at burst frequencies control NAc shell activity , it remains to be established whether GLU cotransmission is necessary for the expression of behaviors dependent on salience attribution and associated with schizophrenia . So we sought to temper GLU release at the higher firing frequency of bursts , independent of DA release . For this we targeted phosphate-activated glutaminase ( PAG ) , encoded by Gls1 , in order to reduce presynaptic glutamate synthesis modestly without affecting DA neuron vesicular dynamics , as well as minimizing effects on DA neuron development . Most presynaptic GLU arises from the action of PAG; once released , GLU is taken up by neighboring astrocytes , metabolized to glutamine , and transferred back to presynaptic terminals where it is converted to GLU by PAG ( Marx et al . , 2015 ) . This GLU–glutamine cycle is particularly important in sustaining GLU release with higher frequency firing ( Billups et al . , 2013; Tani et al . , 2014 ) . Indeed , deletion ( Masson et al . , 2006 ) or heterozygous reduction of Gls1 ( Gaisler-Salomon et al . , 2009b ) decreases GLU neurotransmission at higher firing frequencies selectively . The global heterozygous Gls1 reduction impacts several DA dependent behaviors that underpin a schizophrenia resilience phenotype ( Gaisler-Salomon et al . , 2009b ) , characterized by an attenuated response to psychostimulant challenge , potentiated latent inhibition , procognitive effects ( Hazan and Gaisler-Salomon , 2014 ) , together with CA1 hippocampal hypoactivity inverse to the CA1 hyperactivity seen in patients with schizophrenia ( Gaisler-Salomon et al . , 2009a; Schobel et al . , 2009 ) . Genetic mutations engendering resilience carry strong therapeutic valence as they directly identify therapeutic targets ( Mihali et al . , 2012 ) . Here we show in DAT GLS1 conditional heterozygous ( cHET ) mice — with a DAT ( Slc6a3 ) -driven Gls1 reduction — that DA neuron GLU cotransmission is reduced in a frequency dependent manner , without affecting DA neuron development or DA release , and that behaviors that rely on the motivational salience-encoding function of DA neurons are selectively affected , with implications of DA neuron GLU cotransmission for schizophrenia pharmacotherapy . DA neurons immunoreactive for PAG are found in both the ventral tegmental area ( VTA ) and substantia nigra pars compacta ( SNc ) in rat ( Kaneko et al . , 1990 ) , but this has not been examined in mouse . Moreover , the expression of PAG in DA neurons has never been addressed stereologically . We immunostained ventral midbrain sections of the VTA and SNc for the DA-synthetic enzyme tyrosine hydroxylase ( TH ) and for PAG ( Figure 1A; Figure 1—figure supplement 1A ) . This revealed TH positive ( + ) / PAG+ , PAG only ( TH negative ( — ) / PAG+ ) , or TH only ( TH+ / PAG— ) neurons ( Figure 1B ) . Stereological counts in P25 mice showed that the three cell populations were present in similar proportions in the VTA and SNc ( Figure 1C ) . In contrast , DA neurons expressing VGLUT2 are concentrated in the medial VTA ( Yamaguchi et al . , 2015 ) . 10 . 7554/eLife . 27566 . 003Figure 1 . Expression of phosphate-activated glutaminase ( PAG ) in mouse ventral midbrain DA neurons . ( A ) Confocal mosaic z-projected image of the ventral midbrain showing TH ( green , left ) and PAG ( magenta , right ) immunoreactivity . Merged image ( center ) shows that some TH+ DA neurons co-express PAG ( white ) . The specificity of the PAG antibody was verified in GLS1 KO mice; see Figure 1—figure supplement 1A . ( B ) Magnified confocal images in the VTA ( left ) and SNc ( right ) showing TH+ only ( thin blue arrow ) , PAG+ only ( blue arrow head ) and TH+/PAG+ cells ( thick blue arrow ) . ( C ) Stereological counts of TH+ only ( green ) , PAG+ only ( magenta ) and TH+ / PAG+ ( white ) cells in the VTA and SNc of juvenile ( P25 ) wild type mice ( n = 4 ) . Cell numbers in the VTA ( TH+only = 4681 , PAG+only = 3411 , TH+ / PAG+=3673 ) were greater than in the SNc ( TH+only = 2564 , PAG+only = 2909 , TH+ / PAG+=2595 ) ( two-way ANOVA: main effect of brain region , F ( 1 , 18 ) = 18 . 36; p<0 . 001; effect size ( ES ) partial η2 = 0 . 51 ) , but the relative proportions of cell types did not differ between regions ( main effect of cell type , F ( 2 , 18 ) = 1 . 22; p=0 . 318; cell type X brain region interaction , F ( 2 , 18 ) = 2 . 70; p=0 . 094 ) . ( D ) Single-cell RT-PCR analysis of cells expressing TH mRNA , in the VTA and SNc of juvenile mice ( P25-37 ) , showing the percentage of cells that co-expressed PAG and VGLUT2 mRNA . In the VTA , most cells were either TH+ only ( 7/22 ) or TH+/PAG +/VGLUT2+ ( 8/22 ) ; there were also TH+/PAG+ cells ( 5/22 ) and rarely TH+/VGLUT2+ ( 2/22 ) . In the SNc , most cells were either TH+ only ( 5/12 ) or TH+/PAG+ cells ( 6/12 ) ; and rarely TH+/PAG+/VGLUT2+ ( 1/12 ) . No TH+ cells expressed GAD mRNA . For the full coexpression analysis , including GAD mRNA , see Figure 1—figure supplement 1B and C . ( E ) Comparison of the relative number of TH+ / PAG+ cells in juvenile ( P25 ) and adult ( P60 ) mice . In both the VTA and SNc , there was a significant increase in the number of TH+ / PAG+ cells . # indicates a significant main effect of age ( two-way ANOVA , F ( 1 , 10 ) = 8 . 26; p=0 . 017 , ES partial η2 = 0 . 45 ) ; there was no significant region effect ( F ( 1 , 10 ) = 2 . 154; p=0 . 173 ) , nor interaction , ( F ( 1 , 10 ) = 0 . 846; p=0 . 379 ) . See Figure 1—source data 1 . xlsx for source data and all statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 003 10 . 7554/eLife . 27566 . 004Figure 1—source data 1 . Stereology of TH and PAG positive cells in the VTA and SNc in Juvenile and Adult . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 004 10 . 7554/eLife . 27566 . 005Figure 1—figure supplement 1 . Expression of PAG in dopamine neurons . ( A ) Validation of the phosphate-activated glutaminase ( PAG ) antibody in GLS1 KO mice ( neonates were used since KOs survive only for a few hours ) . Immunoreactivity was absent in GLS1 KO brain . Sagittal sections are shown . Abbreviations: Ctx , cortex; Hipp , hippocampus . ( B ) Sample gel images of single-cell reverse transcription ( RT ) PCR from the VTA ( top ) and the SN ( bottom ) . For each region , the upper gel shows the multiplex result for glutamate decarboxylase ( GAD67 , 702 bp ) , tyrosine hydroxylase ( TH , 377 bp ) , vesicular glutamate transporter ( VGLUT2 , 250 bp ) ; the lower gel shows PAG ( 512 bp ) . Numbers on the top of each image are cell numbers; each lane in the multiplex gel ( top ) and PAG gel ( bottom ) was from the same cell . ( C ) Euler diagrams showing RT-PCR results in the VTA ( top ) and the SN ( bottom ) . Numbers inside each square indicate the number of cells expressing the gene or combination of genes . TH+ cells are grouped ( green square ) in the diagram on the left; TH— cells are divided into those expressing GAD67 ( gray squares ) and VGLUT2 ( blue squares ) , on the right . PAG expressing cells are indicated by magenta-filled magenta squares . Cells expressing TH , VGLUT2 and PAG are indicated by yellow filled squares . There was no overlap of TH and GAD . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 005 Since DA neurons capable of GLU cotransmission express VGLUT2 ( Hnasko et al . , 2010; Stuber et al . , 2010 ) and the majority of neurotransmitter GLU is produced by PAG , DA neurons expressing VGLUT2 should preferentially express PAG . To determine the number of DA neurons expressing both VGLUT2 and PAG mRNA , we performed a single cell reverse transcription ( RT ) -PCR analysis in P25-37 mice ( Figure 1D; Figure 1—figure supplement 1B ) . Since DA neurons also corelease GABA ( Tritsch et al . , 2016 ) , which could derive in part from glutamic acid decarboxylase ( GAD ) metabolism of GLU ( produced by PAG ) , we also examined the expression of GAD67 mRNA ( Figure 1—figure supplement 1B ) . We found that VGLUT2 mRNA was highly concentrated in VTA DA neurons but rarely expressed in SNc DA neurons . Importantly , TH+ / VGLUT2+ neurons preferentially expressed PAG ( 9 out of 11 TH+ / VGLUT2+ cells coexpressed PAG; χ2 =3 . 6 , p=0 . 035 ) , further supporting the role of PAG in GLU cotransmission ( Figure 1D ) . GAD67 was not found in TH+ / PAG+ neurons; while a few DA neurons expressed GAD67 ( Kim et al . , 2015 supplemental information ) , a larger sample would be required to assess the role of PAG in GABA cotransmission . Yet , some TH— / PAG+ neurons in both the VTA ( 2/6 cells ) and SN ( 6/13 cells ) were GAD+ , identifying them as GABA neurons and suggesting that PAG contributes to GABA synthesis in those neurons ( Figure 1—figure supplement 1B , C ) . We also found TH— / PAG+ VTA neurons that coexpress VGLUT2 ( Figure 1—figure supplement 1B , C ) , identifying them as GLU neurons ( Hnasko et al . , 2012; Yamaguchi et al . , 2015 ) . Given that coexpression of VGLUT2 decreases with maturation ( Trudeau et al . , 2014 ) , we compared the number of TH+ / PAG+ neurons in juvenile ( P25 ) and adult ( P60 ) wild-type mice . The number of PAG+ / TH+ neurons in both the VTA and SNc increased modestly with age ( Figure 1E ) . Although DA neurons throughout the ventral midbrain express PAG , only medial DA neurons that also express VGLUT2 are capable of GLU cotransmission , so the impact of a PAG reduction on GLU cotransmission should be further restricted to VGLUT2-expressing DA neurons . To address the specific function of DA neuron GLU cotransmission , we bred DATIREScre mice ( Bäckman et al . , 2006 ) with floxGls1 mice ( Mingote et al . , 2016 ) to reduce Gls1 coexpression selectively ( Figure 2 ) . DAT and other DA neuron specific gene expression is not affected in the ventral midbrain and striatum of DATIREScre HET mice ( Bäckman et al . , 2006 ) , which we confirmed ( Figure 2—figure supplement 1A ) ; the acute locomotor response to amphetamine , a drug that targets DAT function , was also not affected ( Figure 2—figure supplement 1B , C ) . We have shown previously that Gls1 expression from the floxGls1 allele is normal ( Mingote et al . , 2016 ) . 10 . 7554/eLife . 27566 . 006Figure 2 . DA neuron selective PAG deletion . ( A ) PCR screens for the floxGLS1 allele ( left ) and ΔGLS1 allele ( right ) in brain regions from both GLS1lox/lox and DAT GLS cKO mice . The ΔGLS1 allele was present solely in DAT GLS1 cKO ventral midbrain . dStr , dorsal striatum; HIPP , hippocampus; VMB , ventral midbrain; CTX , cortex . Gel is representative of 3 replications . ( B ) Single-cell rtPCR analysis of TH expressing cells in the VTA in DATIREScre/+ and DAT GLS1 cKO mice . In the VTA of DATIREScre/+mice , 11/30 TH cells expressed PAG mRNA , while in DAT GLS1 cKO none did ( 0/38 cells ) . ( C ) Confocal photomicrographs of the VTA from DATIREScre/+ and DAT GLS1 cKO mice showing TH+ only ( thin blue arrow ) and PAG+ only ( blue arrow head ) and TH+/PAG+ cells ( thick blue arrow ) . There were no TH+/PAG+ cells in the DAT GLS1 cKO ventral midbrain . Expression of dopaminergic markers and amphetamine-induced hyperlocomotion were not affected in DATIREScre mice; see Figure 2—figure supplement 1 . These mice were control ( CTRL ) mice in subsequent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 00610 . 7554/eLife . 27566 . 007Figure 2—figure supplement 1 . Expression of dopaminergic markers and amphetamine-induced hyperlocomotion were not affected in DATIREScre mice . ( A ) The relative mRNA expression of dopamine transporter ( DAT ) , tyrosine hydroxylase ( TH ) , vesicular monoamine transporter 2 ( VMAT2 ) and dopamine D2 receptor ( D2R ) in the ventral midbrain ( left ) , and D1R and D2R in dorsal striatum ( dStr , right ) of DATIREScre/+ and wild-type littermates ( CTRL ) . A multivariate ANOVA showed no genotypic effect for any of the dopaminergic markers ( ventral midbrain , DAT , F ( 1 , 8 ) = 0 . 061 , p=0 . 811; TH , F ( 1 , 8 ) = 0 . 320 , p=0 . 587; VMAT2 , F ( 1 , 8 ) = 1 . 742 , p=0 . 223; D2 , F ( 1 , 8 ) = 3 . 903 , p=0 . 084; dStr , D1 = F ( 1 , 10 ) = 0 . 384 , p=0 . 549; F ( 1 , 10 ) = 0 . 851 , p=0 . 004 ) . ( B ) Amphetamine ( Amph ) stimulated locomotion . Total locomotor counts ( i . e . , beam breaks ) in the open field made over 2 . 5 hr following Vehicle ( 0 mg/kg ) or Amph , 3 or 5 mg/kg ( i . p . ) . A two-way ANOVA showed a main effect of drug ( F ( 2 , 28 ) = 83 . 1; p<0 . 001 , ES partial η2 = 0 . 86 ) , but no significant main effect of genotype ( F ( 1 , 28 ) = 0 . 846; p=0 . 366 ) or significant interaction ( F ( 2 , 28 ) = 28 . 2; p=0 . 973 ) . ( C ) Time course of Amph-evoked locomotion . There were no genotypic differences for either the 3 mg/kg ( top ) or 5 mg/kg doses ( bottom ) . The repeated measures ( RM ) ANOVA showed no significant main effect of genotype ( 3 mg/kg dose , F ( 1 , 10 ) = 0 . 003 , p=0 . 960; 5 mg/kg dose , F ( 1 , 9 ) = 1 . 322 , p=0 . 280 ) or time X genotype interaction ( 3 mg/kg dose , F ( 14 , 140 ) = 0 . 784 , p=0 . 685; 5 mg/kg dose , F ( 14 , 126 ) = 1 . 663 , p=0 . 071 ) ; there was a main effect of time ( 3 mg/kg dose , F ( 14 , 140 ) = 20 . 5 , p<0 . 0001 , ES partial η2 = 0 . 67; 5 mg/kg dose , F ( 14 , 126 ) = 20 . 5 , p<0 . 0001 , ES partial η2 = 0 . 69 ) . Numbers of cells are shown above each bar in the graphs . See Figure 2—figure supplement 1—source data 1 . xlsx for source data and statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 007 10 . 7554/eLife . 27566 . 008Figure 2—figure supplement 1—sourcedata 1 . Expression of dopaminergic markers and amphetamine-induced hyperlocomotion in DATIREScre mice . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 008 Conditional targeting was verified in DAT GLS1 cKO mice ( DATIREScre/+::GLS1lox/lox ) mice . PCR screens of genomic DNA showed the non-functional truncated ( Δ ) Gls1 allele in the ventral midbrain of DAT GLS1 cKO mice , but not in forebrain regions that do not contain DA neurons , the dorsal striatum ( dStr ) , frontal cortex and hippocampus ( Figure 2A ) . We used single cell RT-PCR analysis to verify further the Gls1 inactivation in DA neurons ( Figure 2B ) . In DAT GLS1 cKO mice , Gls1 mRNA was absent in VTA cells expressing TH mRNA . There was no impact on the number of DA neurons that expressed VGLUT2 ( 3/38 in DATIREScre/+ vs . 6/30 DAT GLS1 cKO mice , χ2 =1 . 2 , p=0 . 27 ) . To confirm the conditional strategy at the protein level , we examined TH and PAG immunoreactivity in the VTA ( Figure 2C ) . In DAT GLS1 cKO mice , all TH+ cells were PAG— , while neighboring TH— but PAG+ cells were seen , demonstrating the specificity of Gls1 targeting . Since heterozygous reduction in Gls1 is sufficient to attenuate GLU transmission at higher-firing frequencies ( Gaisler-Salomon et al . , 2009b ) , and to minimize compensatory mechanisms seen in KOs ( Bae et al . , 2013 ) , we used DAT GLS1 cHET mice ( DATIREScre/+::GLS1lox/+ ) and DATIREScre/+ mice as controls ( CTRL ) . To measure DA neuron synaptic transmission , we conditionally expressed channelrhodopsin 2 ( ChR2 ) in DA neurons using Ai32 ( RCL-ChR2 ( H134R ) /EYFP ) mice ( Madisen et al . , 2012 ) , to obtain triple mutant DAT GLS1 cHET::ChR2 ( DATIREScre/+::GLS1lox/+::Ai32 ) and double mutant control CTRL::ChR2 ( DATIREScre/+::Ai32 ) littermates . We confirmed that the expression of ChR2-EYFP was specific to DA neurons independent of Gls1 genotype ( Figure 3—figure supplement 1A , B ) . We also confirmed that TH+ / DAT— striatal interneurons ( Xenias et al . , 2015 ) do not express ChR2-EYFP ( Figure 3—figure supplement 1C ) . We then examined the impact of PAG deficiency on GLU cotransmission in recordings from cholinergic interneurons ( ChIs ) and spiny projection neurons ( SPNs ) in the NAc medial shell , the striatal hotspot for DA neuron GLU transmission ( Chuhma et al . , 2014; Mingote et al . , 2015 ) ( Figure 3A ) . ChIs and SPNs were identified by soma size and electrophysiological signature , under current clamp ( Figure 3—figure supplement 2A ) . We confirmed that the intrinsic membrane properties of ChIs and SPNs did not differ between genotypes ( Figure 3—figure supplement 2B , C , D , E ) . 10 . 7554/eLife . 27566 . 009Figure 3 . DA neuron GLU cotransmission is attenuated in DAT GLS1 cHETs at phasic firing frequencies . ( A ) Schematic of a coronal slice ( −1 . 34 mm from bregma ) indicating the location of the patch-clamp recordings in the medial NAc shell . DA neuron excitatory responses evoked by photostimulation ( blue circles ) were measured from ChIs and SPNs ( left ) . See also Figure 3—figure supplement 1 . ( B ) Representative traces ( left ) of EPSCs generated by a single-pulse photostimulation ( blue bar ) at 0 . 1 Hz recorded from ChIs and SPNs . Traces shown are averages of 10 consecutive traces . Comparison is made between responses in CTRL ( black traces ) and DAT GLS1 cHET mice ( gray traces ) ; all responses were completely blocked by CNQX ( 40 µM; red traces ) . Summary of average EPSC amplitude after single-pulse photostimulation ( right ) . # indicates a significant main effect of cell type ( two-way ANOVA , F ( 1 , 36 ) = 25 . 6 , p<0 . 001 , ES partial η2 = 0 . 42 ) ; there was no significant genotype effect ( F ( 1 , 36 ) = 1 . 084 , p=0 . 305 ) , nor interaction ( F ( 1 , 36 ) = 0 . 628 , p=0 . 433 ) . See also Figure 3—figure supplement 2 . ( C ) Representative traces of EPSCs generated by burst photostimulation ( 5 pulses at 20 Hz ) recorded from ChIs ( top ) and SPNs ( bottom ) . Summary of the average EPSC amplitudes after burst photostimulation ( right ) are shown as percentage of the first response , which did not differ between genotypes ( ChIs , CTRL 95 ± 29 pA vs . cHET 107 ± 12 pA , Mann-Whitney , p=0 . 14; SPNs , CTRL 27 ± 4 pA vs . cHET 28 ± 5 pA , Mann-Whitney , p=0 . 88 ) . The shaded violet bar at the bottom of the graphs represents the average baseline noise ( ChIs 3 . 8 ± 0 . 4 pA; SPNs 3 . 5 ± 0 . 3 pA ) . For ChIs , repeated measures ( RM ) ANOVA revealed a significant pulses X genotype interaction ( F ( 3 , 54 ) = 28 . 2 , p=0 . 006 , ES partial η2 = 0 . 27 ) , main effect of pulses ( F ( 3 , 54 ) = 20 . 9 , p<0 . 001 ) , and main effect of genotype ( F ( 1 , 18 ) = 5 . 06 , p=0 . 037 ) . * indicates significant difference from CTRL ( p=0 . 006 ) after applying a Bonferroni correction for 4 comparisons ( α = 0 . 0125 ) . For SPNs , ◊ # indicates a significant main effect of genotype ( F ( 1 , 18 ) = 4 . 6 , p=0 . 047 , ES partial η2 = 0 . 20 ) and main effect of pulses ( F ( 3 , 54 ) = 7 . 7 , p<0 . 001 , ES partial η2 = 0 . 30 ) by RM ANOVA; but no significant interaction ( F ( 3 , 54 ) = 2 . 0 , p=0 . 101 ) . ( D ) Effect of photostimulation mimicking DA neuron bursting ( 5 pulses at 20 Hz ) on ChI firing . Representative traces are shown above ( left ) , with peristimulus histograms summing ten consecutive traces ( 0 . 1 s bin ) below . Ratio of firing during burst photostimulation ( 0–0 . 5 s from onset of train ) and after ( 0 . 5–1 s from onset ) to baseline firing are shown on the right . * indicates significant effect of genotype ( one-way ANOVA , F ( 1 , 33 ) = 7 . 0 , p=0 . 013 , ES partial η2 = 0 . 17 ) . ( E ) Colored-coded tables showing action potential counts in 50 ms intervals , prior to , during and after DA terminal photostimulation for CTRL ( left ) and DAT GLS1 cHET mice ( right ) for all recorded cells . The blue horizontal bar at the bottom of each table indicates the duration of burst photostimulation , with onset at time 0 . In all the graphs , the number of cells is shown above the bars or next to the lines . In this and subsequent figures , error bars represent SEM . See Figure 3—source data 1 . xlsx for source data and statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 009 10 . 7554/eLife . 27566 . 010Figure 3—source data 1 . Slice patch clamp experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 010 10 . 7554/eLife . 27566 . 011Figure 3—figure supplement 1 . Comparison between CTRL::ChR2 and DAT GLS1 cHET::ChR2 mice showing selective ChR2 expression in DA neurons did not differ between genotypes . ( A ) ChR2-EYFP expression ( green ) in the ventral midbrain was restricted to TH+ cells ( magenta ) , with similar colocalization ( white ) in both CTRL::ChR2 mice ( left ) and DAT GLS1 cHET::ChR2 mice ( right ) . ( B ) Stereological counts of ChR2-EYFP and TH+ cells in the VTA and SNc of CTRL::ChR2 ( n = 3 ) and DAT GLS1 cHET::ChR2 mice ( n = 3 ) . Values are presented as percent of the total number of cells counted in each region , for each genotype ( VTA , CTRL::ChR2 = 5097 ± 817 cells and DAT GLS1 cHET::ChR2 = 3891 ± 628 cells; SNc , CTRL::ChR2 = 4782 ± 889 cells and DAT GLS1 cHET::ChR2 = 3345 ± 453 cells; Kruskal-Wallis test in each region showed no genotype effect ) . The percentage of TH+/ChR2-EYFP+ cells did not differ genotypically , in the VTA or SNc ( Kruskal-Wallis test ) . ( C ) TH+ interneurons ( magenta ) in the dorsal striatum ( dStr ) did not express ChR2-EYFP ( green ) . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 01110 . 7554/eLife . 27566 . 012Figure 3—figure supplement 2 . Comparison between CTRL::ChR2 and DAT GLS1 cHET::ChR2 mice showing that intrinsic electrophysiological membrane properties and spontaneous EPSCs measured in NAc shell cells did not differ between genotypes . ( A ) ChIs and SPNs are identifiable based on their electrophysiological signature under current clamp . The ChI ( left ) had a resting membrane potential around −70 mV , fired spontaneously ( black trace ) , and showed a voltage sag with hyperpolarizing current injection ( green trace ) . The SPN ( right ) had a deep resting membrane potential around −100 mV , did not fire spontaneously ( black ) , showed no sag with hyperpolarizing current injection ( green trace ) , and fired rapidly with depolarizing current injection , after a delay ( blue trace ) . ( B ) The average baseline membrane potential ( Vrest ) was more negative in SPNs than in ChIs , but not genotypically different . A two-way ANOVA showed a main effect of cell type ( F ( 1 , 62 ) = 128 . 3 , p<0 . 0001 , ES partial η2 = 0 . 67 ) , indicated by the # , but no main effect of genotype ( F ( 1 , 62 ) = 1 . 67 , p=0 . 201 ) or significant interaction ( F ( 1 , 62 ) = 0 . 138 , p=0 . 711 ) . ( C ) Action potential ( AP ) threshold in ChIs and SPNs . A two-way ANOVA showed no main effect of genotype ( F ( 1 , 62 ) = 0 . 53 , p=0 . 819 ) or cell type ( F ( 1 , 62 ) = 2 . 78 , p=0 . 100 ) , or significant interaction ( F ( 1 , 62 ) = 0 . 480 , p=0 . 491 ) . ( D ) Input impedance was significantly higher in ChIs compared to SPNs , but not statistically different between genotypes . A two-way ANOVA showed a main effect of cell type ( F ( 1 , 62 ) = 15 . 7 , p<0 . 001 , ES partial η2 = 0 . 20 ) , indicated by the # , but no main effect of genotype ( F ( 1 , 62 ) = 0 . 233 , p=0 . 631 ) or significant interaction ( F ( 1 , 62 ) = 1 . 96 , p=0 . 167 ) . ( E ) The hyperpolarization-activated cation current ( Ih ) ratio was lower in the ChIs than SPNs , revealing the presence of an Ih in ChIs but not SPNs . A two-way ANOVA showed a main effect of cell type ( F ( 1 , 62 ) = 15 . 0 , p<0 . 001 , ES partial η2 = 0 . 20 ) , indicated by the # , but no main effect of genotype ( F ( 1 , 62 ) = 0 . 001 , p=0 . 976 ) or significant interaction ( F ( 1 , 62 ) = 0 . 856 , p=0 . 358 ) . ( F ) Characterization of photostimulated DA neuron evoked EPSCs in ChIs and SPNs under voltage clamp revealed that rise time ( from 10% to 90% of peak amplitude ) was faster in SPNs than ChIs , but not genotypically different . A two-way ANOVA showed a significant cell type effect ( F ( 1 , 46 ) = 28 . 4 , p<0 . 0001 , ES partial η2 = 0 . 08 ) indicated by the # , but no main effect of genotype ( F ( 1 , 46 ) = 1 . 02 , p=0 . 305 ) or significant interaction ( F ( 1 , 46 ) = 1 . 02 , p=0 . 305 ) . ( G ) Decay times of evoked EPSCs under voltage clamp . A two-way ANOVA showed no main effect of genotype ( F ( 1 , 46 ) = 2 . 135; p=0 . 151 ) or cell type ( F ( 1 , 46 ) = 0 . 458; p=0 . 502 ) , or significant interaction ( F ( 1 , 46 ) = 1 . 331; p=0 . 255 ) . ( H ) Amplitude of spontaneous EPSCs measured under voltage clamp ( holding potential −70 mV ) in ChIs and SPNs . A two-way ANOVA showed a main effect of cell type ( F ( 1 , 36 ) = 5 . 85 , p<0 . 021 , ES partial η2 = 0 . 14 ) , indicated by the # , but no main effect of genotype ( F ( 1 , 36 ) = 0 . 257 , p=0 . 615 ) or significant interaction ( F ( 1 , 36 ) = 1 . 68 , p=0 . 203 ) . ( I ) Frequency of spontaneous EPSCs measured under voltage clamp ( holding potential −70 mV ) in ChIs and SPNs . A two-way ANOVA showed no main effect of genotype ( F ( 1 , 36 ) = 0 . 308 , p=0 . 582 ) or cell type ( F ( 1 , 36 ) = 0 . 764 , p=0 . 388 ) , or significant interaction ( F ( 1 , 36 ) = 1 . 97 , p=0 . 169 ) . Numbers of cells are shown above each bar or circle in the graph . See Figure 3—figure supplement 2—source data 1 . xlsx for source data and statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 012 10 . 7554/eLife . 27566 . 013Figure 3—figure supplement 2—source data 1 . Intrinsic electrophysiological membrane properties and spontaneous EPSCs measured in NAc shell cells of CTRL and DAT GLS1 cHET mice . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 013 We measured DA neuron GLU cotransmission in DAT GLS1 cHET::ChR2 mice ( P60-P76 ) in the NAc shell with single pulse photostimulation ( 5 ms duration , delivered with a 10 s interval ) and burst photostimulation ( 5 pulses at 20 Hz , delivered with a 30 s interval ) of DA neuron terminals . The burst photostimulation was chosen to mimic in vivo phasic firing of DA neurons ( Paladini and Roeper , 2014 ) . Single photostimulation-evoked EPSCs in both ChIs and SPNs ( Figure 3B ) were blocked by the AMPA-kainate receptor antagonist CNQX , confirming GLU mediation ( n = 4 ChIs per genotype; n = 3 SPNs per genotype ) . As reported previously ( Chuhma et al . , 2014 ) , the amplitude of EPSCs in ChIs ( CTRL 51 ± 6 . 0 pA ) was greater than in SPNs ( CTRL 21 . 5 ± 2 . 2 pA ) . The amplitude of single-evoked EPSCs was unaffected in cHET mice ( ChIs 66 . 1 ± 6 . 7 pA; SPNs 21 . 2 ± 2 . 3 pA ) ( Figure 3B ) , as were EPSC rise and decay time constants ( Figure 3—figure supplement 2F , G ) . Burst-induced EPSCs in ChIs and SPNs showed short-term depression in both genotypes that was significantly greater in cHETs ( Figure 3C ) . This was particularly evident when EPSC amplitudes were normalized to the first EPSC in the burst , which showed no genotypic difference ( Figure 3C , graphs ) . In CTRL mice , EPSCs in ChIs decreased to 48 ± 6 . 0% with the second pulse and to 23 ± 4 . 2% with the fifth , while in cHET mice EPSCs decreased to 20 ± 6 . 3% with the second and to 14 ± 3 . 3% with the fifth . The rundown was apparently faster in SPNs ( Figure 3C , bottom traces and graph ) ; in CTRL mice , EPSCs decreased to 48 ± 6 . 0% with the second pulse and to 24 ± 6 . 4% with the fifth , while in cHET mice EPSC amplitude decreased to 25 ± 4 . 9% with the second , and to 16 ± 1% with the fifth , which was close to baseline . Observing a more rapid frequency-dependent EPSC depression in cHETs in both ChIs and SPNs , and no differences in their intrinsic properties ( Figure 3—figure supplement 2B , C , D , E ) , is consistent with a presynaptic reduction in PAG . The average amplitude and frequency of spontaneous EPSCs , measured in both the SPNs and ChIs , showed no genotypic difference ( Figure 3—figure supplement 2H , I ) , indicating that GLU inputs mostly from forebrain regions , as well as signaling through postsynaptic GLU receptors , was unaffected in DAT GLS1 cHETs . At the striatal circuit level , DA neuron control of ChI firing in the medial NAc shell ( Chuhma et al . , 2014 ) was attenuated in cHET mice ( Figure 3D ) . We quantified this using the firing ratio , the firing frequency during train photostimulation ( 0–0 . 5 s from the onset of train ) divided by the preceding 2 s of baseline firing . There were no genotypic differences in baseline firing frequencies ( CTRL 4 . 7 ± 1 . 1 Hz; cHET 3 . 9 ± 0 . 6 Hz; ANOVA , F ( 1 , 33 ) = 0 . 60 , p=0 . 444 ) . The firing ratio in CTRL mice was 4 . 3 ± 0 . 7 compared to 2 . 1 ± 0 . 2 in cHET mice , which was significantly reduced ( Figure 3D , right ) . In the subsequent half-second window , the firing ratio reversed to below baseline in CTRL ( 0 . 6 ± 0 . 08 ) and cHETs ( 0 . 7 ± 0 . 11 ) , which did not differ ( Figure 3D , right ) . This reduction in firing is mainly mediated by activity-dependent components , and less so by DA D2-mediated inhibition ( Chuhma et al . , 2014 ) . Color-coded tables with a 50 msec window ( Figure 3E ) clearly show greater burst firing in CTRL than in cHET , but little difference in the post-burst period . Dividing the 0 . 5 to 1 s interval into 250 ms windows revealed no significant differences ( one-way ANOVA: 0 . 5–0 . 75 period , CTRL 0 . 6 ± 0 . 12 vs . cHET 0 . 8 ± 0 . 12 , F ( 1 , 34 ) = 1 . 98 , p=0 . 168; 0 . 75–1 period , CTRL 0 . 5 ± 0 . 08 vs . cHET 0 . 8 ± 0 . 1 , F ( 1 , 34 ) = 3 . 510 , p=0 . 070 ) . Thus , PAG plays an important role in sustaining DA neuron GLU cotransmission at higher firing frequencies and determines their ability to drive Chls to fire in bursts . To evaluate the specificity of the reduction in GLU cotransmission in DAT GLS1 cHET mice further , we counted DA neurons by unbiased stereology , at P110 ( Figure 4A ) . We found no reduction in the number of DA neurons in the VTA ( unilateral counts: CTRL 7548 ± 418 , cHET 7310 ± 450 ) or SNc ( CTRL 6595 ± 373 , cHET 6781 ± 518 ) . DA neurons in cHET mice showed no differences in their intrinsic electrophysiological properties ( Figure 4—figure supplement 1 ) . Presynaptic DA content and turnover , in the NAc and dStr of adult mice ( P71-P110 ) , did not significantly differ between genotypes ( Figure 4B ) . We performed fast-scan cyclic voltammetry ( FSCV ) in DAT GLS1 cHET::ChR2 mice ( P71-P85 ) to determine whether DA release dynamics were affected ( Figure 4C ) . We compared DA release evoked by single or burst photostimulation in the NAc medial shell . To challenge DA neuron synapses further , single pulse stimulation was repeated twice followed by a burst , and burst stimulation was repeated twice followed by a single . There were no genotypic differences in DA release with either stimulation pattern ( Figure 4D ) . The decay time constant of DA responses did not differ significantly between genotypes with single ( CTRL 409 ± 30 ms; cHET 362 ± 26 ms; ANOVA , F ( 1 , 23 ) =1 . 42 , p=0 . 245 ) or burst photostimulation ( CTRL 540 ± 28 ms; cHET 482 ± 25 ms; ANOVA , F ( 1 , 22 ) = 2 . 12 , p=0 . 160 ) . Thus , the conditional Gls1 reduction does not affect DA neuron DA release in the NAc medial shell , where GLU cotransmission is strongest . Evoked DA release was not affected in the NAc core ( Figure 4—figure supplement 2A , B , C ) nor in the dStr ( Figure 4—figure supplement 2D , E , F ) , indicating that DA storage and release dynamics throughout the striatum are normal in DAT GLS1 cHETs . The effect sizes for all non-significant F values were small to negligible ( partial η2: Stereology = 0 . 014; DA content = 0 . 004; DA release: range 0 . 0002 to 0 . 011 ) . Thus , DA neuron development and DA transmission are unaffected in DAT GLS1 cHETs . 10 . 7554/eLife . 27566 . 014Figure 4 . PAG reduction in DA neurons does not alter the number of DA neurons or striatal DA function . ( A ) Stereological-estimate of the total number of DA neurons ( TH+ neurons ) in the VTA and SNc in one hemisphere showed no difference between genotypes ( one-way ANOVA: VTA , F ( 1 , 6 ) = 0 . 149 , p=0 . 713; SNc , F ( 1 , 6 ) = 0 . 085 , p=0 . 781 ) . There were no differences in DA neuron intrinsic electrophysiological properties; see Figure 4—figure supplement 1 . ( B ) Tissue DA content in the NAc and dStr ( left ) and DA turnover measured by DOPAC/DA ratio ( right ) did not differ between genotypes by one-way ANOVA ( NAc DA content , F ( 1 , 22 ) = 0 . 070 , p=0 . 794; NAc DOPAC/DA , F ( 1 , 22 ) = 3 . 01 , p=0 . 098; dStr DA content , F ( 1 , 20 ) = 0 . 078 , p=0 . 783; dStr DOPAC/DA , F ( 1 , 20 ) = 1 . 68 , p=0 . 211 ) . ( C ) FSCV recordings in the medial NAc shell . A representative voltammogram is shown above a schematic of a coronal slice ( −1 . 34 mm from bregma ) indicating the recording configuration . ( D ) DA release evoked by three consecutive single photostimulation pulses followed by a burst ( 5 pulses at 20 Hz ) ( above ) , or by three consecutive bursts followed by a single ( below ) . Representative recordings of evoked DA release are shown with dashed boxes indicating initial traces that were enlarged and superimposed on the left , showing that DA release dynamics did not differ between genotypes for the single ( above ) or burst ( below ) responses . DA release dynamics did not differ between genotypes for consecutive singles followed by a burst ( above ) or repeated bursts followed by a single pulse ( below ) . The average evoked DA release is shown on the graph ( right ) . For consecutive single pulses followed by a burst , a RM ANOVA revealed a significant main effect of pulses ( F ( 3 , 69 ) = 135 . 1 , p<0 . 001 , ES partial η2 = 0 . 85 ) ; there was no effect of genotype ( F ( 1 , 23 ) = 0 . 069 , p=0 . 795 ) nor interaction ( F ( 3 , 69 ) = 0 . 247 , p=0 . 864 ) . For the consecutive bursts followed by a single , a RM ANOVA revealed a significant main effect of pulses ( F ( 3 , 66 ) = 124 . 5; p<0 . 001 , ES partial η2 = 0 . 85 ) ; there was no effect of genotype ( F ( 1 , 22 ) = 0 . 004 , p=0 . 948 ) or interaction ( F ( 3 , 66 ) = 0 . 103 , p=0 . 103 ) . Dopamine release in the NAc core and dStr was also not affected in DAT GLS1 cHETs; see Figure 4—figure supplement 2 . Numbers of mice or the number of slices ( FSCV ) are shown in each graph above the bars . See Figure 4—source data 1 . xlsx for source data and statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 014 10 . 7554/eLife . 27566 . 015Figure 4—source data 1 . Dopamine transmission in CTRL and DAT GLS1 cHET mice . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 015 10 . 7554/eLife . 27566 . 016Figure 4—figure supplement 1 . Electrophysiological properties of putative DA neurons in the ventral midbrain . ( A ) DA neuron pacemaker firing recorded in the ventral tegmental area ( VTA , left ) or substantia nigra pars compacta ( SNc , right ) , in CTRL and DAT GLS1 cHET slices . ( B ) Graph of average firing frequency . Numbers of cells recorded are shown above the bars . ( C ) Input impedance . ( D ) Baseline membrane potential ( Vrest ) . ( E ) Action potential threshold . There was no genotypic difference in either the VTA or SNc for any of these measures by one-way ANOVA ( firing frequency , VTA , F ( 1 , 27 ) = 0 . 238 , p=0 . 630; SNc , F ( 1 , 29 ) = 2 . 59 , p=0 . 118; input impedance , VTA , F ( 1 , 27 ) = 0 . 005 , p=0 . 945; SNc , F ( 1 , 29 ) = 1 . 48 , p=0 . 233; baseline membrane potential , VTA , F ( 1 , 27 ) = 0 . 658 , p=0 . 424; SNc , F ( 1 , 29 ) = 0 . 140 , p=0 . 711; action potential threshold , VTA , F ( 1 , 27 ) = 0 . 480 , p=0 . 494; SNc , F ( 1 , 29 ) = 0 . 567 , p=0 . 458 ) . These results indicate that basic DA neuron properties are not affected in DAT GLS1 cHET mice , nor was there evidence for cell deterioration . The number of cells is shown in the graphs above the bars or circles . See Figure 4—figure supplement 1—source data 1 . xlsx for source data and statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 016 10 . 7554/eLife . 27566 . 017Figure 4—figure supplement 1—source data 1 . Dopamine neuron membrane properties in CTRL and DAT GLS1 cHET mice . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 017 10 . 7554/eLife . 27566 . 018Figure 4—figure supplement 2 . Dopamine release in nucleus accumbens core and dorsal striatum , measured by fast-scan cyclic voltammetry ( FSCV ) , is not affected in DAT GLS1 cHET mice . ( A ) Schematic of a coronal slice with recording site in the nucleus accumbens ( NAc ) core . ( B ) Representative FSCV traces , organized as in B . ( C ) Average evoked DA release in the NAc core . Graphs correspond to traces in E . In the upper graph , RM ANOVA showed a significant main effect of pulses ( F ( 3 , 36 ) = 22 . 903 , p<0 . 0001 , ES partial η2 = 0 . 656 ) , but no main effect of genotype ( F ( 1 . 12 ) = 0 . 32 , p=0 . 523 ) or significant interaction ( F ( 3 , 36 ) = 0 . 418 , p=0 . 741 ) . In the lower graph , RM ANOVA showed a significant main effect of pulses ( F ( 3 , 36 ) = 60 . 79 , p<0 . 0001 , ES partial η2 = 0 . 835 ) , but no main effect of genotype ( F ( 1 . 12 ) = 0 . 249 , p=0 . 627 ) or significant interaction ( F ( 3 , 36 ) = 0 . 210 , p=0 . 889 ) . ( D ) Schematic of a coronal slice with recording site in the medial dorsal striatum ( dStr ) . DA release evoked by photostimulation was measured using FSCV . A representative cyclic voltammogram is shown in the upper left . ( E ) Representative FSCV traces of photostimulated DA release . The first two responses in each trace ( dashed box ) are enlarged and superimposed on the left . The upper pair of traces shows responses to 3 single photostimulations followed by a burst; the lower pair to 3 burst photostimulations followed by a single . ( F ) Average evoked DA release in the dStr . Graphs correspond to traces in B . In the upper graph , RM ANOVA showed a significant main effect of photostimulation ( F ( 3 , 36 ) = 48 . 52 , p<0 . 0001 , ES partial η2 = 0 . 802 ) , but no main effect of genotype ( F ( 1 , 12 ) = 0 . 072 , p=0 . 793 ) or significant interaction ( F ( 3 , 36 ) = 0 . 26 , p=854 ) . In the lower graph , RM ANOVA showed a significant main effect of pulses ( F ( 3 , 36 ) = 37 . 257 , p<0 . 0001 , ES partial η2 = 0 . 756 ) , but no main effect of genotype ( F ( 1 , 12 ) = 0 . 084 , p=0 . 777 ) or significant interaction ( F ( 3 , 36 ) = 0 . 083 , p=0 . 969 ) . The numbers of slices recorded are shown above the first pair of bars in the graphs . See Figure 4—figure supplement 2—source data 2 . xlsx for source data and statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 018 10 . 7554/eLife . 27566 . 019Figure 4—figure supplement 2—source data 2 . Dopamine release measured by FSCV in the NAc Core and dStr in CTRL and DAT GLS1 cHET mice . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 019 We examined DA neuron dependent behaviors in DAT GLS1 cHETs ( P90-120 ) . We assessed motor learning and coordination on the rotarod , which is affected following neurotoxic loss of DA neurons ( Rozas et al . , 1998; Beeler et al . , 2010 ) and also variably affected in DAT VGLUT2 cKO mice ( Fortin et al . , 2012 ) ( but see also Birgner et al . , 2010; Hnasko et al . , 2010 ) . DAT GLS1 cHET mice showed robust motor learning , which did not differ from CTRL mice , on the first training day , when rotarod speed was 20 rpm ( Figure 5A ) , and then accelerated to 30 rpm and 40 rpm on subsequent days . Novelty-induced exploration in the open field was unaffected ( Figure 5B ) . Mice used in this experiment belonged to two cohorts that were subsequently used in the amphetamine-induced locomotion and sensitization experiments . The results from the first cohort were replicated in the second cohort; since there was no significant cohort effect ( two-way ANOVA for total locomotion in 60 min: cohort , F ( 1 , 100 ) = 50 . 9 , p=0 . 64; cohort X genotype , F ( 1 , 100 ) = 2 . 6 , p=0 . 11 ) , the cohorts were combined . 10 . 7554/eLife . 27566 . 020Figure 5 . Motor performance , anxiety and amphetamine-induced hyperlocomotion are unaffected in DAT GLS1 cHETs . ( A ) Motor performance on an accelerating rotarod over 3 days showed no difference between genotypes ( RM ANOVA , significant effect of trials , F ( 8 , 520 ) = 22 . 9 , p<0 . 0001 , ES partial η2 = 0 . 26 ) ; there was no effect of genotype ( F ( 1 , 65 ) = 0 . 018 , p=0 . 894; nor interaction F ( 8 , 520 ) = 0 . 562 , p=0 . 809 ) . ( B ) Locomotor activity in the open field for one hour revealed no genotypic difference in novelty-induced locomotion and habituation ( RM ANOVA , main effect of time , F ( 5 , 510 ) = 193 . 0 , p<0 . 0001 , ES partial η2 = 0 . 65 ) ; no effect of genotype ( F ( 1 , 102 ) = 0 . 664 , p=0 . 417 ) nor interaction ( F ( 5 , 510 ) = 0 . 329 , p=0 . 895 ) . ( C ) Exploration in the elevated-plus maze ( 5 min ) showed no genotypic difference in percentage of time spent in the open arms ( left ) ( one way-ANOVA , F ( 1 , 22 ) = 0 . 004 , p=0 . 949 ) or time spent in the open arms per entry ( right ) ( one way-ANOVA , F ( 1 , 22 ) = 0 . 547 , p=0 . 467 ) . ( D ) Fear conditioning to tone ( left ) measured as the average percentage of freezing during the CS ( two tone presentations ) or to a context previously paired with a shock ( right ) showed no genotypic differences ( one-way ANOVA , tone fear conditioning , F ( 1 , 16 ) = 1 . 145 , p=0 . 300; context fear conditioning , F ( 1 , 16 ) = 0 . 207 , p=0 . 655 ) . ( E ) Amphetamine-induced locomotor activity recorded over 90 min post injection showed no genotypic difference in the dose-dependent responses ( two-way ANOVA , main effect of drug treatment , F ( 2 . 66 ) = 34 . 8 , p<0 . 0001 , ES partial η2 = 0 . 51; no effect of genotype , F ( 2 . 66 ) = 0 . 068 , p=0 . 795; nor interaction , F ( 2 . 66 ) = 0 . 18 , p=0 . 836 ) . The number of mice is shown in the graphs above the bars or next to the lines . See Figure 5—source data 1 . xlsx for source data and statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 020 10 . 7554/eLife . 27566 . 021Figure 5—source data 1 . Dopamine neuron dependent behaviors in CTRL and DAT GLS1 cHET mice . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 021 DA neuron loss can have anxiogenic effects ( Drui et al . , 2014 ) , and DAT VGLUT2 cKO mice showed decreased time spent in the center of the open field , indicative of increased anxiety ( Birgner et al . , 2010 ) . DAT GLS1 cHET and CTRL mice spent the same time in the center of the open field ( CTRL = 256± 43 s; DAT GLS1 cHET = 301 ± 23 s; one-way ANOVA , no genotype effect , F ( 1 , 102 ) = 0 . 551 , p=0 . 46 ) . We tested the mice in the elevated plus maze , another test of anxiety . A large cohort of mice ( CTRL = 30 mice , cHET mice = 37 mice ) was tested in an elevated plus maze with short arms . DAT GLS1 cHET and CTRL mice spent the same time in the open arms ( CTRL = 31 . 7 ± 3 . 8 s , cHET = 25 . 4 ± 3 . 5 s; one-way ANOVA , no genotype effect , F ( 1 , 65 ) =1 . 11 , p=0 . 30 ) . A small effect size of 0 . 022 ( partial η2 ) was detected . So , we tested a second cohort in a more anxiogenic elevated plus maze with longer arms ( Figure 5C ) . We found no difference between genotypes in the time spent in the open arms , nor was there a difference between time spent in the open arms per entry ( Figure 5C ) , or the time spent in the proximal and distal portions of the longer arms ( proximal time , CTRL = 47 . 4 ± 4 . 4 s , cHET = 40 . 9 ± 4 . 9 s , one-way ANOVA , no genotype effect , F ( 1 , 24 ) = 0 . 887 , p=0 . 36; distal time , CTRL = 43 . 7 ± 8 . 5 s , cHET = 51 . 27 ± 7 . 90 s , F ( 1 . 24 ) = 0 . 41 , p=0 . 53 ) . DA neurons play a role in fear conditioning ( Fernandez Espejo , 2003; Wen et al . , 2015 ) . Moreover , stopGls1 HET mice , with a global Gls1 reduction , show reduced contextual fear conditioning ( Gaisler-Salomon et al . , 2009b ) . However , DAT GLS1 cHET mice showed normal tone- and context-dependent fear conditioning ( Figure 5D ) . DA neurons are the substrate for psychostimulant-induced behaviors ( Lüscher and Malenka , 2011 ) , and DAT VGLUT2 cKO mice show a blunted locomotor response to amphetamine ( Birgner et al . , 2010 ) and cocaine ( Hnasko et al . , 2010 ) . stopGLS1 HET mice also show a reduced response to acute amphetamine ( Gaisler-Salomon et al . , 2009b ) , revealing a role of PAG in amphetamine-induced responses . DAT GLS1 cHET mice responded to low ( 2 . 5 mg/Kg ) and high ( 5 mg/Kg ) doses of amphetamine indistinguishably from CTRL mice ( Figure 5E ) . For all these behavioral experiments , effect sizes were negligible for nonsignificant F values ( partial η2: Rotarod genotype effect = 0 . 0002 and interaction = 0 . 0085; Open Field genotype effect = 0 . 0064 and interaction = 0 . 0032; Center Time genotype effect = 0 . 0053 , Elevated Plus Maze genotype effect = 0 . 0002 , Context Fear Conditioning genotype effect = 0 . 013 , Acute Amphetamine genotype effect = 0 . 0010 and interaction = 0 . 0054 ) . Tone fear conditioning did show a medium effect size ( partial η2 = 0 . 067 ) , but a significant genotypic effect was not seen in a replication experiment ( Figure 6F ) . Thus attenuation of phasic GLU cotransmission does not affect motor performance , exploratory behaviors , anxiety regulation , fear conditioning or responses to acute amphetamine , revealing that several DA neuron VGLUT2-dependent and PAG-dependent behaviors were normal in DAT GLS1 cHET mice . 10 . 7554/eLife . 27566 . 022Figure 6 . DAT GLS1 cHET mice showed attenuated amphetamine sensitization and potentiated latent inhibition . ( A ) Schematic of amphetamine sensitization protocol . ( B ) Locomotor activity in the open field after vehicle ( Veh ) or Amphetamine ( Amph ) injection . There were no between group differences in activity on the habituation days ( Days 1 and 2 ) . Over the subsequent 5 treatment days , CTRL mice showed sensitization to Amph while DAT GLS1 cHET mice did not ( RM ANOVA , significant genotype X treatment X day interaction , F ( 4 , 296 ) = 4 . 4 , p=0 . 002 , ES partial η2 = 0 . 06; RM ANOVA within Amph-treated mice , significant genotype X day interaction , F ( 4 , 160 ) = 5 . 9 , p<0 . 001 , ES partial η2 = 0 . 112 ) . p<0 . 016 indicates significantly different from CTRL Amph-treated mice , after Bonferroni correction for 3 comparisons ( α = 0 . 016 ) . On the Veh challenge day ( day 18 ) , Amph-treated mice showed a modest increase in locomotion relative to Veh-treated mice independent of genotype . # indicates significant treatment effect ( F ( 1 , 74 ) = 4 . 03 , p=0 . 048; partial η2 = 0 . 052 ) , but no main effect of genotype ( F ( 1 , 74 ) < 0 . 001 , p=1 ) or significant interaction ( F ( 1 , 74 ) = 0 . 163 , p=0 . 688 ) . On the challenge day ( Day 19 ) , Amph-treated mice showed increased locomotion relative to Veh-treated mice independent of genotype . # indicates significant treatment effect ( two-way ANOVA: F ( 1 , 74 ) = 13 . 7 , p<0 . 001 , ES partial η2 = 0 . 112 ) , with no significant genotype effect ( F ( 1 , 74 ) = 2 . 76 , p=0 . 101 ) , but a trend for interaction ( F ( 1 , 74 ) = 3 . 18 , p=0 . 078 ) . ( C ) On the Amph challenge day Veh-treated ( left ) and Amph-treated mice ( right ) received Amph and activity was monitored for 90 min . Veh-treated mice showed no genotypic difference in their response to Amph ( RM ANOVA genotype effect , F ( 1 , 74 ) = 0 . 012 , p=0 . 91; genotype X time interaction , F ( 1 , 74 ) = 0 . 53 , p=0 . 83 ) . Amph-treated CTRL mice showed a sensitized response to Amph while DAT GLS1 cHET did not . ♢ # indicate a significant genotype difference ( RM ANOVA , F ( 1 , 40 ) = 89 . 3 , p=0 . 034 , ES partial η2 = 0 . 107 ) , and significant effect of time ( F ( 8 , 320 ) = 12 . 8 , p<0 . 0001 , ES partial η2 = 0 . 243 ) , but no significant interaction ( F ( 8 , 320 ) = 0 . 576 , p=0 . 798 ) . stopGLS1 mice , with a global GLS1 HET reduction , show attenuated amphetamine sensitization; see Figure 6—figure supplement 1 . ΔGLS1 HET mice , generated by breeding floxGLS1 mice with mice expressing cre under the control of the ubiquitous tamoxifen-inducible ROSA26 promoter ( Figure 6—figure supplement 2 ) , also show attenuated amphetamine sensitization ( Figure 6—figure supplement 3 ) . ( D ) Schematic of latent inhibition protocol . ( E ) On the tone test day ( Day 3 ) , the percent time freezing for the 3 min before and 8 min after CS ( tone ) presentation are shown for CTRL ( left ) and DAT GLS1 cHET mice ( right ) . CTRL non-preexposure ( NPE ) and preexposure ( PE ) groups did not differ , evidencing no LI ( RM ANOVA during CS , no preexposure effect , F ( 2 , 12 ) = 0 . 127 , p=0 . 728; nor preexposure X time interaction , F ( 7 , 84 ) = 1 . 66 , p=0 . 129 ) . DAT GLS1 cHET NPE and PE groups did not differ before CS presentation ( PE effect , F ( 1 , 20 ) = 0 . 646 , p=0 . 431; interaction , F ( 2 , 40 ) = 2 . 12 , p=0 . 132 ) ; during CS presentation , PE mice showed less freezing than NPE mice , evidencing potentiated LI ( RM ANOVA , significant time X PE treatment interaction , F ( 7 , 140 ) = 2 . 88 , p=0 . 008 , ES partial η2 = 0 . 126 ) . p<0 . 006 indicates significant different between PE and NPE groups , after Bonferroni correction for 8 comparisons ( α = 0 . 006 ) . ( F ) Percent total time freezing during 8 min CS presentation on the tone test ( Day 3 ) . DAT GLS1 cHET PE mice , but not CTRL mice , showed less freezing during CS presentation , evidencing potentiated LI ( two-way ANOVA , significant genotype X PE treatment interaction , F ( 1 , 32 ) = 5 . 3 , p=0 . 028 , ES partial η2 = 0 . 334; no significant genotype effect , F ( 1 , 32 ) = 0 . 145 , p=0 . 71 , nor PE effect , F ( 1 , 32 ) = 1 . 52 , p=0 . 227 ) . Within the NPE group , there was no genotype effect , showing that learning was not affected in DAT GLS1 cHETs ( F ( 1 , 15 ) = 1 . 56 , p=0 . 23 ) . * indicates significant pre-exposure effect within the DAT GLS1 cHET group by ANOVA ( F ( 1 , 20 ) = 10 . 03 , p=0 . 005 , ES partial η2 = 0 . 334 ) . stopGLS1 mice ( Gaisler-Salomon et al . , 2009b ) , as well as ΔGLS1 HET mice ( Figure 6—figure supplement 3 ) , both with a global GLS1 reduction , show potentiation of LI . ( G ) Schematic of the EMX1 GLS1 cHET mouse brain ( sagittal view ) illustrating the GLS1 cHET genotype in forebrain . See Figure 6—figure supplement 4 . ( H ) Novelty-induced locomotion and habituation to the open field did not differ between CTRL ( white circles ) and EMX1 GLS1 cHET mice ( grey circles ) . RM ANOVA showed a significant time effect ( F ( 5 , 170 ) = 138 . 1 , p<0 . 0001 , ES partial η2 = 0 . 802 ) ; no significant genotype effect ( F ( 1 , 34 ) = 0 . 599 , p=0 . 44 ) ; and no significant interaction ( F ( 5 , 170 ) = 0 . 820 , p=0 . 537 ) . ( I ) Both CTRL and EMX1 GLS1 cHET mice showed sensitization to Amph during the 5 treatment days ( RM ANOVA: days X drug treatment effect , F ( 4 , 128 ) = 11 . 33 , p<0 . 0001 , ES partial η2 = 0 . 259; there was no significant day X drug treatment X genotype interaction , F ( 4 , 128 ) = 0 . 161 , p=0 . 96 ) . On the Veh challenge day , there were no significant differences between genotypes of drug-treatment groups . On the Amph challenge day , Amph-treated mice showed a sensitized response relative to Veh-injected mice , independent of genotype . # indicates a significant main effect of drug treatment ( F ( 1 . 32 ) = 16 . 83 , p<0 . 0001 , ES partial η2 = 0 . 330 ) . ( J ) EMX1 GLS1 cHET mice did not show potentiation of LI . Percent time freezing during the 8 min CS presentation on the tone test day ( Day 3 ) did not differ between NPE and PE groups , independent of genotype ( two-way ANOVA: no significant main effect of genotype , F ( 1 , 35 ) = 0 . 281 , p=0 . 60; PE , F ( 1 , 35 ) = 0 . 163 , p=0 . 69; or interaction , F ( 1 , 35 ) = 0 . 586 , p=0 . 45 ) . EMX1 GLS1 cHET mice , as well as ΔGLS1 HET and DAT GLS1 cHET mice , showed clozapine-induced potentiation of LI ( Figure 6—figure supplement 5 ) . In all graphs , the number of mice is shown above the bars or next to the lines . See Figure 6—source data 1 . xlsx for source data and statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 022 10 . 7554/eLife . 27566 . 023Figure 6—source data 1 . Amphetamine sensitization and latent inhibition in DAT GLS1 cHET and EMX1 GLS1 cHET mice . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 023 10 . 7554/eLife . 27566 . 024Figure 6—figure supplement 1 . stopGLS1 HET with a global PAG reduction show attenuated amphetamine sensitization . ( A ) While potentiation of LI is seen in stopGLS1 mice ( Gaisler-Salomon et al . , 2009b ) , amphetamine sensitization had not been tested . To test for amphetamine sensitization in stopGLS1 HET mice , Amph ( 4 mg/kg ) or Veh was administered over 4 consecutive days . Amph-treated CTRL mice showed a sensitized response to Amph while Amph-treated stopGLS1 HET mice did not . A three-way RM ANOVA revealed a significant day X genotype interaction ( F ( 3 , 120 ) = 3 . 4 , p=0 . 021 , ES partial η2 = 0 . 078 ) , a treatment X genotype interaction ( F ( 1 , 40 ) = 5 . 85 , p=0 . 020 , ES partial η2 = 0 . 128 ) , and a trend for a day X treatment X genotype interaction ( F ( 3 , 120 ) = 2 . 6 , p=0 . 058 , ES partial η2 = 0 . 060 ) . Analysis of genotype and treatment effects on each day revealed significant genotype X treatment interactions on Day 3 ( F ( 1 , 40 ) = 7 . 68 , p=0 . 008 , ES partial η2 = 0 . 161 ) and Day 4 ( F ( 1 , 40 ) = 7 . 00 , p=0 . 012 , ES partial η2 = 0 . 149 ) , but not on Days 1 or 2 . Analysis of simple effects on Days 3 and 4 revealed a genotype effect within the Amph-treated groups indicated by * ( Day 3 , F ( 1 , 20 ) = 8 . 29 , p=0 . 009 , ES partial η2 = 0 . 313; Day 4 , F ( 1 , 20 ) = 9 . 13 , p=0 . 007 , ES partial η2 = 0 . 616 ) . One week later ( Day 11 ) , all mice received a lower challenge dose of Amph ( 2 mg/kg; gray shading ) . Amph-treated CTRL mice showed a significantly increased response to Amph compared to Veh-treated CTRL mice , revealing sensitization . In contrast , Amph-treated stopGLS1 HET mice showed a slightly increased response to Amph compared to Veh-treated stopGLS1 HET mice , showing reduced expression of sensitization . A two-way ANOVA revealed a significant treatment X genotype interaction ( F ( 1 , 40 ) = 4 . 5 , p=0 . 039 , ES partial η2 = 0 . 103 ) . Analyses within each genotype , showed a significant effect of drug treatment for CTRL mice ( F ( 1 , 19 ) = 24 . 8 , p<0 . 001 , ES partial η2 = 0 . 566 ) , and a trend for treatment in stopGLS1 HET mice ( F ( 1 , 21 ) = 4 . 1 , p=0 . 055 , ES partial η2 = 0 . 164 ) . * indicates significantly different from Amph-treated CTRL mice , analyses of simple main effects . ( B ) Time course of Amph-induced locomotion for CTRL and stopGLS1 HET mice on the challenge day ( Day 11 ) . There were no genotypic differences in baseline activity , prior to Amph injection . After Amph injection , Veh-treated mice — receiving Amph for the first time — showed a modest locomotor response that did not differ genotypically ( left graph ) . Amph-treated mice showed a robust locomotor response to Amph , greater in CTRL than stopGLS1 HET mice ( right graph ) . These differences were supported by a RM ANOVA within each treatment , showing no time X genotype interaction for Veh-treated mice ( F ( 8 , 160 ) = 0 . 782 , p=0 . 619 ) , but a significant time X genotype interaction for Amph-treated mice ( F ( 8 , 160 ) = 3 . 9 , p<0 . 0001 , ES partial η2 = 0 . 165 ) . Taken together , these results indicate that stopGLS1 HET mice show attenuated amphetamine sensitization . Numbers of mice are shown above the bars . Abbreviations: Amph - amphetamine; Veh - vehicle . See Figure 6—figure supplement 1—source data 1 . xlsx for source data and statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 024 10 . 7554/eLife . 27566 . 025Figure 6—figure supplement 1—source data 1 . Amphetamine sensitization in stopGLS1 HET mice . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 025 10 . 7554/eLife . 27566 . 026Figure 6—figure supplement 2 . Breeding ΔGLS1 HET mice ( with a global GLS1 reduction ) from floxGLS1 mice . ( A ) Inducible Rosa26creERT2: : GLS1lox/+ mice ( pink outline ) were used to produce a global heterozygous GLS1 inactivation in adulthood by tamoxifen-induced recombination of the floxGLS1 allele . These Rosa26creERT2: : GLS1Δ/+ mice ( gray with pink outline ) were bred with wild-type ( WT ) C57BL6 mice ( white ) to generate ΔGLS1 HET mice ( gray ) . ( B ) Expression of PAG in the hippocampus ( HIPP ) of Rosa26creERT2: : GLS1Δ/+ mice after tamoxifen revealed that the protein was reduced to 55 . 5% of control levels measured in Rosa26creERT2 mice . These mice were bred with WT mice to generate ΔGLS1 HET mice . * indicates significant difference from CTRL ( Rosa26creERT2 ) ( ANOVA , F ( 1 , 7 ) = 20 . 15 , p=0 . 003 , ES partial η2 = 0 . 742 ) . ( C ) GLS1 allelic abundance for WT and floxGLS1 alleles in ΔGLS1 HET mice showed one WT allele and the absence of the floxGLS1 allele ( blue bars ) in the hippocampus ( HIPP ) , prefrontal cortex ( PFC ) , dorsal striatum ( dStr ) , thalamus ( Thal ) and ventral midbrain ( VMB ) , further validating the global heterozygous GLS1 deletion . Allelic abundance data were normalized to CTRL values in GLS1 lox/+ ( gray line ) . See Figure 6—figure supplement 2—source data 2 . xlsx for source data and statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 026 10 . 7554/eLife . 27566 . 027Figure 6—figure supplement 2—source data 2 . PAG protein determinations in Rosa26ERT2cre GLS1 mice . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 027 10 . 7554/eLife . 27566 . 028Figure 6—figure supplement 3 . ΔGLS1 HET mice show reduced novelty-induced locomotion , attenuated amphetamine sensitization and potentiated latent inhibition . ( A ) Novelty-induced locomotion , but not habituation , was reduced in ΔGLS1 HET mice . ◊ indicates a significant main effect of genotype ( RM ANOVA: F ( 1 , 33 ) = 5 . 98 , p<0 . 020 , ES partial η2 = 0 . 153 ) . # indicates a significant main effect of time ( F ( 5 , 165 ) = 91 . 92 , p<0 . 001 , ES partial η2 = 0 . 736 ) ; there was no significant interaction ( F ( 5 , 165 ) = 0 . 942 , p=0 . 455 ) . ( B ) ΔGLS1 HET mice were tested for amphetamine sensitization following a similar protocol and drug dose to that used for DAT GLS1 cHET mice ( Figure 6A ) . After Veh injections ( Days 1 and 2 ) , ΔGLS1 HET mice were overall less active that CTRL mice . This was supported by a 2 ( genotype ) x 2 ( treatment ) x 2 ( days ) ANOVA that showed a significant main effect of genotype ( F ( 1 , 31 ) = 17 . 03 , p<0 . 001 , ES partial η2 = 0 . 355 ) indicated by ◊ , but no other significant main effects or interactions . During the 5 consecutive treatment days , Amph-treated ΔGLS1 HET mice showed both a blunted response to acute amphetamine and no sensitization . This is reflected in the 2 ( genotype ) x 2 ( treatment ) x 5 ( days ) ANOVA by the absence of a significant genotype X drug treatment X days interaction ( F ( 4 , 124 ) = 0 . 733 , p=0 . 585 ) , but a significant genotype X drug treatment interaction ( F ( 1 , 31 ) = 13 . 2 , p=0 . 001 , ES partial η2 = 0 . 299 ) . Analysis of genotype and treatment effects on each day revealed significant genotype X treatment interactions on all days except Day 4 ( Day 3 , F ( 1 , 31 ) = 10 . 14 , p=0 . 03; Day 4 , F ( 1 , 31 ) = 3 . 73 , p=0 . 063; Day 5 , F ( 1 , 31 ) = 15 . 00 , p=0 . 001; Day 6 , F ( 1 , 31 ) = 11 . 64 , p=0 . 002; Day 7 , F ( 1 , 31 ) = 10 . 56 , p=0 . 003 ) . Analysis of simple effects on Days 3 , 5 , 6 and 7 revealed a genotype effect within the Amph-treated groups indicated by * ( Day 3 , F ( 1 , 19 ) = 18 . 21 , p<0 . 001; Day 5 , F ( 1 , 19 ) = 21 . 57 , p<0 . 001; Day 6 , F ( 1 , 19 ) = 18 . 97 , p<0 . 001; Day 7 , F ( 1 , 19 ) = 17 . 82 , p<0 . 001; ES partial η2 > 0 . 400 for all ) . After a withdrawal period , on Day 19 , all mice received Amph ( 2 . 5 mg/kg ) . Amph-treated CTRL mice showed a sensitized locomotor response compared to ΔGLS1 HET mice , yet due to a ceiling effect the responses of Amph-treated and Veh-treated CTRL mice did not differ . ◊ indicates significant main effect of genotype ( two-way ANOVA: F ( 1 , 31 ) = 10 . 6 , p=0 . 003 , ES partial η2 = 0 . 254 ) . There was no significant drug treatment effect ( F ( 1 , 31 ) = 3 . 64 , p=0 . 066 ) or interaction ( F ( 1 , 31 ) = 2 . 56 , p=0 . 120 ) . Four days later ( Day 23 ) , mice received a low-dose Amph challenge ( 1 . 25 mg/kg ) . Amph-treated CTRL mice showed a sensitized response compared to Veh-treated mice and Amph-treated ΔGLS1 HET mice ( two-way ANOVA: significant genotype X drug treatment interaction , F ( 1 , 31 ) = 4 . 22 , p=0 . 048 , ES partial η2 = 0 . 120 ) . * indicates significant genotype effect for Amph-treated mice ( ANOVA , F ( 1 , 31 ) = 5 . 20 , p=0 . 034 , ES partial η2 = 0 . 215 ) . ( C ) ΔGLS1 HET mice were tested for potentiation of LI , following the same protocol as used for the DAT GLS1 cHETs ( Figure 6D ) . The graph shows percent time during the 8 min CS presentation on the test day ( Day 3 ) . There was no difference between NPE and PE CTRL groups . ΔGLS1 HET PE mice froze less during the CS exposure revealing a potentiated LI response ( two-way ANOVA: significant genotype X PE interaction , F ( 1 , 24 ) = 5 . 40 , p=0 . 029 , ES partial η2 = 0 . 183 ) . * indicates a significant PE effect for ΔGLS1 HETs ( ANOVA: F ( 1 , 10 ) = 11 . 2 , p=0 . 007 , ES partial η2 = 0 . 530 ) . Numbers of mice are shown either next to the lines or above the bars . Abbreviations: Amph - amphetamine; Veh - vehicle; PE - preexposed group; NPE - non-preexposed group . See Figure 6—figure supplement 3—source data 3 . xlsx for source data and statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 028 10 . 7554/eLife . 27566 . 029Figure 6—figure supplement 3—source data 3 . Novelty-induced locomotion and amphetamine sensitization in ΔGLS1 HET mice . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 029 10 . 7554/eLife . 27566 . 030Figure 6—figure supplement 4 . Conditional forebrain PAG reduction in EMX1 GLS1 cHET mice . ( A ) Validation of the forebrain-specific GLS1 deletion in EMX1 GLS1 cKO mice using PAG immunoreactivity . P18 mice were used , as EMX1 GLS1 cKO mice die by P21 . PAG immunoreactivity in EMX1 GLS1 cKO mice was absent in HIPP and PFC , but not Thal . ( B ) GLS1 allelic abundance for WT and floxGLS1 alleles in EMX1 GLS1 cHET mice showed that the floxGLS1 allele was reduced to 38% in the HIPP and 44% in the PFC , but not affected in the dStr and Thal , further validating the regional specificity of the EMX1cre-induced heterozygous GLS1 reduction . Allelic abundance data were normalized to CTRL values in GLS1 lox/+ ( gray line ) . ( C ) PAG protein expression in the hippocampus of EMX1 GLS1 cHET mice . PAG protein was reduced to 52% of CTRL . * indicates significant difference from CTRL ( EMX1cre ) ( one-way ANOVA , F ( 1 , 19 ) = 51 . 38 , p<0 . 0001 , ES partial η2 = 0 . 730 ) . The number of mice is shown above the bars . See Figure 6—figure supplement 4—source data 4 . xlsx for source data and statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 030 10 . 7554/eLife . 27566 . 031Figure 6—figure supplement 4—source data 4 . PAG protein determinations in EMX1 GLS1 cHET mice . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 031 10 . 7554/eLife . 27566 . 032Figure 6—figure supplement 5 . Clozapine-induced potentiation of latent inhibition in EMX1 GLS1 cHET , ΔGLS1 HET and DAT GLS1 cHET mice . ( A ) Schematic of the LI protocol to test the effect of clozapine using the same protocol used for the DAT GLS1 cHETs ( Figure 6D ) . Both NPE and PE groups received a single injection of clozapine ( 1 . 5 mg/kg ) on Day 1 , 30 min before being put in the conditioning boxes . ( B ) EMX1 GLS1 cHET mice were tested for potentiation of LI following pretreatment with clozapine . Graph shows percent freezing on the tone test ( Day 3 ) during the 8 min CS presentation . Clozapine decreased freezing in CTRL and EMX1 GLS1 cHET PE groups , revealing potentiation of LI . # indicates a significant main PE effect ( two-way ANOVA , F ( 1 , 23 ) = 13 . 13 , p=0 . 001 , ES partial η2 = 0 . 363 ) ; there was no significant main effect of genotype ( F ( 1 , 23 ) = 0 . 452 , p=0 . 508 ) or interaction ( F ( 1 , 23 ) = 0 . 002 , p=0 . 967 ) . The percent time freezing of clozapine-treated NPE groups was similar to the freezing reported for vehicle-treated NPE groups ( dashed grey line ) indicating that clozapine did not affect aversive associative learning . ( C ) ΔGLS1 HET ( left ) and DAT GLS1 cHET ( right ) mice were tested for potentiation of LI following pretreatment with clozapine . Clozapine pretreatment selectively decreased freezing in the PE groups , revealing potentiation of LI , independent of genotype . # indicates a significant main PE effect ( two-way ANOVA: for ΔGLS1 HETs , F ( 1 , 23 ) = 6 . 74 , p=0 . 016 , ES partial η2 = 0 . 227; DAT GLS1 cHETs , F ( 1 , 24 ) = 6 . 06 , p=0 . 021 , partial η2 = 0 . 202 ) ; there was no significant main effect of genotype ( ΔGLS1 HETs , F ( 1 , 23 ) = 0 . 120 , p=0 . 732; DAT GLS1 cHETs , F ( 1 , 24 ) = 0 . 069 , p=0 . 794 ) or interaction ( ΔGLS1 HETs , F ( 1 , 23 ) = 0 . 051 , p=0 . 824; DAT GLS1 cHETs , F ( 1 , 24 ) = 1 . 978; p=0 . 172 ) . The percent time freezing of clozapine-treated NPE groups was similar to the freezing reported for Veh-treated NPE groups ( dashed gray line ) indicating that clozapine did not affect aversive associative learning . The number of mice is shown above the bars . See Figure 6—figure supplement 5—source data 5 . xlsx for source data and statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 032 10 . 7554/eLife . 27566 . 033Figure 6—figure supplement 5—source data 5 . Clozapine-induced potentiation of latent inhibition . DOI: http://dx . doi . org/10 . 7554/eLife . 27566 . 033 stopGLS1 HET mice manifest a schizophrenia resilience phenotype characterized behaviorally by reduced amphetamine sensitization and potentiated LI ( Figure 6—figure supplement 1 , and Gaisler-Salomon et al . , 2009b ) ; as do ΔGLS1 HET mice , with a global Gls1 reduction , generated by breeding floxGLS1 mice with mice expressing cre under the control of the ubiquitous tamoxifen-inducible ROSA26 promoter ( Figure 6—figure supplements 2 and 3 ) . The activity of DA neurons projecting to the NAc shell , the majority of which are capable of GLU cotransmission , play a crucial role in both amphetamine sensitization and LI ( Ikemoto , 2007; Nelson et al . , 2011 ) , so we asked whether DAT GLS1 cHETs display similar behavioral phenotypes . We tested DAT GLS1 cHET mice ( P90-P120 ) for amphetamine sensitization , following the protocol schematized in Figure 6A . Two cohorts were tested , since there was no difference between the cohorts ( ANOVA cohort effect: CTRL veh , F ( 1 , 17 ) = 0 . 37 , p=0 . 872; cHET Veh , F ( 1 , 15 ) = 0 . 49 , p=0 . 494; CTRL Amph , F ( 1 , 19 ) = 0 . 94 , p=0 . 346; cHET Amph , F ( 1 , 19 ) = 3 . 752 , p=0 . 068 ) they were combined . With daily amphetamine injections ( 2 . 5 mg/kg ) over 5 days , CTRL mice showed an increase in drug-induced hyperlocomotion , characteristic of a sensitized response ( Figure 6B ) , while cHET mice showed no increase in hyperlocomotion . Ten days later , all mice were tested , first with a vehicle challenge ( Day 18 ) and then with amphetamine ( 2 . 5 mg/kg; Day 19 ) . The vehicle challenge revealed a modest but significant conditioned response in the Amph-treated groups . During the amphetamine challenge , amphetamine-treated CTRL mice showed a significant sensitized response , while amphetamine-treated cHET mice showed a significant but smaller sensitized response ( Figure 6B , gray area ) . Further comparison of the locomotor response during the 90 min post-amphetamine ( Figure 6C ) showed no difference between vehicle-treated cHET and CTRL mice , but a significantly smaller sensitized response in amphetamine-treated cHET mice compared to amphetamine-treated CTRL mice . Thus attenuating phasic GLU cotransmission blocks the induction of amphetamine sensitization and reduces the expression of sensitization , after a withdrawal period . LI is characterized by an attenuated response to a conditioned stimulus ( CS ) presented without reinforcement prior to being paired with an unconditioned stimulus ( US ) ( Weiner , 2003 ) . LI is potentiated by neurotoxin-induced loss of DA neurons projecting to the NAc shell ( Joseph et al . , 2000; Nelson et al . , 2011 ) , which would affect GLU cotransmission . We asked whether DAT GLS1 cHETs show potentiated LI , using the protocol schematized in Figure 6D . On Day 1 , mice ( P90-120 ) were assigned either to a preexposure ( PE ) group that received 20 tone exposures prior to tone ( CS ) - shock ( US ) pairing , or to a non-preexposure ( NPE ) group that received only the CS-US pairing . The number of CS pre-exposures was limited so as not to elicit LI in the PE group , enabling detection of potentiated LI . On Day 2 , freezing to context was tested in the same chamber; there was no genotypic difference between the NPE and PE groups ( CTRL NPE = 20 ± 4 . 5 s; cHET NPE = 35 ± 5 . 4 s; CTRL PE = 39 . 9 ± 5 . 9 s; cHET PE = 36 . 8 ± 5 . 7 s; two-way ANOVA; genotype factor , F ( 1 , 32 ) = 1 . 00 , p=0 . 323; preexposure factor , F ( 1 , 32 ) = 3 . 46 , p=0 . 074; interaction , F ( 1 , 32 ) = 2 . 358 , p=0 . 134 ) . On Day 3 , mice were put in a different context and presented with the CS . Less freezing during CS presentation in PE compared to NPE groups reflects potentiation of LI . During the 3 min before CS presentation , both CTRL and cHET mice showed less than 20% freezing , and there was no difference between the NPE and PE groups ( Figure 6E ) . During CS presentation , CTRL mice showed increased freezing with no difference between the NPE and PE groups ( Figure 6E , left graph ) , revealing the learned tone-fear association and no LI . In contrast , the cHET PE group showed less freezing in comparison to the NPE group , revealing potentiated LI ( Figure 6E , right graph ) . Importantly , when analyzing the total freezing during the CS presentation and comparing responses between genotypes directly , the cHET NPE group did not differ from the CTRL NPE group , showing that aversive associative learning per se was not affected in cHETs ( Figure 6F ) , replicating previous findings ( Figure 5D ) . Thus , the restricted Gls1 reduction in DA neurons is sufficient to reduce amphetamine sensitization and potentiate LI . It is striking that the behavioral phenotypes seen in GLS1 HETs were engendered by the restricted Gls1 reduction in DA neurons , and apparently do not depend on Gls1 reductions in forebrain where GLS1 mRNA and PAG are highly expressed ( Kaneko , 2000; Gaisler-Salomon et al . , 2012 ) . To verify this , we made a forebrain-restricted Gls1 reduction by breeding EMX1IREScre mice with floxGLS1 mice to generate EMX1 GLS1 cHET progeny ( Figure 6G and Figure 6—figure supplement 4 ) . EMX1 GLS1 cHETs ( P85-107 ) did not differ from CTRL mice in their novelty-induced locomotion in the open field ( Figure 6H ) and amphetamine sensitization ( Figure 6I ) . The effect size for the nonsignificant drug treatment X time X genotype interaction was negligible ( partial η2: 0 . 005 ) . EMX1 GLS1 cHETs ( P80-96 ) did not show potentiation of LI ( ES for nonsignificant PE X genotype interaction = 0 . 016 ) ( Figure 6J ) . To confirm in EMX1 GLS1 cHETs that the limited number of pre-exposures did not elicit LI and yet was sufficient to reveal potentiation of LI , we tested for clozapine-induced potentiation of LI ( Gaisler-Salomon et al . , 2009b ) ( Figure 6—figure supplement 5A ) . In both CTRL and EMX1 GLS1 cHETs , clozapine treatment on Day 1 , potentiated LI in the PE groups ( Figure 6—figure supplement 5B ) , but had no effect in the NPE groups showing that it did not affect learning . Similar clozapine effects were seen in ΔGLS1 HET and DAT GLS1 cHET mice ( Figure 6—figure supplement 5C ) . The lack of further potentiation of LI in ΔGLS1 HET and DAT GLS1 cHET mice suggests that clozapine treatment and Gls1 deficiency in DA neurons each either induce maximal potentiation of LI , or involve shared mechanisms so that Gls1 deficiency occludes clozapine-induced potentiation of LI . In summary , our results argue that reducing Gls1 in DA neurons is not only sufficient but also necessary for the reduction of amphetamine sensitization and potentiation of LI . A stereological analysis of PAG expression in DA neurons revealed that about half of DA neurons express PAG in both the VTA and SNc , in contrast to VGLUT2 expression , which is mostly restricted to DA neurons in the VTA ( Yamaguchi et al . , 2015 ) . The function of PAG in SNc DA neurons incapable of GLU cotransmission is still uncertain , although we show that a minor reduction of PAG expression in those neurons had no impact on their survival or intrinsic physiology , nor did it affect DA transmission in the dStr or motor behaviors controlled by the dStr . In contrast , DA neurons capable of GLU cotransmission ( TH+ / VGLUT2+ cells ) preferentially express PAG , and a reduction of PAG in those VTA DA neurons was sufficient to attenuate phasic GLU cotransmission in the NAc shell and impact behaviors controlled by the NAc , revealing the important role of PAG in DA neuron GLU cotransmission . The reduction in PAG activity of about 20% seen in stopGLS1 HET brain slices ( El Hage et al . , 2012 ) is associated with about a 15% reduction in GLU content ( Gaisler-Salomon et al . , 2009b ) that presumably reflects a presynaptic diminution , since the highest concentrations of GLU are intracellular ( Danbolt , 2001 ) . Decreases in presynaptic GLU lead to decreases in vesicular GLU content and synaptic efficacy ( Ishikawa et al . , 2002 ) . In DAT GLS1 cHETs , the first EPSC elicited by burst photostimulation was unaffected , as was observed in cultured GLS1 KO neurons ( Masson et al . , 2006 ) , indicating that the readily releasable vesicle pool is replete . Smaller subsequent responses may reflect either diminished filling of the recycling pool ( Alabi and Tsien , 2012 ) , or decreased probability of release of vesicles with diminished GLU content ( Iwasaki and Takahashi , 2001 ) . PAG expression in VTA DA neurons is weak to moderate relative to other brain regions ( Kaneko , 2000 ) . The fact that a heterozygous GLS1 reduction in DA neurons is sufficient to decrease synaptic efficacy indicates that PAG levels are not only lower but rate limiting . Single cell RT-PCR studies show that DA neurons also have low VGLUT2 mRNA copy numbers ( Trudeau et al . , 2014 ) . Lower VGLUT2 expression would place further demands on the GLU-glutamine cycle to sustain synaptic transmission during periods of high activity , given that vesicular loading depends both on cytosolic GLU concentration and vesicular transporter number ( Wilson et al . , 2005 ) . This indicates that the DA neuron GLU cotransmission in DAT GLS1 cHETS is highly dependent on PAG activity , and suggests that the global reduction in PAG activity in global GLS1 HETs affects DA neuron GLU cotransmission preferentially . Discerning the behavioral role of DA neuron GLU cotransmission has been challenging because of the impact of knocking out VGLUT2 in DA neurons on DA function . In DAT VGLUT2 cKOs , DA neuron function is affected profoundly due to the developmental role of VGLUT2 in DA neurons ( Fortin et al . , 2012 ) . In DAT GLS1 cHETs , DA neuron DA functions appear normal; Gls1 reduction affects neither the survival of DA neurons nor their intrinsic electrophysiological properties . VGLUT2 also plays an important role in vesicular DA uptake ( Hnasko et al . , 2010 ) , but there was no impact of Gls1 deficiency on DA content or release , even when DA terminals were stimulated repeatedly to increase the demand on DA release . Since DAT GLS1 cHET DA neurons show normal GLU cotransmission with low-frequency activity , our results suggest that modestly reduced presynaptic GLU is sufficient for the maintenance of normal vesicular DA dynamics in adulthood . Alternately , DA neuron GLU release may arise from segregated release sites ( Zhang et al . , 2015 ) , so reduced vesicular GLU filling would not affect DA release . In the absence of a direct effect on synaptic DA transmission , finding that GLU signaling with high-frequency activity was affected selectively in DAT GLS1 cHETs allowed us to focus on the function of GLU cotransmission . DAT GLS1 cHET mice do not show several behavioral phenotypes of DAT VGLUT2 cKOs , such as decreased novelty-induced locomotion , motor deficits on the rotarod , an anxiety phenotype , or blunted responses to psychostimulants ( Birgner et al . , 2010; Hnasko et al . , 2010; Fortin et al . , 2012 ) . Presumably the behaviors not affected by a mild disruption in GLU cotransmission , are sensitive to manipulations that affect both DA and GLU transmission , such as in DAT VGLUT2 cKO mice . Strikingly , the subtle activity-dependent reduction in DA neuron GLU cotransmission in DAT GLS1 cHETs had major effects on amphetamine sensitization and LI , arguing that DA neuron GLU cotransmission is a key regulator of these behaviors . DA signaling increases with psychostimulant sensitization ( Vezina , 2004; Bocklisch et al . , 2013; Covey et al . , 2014 ) . While DA neuron DA signaling was not affected in DAT GLS1 cHETs , changes in DA signaling with repeated psychostimulant administration are likely , although attenuated due to reduced DA neuron GLU cotransmission . DA neuron excitatory connections to SPNs in the NAc core are modestly but significantly increased weeks after chronic psychostimulant administration ( Ishikawa et al . , 2013 ) ; psychostimulant-induced plasticity may be even stronger at DA neuron excitatory connections to ChIs in the NAc shell ( Chuhma et al . , 2014 ) . At the VTA-NAc circuit level , reducing GLU cotransmission may attenuate increases in DA neuron activity associated with sensitization ( Bocklisch et al . , 2013 ) . While subsequent circuit effects impacted by reduced GLU cotransmission involve DA signaling , we show here for the first time that the attenuation of GLU cotransmission in the absence of developmental alterations and direct effects on DA transmission has strong and selective behavioral effects , revealing a new mechanism through which DA neurons control behavior . DA neuron activity mediates both amphetamine sensitization and LI by encoding motivational salience of relevant events ( Young et al . , 2005; Bromberg-Martin et al . , 2010; Robinson et al . , 2016 ) . Our results suggest that DA neuron GLU signaling plays a key role in salience attribution . In amphetamine sensitization , increases in DA neuron firing are restricted to medial VTA DA neurons ( Lodge and Grace , 2012 ) , the majority of which are capable of GLU cotransmission ( Yamaguchi et al . , 2015 ) . Recent evidence suggests that all abused drugs increase DA neuron activity to strengthen the motivational salience of drug exposure or associated events ( Covey et al . , 2014 ) . DAT GLS1 cHET mice do not show sensitization to amphetamine with repeated administration , and after a withdrawal period show reduced expression of sensitization . Similar results were found in mice with a conditional NR1 deletion in their DA neurons that resulted in a dramatic reduction in phasic firing ( Zweifel et al . , 2009 ) . While the development of sensitization was unaffected in DAT NR1 cKO mice , the mice showed reduced expression of sensitization weeks after withdrawal ( Zweifel et al . , 2008 ) . Taken together , several lines of evidence suggest that phasic DA neuron GLU signals facilitate sensitization by determining how rapidly and efficiently pathological levels of salience are attributed to drug exposure . In contrast , DAT NR1 cKO showed a reduction in conditioned responses to context not seen in the present study , suggesting that the abrogation of both DA and GLU phasic transmission must be affected to impact the development of drug-induced conditioned responses . In LI , it is thought that the activity of DA neurons in the NAc updates the salience of a preexposed stimulus during the conditioning phase by integrating previous with current behavioral experiences ( Young et al . , 2005 ) . Thus , the potentiation of LI seen in DAT GLS1 cHET mice represents a failure of DA neurons to increase the salience of the inconsequential preexposed stimulus under changed reinforcement contingencies during conditioning . The temporal precision of the DA neuron GLU signal makes it particularly suitable for updating salience . In the NAc medial shell , a structure known to regulate motivational salience ( Ikemoto , 2007 ) , DA neuron GLU connections to ChIs drive them to fire in bursts ( Chuhma et al . , 2014 ) . Direct optogenetic excitation of ChIs in the NAc shell — as would result from DA neuron GLU actions — does not drive reinforcement learning on its own but instead modulates learning ( Lee et al . , 2016 ) . It has been recently shown that GLU cotransmission is not required for self-administration reinforced by DA neuron activation ( Wang et al . , 2017 ) . Taken together with our behavioral results , this suggests that DA neuron GLU signals modulate learning by regulating the attribution of motivational salience to relevant events via their direct control of ChI activity . In the context of schizophrenia , the behaviors affected in DAT GLS1 cHETs align with the schizophrenia resilience phenotype of stopGLS1 HET ( Gaisler-Salomon et al . , 2009b ) , as well as ΔGLS1 HET mice , both with a global GLS1 heterozygous reduction . Several other phenotypes of GLS1 HETs , a blunted locomotor response to novelty , diminished sensitivity to acute amphetamine or reduced contextual fear conditioning were not seen in DAT GLS1 cHETs and so apparently do not depend on GLU cotransmission . Furthermore , none of these behavioral deficits were recapitulated in EMX1 GLS1 cHETs , with a forebrain-restricted GLS1 reduction , demonstrating that PAG in DA neurons is necessary for amphetamine sensitization and potentiation of LI , and reinforcing the likelihood that DA neuron GLU cotransmission is particularly sensitive to PAG reduction . Modeling resilience in mice using transgenic approaches offers a direct path to intervention , as resilience mutations point directly to therapeutic targets ( Mihali et al . , 2012 ) . Supported by the recent demonstration of VGLUT2 expression — and thus of GLU cotransmission — in primate DA neurons ( Root et al . , 2016 ) , the therapeutic potential of PAG inhibition as a pharmacotherapy for schizophrenia ( Mingote et al . , 2016 ) may involve tempering DA neuron GLU cotransmission . Finally , our findings put forward the possibility that an increase in GLU cotransmission in the NAc may contribute to the pathophysiology of schizophrenia , in particular to aberrant salience leading to psychosis . Increased activity in the midbrain and NAc has been associated with aberrant salience attribution to irrelevant stimuli in patients with psychosis or individuals at high risk ( Romaniuk et al . , 2010; Roiser et al . , 2013 ) , while increased NAc activity does not correlate with increased dopamine synthesis capacity ( Roiser et al . , 2013 ) . This inconsistency would be reconciled if increased activity in the VTA and NAc in SCZ is associated with a greater pathological increase in GLU cotransmission than in DA transmission . We used stopGLS1 ( JIMSR Cat# JAX:017956 , RRID:IMSR_JAX:017956 , Gaisler-Salomon et al . , 2009b ) and floxGLS1 mice ( IMSR Cat# JAX:017894 , RRID:IMSR_JAX:017894 , Mingote et al . , 2016 ) , both on a 129SVE-F background , and DATIREScre ( RRID:IMSR_JAX:006660 ) , EMX1IREScre ( RRID:IMSR_JAX:005628 ) and Rosa26creERT2 mice ( RRID:IMSR_JAX:008463 ) on a C57BL/6 background . These mice were used to generate DAT GLS1 cHET or cKO mice , EMX1 GLS1 cHET or cKO mice , and ΔGLS1 HET mice , all on a mixed 129SVE-F and C57BL/6 background . For optogenetic stimulation , mice were bred with Ai32 ( RCL-ChR2 ( H134R ) /EYFP ) mice ( RRID:IMSR_JAX:024109 ) mice to confer conditional expression of ChR2 in their DA neurons . Inducible Rosa26creERT2::GLS1lox/+ mice were used to produce a global heterozygous GLS1 deletion in adulthood by administration of tamoxifen . Tamoxifen ( Sigma-Aldrich , T5648 ) was dissolved in a peanut oil/ethanol ( 9:1 mixture ) at 25 mg/ml , solubilized by vortexing for 5 min and warming to 37°C for several hours . Mice received 0 . 2 mL i . p . ( 5 mg Tamoxifen ) daily for 5 successive days . Tamoxifen-treated Rosa26creERT2::GLS1Δ/+ mice were then crossed with wild-type C57BL/6 mice ( RRID:IMSR_JAX:000664 ) to generate ΔGLS1 HETs . Mice were anesthetized with ketamine ( 90 mg/kg ) + xylazine ( 7 mg/kg ) and perfused with cold PBS followed by 4% paraformaldehyde ( PFA ) , the brains removed , post-fixed overnight in 4% PFA , and cut at 50 µm with a vibrating microtome ( Leica VT1200S ) . Coronal slices were collected into a cryoprotectant solution ( 30% glycerol , 30% ethylene glycol in 0 . 1 M Tris HCl [pH 7 . 4] ) and kept at −20°C until processing . Sections were washed in PBS ( 100 mM; pH 7 . 4 ) and incubated in glycine ( 100 mM ) for 30 min to quench aldehydes . Non-specific binding was blocked with 10% normal goat serum ( NGS; Millipore ) in 0 . 1% PBS Triton X-100 for 2 hr ( PBS-T ) . Primary antibodies used were anti-TH ( 1:10 , 000 dilution , mouse monoclonal , Millipore Cat# MAB318 RRID:AB_2201528 ) , anti-PAG ( 1:10 , 000 dilution , rabbit polyclonal , Norman Curthoys , Colorado State ) , and anti-GFP ( 1:2000 dilution; rabbit polyclonal , Millipore Cat# AB3080 - RRID:AB_91337 ) . Secondary antibodies were: anti-rabbit Alexa Fluor 488 ( 1:200 dilution , ThermoFisher Scientific Cat# A-21206 RRID:AB_2535792 ) and anti-mouse Alexa Fluor 594 ( ThermoFisher Scientific Cat# A-21203 RRID:AB_2535789 ) . Primary antibodies in 0 . 02% PBS-T and 2% NGS were applied for 24 hr at 4°C . Sections were then washed with PBS and secondary antibodies applied for 45 min in 0 . 02% PBS-T at room temperature . Sections were mounted on slides and cover slipped with Prolong Gold aqueous medium ( ThermoFisher Scientific ) and stored at 4°C . Fluorescence images were acquired with a Fluoview FV1000 ( Olympus ) or A1 ( Nikon ) confocal laser scanning microscope , or a Axiovert 35M ( Zeiss ) epifluorescence microscope . The SNc and VTA were delineated based on low-magnification images of TH immunostaining . Stereological counts were made of DA neurons using the Optical Fractionator Probe in Stereo Investigator ( MBF Bioscience ) at regular predetermined intervals ( grid size: x = 170 μm , y = 120 μm ) with an unbiased counting frame ( x = 55 μm , y = 33 . 6 μm; dissector height , z = 33 . 6 μm ) . The actual mounted section thickness averaged 24 μm ( 50% shrinkage from the unprocessed section thickness ) . Sampling was done from acute ventral midbrain slices . Anesthetized mice ( male or female WT or DAT GLS1 cKO and littermate control mice ) were decapitated and brains quickly removed in ice-cold high-glucose artificial cerebrospinal fluid ( aCSF; in mM: 75 NaCl , 2 . 5 KCl , 26 NaHCO3 , 1 . 25 NaH2PO4 , 0 . 7 CaCl2 , 2 MgCl2 and 100 glucose , adjusted to pH 7 . 4 ) . 300 µm coronal midbrain sections were cut on a vibrating microtome ( Leica VT1200S ) . Sections were preincubated for at least one hour at room temperature in high sucrose aCSF saturated with carbogen ( 95% O25% CO2 ) , then mounted in a chamber on the stage of an upright microscope ( Olympus BX61WI ) continuously perfused with standard aCSF ( in mM: 125 NaCl , 2 . 5 KCl , 25 NaHCO3 , 1 . 25 NaH2PO4 , 2 CaCl2 , 1 MgCl2 and 25 glucose , pH 7 . 4; perfusion 1 ml/min ) saturated with 95% O25% CO2 . Sampling was done from the VTA and SN , using the medial lemniscus as the dividing boundary . Glass pipettes for sampling were fabricated from thin wall glass capillaries ( Harvard Apparatus ) , which were cleaned with water and ethanol and then treated at 200°C for 4 hr to inactivate RNase . Pipettes were filled with 5 µl DEPC treated water . Whole cell recordings were made using digitally enhanced DIC optics , at room temperature ( 21–23°C ) . The cytosol of single neurons was aspirated using a glass pipette . In most cases , the nucleus was aspirated along with the cytosol . The sampled single-cell cytosol was ejected in a 0 . 2 ml PCR tube with a sample mixture of 0 . 5 µl dithiothreitol ( DTT; 0 . 1 M , Invitrogen ) , 0 . 5 µl RNase inhibitor ( RNaseOUT , 40 U/ml , Invitrogen ) , 1 µl random hexamers ( 50 µM , Applied Biosciences ) and 5 µl DEPC treated water . Sampling was done and the tubes with sample mixture were kept on ice until reverse transcription . The sample mixture was treated at 70°C for 10 min . The second mixture ( 4 µl x5 ) was added to the sample mixture . First strand buffer ( Invitrogen ) , 0 . 5 µl RNase inhibitor , 1 µl dNTP mix ( 10 mM , Invitrogen ) , 1 . 5 µl DTT , and 1 µl reverse transcriptase ( SuperScript III , 200 U/µl , Invitrogen ) . Reverse transcription was done at 50°C for 50 min , and stopped by raising the temperature to 85°C for 5 min . Subsequently , 0 . 5 µl RNase ( 2 u/µL , Invitrogen ) was added to each tube and incubated at 37°C for 20 min to eliminate RNA contamination . The cDNA produced by reverse transcription was frozen at −80°C pending PCR analysis . After reverse transcription , cDNA was amplified by nested PCR . First round PCR primers spanned at least one intron to preclude amplification of genomic DNA . TH and GAD67 primer sequences for both first and second round PCR were obtained from Liss et al . ( 1999 ) ; VGLUT2 primer sequences for the second round were obtained from Mendez et al . ( 2008 ) . VGLUT2 first round primers and GLS1 primers for both the first and second round PCR were custom designed , with the following sequences ( 5’ to 3’ ) : VGLUT2 first round upper cacccgcccaaataccacgg and lower gccccaaagacccggttagc; GLS1 first round upper ttgttgtgacttctctaat and lower atggtgtccaaagtgtag; GLS1 second round upper gtggcatgtatgacttct and lower atggtgtccaaagtgtag . Products of the second round PCR were confirmed by sequencing , and had the following sizes ( in bp ) : TH 377 , GAD67 702 , VGLUT2 250 and GLS1 512 . Both first and second round amplifications was done with the following temperature cycle: 3 min at 94°C , 35 cycles of 30 s at 94°C , 1 min at 58°C , 3 min at 72°C , followed by 7 min at 72°C . 2 µl of the first round PCR product was used for the second round . PCR products were separated on 1 . 5% agarose gels . Only clear bands were counted as positive; runs with unclear bands or bands of incorrect size were discarded . We used male and female juvenile ( P30 ) DATIREScre/+ and littermate controls . Mice were anesthetized with ketamine/xylazine . The ventral midbrain and dorsal striatum were dissected and put in tubes with 300 μl Qiazol ( Qiagen ) , a RNase-inhibitor buffer , and rapidly frozen on dry ice . RNA extraction was done using the RNeasy Lipid Mini Kit ( Qiagen ) , according to the manufacturer’s instructions , and stored in RNase-free water at −80°C until further processing . RNA concentrations were standardized to 1 μg per 10 μl water using a NanoDrop 1000 Spectrophotometer ( ThermoScientific ) . The 260:280 nm absorbance ratio was measured to assess RNA quality; samples were excluded if the ratio was outside the range 2 . 0 ± 0 . 2 , or if the RNA concentration was too low . Genomic DNA elimination was performed using RNase-free DNase set ( Qiagen ) . Reverse transcription was carried with the RT2 first-strand kit ( Qiagen ) . Reverse transcription product ( cDNA ) was diluted to a volume of 1 ml in water . The real time quantitative PCR ( RT-qPCR ) was performed using an Opticon 2 DNA Engine ( Bio-Rad ) and microprofiler plates with primers designed by SuperArray Biosciences ( Qiagen ) . The primers were custom designed to recognize cDNA for DAT , D1 and D2 receptors , TH , VMAT2 . The cycle threshold ( Ct ) values were normalized to GAPDH ( ΔCt ) . Relative copy number was obtained by exponentiation of ΔCt values ( function 2-ΔCT ) multiplied by 1000 . We used male and female adult ( P90-150 ) ΔGLS1 HET mice and littermate controls , or EMX1 GLS1 cHETs and littermate controls . Mice were anesthetized with ketamine+xylazine , decapitated and brains quickly removed to ice-cold saline for dissection . The hippocampus , prefrontal cortex , striatum , thalamus and ventral midbrain , dissected from one hemisphere , were put in 96-well plates and sent to Transnetyx ( Cordova , TN ) for quantitative genotyping using probe-based quantitative PCR ( qPCR ) . Allelic abundance was obtained from the mean of 4 qPCR determinations ( 2 runs done in duplicate ) . The floxGLS1 and WT allele signals were normalized to the one-allele signal from floxGLS1 heterozygous mice . Recordings in the NAc shell were made from 300 µm coronal striatal slices , as described previously ( Chuhma et al . , 2011 ) . Animals were anesthetized with ketamine+xylazine . Brains removed into ice-cold high-glucose aCSF saturated with carbogen ( 95% O25% CO2 ) . The composition of the high-glucose aCSF was , in mM: 75 NaCl , 2 . 5 KCl , 26 NaHCO3 , 1 . 25 NaH2PO4 , 0 . 7 CaCl2 , 2 MgCl2 and 100 glucose , adjusted to pH 7 . 4 . After 1 hr incubation in high-sucrose aCSF at room temperature to allow slices to recover , slices were placed in a recording chamber with continuous perfusion of standard aCSF equilibrated with carbogen , and maintained at 30–32°C ( TC 344B Temperature Controller , Warner Instruments ) . Expression of ChR2 was confirmed by visualization of EYFP fluorescence in DA neuron axons and varicosities . Whole-cell patch recording followed standard techniques using glass pipettes ( 5–8 MΩ ) . For voltage clamp experiments , a cocktail of antagonists was included in the perfusate to isolate AMPA-mediated responses: SR95531 10 µM ( GABAA antagonist ) , CGP55345 3 µM ( GABAB antagonist ) , SCH23390 10 µM ( D1 antagonist ) , ( - ) -sulpiride 10 µM ( D2 antagonist ) , scopolamine 1 µM ( muscarinic antagonist ) and D-AP5 50 µM ( NMDA antagonist ) ( all from Tocris Bioscience ) . Patch pipettes were filled with intracellular solution containing ( in mM ) 140 Cs+-gluconate ( voltage clamp recordings ) or 140 K+-gluconate ( current clamp recordings ) , 10 HEPES , 0 . 1 CaCl2 , 2 MgCl2 , 1 EGTA , 2 ATP-Na2 and 0 . 1 GTP-Na2 ( pH 7 . 3 ) . The Na+-channel blocker lidocaine N-ethyl bromide ( QX-314 , 5 mM , Sigma-Aldrich ) was added to the intracellular solution in voltage clamp experiments to block active currents . For current clamp experiments , no drugs were added to the perfusate . Recordings were made with an Axopatch 200B ( Molecular Devices ) ; for voltage clamp recordings ( holding potential −70 mV ) , series resistance ( 6–35 MΩ ) was compensated online by 70–80% . Liquid junction potentials ( 12–15 mV ) were adjusted online . ChR2 responses were evoked by field illumination with a high-power blue ( 470 nm ) LED ( ThorLabs ) . GLU mediation was confirmed by blockade with 40 µM 6-cyano-7-nitroquinoxaline-2 , 3-dione ( CNQX , Tocris Bioscience ) . Data were filtered at 5 kHz with a 4-pole Bessel filter , digitized ( InstruTECH ITC-18 Interface , HEKA ) at 5 kHz , and analyzed using Axograph X ( Axograph Scientific ) . Recordings from putative DA neurons in adult ( P59–P64 ) DAT GLS1 cHET mice and CTRL littermates were made in 300 µm VTA/SNc horizontal slices , blinded to genotype . The medial optic tract defined the boundary between the SNc and the VTA . SNc neurons showing slow pacemaker firing and a prominent Ih were identified as DA neurons; in the lateral VTA , large neurons with slow pacemaker firing and a prominent Ih were always DAT-driven reporter positive ( Chuhma , unpublished observation ) . VTA neurons in the medial VTA with these properties are not always TH+ ( Margolis et al . , 2010 ) , so VTA recordings were restricted to the lateral VTA . Whole-cell patch recordings were made with borosilicate glass pipettes ( 3–6 MΩ ) with intracellular solution containing ( in mM ) : 135 K+-methanesulfonate , 5 KCl , 2 MgCl2 , 0 . 1 CaCl2 , 10 HEPES , 1 EGTA , 2 Na2-ATP , 0 . 1 GTP ( pH 7 . 3 ) , using an Axopatch 200B in fast current clamp mode . Since DA neurons were spontaneously active , resting membrane potential was measured as the average of the pacemaker fluctuation of the membrane potential after action potentials were truncated . Input impedance was measured with −100 pA current pulses . Action potential threshold was determined as the point where membrane potential change exceeded 10 mV/ms , using AxographX automatic detection . Recordings were done in adult ( P71–P85 ) DAT GLS1 cHET::ChR2 and CTRL::ChR2 mice , in 300 µm coronal slices through the striatum , as described previously for the electrophysiology experiments . DA release was evoked by photostimulation ( blue high-power LED ) and measured using carbon fiber electrodes , calibrated to 1 µM DA , post-experiment . A triangle wave ( −450 to +800 mV at 312 . 5 V/sec vs . Ag/AgCl ) was applied to the electrode at 10 Hz . Fibers were conditioned in the brain slice by cycling the fiber for 20–30 min or until the current stabilized . Current was recorded using an Axopatch 200B filtered at 10 kHz with a 4-pole Bessel filter , digitized at 25 kHz ( ITC-18 ) using Igor Pro 6 ( WaveMetrics ) and analyzed with MATLAB R2014b ( MathWorks ) . The apparent DA oxidation peak in response to the applied triangle wave in cyclic voltammetry is determined by the pipette offset of the amplifier , which is used to correct for resistance between the ground and carbon fiber electrodes . In this study , we made no pipette offset adjustment , which decreased the instances that the current signal overloaded the amplifier . However , the DA oxidation peak , which is normally reported at ~+600 mV with a typical offset of ~+150 mV , was left shifted and occurred at ~+400 mV without an offset . We found in calibrations that the absence of pipette offset did not affect the peak amplitude of the DA response . To measure tissue DA and DOPAC content , mice underwent cervical dislocation; brains were removed rapidly and flash frozen in isopentane . Tissue samples were obtained from 1 mm circular punches from approximately 1 mm thick coronal sections , weighed , placed in 200 µl of HeGA preservative solution ( 0 . 1 M Acetic Acid , 0 . 105% EDTA , 0 . 12% Glutathione , pH 3 . 7 ) , homogenized ( 150 VT Ultrasonic homogenizer; Homogenizers . net ) , centrifuged and supernatant frozen at −80°C pending analysis . Samples were separated by HPLC coupled to an electrochemical detector . DA and DOPAC were separated with a reverse phase C18 column ( ChromSep SS 100 × 3 . 0 mm , Inertsil 3 ODS-3; Varian , Palo Alto , CA ) and a mobile phase containing: 75 mM NaH2PO4 , 25 mM citric acid , 25 µM EDTA , 100 µl/L tetraethylamine , 2 . 2 mM octanesulfonic acid sodium salt , 10% acetonitrile , 2% methanol , pH 3 . 5 . DA was oxidized with a coulometric electrode ( Model 5014; ESA , Chelmsford , MA ) , with conditioning cell set to a potential of −150 to −200 mV and the analytical cell set to a potential of 350 mV . The concentration of DA and its metabolites was quantified using an external standard curve from standards prepared in the same aCSF/preservative mixture as the brain dialysates . Protein analysis was performed using the Simon Simple Western assay ( ProteinSimple ) . Hippocampal tissue samples were dissected and homogenized in 100 µL lysis solution . Lysis solution was prepared by mixing 1 mL of 1x lysis buffer ( Cell Signalling Technology , 9803 ) containing 1 µl calyculin A and 0 . 5 µl okadaic acid ( protein phosphatase inhibitors from Sigma-Aldrich , C5552 and 08010 respectively ) and 5 µl of protease inhibitor cocktail ( Sigma-Aldrich , P8340 ) . After homogenization , the lysate was centrifuged at 12 , 000 rpm for 30 min at 4°C . The supernatant was transferred to new tubes and frozen at −80°C pending subsequent analysis . Tissue samples were diluted to a concentration of 0 . 2 mg/mL in ProteinSimple sample buffer . A master mix containing 10x sample buffer , 1M DTT , and 10x fluorescent standard was added to the samples , which were then loaded in the first row of a ProteinSimple cassette . A mixture of two rabbit polyclonal antibodies was loaded in the second row: PAG antiserum ( Curthoys et al . , 1976 ) diluted 1:200 and GAPDH ( 14C10 ) ( Cell Signalling Technology , 2118S; AB Registry ID: AB_2107301 ) diluted 1:25 . The luminol-S/peroxide chemiluminescent detection mixture was loaded in the third row . Size-based separation , immunoprobing , washing , and detection were done automatically by the Simon , which in an automated sequence drew up the sample mixture , the antibodies , and then the detection reagent into a capillary array . Chemiluminescence was measured along the length of the capillary over time , and analyzed using ProteinSimple Compass software . Sample size estimates were made using G*Power ( Faul et al . , 2007 ) . Sample sizes were calculated using a power of 0 . 80 and an α of 0 . 05 , as we assumed that a 4:1 ratio between type 1 and type 2 errors was appropriate for all our experiments ( Keppel , 1991 p . 75 ) . The predicted effects sizes were different for the behavioral , electrophysiology/voltammetry , and stereology experiments . Since we were assessing the effects of a conditional heterozygous manipulation , for the behavioral studies we predicted a medium effect size of 0 . 06 ( partial η2 ) , which resulted in an estimated sample size range between 17 to 51 mice per group ( rotarod = 17; elevated plus maze short arms = 51; novelty-induced locomotion = 22; amphetamine sensitization = 21 ) . After running these first experiments in sequence using samples sizes within the estimated range , we obtained significant F values with effects sizes ranging from 0 . 06 to 0 . 15 and a better than predicted power of 0 . 9 , which led us to use smaller samples size in subsequent experiments ( elevated plus maze longer arms , acute amphetamine , fear conditioning , latent inhibition ) . For the electrophysiology and voltammetry studies , which measured the direct effects of the conditional heterozygous manipulation on synaptic release , we predicted an effect size of 0 . 1 , which resulted in an estimated sample size of 12 per group . For the stereology experiments , we estimated a larger effect size of 0 . 2 based on previous experiments and pilot studies , for a sample size of 4 per group . In Figure 1 , the stereological estimate of the number of TH+ only , PAG+ only and TH+ / PAG+ cells in the VTA and SNc of juvenile wild-type mice was analyzed using a 3 ( cell type ) X 2 ( brain region ) ANOVA . For the comparison between the relative number of TH+ / PAG+ cells in juvenile ( P25 ) and adult ( P60 ) mice , a 2 ( age ) X 2 ( brain region ) ANOVA was used . For the single cell RT-PCR data , the Chi-Square test was used to determine whether TH+ / VGLUT2+ neurons preferentially expressed PAG . In Figure 3 , comparison of response amplitudes to single photostimulation was analyzed using a 2 ( genotype ) X 2 ( cell type ) ANOVA . Comparison between genotypes of first response amplitude to burst photostimulation was done for each cell type separately using the nonparametric Mann-Whitney test , since samples were not normally distributed . For the analysis of the amplitude of EPSCs induced by repeated burst photostimulation , data was converted to percent of the first response amplitude and analyzed for each cell type separately using a 2 ( genotype ) X 4 ( pulses , repeated measures factor ) ANOVA . Only the results obtained from ChIs revealed a significant genotype X pulses interaction , which led us to conduct further analysis of simple effects involving the non-repeated measures factor ( genotype ) to detect the source of the interaction . To control for increased family-wise type 2 errors due to multiple comparisons , we applied the Bonferroni correction for simple effects and using α = 0 . 0125 . Finally , the analysis of the ratio of firing during burst ( 0–0 . 5 s from onset of train ) and just after burst photostimulation ( 0 . 5–1 s from onset ) was done using a one-way ANOVA . In Figure 4 , genotypic differences in numbers of TH+ neurons , DA content and DOPAC/DA ratio values were evaluated with one-way ANOVAs . For voltammetry data , the peak amplitude of DA release evoked by consecutive bursts of photostimulation followed by a single , or consecutive single pulses followed by burst , was analyzed using a 2 ( genotype ) x 4 ( pulses , repeated measures factor ) ANOVA . In Figure 5 , the latency to fall from the rotarod was analyzed using a 2 ( genotype ) x 9 ( trials , repeated measures factor ) ANOVA . Locomotor counts in the open field were analyzed using a 2 ( genotype ) x 6 ( time , bins of 10 mins , repeated measures factor ) ANOVA . Behavior in the elevated plus maze and fear conditioning chambers was analyzed using a one-way ANOVA to evaluate genotypic effects . Dose effects in amphetamine-induced locomotion were analyzed using a 2 ( genotype ) x 3 ( dose ) ANOVA . In Figure 6 , for the sensitization experiment , locomotor activity during the first 2 habituation days ( vehicle injections ) was analyzed separately using a 2 ( genotype ) x 2 ( drug treatment ) x 2 ( days , repeated measure factor ) ANOVA . Locomotor activity during the subsequent 5 test days ( vehicle or amphetamine injections ) was analyzed using a 2 ( genotype ) x 2 ( drug treatment ) x 5 ( days , repeated measure factor ) ANOVA . A significant three-way interaction was further analyzed for simple effects . Within each drug treatment , a 2 ( genotype ) x 5 ( days , repeated measure factor ) ANOVA was used . Only within the amphetamine-treated groups was there a significant genotype X day interaction , which allowed us to conduct a further analysis of simple effects involving the non-repeated measures factor ( genotype ) . Comparisons during the last 3 days of injections were corrected by a Bonferroni adjustment ( α = 0 . 016 ) . The data from the challenge day were analyzed separately using a 2 ( genotype ) X 2 ( drug treatment ) ANOVA . In addition , data obtained during the 90 min following injections was analyzed separately for each amphetamine- and vehicle-treated group using a 2 ( genotype ) x 9 ( time , bins of 10 min , repeated measure factor ) ANOVA . For the latent inhibition experiment ( in Figure 6 ) , freezing before CS presentation was analyzed separately for each genotype using a 2 ( preexposure treatment ) X 2 ( time , bins of 1 min , repeated measure factor ) ANOVA . After CS presentation data were analyzed using a 2 ( preexposure treatment ) X 8 ( time , bins of 1 min , repeated measure factor ) ANOVA . A significant preexposure X time interaction was found for DAT GLS1 cHET mice , allowing us to examine simple effects . The multiple comparisons for each 1 min time bin after CS presentation were corrected by a Bonferroni adjustment ( α = 0 . 006 ) . The data for total amount of freezing during the CS were analyzed using a 2 ( genotype ) X 2 ( preexposure treatment ) ANOVA . We found a significant genotype X preexposure interaction , allowing us to explore further the source of the interaction within each genotype using one-way ANOVAs . A few mice were removed from experiments because of procedural errors ( mice were put in the wrong treatment group , or tested in the wrong operant box ) .	A small cluster of neurons found in the midbrain use dopamine to send signals to neurons involved in many processes including motivation and attention . Drugs of abuse such as amphetamine co-opt motivation by increasing dopamine signaling . When used excessively , the drugs can engender delusional thinking , as is seen in schizophrenia . In contrast , the drugs used to treat schizophrenia block excess dopamine signaling . Recently it has been shown that dopamine neurons in the middle part of the midbrain release both dopamine and glutamate . The exact role of this dopamine neuron glutamate signaling has been difficult to find out . Previous experiments involved genetically modifying dopamine neurons so that they would not release glutamate . However , this affected how the neurons develop , making it difficult to discern the effects of glutamate signaling . Now , in genetically modified mice that have less glutaminase in their dopamine neurons than normal , Mingote et al . find that glutamate signaling is reduced just when dopamine neurons fire more rapidly . This did not change how dopamine neurons develop or how they use dopamine to signal . This reduction in dopamine neuron glutamate signaling affects two behaviors that are driven by the activity of dopamine neurons . First , it reduces the effects of a process called amphetamine sensitization , in which repeated doses of amphetamine increase dopamine neuron signaling so that events associated with drug use take up more attention than they normally would . Second , the modified mice were better able to ignore familiar , irrelevant sounds in their environment; the mice continued to pay less attention to a familiar sound , even when it was paired with a shock and came to predict an unpleasant event – a process known as potentiation of latent inhibition . The effects on both of these processes suggest that dopamine neuron glutamate signaling helps animals decide which features of their environment are most important . This result suggests a new way of treating schizophrenia . When humans take amphetamine repeatedly , which produces sensitization , they can develop psychosis , a principal symptom of schizophrenia . During a period of psychosis , thoughts and perceptions are disturbed , making it difficult to distinguish between relevant or irrelevant things in the environment . By reducing amphetamine sensitization and potentiating latent inhibition , blocking dopamine neuron glutamate signaling might help to treat the symptoms of schizophrenia .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Dopamine neuron dependent behaviors mediated by glutamate cotransmission
The SWI/SNF complex is a critical regulator of pluripotency in human embryonic stem cells ( hESCs ) , and individual subunits have varied and specific roles during development and in diseases . The core subunit SMARCB1 is required for early embryonic survival , and mutations can give rise to atypical teratoid/rhabdoid tumors ( AT/RTs ) in the pediatric central nervous system . We report that in contrast to other studied systems , SMARCB1 represses bivalent genes in hESCs and antagonizes chromatin accessibility at super-enhancers . Moreover , and consistent with its established role as a CNS tumor suppressor , we find that SMARCB1 is essential for neural induction but dispensable for mesodermal or endodermal differentiation . Mechanistically , we demonstrate that SMARCB1 is essential for hESC super-enhancer silencing in neural differentiation conditions . This genomic assessment of hESC chromatin regulation by SMARCB1 reveals a novel positive regulatory function at super-enhancers and a unique lineage-specific role in regulating hESC differentiation . Given the near complete uniformity of the DNA sequence across cell types , the regulation of chromatin accessibility , and thereby gene activity , is critical for all stages of development , including the initial fate decisions of embryonic stem cells ( ESCs ) . Accordingly , it has been observed that widespread chromatin re-organization occurs during ESC differentiation via the activity of multiple chromatin-remodeling complexes ( Alexander et al . , 2015; Ahn et al . , 2011; Lessard et al . , 2007; Dixon et al . , 2015 ) . One of the most studied of these complexes in terms of the regulation of pluripotency and differentiation is the SWItch/Sucrose Non-Fermentable ( SWI/SNF ) complex . In mESCs and hESCs , the SWI/SNF complex is composed of SMARCA4 and several core subunits , including SMARCB1 ( Ho et al . , 2009; Zhang et al . , 2014 ) . Complex activity as a whole is essential for the full complement of pluripotency in ESCs and appears to function widely , regulating accessibility and transcription at promoters , active and poised enhancers , as well as at pluripotency factor binding sites ( Ho et al . , 2009; Rada-Iglesias et al . , 2011; Zhang et al . , 2014; Alexander et al . , 2015; King and Klose , 2017; Hodges et al . , 2018 ) . The roles of certain SWI/SNF subunits have been examined in the context of hESC pluripotency and differentiation; for example , separate studies have detailed the requirement of SMARCC2 in the maintenance of self-renewal and revealed that its negative regulation by the microRNA mir-302 is essential for efficient definitive endodermal differentiation ( Zhang et al . , 2014; Wade et al . , 2015 ) . However , the functions of most subunits in hESCs have not been considered , and no data are available on how they regulate chromatin accessibility in this cell type . The core SWI/SNF subunit SMARCB1 is required for embryonic survival past implantation in mice and therefore likely has essential roles in early cell populations ( Roberts et al . , 2000 ) . SMARCB1 is also a potent tumor suppressor , being mutated or deleted in nearly all atypical rhabdoid/teratoid tumors ( AT/RTs ) , aggressive cancers that primarily affect the central nervous system and which can be diagnosed at very young ages , even prenatally ( Hoot et al . , 2004; Pawel , 2018; Negahban et al . , 2010 ) . Whether AT/RT-like tumors develop in mouse models is highly sensitive to the timing of SMARCB1 inactivation . Specifically , deletion at embryonic day ( E ) 6 – E7 leads to highly penetrant CNS tumors , whereas deletion at subsequent time points has lower penetrance or no effect ( Han et al . , 2016 ) , suggesting that a more undifferentiated state is necessary for tumorigenesis . Recent studies in SMARCB1-null cell lines have revealed that its reintroduction results in the widespread recruitment of the SWI/SNF complex to previously unoccupied enhancers , the activation of these enhancers , and the resolution of bivalency at promoters toward an active state ( Alver et al . , 2017; Wang et al . , 2017; Nakayama et al . , 2017 ) . However , these findings are somewhat in disagreement as regards SMARCB1 activity at super-enhancers ( Hnisz et al . , 2013 ) , with different groups reporting either a requirement or dispensability of SMARCB1 in targeting the SWI/SNF complex to super-enhancers and maintaining the active H3K27ac histone marks ( ( Nakayama et al . , 2017; Alver et al . , 2017; Wang et al . , 2017; Hnisz et al . , 2013 ) . We sought to explore the transcriptional and genomic impact of SMARCB1 loss in steady state and differentiating hESCs , focusing on the role of this subunit in enhancer architecture and differentiation . We find that SMARCB1 knockdown ( KD ) leads to widespread transcriptional upregulation in hESCs , with an enrichment in bivalent genes , as well as differential effects on enhancer and superenhancer accessibility . Directed differentiation assays subsequently revealed that loss of SMARCB1 activity strongly inhibits neural induction in a lineage-specific manner . These findings reveal a precise requirement for SMARCB1 in the earliest stages of development and indicate a complex , state-specific role in enhancer regulation . These results will be relevant to additional developmental stages and pathological processes , including oncogenesis . To assess the function of SMARCB1 in steady state hESCs , H1 cells were transduced with lentiviral constructs carrying a doxycycline-inducible shRNA against SMARCB1 or a non-targeting control ( NTC ) region ( Supplementary file 1 ) , followed by treatment with 1 µM doxycycline ( dox ) for two or three days ( Meerbrey et al . , 2011; Silva et al . , 2005 ) . qPCR analysis and western blotting revealed strong downregulation of SMARCB1 at the transcript and protein levels , whereas dox-treated cells expressing the NTC control shRNA showed no such reduction ( Figure 1A ) . Several previous reports have demonstrated that SWI/SNF complex stoichiometry is tightly regulated ( Chen and Archer , 2005; Keppler and Archer , 2010; Sohn et al . , 2007 ) , raising the possibility that SMARCB1 KD may induce instability in other complex members . Notably , SMARCB1 KD did not decrease the protein levels of other SWI/SNF subunits , including SMARCA4 , SMARCC1 , SMARCC2 , SMARCD1 , or SMARCE1 ( Figure 1A , Figure 1—figure supplement 1A ) . SMARCB1 KD cells did not exhibit dramatic morphological differences from untreated controls , and no decreases in the transcript levels of SOX2 , OCT4 , or NANOG were detected , indicating a maintenance of the pluripotency transcriptional program at the analyzed time point ( Figure 1—figure supplement 1B , C ) . To assess SMARCB1 transcriptional regulatory functions , SMARCB1 KD cells were assayed using RNAseq , with significantly affected genes being called at q > 0 . 05 and FC >1 . 5 . The results showed a strong bias towards upregulation following SMARCB1 KD ( 1785 up vs . 95 down ) ( Figure 1B ) , with the most upregulated genes including the transcription factors ZIC1 ( Fold change [FC] ) =13 . 7 ) and SOX21 ( FC = 13 . 3 ) , and the most downregulated genes including the MYC target LINC00176 ( FC = −7 . 9 ) and the receptor MCHR1 ( FC = −3 . 9 ) ( Supplementary file 1 ) ( Pérez-Morales et al . , 2018; Tran et al . , 2018 ) . Ingenuity Pathway Analysis ( IPA ) showed the top Physiological System Development and Function categories in SMARCB1 KD cells were general developmental programs , with the most enriched category being Organismal Development ( Figure 1—figure supplement 1D ) . The component functions within this category consisted of processes involved in multiple developmental processes , including neural development , angiogenesis , and genitourinary system development ( Figure 1C ) . Consistent with the pathway analysis , qPCR for several early markers of the three germ layers all showed upregulation following SMARCB1 KD , including PAX6 , SOX1 , BRN2 ( ectoderm ) , BRACHYURY , GOOSECOID ( mesoderm ) , and CXCR4 ( endoderm ) ( Figure 1D ) ( Walther and Gruss , 1991; Bylund et al . , 2003; Castro et al . , 2006; D'Amour et al . , 2005; Ro and Dawid , 2010; Smith et al . , 1991 ) . As SMARCB1 was recently reported to activate bivalent gene transcription upon reintroduction into SMARCB1-/- cells , we assessed the set of differentially affected genes based on their histone modification characteristics , using previously defined genes sets ( Pan et al . , 2007 ) . Unexpectedly , we found that genes upregulated by SMARCB1 KD were more than 50% more enriched in bivalent genes than the total considered gene set ( 29% vs . 18% , respectively ) ( Figure 1E ) . A similar percentage ( 28% ) of the small number of downregulated genes were also bivalent ( Figure 1—figure supplement 1E ) . Notably , when considering the most highly upregulated genes , the percentage of bivalent genes was nearly 3-fold that of the considered gene set ( 51% vs . 18% ) ( Figure 1E ) . Consistent with this result , we found that the TSS ( ±2 . 5 kb ) of all upregulated genes significantly overlapped with the hESC ChIPseq binding sites of several members of the Polycomb Repressive Complex 2 ( PRC2 ) , including EZH2 ( 4 . 7e-12 ) and SUZ12 ( 1 . 0e-4 ) , as well as the repressive histone marks H3K27me3 ( 1 . 5e-25 ) ( ENCODE Project Consortium , 2012 ) ( Figure 1F , Supplementary file 1 ) . We also tested for enrichment for the only SWI/SNF subunit for which ChIPseq data are available in this cell type , the catalytic subunit SMARCA4 . Although these peaks did not emerge as significantly enriched near the TSS of differentially expressed genes ( DEGs ) , this is largely due to SMARCA4 being highly biologically enriched at most promoters in hESCs . Specifically , SMARCA4 peaks are present at 16 , 031/17 , 462 ( 92% ) of the considered genes in the RNAseq analysis , whereas these values were 90% and 88% for up- and downregulated genes , respectively ( Supplementary file 1 ) ( Rada-Iglesias et al . , 2011 ) . SMARCB1 therefore appears to function largely as a transcriptional repressor in hESCs , particularly at bivalent genes , in contrast to what has been reported in experiments in mouse embryonic fibroblasts ( MEFs ) and several null tumor cell lines , including TTC-1240 ( Wang et al . , 2017; Nakayama et al . , 2017; Wilson et al . , 2010 ) . To interrogate the chromatin effects of decreased SMARCB1 levels , we performed the chromatin accessibility assay ATACseq on cells subjected to SMARCB1 KD for 48 and 72 hr , with a total of 163 , 782 peaks being called ( Buenrostro et al . , 2013 ) ( Figure 2—figure supplement 1A ) . As for the RNAseq analysis , only 3day knockdown data were considered for subsequent analysis , with 15 , 318 peaks being lost between the untreated and knockdown conditions , and 9949 peaks being gained ( Figure 2—figure supplement 1B ) . Following statistical thresholding at q < 0 . 05 , fold change >1 . 5 , SMARCB1 KD cells were found to exhibit 4186 peaks with significantly lower accessibility and 3121 peaks with higher accessibility ( Materials and methods ) ( Figure 2A ) . Given SMARCB1’s core membership in the SWI/SNF complex , we would expect that the chromatin regions altered by loss SMARCB1 activity would significantly overlap known binding sites of the catalytic subunit SMARCA4 . Indeed , both lower and higher accessibility peaks were enriched in known ChIPseq peaks for SMARCA4 ( 89% of lower accessibility peaks , p=2 . 6e-155; 77% of higher accessibility peaks , p=1 . 4e-12 ) ( Supplementary file 2 ) ( Rada-Iglesias et al . , 2011 ) . An important readout of modified chromatin accessibility is differential gene expression , although it is difficult to assign changes in a particular ATAC peak to changes in transcriptional output . We therefore utilized an approach in which the number of differentially expressed genes ( DEGs ) within a given distance of all differentially accessible ATAC peaks was compared to the total number of genes within those ranges . A hypergeometric test was then used to assess whether more DEGs fell within this range than would be expected by chance . For example , when ranges of 20 kb were made around all HA peaks , 955 TSS are encompassed , 132 of which were upregulated following SMARCB1 KD . These values respectively correspond to 5 . 3% of the total gene set and 7 . 5% of the upregulated gene set , indicating a significant enrichment of upregulated genes , with a p-value of 7 . 6e-5 ( Also see Figure 2—figure supplement 1C ) . Plotting these significance values over ranges from 5 kB to 2 MB revealed that SMARCB1 KD higher and lower accessibility peaks were significantly associated with up- and downregulated genes , respectively . To confirm that this association derived from the fact that these peaks were differentially accessible and was not a feature of any similar set of ATAC peaks , 1000x sets randomly selected hESC ATAC peaks were matched to the differential peaks in size and number and subjected to the same analysis . The shaded regions in Figure 2B correspond to the significance values for these random peaks and indicate no meaningful association with gene expression compared to differential peaks . SMARCB1 therefore both positively and negatively regulates gene expression over a range that is consistent with activity at enhancers ( Figure 2B ) . Moreover , and consistent with SMARCB1’s regulation of bivalent genes , the closest genes to differential peaks were significantly enriched in those classified as bivalent ( Figure 2—figure supplement 1D ) . Based on the association between differentially accessible peaks and differentially expressed genes at distances of 500 kb , ( Figure 2B ) and given previous reports of SMARCB1 activity at enhancers and super-enhancers , we assessed how SMARCB1 KD affected accessibility of the enhancer landscape in hESCs . For this analysis , we considered hESC active and poised enhancers as well as defined human super-enhancer data sets for 98 other cell types ( Rada-Iglesias et al . , 2011; Wang et al . , 2017; Nakayama et al . , 2017; Khan and Zhang , 2016 ) . The super-enhancers of other cell types were included in this analysis given previous results indicating that manipulation of the SWI/SNF complex can promote or repress particular fates ( Wade et al . , 2015; Zhang et al . , 2014 ) . As super-enhancers are key regions associated with cell identity , the preferential localization of differential ATAC peaks in these regions for other cell types would inform any effects of SMARCB1 KD on differentiation . SMARCB1 KD lower accessibility peaks were enriched in active hESC enhancers , such as an enhancer located 12 kB upstream of KCNQ2 , as well as poised hESC enhancers , with limited enrichment in other enhancer sets ( Figure 2C and D , Supplementary file 2 ) ( Rada-Iglesias et al . , 2011 ) . In agreement with this result , enrichment analysis using publicly available hESC ChIPseq peaks for histone marks showed strong enrichment for SMARCB1 KD lower accessibility peaks in the enhancer markers H3K4me1 ( p=2 . 1e-218 ) , H3K27ac ( p=4 . 4e-36 ) , and H3K4me2 ( p=1 . 2 e −153 ) ( Figure 2—figure supplement 1E , Supplementary file 2 ) ( ENCODE Project Consortium , 2012 ) . These results are consistent with the previously reported role of SMARCB1 in maintaining enhancer accessibility ( Wang et al . , 2017; Nakayama et al . , 2017 ) . Of all considered enhancers sets , SMARCB1 KD higher accessibility peaks were enriched only in H1 hESC super-enhancers ( p=4 . 6e-42 ) , being present in ~19% ( 127/684 ) of these regions ( Hnisz et al . , 2013 ) ( Figure 2C ) . A subset of these super-enhancers ( 34/127 , 27% ) contained multiple higher accessibility peaks , including the super-enhancer 40 kB downstream of the phospholipase PLA2G16 , indicating that some super-enhancers may be more strongly regulated by SMARCB1 than others ( Figure 2D , Figure 2—figure supplement 1F , Supplementary file 2 ) . These data are the first to show that SMARCB1 negatively regulates super-enhancers in any cell type and indicate that this subunit differentially regulates accessibility across the hESC enhancer landscape . To assess how the loss of SMARCB1 affects accessibility at smaller-scale hESC regulatory regions , we analyzed the distribution of higher and lower peaks in terms of all ENCODE ChIPseq datasets for hESCs as well as a previously published SOX2 ChIPseq dataset ( Zhou et al . , 2016; ENCODE Project Consortium , 2012 ) . In addition to SMARCA4 binding sites , SMARCB1 KD lower accessibility peaks were significantly enriched in SOX2 ( p=1 . 8E-156 ) and OCT4 ( p=8 . 0e-15 ) ChIPseq peaks , indicating that SMARCB1 is necessary to maintain chromatin accessibility at key pluripotency factor binding sites ( Figure 2E , Supplementary file 2 ) . In contrast , SMARCB1 KD higher accessibility peaks were enriched in binding sites for the cohesin complex member RAD21 ( p=7 . 3E-174 ) and the cohesin-interacting protein CTCF ( p=2 . 3E-155 ) . In fact , 50% of SMARCB1 KD HA peaks overlapping hESC RAD21 and CTCF binding sites , consistent with the enrichment of these sites near the TSS of upregulated genes ( Figure 1F , Figure 2E , Supplementary file 1 , Supplementary file 2 ) ( Parelho et al . , 2008 ) . In agreement with the ChIPseq peak enrichment findings , motif analysis showed that lower accessibility peaks were enriched in OCT4 and SOX2 binding motifs , whereas higher accessibility peaks were associated with CTCF and CTCF-like binding motifs ( Figure 2F , Supplementary file 2 ) . These data indicate reveal a complex picture of chromatin regulation by SMARCB1 at pluripotency-associated regions and are the first to reveal a negative regulatory role at super-enhancers , a significant result in terms of this subunit’s role in hESC differentiation . We next performed directed differentiation assays to assess how the loss of SMARCB1 activity would affect hESC differentiation down the three germ lineages . We found that SMARCB1 KD cells successfully executed definitive endodermal differentiation , based on the expression of the transcription factor SOX17 and the surface marker CXCR4 ( Figure 3A , Figure 3—figure supplement 1A–C ) ( D'Amour et al . , 2005; Liu et al . , 2007 ) . Mesodermal induction was similarly unimpaired , with SMARCB1 KD cells expressing the early mesodermal marker EOMES/TBR2 as well as several other markers , including HAND1 , GOOSECOID , and FOXF1 ( Russ et al . , 2000; Barnes et al . , 2010; Niehrs et al . , 1994 ) ( Figure 3B ) . In contrast to the results for endodermal and mesodermal induction , SMARCB1 KD cells exhibited a robust resistance to neural induction . Control cultures exhibited large areas of PAX6+/OCT4- cells , a robust upregulation of PAX6 transcript levels and increased expression of several neural stem cell markers ( Figure 3C ) . In striking contrast , SMARCB1 KD cells showed few PAX6+ cells , numerous strongly positive OCT4+ cells , and a greatly attenuated upregulation of PAX6 transcript levels . The upregulation of other NSC markers was similarly blunted ( Figure 3C ) . Similar results were obtained in a second SMARCB1 KD line , while NTC-shRNA-expressing cells showed no such resistance to induction on dox treatment ( Figure 3—figure supplement 1D ) . To test the robustness of this resistance of SMARCB1 KD cells to neural induction , we generated embryoid bodies ( EBs ) from control and SMARCB1 KD cells and subjected them to a ‘4-/4+’ protocol consisting of four days of growth without retinoic acid ( RA ) and 4 days with 1 µM RA ( Bain et al . , 1995 ) . While control EBs were large and globular , as has previously been observed in neural induction protocols , SMARCB1 KD EBs remained small and spherical ( Stanslowsky et al . , 2016 ) . Moreover , whereas several neural differentiation-related genes , including PAX6 , were highly upregulated in the control condition , SMARCB1 KD EBs exhibited undetectable PAX6 levels and either attenuated upregulation or downregulation of other NSC markers compared to steady state conditions ( Figure 3D ) . To obtain a more comprehensive picture of the transcriptional effects of SMARCB1 KD in cells subjected to neural induction , we performed RNAseq on control and SMARCB1 KD cells subjected to the monolayer neural induction protocol . We identified 1207 genes that were more highly expressed in the SMARCB1 KD condition and 1180 genes with lower expression ( q < 0 . 01 , FC > 2 . 0 ) ( Figure 3E ) . When the statistical constraints were relaxed ( q < 0 . 05 , FC > 1 . 5 ) , the numbers of higher and lower expressed genes more than doubled to 2677 and 3 , 638 , respectively . Some of the genes with the highest expression in SMARCB1 KD vs . control cells included the pluripotency markers TDGF1 ( Fold difference [FD]=35 . 9 ) , FGF4 ( FD = 26 . 8 ) and NANOG ( FD = 23 . 5 ) . In contrast , some of the most highly repressed genes in SMARCB1 KD vs . control cells included the neural differentiation markers FOXG1 ( FD = −28 . 2 ) , DLK ( FD = −25 . 9 ) , and LHX2 ( FD = −14 . 2 ) ( Supplementary file 3 ) ( Watanabe et al . , 2005; Porter et al . , 1997; Tedeschi and Bradke , 2013 ) . Moreover , consistent with the qPCR results , the RNAseq data confirmed that PAX6 was expressed at appreciable levels only in control cells ( Figure 3E ) . As expected , IPA analysis revealed that the genes expressed at lower levels in SMARCB1 KD cells were enriched in several categories related to neural development . In contrast , more highly expressed genes were enriched in categories related to the self-renewal of stem cells and the maintenance of pluripotency ( Figure 3F ) . Together , these results show that SMARCB1 KD cells have a lineage-specific requirement for neural induction and that loss of SMARCB1 activity results in both a failure to upregulate neural genes as well as a failure to downregulate pluripotency-related pathways . To address mechanisms by which SMARCB1 KD prevents neural induction , we subjected control and SMARCB1 KD cells to ATACseq midway through ( Day 3 ) and at the completion of the neural induction protocol ( Day 6 ) . Here , we focus on differences between the control and SMARCB1 KD cells at Day 6 , although regions of interest were defined using data from all starting , Day 3 , and Day six conditions . Of the considered 88 , 749 ATAC peaks , 7801 exhibited higher accessibility ( HA ) in SMARCB1 KD cells , whereas 13 , 614 exhibited lower accessibility ( LA ) ( q < 0 . 01 , FC >2 . 0 ) ( Figure 4A , Figure 4—figure supplement 1A , Materials and methods ) . To assess the effects of SMARCB1 KD on the accessibility of regions that normally exhibit open chromatin in neural stem cells ( NSCs ) , we compared differential peaks with over 80 , 000 previously published regions known to be accessible in human cortical NSCs but not in pluripotent stem cells ( Forrest et al . , 2017 ) . We found that 30% ( 4 , 107/13 , 614 ) of lower accessibility peaks overlapped NSC-specific regions ( p<2 . 2E-308 ) ( Figure 4—figure supplement 1B ) . No such enrichment was observed for higher accessibility peaks , with only 555/7 , 801 ( 7% , p=1 ) overlapping NSC-specific accessible regions ( Figure 4—figure supplement 1B ) . The accessibility signal over these 4107 lower accessibility peaks in SMARCB1 KD cells following the neural induction protocol was virtually indistinguishable from the signal for hESCs , indicating that the increase in accessibility normally observed in these regions during neural induction was completely abrogated ( Figure 4B ) . Moreover , several genes important for neural differentiation had multiple nearby regions that normally gain accessibility during differentiation but that failed to do so in SMARCB1 KD cells . For example , within the multiple ATAC-sensitive peaks in and around the PAX6 locus , one can clearly identify six such regions were located upstream and down of the PAX6 locus that are substantially depressed ( Figure 4C ) . We next asked whether these lower accessibility peaks were associated with differentially expressed genes as determined by RNAseq . The results indicated that lower accessibility peaks were significantly associated with differential gene expression to nearly 1 Mb ( Figure 4D ) , after which a significant effect was not observed . Importantly , an analysis of 1000x sets of randomly selected size-matched ATAC peaks showed no such association ( Figure 4D ) . This association between lower accessibility peaks and gene expression are suggestive of the activity of distal enhancers . Moreover , several promoters of neural/neural differentiation-related genes contained lower accessibility peaks ( e . g . , CER1 , CRX , DPP6 , SIX1 ) , while multiple promoters of genes related to pluripotency contained higher accessibility peaks ( e . g . , including OCT4 , mir302 , DNMT3B , ZIC3 , and DPPA2/4 ) ( Supplementary file 4 ) . IPA analysis showed that differentially affected genes within 500 kb of lower accessibility peaks were enriched in genes within the Neuron Development pathway , for which a strongly negative activation score was obtained ( p=2 . 6E-40 , z = −2 . 935 , Supplementary file 4 ) . Many of these genes are regulators of neural differentiation and are early markers of NSCs , including PAX6 and FOXG1 ( Figure 4E , Supplementary file 4 ) ( Walther and Gruss , 1991; Watanabe et al . , 2005 ) . Consistent with the above data , HOMER motif analysis of lower accessibility peaks revealed enrichment in motifs for transcription factors that regulate neural development , including Otx2 , Lhx1 , and Lhx2 ( Figure 4F , Figure 4—figure supplement 1D ) ( Ang et al . , 1996; Porter et al . , 1997; Shawlot and Behringer , 1995 ) . Together , these data indicate that SMARCB1 is required for the positive changes in chromatin accessibility , and thereby the necessary increases in gene expression , that are required during the early stages of neural induction . We next determined whether SMARCB1 KD cells maintained hESC chromatin characteristics , including accessibility at enhancers , super-enhancers , and pluripotency factor binding sites during the initial stages of neural induction . We found enrichment of these powerful regulatory stretches among higher accessibility peaks , with 282 ( 41% ) hESC super-enhancers intersecting at least one higher accessibility peak ( p=5 . 8E-65 ) ( Figure 5A , Supplementary file 5 ) . In contrast to what was observed in the steady state , active hESC enhancers were also enriched in higher accessibility peaks , with 562 higher accessibility peaks intersecting this set ( p=1 . 2E-35 , Figure 5A ) ( Rada-Iglesias et al . , 2011 ) . Importantly , no enrichment in enhancers or super-enhancers was observed for lower accessibility peaks ( both p=1 ) ( Figure 5—figure supplement 1 ) . The maintenance of accessibility over hESCs super-enhancers was even more evident when the normalized accessibility signal was used to generate metaplots over these regions . Whereas the accessibility signal for control cells exhibited a ≈30% decrease relative to steady state hESCs , the accessibility signal of SMARCB1 KD cells was essentially unchanged from hESCs not subjected to the induction protocol ( Figure 5B ) . An example of this was observed with the super-enhancer near the kinase DAPK1 , which contains three higher accessibility regions with comparable accessibility to what is seen in steady state hESCs ( Figure 2B ) . This result indicates that the accessibility of hESC super-enhancers was strongly maintained throughout the neural induction protocol . Notably , there was no enrichment among higher accessibility peaks for any of the 98 other analyzed super-enhancer sets for different cell types ( Supplementary file 5 ) ( Khan and Zhang , 2016 ) . As super-enhancers are drivers of cell identity and have a strong effect on gene expression , we calculated the association between hESC super-enhancers with higher accessibility peaks and differentially expressed genes by RNAseq ( Hnisz et al . , 2013 ) . We found that these higher accessibility peaks were correlated with differential gene expression out to 1 Mb ( Figure 5C ) . The strongest association was seen within 500 kB of higher accessibility peaks within super-enhancers , and a slowly declining association was observed for more distal genes . As before , 1000x sets of randomly selected sized-matched ATAC peaks did not show a significant association with differential gene expression ( Figure 5C ) . As expected based on the previously described localization of super-enhancers near genes related to cell identity , several of the differentially expressed genes near HA-super-enhancer regions were found to be powerful positive regulators of pluripotency , including OCT4 and the microRNA mir-302 ( Figure 5C ) ( Hnisz et al . , 2013 ) . To assess whether these enhancers exhibited maintained activity , we analyzed levels of transcription in these regions . Consistent with maintained activity , higher levels of transcription were detected in 89 hESC super-enhancers in SMARCB1 KD cells compared to the control condition , whereas lower levels of transcription were observed at only 23 super-enhancers ( Figure 5D , Supplementary file 5 ) . Moreover , of the 89 with higher eRNA expression , 58 ( 65% ) had closest genes that were differentially expressed by RNAseq , most of which ( 51/58 , 88% ) were expressed at higher levels in SMARCB1 KD cells ( Supplementary file 5 ) . To assess whether smaller-scale features of the hESCs chromatin landscape might also be maintained in SMARCB1 KD cells , we evaluated the accessibility of known hESC pluripotency factor binding sites . Indeed , higher accessibility peaks were enriched in the binding sites for several key hESC TFs , including OCT4 ( p=5 . 3e-125 ) , NANOG ( p=6 . 7e-40 ) , and SOX2 ( p=1 . 2e-29 ) ( Figure 5E , Supplementary file 5 ) . This overlap comprised a significant percentage of these binding sites , including 14% , 10% and 7% of hESC SOX2 , OCT4 , and NANOG , binding sites , respectively . Moreover , higher accessibility peaks overlapped 14% of hESC peaks for the SWI/SNF-associated factor BCL11A ( p=2 . 1e-37 ) , suggesting that the SWI/SNF complex mediates the repression of these regions during the early stages of neural induction in a SMARCB1-dependent manner ( Figure 5E , Supplementary file 5 ) . In all of these pluripotency factor binding sites , the normalized accessibility signal in SMARCB1 KD cells subjected to neural induction was similar to that of untreated hESCs , whereas the signal for control cells was strongly attenuated compared to the hESC accessibility signal , indicating a failure to repress accessibility in these regulatory regions in SMARCB1 KD conditions ( Figure 5F ) . Several SWI/SNF subunits are known to be essential for hESC to maintain their full complement of pluripotency ( Zhang et al . , 2014; Ho et al . , 2009; Schaniel et al . , 2009 ) . However , there are few data on the specific role of core subunits in genome-wide chromatin accessibility in hESCs , nor has the role of SMARCB1 been explored in terms of its role in regulating hESC differentiation . An inducible knockdown strategy permitted the assessment of SMARCB1’s regulation of transcription and chromatin accessibility in steady state hESCs and under differentiation conditions . These data revealed surprising insights into SMARCB1’s regulation of the hESC enhancer landscape and its specific requirement for neural induction . The observed widespread upregulation of bivalent genes in SMARCB1 KD hESCs was unexpected given previous reports that SMARCB1 loss leads to PCR2-mediated repression in MEFs and that reintroduction leads to bivalent gene activation in SMARCB1-null cell lines ( Nakayama et al . , 2017; Wilson et al . , 2010 ) . Given the highly euchromatic nature of embryonic stem cells and the developmental repercussions of premature differentiation , it is possible that the SWI/SNF complex has a more repressive role in hESCs than in differentiated or tumor lines ( Meshorer and Misteli , 2006 ) . In line with this interpretation , a previous microarray analysis of SMARCA4 KD in hESCs revealed a bias in gene upregulation ( 472/529 , 74% ) , a result that is consistent with our unpublished observations ( Zhang et al . , 2014 ) . It is also worth noting that while there were a greater number of lower accessibility peaks in the steady SMARCB1 KD condition , there was a strong bias towards transcriptional upregulation . We hypothesize that SMARCB1 positively affects transcription by other mechanisms than altered accessibility . For example , SMARCB1 may be required for the recruitment of transcription factors or transcriptional machinery to promoters hESCs , which would result in decreased transcription but not a dramatic change in accessibility . Evidence for transcriptional control independent of accessibility changes include the established antagonism between the SWI/SNF and PRC2 complexes , the interaction between the complex and the tumor suppressor p53 , and the observed associations between SMARCB1 and RNA Pol I and RNA Pol II ( Kadoch et al . , 2017; Lee et al . , 2002; Cho et al . , 1998; Zhai et al . , 2012 ) . The role of SMARCB1 at enhancers has received significant attention in recent years , and the results are not wholly in agreement . Specifically , it was reported that SMARCB1 deletion in MEFs decreases levels of the active markers H3K27ac and H3Kme1 at enhancers , whereas super-enhancers were relatively spared ( Wang et al . , 2017 ) . In contrast , others have found that both enhancers and super-enhancers show increased levels of H3K27ac upon SMARCB1 reintroduction into null tumor cell lines ( Nakayama et al . , 2017 ) . Our data are consistent with both sets of previous results in that SMARCB1 KD in steady state hESCs leads to widespread loss of enhancer accessibility . However , we find that SMARCB1 has a repressive role at hESC super-enhancers , a previously undescribed effect and one not observed for any of the 98 other analyzed human super-enhancer datasets . This is a salient difference in the context of AT/RT tumorigenesis in that the cell of origin is likely an undifferentiated NSC , indicating that current models of SMARCB1 activity at enhancers that are based on differentiated cells or reintroduction experiments may not fully capture the functions that contribute its tumorigenic role ( Han et al . , 2016 ) . That SMARCB1 KD leads to elevated accessibility at CTCF binding sites is particularly noteworthy given recent findings that a SMARCB1-excluding non-canonical SWI/SNF complex ( ncBAF , also termed GBAF ) preferentially targets CTCF sites in mESCs and several tumor cell lines ( Gatchalian et al . , 2018; Michel et al . , 2018 ) . That CTCF sites gain accessibility in SMARCB1 KD conditions suggests that an absence of SMARCB1 may promote the formation of the ncBAF . This is an intriguing possibility given that the ncBAF complex positively regulates naïve pluripotency in mESCs and may provide a mechanism by which SMARCB1-deficient cells resist differentiation ( Gatchalian et al . , 2018 ) . The lineage specificity of SMARCB1’s requirement in neural differentiation is relevant in the context of its role as a tumor suppressor . Although loss of SMARCB1 in adult mice leads to lymphoma , SMARCB1 mutation-associated pediatric AT/RTs are found in the CNS , a finding that has been recapitulated in a conditional SMARCB1 KO mouse model ( Han et al . , 2016; Roberts et al . , 2000; Babgi et al . , 2018 ) . Neural differentiation has been reported to be the default lineage choice of ESCs , and recent work indicates that this characteristic is cell-intrinsic and driven in part by expression of the zinc finger protein Zfp521 ( Muñoz-Sanjuán and Brivanlou , 2002; Kamiya et al . , 2011 ) . It is possible that loss of SMARCB1 activity renders cells deficient in the mechanisms involved in this intrinsic process , leaving differentiation pathways that normally require extrinsic stimuli unimpaired . The above results provide critical insights into how a core SWI/SNF subunit regulates both the hESC enhancer landscape as well as differentiation down a lineage where it is strongly implicated as a developmental tumor suppressor . The complex roles that SMARCB1 plays at different enhancer subtypes should be borne in mind when assessing both subsequent stages of development as well as the initial stages of cellular transformation . H1 hESCs were cultured in TeSR-E8 growth medium ( STEMCELL Technologies , #05990 ) at 37°C in a 5% CO2 atmosphere . Cells were grown feeder-free on a substrate of hESC-qualified Matrigel Matrix ( Corning #354277 ) . Cells were passaged at or before 80% confluency and at densities of 1:6-1:24 , unless otherwise specified . Generation and use of doxycycline-inducible KD lines shRNAs were a kind gift from the laboratory of Dr . Guang Hu and were described in Silva et al . ( 2005 ) . Sub-cloning of the non-targeting shRNA and those against SMARCB1 and SMARCA4 were performed by digestion of the pINDUCER backbone vector with XhoI ( New England BioLabs , #R0146 ) and MluI ( New England BioLabs , #R0198 ) , following standard procedures ( Meerbrey et al . , 2011 ) . Lentiviruses carrying the respective shRNAs were produced at the NIEHS Viral Vector Core Laboratory according to a previously established protocol ( Salmon and Trono , 2007 ) . H1 cells were infected at MOI8 and selected using 1 µg/ml puromycin for 24 hr . To further select for high-expressing cells , target and control shRNA-carrying cultures were treated for 18 hr with 1 µg/ml doxycycline collected in a single cell solution using Gentle Cell Dissociation Reagent ( STEMCELL Technologies #07174 ) for 10 min at room temperature and sorted on a BD FACSAria II to obtain the top 20% of RFP- expressing cells . Doxycycline was immediately removed thereafter , and the cells were cultured for in the presence of the ROCK inhibitor Y-27632 ( STEMCELL Technologies , #72304 ) for 48 hr to promote survival . For all KD experiments using steady state hESCs , high shRNA-expressing cells were split at a density of 1:24 and allowed to recover for 48 hr prior to the initiation of 1 µg/ml of doxycycline treatment . Treatment was continued for 3 days prior to collection for KD validation , RNAseq , or ATACseq . RNA was isolated using Norgen Total RNA Purification Plus kits ( #48300 ) , and for all qPCR experiments , cDNA was generated using an iScript cDNA Synthesis Kit ( BioRad , #1708891 ) . The primers used for qPCR analysis are available in Supplementary file 6 . Cells carrying inducible shRNAs against SMARCB1 were cultured for 3 days in the presence or absence of 1 µg/ml doxycycline . Cells were then collected in Gentle Cell Dissociation Reagent for 10 min at 37°C and dissociated into single cells . The remainder of the protocol was performed as per the STEMCELL Technology instructions for neural induction using STEMDiff Neural Induction Media ( #05835 ) , with minor modifications . Specifically , cells were plated at a density of 1 . 5 × 105 cells/cm2 , as this density provided the highest efficiency of induction . At 6 days , cells were collected for RNAseq , ATACseq , or processed for immunohistochemistry . The monolayer protocol was performed three times with three biological replicates each , with similar results for each performance . EBs were formed from control and SMARCB1 KD cells ( following 3 days of dox treatment ) . The EBs were 1 , 000 cells each and were generated in AggreWell400 plates ( StemCell Technologies #34421 ) . After 24 hr , the EBs were transferred to ultra-low-adherence dishes ( Corning #3471 ) , following the StemCell Technologies protocol . The EBs were cultured in AggreWell EB Formation Medium ( StemCell Technologies , #05893 ) , with media replaced daily by allowing the EBs to settle in 15 ml conical tubes for 10 min prior to media aspiration . On Day four following EB formation , 1 µM all-trans retinoic acid was added to the media , whereupon the EBs were cultured for an additional 4 days prior to collection for isolation of RNA . The EB experiment was performed 5 times with 2–3 replicates per control/SMARCB1 KD condition for each experiment . Similar morphological effects of SMARCB1 KD were observed for each performance , and qPCR analysis was performed on replicates from one performance . For both monolayer and EB experiments , RNA was isolated using Qiagen RNAeasy kits ( #74104 ) , and cDNA was generated using an iScript cDNA Synthesis Kit ( BioRad , #1708891 ) . The primers used for the analysis of neural and pluripotency markers are available in Supplementary file 6 . Cells carrying inducible shRNAs against SMARCB1 were cultured for 3 days in the presence or absence of 1 µg/ml doxycycline . Cells were then collected in Gentle Cell Dissociation Reagent for 10 min at 37°C and dissociated into single cells . The remainder of the protocol was performed as per the STEMCELL Technology instructions for endodermal differentiation using STEMDiff Definitive Endoderm Kit ( #05110 ) . The experiment was performed three times , with results being tested by qPCR ( n = 3 biological replicates/condition ) , flow cytometry ( total n = 5 biological replicates/condition ) , and/or immunohistochemistry ( n = 1 biological replicate/condition ) . Cells carrying inducible shRNAs against SMARCB1 were cultured for 4 days in the presence or absence of 1 µg/ml doxycycline until ≈90% confluency . Cells were then cultured for 2 days in RPMI1640 supplemented with NeuroCult without insulin ( StemCell Technologies #05733 ) and 5 µM CHIR99021 ( GSK inhibitor , StemCell Technologies , #72052 ) , after which they were treated for two additional days in RPMI1640 supplemented with NeuroCult without insulin . The mesodermal induction experiment was performed twice , with qPCR results being based on three biological replicates per condition and immunohistochemistry results being based on one biological replicate per condition . Cells subjected to the neural induction protocol were processed for IHC as described it the STEMdiff Human Neural Progenitor Antibody Panel ( StemCell Technologies , #69001 ) . Briefly , cells in glass-bottom plates were fixed for 15 min at room temperature in 4% formaldehyde , permeabilized with 0 . 1% Tween for 10 min , blocked with 5% FBS in PBS , and stained with antibodies against PAX6 ( anti-rabbit , StemCell Technologies #60094 ( 1:500 , Lot# SC09342 ) , or BioLegend #901301 , ( 1:300 , Lot# B235967 ) , NESTIN ( anti-mouse , StemCell Technologies #60091 ( 1:1000 , Lot# SC09341 ) , and OCT4 ( anti-mouse , StemCell Technologies #60093 , 1:1000 , Lot# SC09338 ) . After 3x rinses with PBS , the cells were incubated with secondary antibodies for 1 hr at room temperature ( secondaries available upon request ) and counterstained with DAPI ( ProLong Diamond Antifade Mountant , ThermoFisher #P36971 ) . Images were obtained on a Zeiss LSM 710 inverted confocal microscope and analyzed in ImageJ ( Schneider et al . , 2012 ) . Endodermal and mesodermal differentiation cultures were processed in the same manner and probed with antibodies against SOX17 ( AF1924 , R and D Systems , 1:500 ) and EOMES ( MAB6166 , R and D Systems , 10 µg/ml , Lot# CEDQ0218011 ) , respectively . For detection of SWI/SNF subunit protein levels , cells were lysed in Buffer X ( 100 mM Tris-HCL , pH 8 . 5 , 250 mM , 1% NP-40 , 1 mM EDTA ) containing 1:100 protease inhibitor cocktail ( Pierce Technology , #PI78442 ) and 1:100 PMSF on ice for 20 min . The lysates were then homogenized by vortexing for 30 s and centrifuged for 12 , 800 x g for 15 min . The protein concentration in the supernatant was quantified using the Bradford assay , and 30–50 µg protein was separated on 4–12% Tris-glycine gels ( ThermoFisher , XP04122BOX ) at 100 V for 2 hr . The gels were then transferred to PVDF membranes for 2 hr at 400 mA at 4 ˚C . The blots were then blocked for 1 hr at room temperature or 4° overnight in TBS containing 5% milk . The membranes were then incubated overnight at 4° with the following primary antibodies diluted in TBS-Tween containing 5% milk: SMARCA4 ( lab-generated antibody targeting aa437–678 , anti-rabbit , 1:2000 [Wade et al . , 2015] #133 ) ) , SMARCC1 ( H76 , Santa-Cruz , anti-rabbit , 1:500 ) , SMARCC2 ( E-6 , Santa-Cruz , anti-mouse , 1:200 ) , SMARCD1 ( 23 , Santa-Cruz , anti-mouse 1:1000 ) , SMARCE1 ( lab-generated , anti-rabbit , 1:2000 [Chen and Archer , 2005] ) , Lamin A/C ( H-110 , Santa-Cruz , anti-rabbit ) , GAPDH ( 6C5 , anti-mouse , 1:10 , 000 , Abcam ) . Secondary antibody staining was performed with Li-Cor IRDye 800CW and IRDye 680RD antibodies against the appropriate species . Image acquisition was performed using an Odyssey CLx Infrared Imaging System and analyzed using ImageStudio Lite software . H1 hESCs subjected to the endodermal differentiation protocol were washed with PBS and treated with trypsin for 3 min . The cells were spun at 300xg for 5 min , resuspended in TeSR-E8 , and stained with BV421 Mouse Anti-Human CD184 ( BD Biosciences , cat# 566282 ) , as per the manufacturer’s instructions . Cells were then re-pelleted and washed 2x with TeSR-E8 and assayed on a Becton Dickinson LSR II Flow Cytometer ( BD Biosciences ) or a Becton Dickinson LSRFortessa following an additional stain with propidium iodide as a vital dye . Only live cells were considered for the analysis , with data being analyzed using FACSDiva software v8 . 0 . 2 . Three replicates were used for all RNAseq experiments . RNA was isolated from H1 hESCs and neural/induction cells using Norgen Total RNA Purification Plus kits ( #48300 ) , and RNAseq with rRNA removal was performed by Expression Analysis , an IQVIA company . The raw data were filtered for quality using sickle ( default parameters ) , adapters were removed , and reads were aligned to hg19 using STAR , keeping only unique alignments . Feature counts were obtained using bedTools featureCounts ( steady state hESCs ) or Salmon ( neural induction ) . The R package limma was used to call regions with differential read counts among genes/regions with a max group mean ( mgm ) of 8 ( hESCs ) or 32 ( neural induction ) . Two replicates were used for all ATAC experiments . Cells were dissociated by incubation with Gentle Cell Dissociation Reagent ( steady state hESCs ) or Accutase ( neural induction ) ( StemCell Technologies , #07920 ) for 10 min at 37°C . Cells were then collected by pipetting , counted , and spun for 5 min at 300 x g , after which they were resuspended in to 1e6 cells/ml . Cells were then subjected to the ATACseq protocol as described by Buenrostro et al . ( 2015 ) , with the following parameters . Buffer: CSK buffer ( 10 mM PIPES pH 6 . 8 , 100 mM NaCl , 300 mM sucrose , 3 mM MgCl2 , 0 . 1% Triton X-100 ) . Transposase volume/25 µl cell suspension: 5 µl . Transposase treatment time: 30 min , with mixing every 10 min . Libraries were sequenced at the NIEHS Epigenomics Core Facility on the NextSeq 500 platform , and reads were trimmed using default parameters ( steady state hESC experiments ) or with -q 26 ( neural induction experiment ) , with the latter modification made due to a technical issue with the sequencing run . Reads were aligned to hg19 with bowtie2 with the following parameters: -X 2000 N 1 --no-unal --no-mixed --dovetail --no-discordant . Uniquely aligned reads were filtered using an in-house python script , which was also used to remove mitochondrial reads . Reads < 135 bp were retained for analysis of accessible regions and deduplicated with picard , using the following parameters: MAX_RECORDS_IN_RAM = 5000000 , REMOVE_DUPLICATES = true . Peaks were called for each replicate using macs2 , with the following parameters: callpeak --nomodel --nolambda --keep-dup all --slocal 10000 -q 0 . 005 [https://github . com/taoliu/MACS] . After the top 0 . 1% of Regions of Exceptionally High Depth of Aligned Short Reads were removed from the called peak files , Untreated , Day 2 , and Day 3 KD peaks were merged and used as the considered regions for statistical analysis ( Pickrell et al . , 2011 ) . Differential peaks were called using the R package limma , with cut-offs of q < 0 . 01 , log2 ( FC ) >2 , and a max group mean ( mgm ) of >32 reads . Peaks in which this mgm criterion was not met in any condition were excluded from further analyses . The signal normalization factors required to permit comparison of conditions were determined by finding the scale factors required to equalize the median read counts for all considered regions of interest . For these determinations , to exclude the use of regions with signals that could not be distinguished from noise , only regions with signals above the median signal value were considered . For each of the differential ATAC-seq peak sets , ranges of increasing size were made around the peaks ( from 5 kb to 2 Mb beginning with 5 kb increments ) . Analyses were performed using R version 3 . 5 . 1 and visualized with ggplot2 version 3 . 1 . 0 ( R Core Team , 2018; Wickham , 2016 ) . Regions and overlaps were determined using Bioconductor 2 . 42 . 0 packages rtracklayer 1 . 42 . 1 and GenomicRanges 1 . 34 . 0 ( Huber et al . , 2015; Lawrence et al . , 2009; Lawrence et al . , 2013 ) . Ranges were intersected with the TSS of all RNA-seq genes in the given analysis ( i . e . , steady state or neural induction ) , and a hypergeometric test was used to determine if differential RNA-seq TSS gene hits were enriched in the regions overlapping the differential ATAC-seq peaks . In addition , for each analysis , an equivalent random subset of observed ATAC-seq peaks were selected 1000 times and tested for enrichment using the same procedure . The data were collected to visualize the P-value of observed differential ATAC-seq peaks , the median and 5–95% quantile range of random ATAC-seq peaks . The detailed workflow is available in a supplementary file ( Peak_gene_enrichment_analysis . html ) . To assess the significance of overlaps between differentially accessible peaks and genomic regions of interest , the following command was performed in R: phyper ( q , m , n , k , lower . tail = FALSE ) , where q = the number of intersection between the differential peak set and the genomic regions of interest , m = the intersection between all considered ATAC peaks , regardless of condition , and the genomic regions of interest , n = the number of all considered ATAC peaks not intersecting the genomic regions of interest , and k = the number of differential peaks in the considered set . HOMER motif analysis was used with findMotifsGenome . pl , using the following parameters: -size given , background regions defined as non-differential peaks that were similar to the considered differential peak set in terms of size , ATAC signal , and distance to the nearest promoter ( Heinz et al . , 2010 ) . Motifs were considered statistically enriched at p<1E-50 . All qPCR results were calculated using the ΔΔCt method relative to control cell populations , as indicated in the figure legends , and were normalized to the geometric mean of 18S and GAPDH levels , with 3–4 technical replicates per sample . The bar heights represent mean relative expression for three biological replicates , and error bars represent standard deviations . All raw RNAseq and ATACseq data have been made available in NCBI’s Gene Expression Omnibus ( Edgar et al . , 2002 ) , with accession number GSE128351 . Several of the described bioinformatic analyses were performed using samtools: ( v0 . 1 . 20 ) , bedTools ( v2 . 21 . 0 ) , deepTools ( v2 . 5 . 3 ) , Picard ( v2 . 9 . 2 ) , GNU Parallel ( v20170522 ) , and RStudio ( v1 . 1 . 383 ) . Pathway analysis was performed using Ingenuity Pathway Analysis: v 01–10 . Flow cytometry data were acquired and analyzed using BD FACSDiva: 8 . 0 . 2 .	Our bodies contain trillions of cells that play a wide variety of roles . Despite looking and behaving very differently to one another , all of these ‘mature’ cells somehow descend from a single fertilized egg that contains just one set of genes . This process is partially controlled by how ‘accessible’ genetic material is to the cell machinery that switches genes on or off . For example , in immature brain cells , genes required for memory are accessible , but genes needed to produce bone are not . The developing embryo needs to control gene accessibility carefully to ensure that the right genes become available at the right time , and that crucial genes are not incorrectly ‘hidden’ . In humans , the protein SMARCB1 plays an important role in this process: if damaged or deleted , development will be severely disrupted , sometimes causing brain cancer early in life . However , it remains unclear how exactly SMARCB1 regulates the accessibility of its ‘target’ genes . Now , Langer et al . set out to answer this question , and also to determine which parts of the body need SMARCB1 to develop properly . Human stem cells can develop into multiple mature cell types if given the right signals . Langer et al . found reducing levels of SMARCB1 prevented stem cells from maturing into brain cells , but not other kinds of cells . This suggests that SMARCB1 has a specific role in brain development , which is consistent with its devastating effect on brain health when damaged . A detailed analysis of genetic activity and DNA accessibility showed that SMARCB1 was doing this by switching off specific regions of DNA , called stem cell super-enhancers . These regions normally enhance the activity of genes that maintain stem cells in their immature state: when certain super-enhancers are turned off by SMARCB1 , this allows stem cells to progress towards a brain cell fate . These results help us understand why damage to SMARCB1 during development causes brain cancer more often than other kinds of cancer . In the future , they could also help explain how certain types of cancer form , which would be the first step towards knowing how to treat them .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression" ]	2019	Tumor suppressor SMARCB1 suppresses super-enhancers to govern hESC lineage determination
Systems vaccinology approaches have been used successfully to define early signatures of the vaccine-induced immune response . However , the possibility that transcriptomics can also identify a correlate or surrogate for vaccine inflammation has not been fully explored . We have compared four licensed vaccines with known safety profiles , as well as three agonists of Toll-like receptors ( TLRs ) with known inflammatory potential , to elucidate the transcriptomic profile of an acceptable response to vaccination versus that of an inflammatory reaction . In mice , we looked at the transcriptomic changes in muscle at the injection site , the lymph node that drained the muscle , and the peripheral blood mononuclear cells ( PBMCs ) isolated from the circulating blood from 4 hr after injection and over the next week . A detailed examination and comparative analysis of these transcriptomes revealed a set of novel biomarkers that are reflective of inflammation after vaccination . These biomarkers are readily measurable in the peripheral blood , providing useful surrogates of inflammation , and provide a way to select candidates with acceptable safety profiles . Systems biology approaches are increasingly being used to describe and define signatures of immunity , initially in the setting of infection but more recently in vaccine-induced responses , leading to the development of systems vaccinology ( Querec et al . , 2009; Olafsdottir et al . , 2015 ) . These transcriptomic analyses have primarily focused on the prediction of vaccine efficacy and immune outcome with the assessment of vaccine safety and potential reactogenicity relying on clinical reactive scores of adverse events ( Bucasas et al . , 2011; Furman et al . , 2013; Obermoser et al . , 2013; Vahey et al . , 2010; Li et al . , 2014; Leonardi et al . , 2015; Vesikari et al . , 2009 ) . However , the transcriptome can also reveal the intricate details of the very early events after vaccination . We have utilized this approach to examine this initial innate response and any subsequent inflammation resulting from a vaccination , to identify biosignatures that have correlative or surrogate potential for vaccine-related inflammation and safety . It is likely that the activation and maturation of the fundamental responders within the immune system will follow set developmental patterns , which may be revealed by an examination of the transcriptome . Thus any potential differences between each vaccine would probably be due to the degree of response , and to the involvement of different populations of central immune system players and of accessory cells . The dominant determinants of such responses are the nature of the vaccine antigen , its formulation , and the presence or absence of molecularly defined adjuvants ( Li et al . , 2017; O'Gorman et al . , 2015 ) . The blood and lymphatic systems are the primary routes of transport of cells and chemical signals from one immunological organ to another , with immune cells and soluble mediators trafficking between the injection site and the local draining lymph node ( Girard et al . , 2012; Jackson , 2014 ) . Indeed , proteins and molecules that can be measured in the blood after vaccination are likely to be useful biomarkers of such responses as blood draws can be readily taken from most individuals . Although a particular cell or inflammatory molecule may be present in relatively high numbers and concentrations in both the lymph node and the injection site , it is apparent that the large volume of blood in the circulatory system may dilute this molecule and prevent detection . In this study , we examined the molecular signatures generated in the blood , the draining lymph node and the injection site in the muscle of mice after the administration of a number of licensed vaccines , which set the benchmark for what is safe , and of a set of known agonists of Toll-like receptors ( TLRs ) or adjuvants , to set an upper boundary of inflammatory response ( Haziot et al . , 1998; Qureshi et al . , 1999; Akira and Takeda , 2004; Fitzgerald et al . , 2004; Matsumoto and Seya , 2008 ) . The four licensed vaccines used in this study have a substantial history of safe use in humans , with large datasets supporting their safety and tolerability . They ranged from protein alone to vaccines that included adjuvants or a known component that have an acceptable degree of reactogenic potential ( whole cell pertussis ) ( Frey et al . , 2003; Tregnaghi et al . , 2012; Mohanty et al . , 2018; Van Den Ende et al . , 2017 ) . We also used the water-in-oil emulsion incomplete Freunds adjuvant ( IFA ) ( Montanide ISA 51 ) and two TLR agonists ( lipopolysaccharide ( LPS ) and Poly I:C ) ( Table 1 ) . A set of indicators of vaccine-elicited inflammation in a small animal model will provide an early warning system that can be used to inform decisions on the potential tolerability and ultimately the utility of vaccines undergoing selection and development , and will provide additional tools to de-risk late-stage failure . High-quality RNA samples , isolated from muscle tissue excised around the injection site , the draining medial iliac lymph nodes ( MLN ) and total peripheral blood mononuclear cells ( PBMCs ) , were taken at 4 , 8 , 24 , 48 , 72 , 168 hr after the vaccination or treatment with LPS , polyI:C or IFA ( Table 1 ) and subjected to genome-wide transcriptome analysis . The unvaccinated mice , and mice receiving saline alone , were used as control groups . Differentially expressed genes were defined as those with a significant ( Benjamini-Hochberg ( BH ) adjusted p-value<0 . 01 ) change in expression in a vaccinated group when compared to the unvaccinated group and the saline group at a given time point . Violin plots that chart the numbers of upregulated and downregulated differentially expressed genes revealed that the immunisations differed in terms of the magnitude and the kinetics of the transcriptomic responses induced in muscle , MLN , and blood ( Figure 1 ) . Strong responses were observed in the injected muscle tissue for all of the immunisations , with the highest levels of fold change and of absolute numbers of differentially expressed genes being detected at 72 and 168 hr after treatment . Pentavac SD , Poly I:C and particularly LPS also induced changes in differential gene expression in the muscle , MLN and blood at 4 , 8 , 24 and 48 hr post-injection . Engerix B and IFA elicited modest changes in the differential gene expression in the muscle at 48 , 72 and 168 hr and at the later time points of 72 and 168 hr in the MLN , but very few differentially expressed genes were observed in the blood . Agrippal and Fluad exhibited similar patterns of differential gene expression at the later time points of 72 and 168 hr , but Fluad had an additional earlier signal at 8 , 24 and 48 hr in the muscle ( Figure 1 ) . Gene-set enrichment analysis was performed to identify the functional patterns of the biological processes that are induced by the different immunisations . Gene modules that are specific to the context of immune responses in blood tissue were previously defined by Li et al . ( 2014 ) . These blood transcriptional modules were used to assess the signatures induced by the immunisations at the different time points . The blood transcriptional module enrichment profiles for every tissue are compared and presented in Figure 2 and Figure 2—figure supplements 1–3 , where the gene sets with significant enrichment ( lower than p<10−6 ) in each tissue are shown . In the injected muscle tissue , each vaccine or TLR agonist ( with the exception of Agrippal ) primarily elicited an upregulation of expression , as compared with the expresson levels seen after injection of the saline control , of genes related to inflammation , growth factors , innate immunity and cell damage ( Figure 2—figure supplement 1 ) . The induced transcriptomic response in the MLN was more refined , probably reflecting the effect of the migration of activated immune cells and local generalised stimulation by cyto/chemokine gradients . In these draining iliac lymph nodes , the injection of LPS , Pentavac SD and Poly I:C promoted an elevation of the interferon response genes as well as of cyto/chemokines ( Figure 2—figure supplement 2 ) . Finally , in the PBMCs , the transcriptomic analysis revealed that LPS , Pentavac SD and Poly I:C elicited an increase in gene transcripts that was associated with the interferon response and with activated monocytes and neutrophils ( Figure 2—figure supplement 3 ) . To identify transcriptional patterns in an unbiased way , gene modules were constructed using Weighted Gene Co-expression Network Analysis ( WGCNA ) ( Langfelder and Horvath , 2008 ) . Briefly , WGCNA discovers interacting genes by computing and matching the correlation patterns in their expression profiles over the entire time period , extracting clusters ( modules ) of highly correlated genes , and thereby categorizing the association between the modules and the traits ( different vaccines ) . WGCNA was applied to the gene expression datasets for muscle , lymph nodes , and blood separately . The co-expression analyses identified 40 modules in muscle with numbers of genes ranging from 27 to 4979 ( Figure 3—figure supplement 1A ) , 47 modules in lymph nodes with numbers of genes ranging from 26 to 3456 ( Figure 3—figure supplement 1B ) , and 30 modules in blood with numbers of genes ranging from 24 to 9789 ( Figure 3—figure supplement 1C ) . WGCNA assigns colours to name each module . We also added ‘mu’ for muscle , ‘ln’ for lymph nodes , or ‘bl’ for blood to each module name to avoid modules having the same name across different tissue datasets ( such as muscle module ( muM1_turquoise ) , lymph node module ( lnM1_turquoise ) and blood module ( blM1_turquoise ) ) . Figure 3—figure supplement 1 shows the correlations ( r >0 . 3 positive correlations are in blue , r <−0 . 3 negative correlations are in red ) between the modules and the immunisations . In muscle , co-expression analysis was able to identify modules that correlated with each immunisation , strongly ( both positive and negative ) for Pentavac SD and to a lesser degree for LPS , Fluad and Agrippal . These modules were different to those that correlated with a saline-only injection . In lymph nodes and blood , the method again revealed that Pentavac SD , LPS , Fluad and Agrippal treatments were associated with a number of positively and negatively significantly correlated modules . The connections between these highly correlated genes modules , which reveal whether the same sets of genes are co-regulated in the different tissues , are shown in Figure 3—figure supplement 1D , which identifies consensus modules associated with Agrippal , LPS and Pentavac SD . To interpret the extracted modules functionally , we compared the WGCNA modules with the reference blood transcriptional modules . A hypergeometric test was performed to quantify the overlap between the two sets of modules and the significant enrichments are reported in a heatmap ( Figure 3 ) . In muscle , the muM1-turquiose module contains genes that overlap with most of the reference modules and is associated positively with LPS and Pentavac SD; the muM2-blue module overlaps with reference modules of extracellular matrix ( ECM ) , migration and mitochondria and is negatively associated with LPS and Pentavac SD ( Figure 3 and Figure 3—figure supplement 1A ) . In lymph nodes , the lnM6-red module exhibits significant overlap with the reference modules associated with the cell cycle , whereas the lnM8-pink module is associated with interferon or antiviral sensing and this module contains genes that are co-regulated in LPS and Poly I:C immunisation ( Figure 3 and Figure 3—figure supplement 1B ) . In blood , the blM1-turquoise module overlaps with reference modules of T cell , whereas the M4-yellow module overlaps with reference modules of monocytes , neutrophils and inflammatory/TLR/chemokines . The M1-turquoise and M4-yellow modules contain genes of similar transcriptional patterns that are significantly associated with LPS and Pentavac SD immunisation ( Figure 3 and Figure 3—figure supplement 1C ) . We developed an interactive web interface ( available at https://vaccinebiomarkers . com ) to facilitate data access and further discovery . This website allows users to ( 1 ) query genes and visualise their transcriptional profiles for each condition , ( 2 ) filter the differentially expressed genes by their functional groups and visualise the fold changes , and ( 3 ) analyse WGCNA modules by visualising the functional enrichments and listing the genes of each module . In order to reveal potential connections between the different tissues , the differentially expressed genes and each sampling time point , we created a circular heatmap diagram showing genes that were common between the three tissues ( Figure 4 ) . We first selected the top 100 genes that were differentially regulated in both the injected muscle tissue and the MLN , then we assessed whether any of these genes were also differentially regulated in the circulating PBMC's RNA expression profiles . This analysis identified a set of genes that coded for soluble or cell-associated proteins present in each of the compartments that were differentially regulated by immunisation . We focused on soluble markers to simplify the sampling and quantification of potential blood biomarkers . The outside rim of the circular heatmap indicates the common individual genes , and the strength of the correlation for each tissue and each immunisation treatment is colour- and sized-coded . ( Black indicates positive correlation , red indicates negative correlation; the thickness of the lines corresponds to the magnitude of the correlation coefficient . ) Many of these genes encoded chemokines and/or cytokines but the analysis also identified proteins that collectively have been termed acute-phase proteins . These proteins are typically present at high levels during inflammatory events , and they include serum amyloid A-3 ( SAA3 ) and pentraxin 3 ( PTX3 ) . Murine SAA3 is an ortholog of the human SAA3 pseudogene and is involved in the murine response to bacterial endotoxins , often acting in combination with TLR2 ( Ather and Poynter , 2018; He et al . , 2009 ) , whereas the long pentraxin PTX3 facilitates pathogen recognition by macrophages and dendritic cells ( Diniz et al . , 2004 ) . SAA3 was strongly induced in muscle tissue after most of the immunisations ( the exception being Agrippal , whereas in the draining lymph node and the PBMC transcriptome , only LPS , Pentavac SD and poly I:C enhanced RNA expression levels . LPS caused the greatest alteration in gene expression profiles of cyto/chemokine genes in the blood , enhancing CCL2 , CCL3 , CCL4 , CXCL1 , CXCL2 , CXCL3 , CXCL9 , CXCL10 , CXCL13 and TNF-alpha expression but not affecting CXCL5 . These differences were seen mainly at the very early time points of 4 and 8 hr and were not maintained at 24 hr post intramuscular immunisation with the LPS . Notably , in the draining lymph node , the TNF-alpha levels are not elevated . In the muscle , there is strong general upregulation of IL-6 and elevated expression of IL-6 also occurs in the lymph node after LPS and Pentavac SD immunisation , whereas there is a generalised downregulation of IL-6 in the PBMCs . The licensed vaccines , surprisingly including the Pentavac SD which contained the whole cell pertussis , induced minor perturbations of the RNA expression levels in the PBMC . Poly I:C strongly upregulated CCL2 and CXCL10 and mildly enhanced CXCL1 , CXCL9 and CXCL13 in the blood , and interestingly this pattern was slightly different from the expression in the draining lymph node where CXCL9 , CXCL10 and CCL2 were enhanced while CXCL1 and CXCL13 were unaffected ( Figure 4 ) . We next examined whether a marker identified from the transcriptomics could be measured in the blood . We measured a panel of cyto/chemokines by Luminex and SAA3 by ELISA in sera harvested from mice vaccinated with saline control , the two licensed vaccines ( Pentavac SD and Fluad ) and the two potent TLR agonists ( LPS and Poly I:C ) . We selected these treatments on the basis of the range of responses observed from the transcriptomic analysis . Pentavac SD , LPS and Poly I:C had signatures of cytokine responses in the MLN , Fluad much less so , but such responses were still detectable , suggesting that these were good candidates for enabling cyto/chemokine detection in the peripheral blood sera ( Figures 2 and 4 ) . Serum was collected from five mice per itreatment and per analysis time point . LPS was clearly inflammatory at early time points after injection , eliciting strong expression of CCL2 , CCL3 , CCL4 , CCL5 , CXCL1 , CXCL2 and CXCL10 . Following treatment with LPS , TNF-α and IL-6 proteins were significantly above baseline and saline control levels at 4 and 8 hr , with a rapid return to much lower expression levels by 24 hr and to basal levels by 48–72 hr ( Figure 5A ) . Levels of CXCL10 were still significantly above those in controls at 24 hr after LPS injection , having reached a peak at 4 hr after treatment which then declined to almost half by 8 hr but was still significantly ( p<0 . 05 ) above controls and baseline at 503 pg/mL by 24 hr post immunisation . In addition , polyI:C elicited measurably elevated levels of a number of these cyto/chemokines , specifically CXCL10 and CCL5 at 4 hr ( CXCL10 – 6 , 263 pg/mL; CCL5 – 8 , 727 pg/mL , p<0 . 0001 ) and 8 hr ( CXCL10 – 1 , 462 pg/mL; CCL5 – 3 , 642 pg/mL , p<0 . 0001 ) and CCL2 and CCL4 at 4 hr ( CCL2 – 3 , 944 pg/mL; CCL4 – 1 , 249 pg/mL , p<0 . 0001 ) . Strikingly , all treatments apart from the saline control elicited very high levels of expression of the SAA3 protein , which were at least 1000-fold greater than those of any other measured analyte , and moreover the kinetics of SAA3 expression were also of a longer duration than those for expression of the cyto/chemokines ( Figure 5A ) . The TLR4 agonist LPS elicited the highest peak SAA3 response of 492 . 8 µg/mL at 24 hr ( p<0 . 0001 ) , which reduced considerably by 48 hr ( 99 . 74 µg/mL , p<0 . 0001 ) , reaching a level of 5 . 96 µg/mL by 72 hr and baseline levels after 168 hr . Although the Pentavac SD vaccination did not achieve the same peak level of SAA3 expression as the LPS injection , the levels of SAA3 continued to increase until 48 hr post-immunisation with Pentavac SD ( 332 . 6 µg/mL , p<0 . 0001 ) and were maintained until 72 hr ( 245 . 6 µg/mL , p<0 . 0001 ) , remaining significantly above baseline levels at the final analysis time point of 168 hr ( 104 . 8 µg/mL , p<0 . 0001 ) . A comparison of the total accumulation of SAA3 after the LPS or Pentavac SD vaccinations revealed that the AUC for Pentavac SD was more than twice that of LPS , being 31 , 455 µg . hr/mL and 14 , 572 µg . hr/mL , respectively . The TLR3 agonist Poly I:C invoked an SAA3 expression profile in which the molecule reached 33 . 11 µg/mL at 4 hr before falling back to 15 . 04 µg/mL at 8 hr and rising again to 33 . 54 µg/mL at 24 hr , although these differences did not reach statistical significance when compared to the saline control . Interestingly , treatment with Fluad , which contains the oil-in-water emulsion adjuvant MF59 , did not generate a peak in the expression level of SAA3 in the sera until 24 hr post-injection ( 81 . 82 µg/mL , p=0 . 0014 ) , suggesting that a delayed mechanism of action is induced by this emulsion . Figure 5B shows the fold changes over saline alone of measured proteins that are induced by different vaccines at the different time points , giving an indication of the degree and duration of expression over the background levels . In the case of the LPS immunisation , this analysis showed that an ‘expression set’ of cyto/chemokines and SAA3 can be defined to include CCL2 , CCL3 , CCL4 , CCL5 , CXCL1 , CXCL2 , CXCL10 , IL-6 and TNF-α but not SAA3 at 4 hr , then the same set of cyto/chemokines but including SAA3 at 8 hr . These comparisons of the differential transcriptomic expression in the blood with the actual levels of expressed proteins measurable in the animal sera revealed that many of the proteins closely matched . The strong expression of CCL2 , CCL3 , CCL4 , CXCL2 , CXCL10 , and TNF-α proteins that was elicited by LPS immunisation was in line with the measured transcriptomic changes at early time points ( 4 , 8 and 24 hr ) . By contrast , there is a downregulation in transcript levels for CCL5 , but CCL5 protein levels are elevated at 4 , 8 , and 24 hr . The sustained upregulation of SAA3 following the Pentavac SD vaccination was reflected in both transcript and protein levels . We next quantified the correlations between blood transcript and protein fold changes across all time points for all measured cyto/chemokines in the LPS immunisation group ( Figure 5—figure supplement 1 ) . CXCL1 , CXCL10 , and SAA3 showed strong correlations that were statistically significant ( r >0 . 6 , p-value <0 . 05 ) . Although LPS generated the highest levels of cytokine expression , the transcriptomic data also revealed different patterns of differential gene expression for the other treatments . To reveal any potential relationship between the differential gene expression in the different tissues and the protein signal that is detected in the serum , we compared the differential gene expression for each vaccine , from each tissue , over the whole period of study and correlated these gene expression levels to the serum protein expression levels at the same time points ( Figure 6 ) . Each vaccine or TLR agonist had different correlation profiles . The correlation between gene expression in the lymph node and blood and the blood serum proteins was most pronounced after LPS injection , with only the differential expression of CCL5 in the blood and the level of CCL5 serum protein being significantly negatively correlated . The muscle tissue had only one significant correlation with blood SAA3 levels in the LPS condition . By contrast , after Poly I:C injection , we identified a significant correlation between IL-6 transcript levels in the muscle tissue and IL-6 protein in the blood , although a number of other blood proteins correlated well but not significantly with the differential expression of their respective genes . Interestingly , the Fluad vaccine produced the highest number of significant correlations between gene expression in the muscle and the proteins present in the blood , indicating that most of the biomarker proteins present in the sera after Fluad vaccination may be derived from the injected muscle tissue . The Pentavac SD vaccine induced no significant correlations between the differentially expressed genes in the blood and the proteins in the blood , correlations were primarily noted between changes in gene expression levels in the MLN , and partially from the injected muscle , and serum protein levels . Despite the widespread use of vaccines and an enormous quantity of data from both animals and humans , the development , degree and mechanism of vaccine sensitivity or potential reactogenicity are still unclear . This is perhaps not surprising because the immune system is a complex network in which robust regulation to prevent run-away activation or self-attack is essential . We used a systems vaccinology approach combining multi-tissue transcriptomics with a simple serum analysis to define a set of biomarkers for potential vaccine-related inflammation in mice . We examined the transcriptomic profile of the injected muscle tissue , the draining lymph nodes and the peripheral blood after injection of four well-tolerated and safe licensed vaccines . These included a safe vaccine that is known to elicit moderate reactogenicity in many people because it contains whole cell Bordetella pertussis , as well as a number of molecularly defined TLR agonists that should generate significant inflammatory reactions . The added value of examining the transcriptomic response in the injected muscle tissues and then in the draining lymph node , and then finally determining whether there is a cellular signature in the blood , was that we could potentially follow the development of the response through these compartments . Of course , there will be cross-talk between these compartments because cells and inflammatory mediators travel from the injection site into the blood and to the lymph nodes and visa-versa , either early on as an indicator of an innate response to a danger signal ( e . g . vaccines plus any adjuvant ) or later as an indicator of cells arriving at the site of injection in response to innate signals generated in the muscle tissue and adaptive responses in the draining lymph node . From the violin plots that visualize the absolute numbers of genes that have significant differential expression when compared to the saline control or unvaccinated animals , it is clear that the Engerix B , Agrippal or Fluad vaccines produce a very low signal in any of the three compartments , the peripheral blood , the draining lymph nodes or the injected muscle . This is reassuring given that these are licensed vaccines with substantial history of safe use . However , for three investigational conditions — injection with the licensed vaccine Pentavac SD ( a safe but more reactogenic vaccine ) , with the TLR4 agonist LPS , or with the TLR3 agonist Poly I:C — there were a large number of differentially expressed genes that were measurable in the muscle , lymph nodes and the peripheral blood . Interestingly , these differences were particularly evident at the very earliest time points of 4 and 8 hr after immunisation , with many differentially expressed genes returning to baseline levels by 24 hr post immunisation . These extremely rapid changes are difficult to capture in studies investigating human responses to vaccination , with much published work only looking at transcriptomic changes at days 1 , 3 or 7 or after a much longer time period because the focus of their studies was to investigate and correlate the transcriptomic responses with the development and maturation of an adaptive immune response ( Querec et al . , 2009; Li et al . , 2014; Kazmin et al . , 2017; Haks et al . , 2017; Davis et al . , 2017 ) . In order to rationalize the large datasets generated by transcriptomic analysis , we used a gene set enrichment technique to identify groups of genes that were assigned to specific groups on the basis of the blood transcriptional modules defined by Li et al . ( 2014 ) and that were significantly enriched over the saline-alone injection . Our analyses revealed that the injected muscle tissue exhibited a large number of transcriptionally enriched modules , interestingly focused mainly on the innate and adaptive immune response , dendritic cells and T cells . As expected , the innate-sensing modules were elevated at earlier time points whereas the cell cycle module enrichments came later , mainly after 48 hours post injection . The response to Poly I:C was notable because this injection caused a significant enrichment of innate modules , but this did not follow through to generate any downstream cell-cycle response . Lower signals in the blood could be due to a number of potential reasons . Genes or sets of genes may be expressed only in the muscle , only in the draining lymph node or indeed only in the blood . The blood signature could also be a downstream result of the increased expression of a gene in the muscle or lymph node tissues , or alternatively might be caused directly by the systemic dissemination of the injected material . Alternatively , there could be an element that is expressed strongly in both the muscle and draining lymph node that we don’t see in the blood due to dilution of the responding cells in the large number of cells in the circulation , compared to , for example , a concentrated accumulation of responsive cells or expanding clones in the lymph node . A primary aim of this work was to identify a signal in the peripheral blood that reflected events occurring in the injected muscle site and/or in the draining lymph nodes; therefore , we next looked at potential connections between the different tissues . An analysis that created a circular heatmap diagram showing the interconnections between the three compartments revealed a strong positive correlation between the injected muscle and the lymph nodes for CCL3 , CCL4 , CXCL1 and IL-6 , with a much weaker connection for these molecules with the peripheral blood compartment . However , there was a positive correlation between the upregulation of CCL2 , CXCL10 and SAA3 gene expression in the tissues and the blood compartment , with the latter being particularly high in the first 48–72 hr . We then examined the serum protein levels of specific chemokines , cytokines and SAA3 to determine the relationship between the level of the transcript and the protein product . The expression of transcripts can be downregulated while the proteins are still being made , a level of control that isn’t unexpected , particularly for highly pleiotrophic proteins such as cyto/chemokines that need to be carefully regulated to prevent dysregulation of the immune and other systems ( Kovarik et al . , 2017 ) . It is also possible that high levels of transcripts may not result in high levels of protein because of the modulation of translation in highly active or stressed cells ( Grootjans et al . , 2016 ) . Luminex and ELISA analysis of the serum samples surprisingly showed that for many of the cyto/chemokines , the levels of proteins peaked within 4 hr of injection , returning to baseline levels by 24–48 hr post injection , with clear differences in the patterns of expression between the different treatments . LPS was by far the most potent inducer of all of the cyto/chemokines measured , with only Poly I:C eliciting similar levels of CCL5 and CXCL10 . In the case of the reactive protein SAA3 , both LPS and Pentavac SD caused a marked increase in the protein levels in the sera , with expression kinetics that were very different from those of cyto/chemokine expression and in quantities that were 100–1000-fold higher than those of the cyto/chemokines . When graphed in a volcano plot , showing the highly significant fold-change differential expression of the proteins , it is clear that groups of cyto/chemokines and SAA3 allow us to differentiate specific signatures for each individual vaccine condition . A comparison of blood transcript and serum protein data revealed significant and informative correlations . For LPS vaccination , there was a strong relationship between the differential gene expression of CXCL1 , CXCL10 and SAA3 and the corresponding proteins in the blood . However , a number of other cyto/chemokine proteins that were significantly elevated in the serum did not correlate with the expression of differentially regulated genes in the blood , posing the question of the source of the protein products of these differentially expressed genes . A further comparison of potential correlations revealed that there were marked relationships between differentially expressed genes in the muscle and draining lymph nodes and the serum proteins , indicating that chemokines produced in the muscle and lymph nodes were released into the circulating blood . In the current study , we have identified a set of biomarkers , or biological indicators , that are measurable in the peripheral blood and that reflect an inflammatory response resulting from an injection into a distal muscle . The biomarker identity was nuanced , with different immunisations generating identifiably distinct levels of serum biomarker proteins . These serum biomarkers were reflective of events in the injected muscle tissue , the draining lymph nodes or the peripheral blood . Our analysis demonstrates that the serum biomarker samples should be taken soon after immunisation , certainly within the first 12 hr , because as the signature is much reduced by 24 hr and an early sampling will be both practically convenient and will provide an early signal of adverse reactions . By looking only at those serum proteins that exhibited clear differentiation between the licensed and safe vaccines and those that were correlated with inflammation caused by the tested TLR agonists , we can create a panel of recommended biomarkers of potential inflammation . This panel would include SAA3 , CCL2 and CXCL10 , with the presence and expression level of all three of these proteins being indicative of adverse or potentially harmful inflammation . The presence of an acute-phase reactive protein such as SAA3 is perhaps not surprising , although these proteins are primarily produced by hepatocytes ( Tannock et al . , 2018 ) . Interestingly , this specific SAA-family member was identified through our analyses as being differentially regulated in the transcriptomes of each of the three tissues , whereas the more highly expressed SAA1 and SAA2 were not . Assessment of these biomarkers of potential inflammatory signals in future pre-clinical studies may give an early indication of unwanted or excessive inflammation , and could potentially identify vaccine candidates that have harmful or highly inflammatory profiles . With the aim of facilitating future discovery , we have created an interactive web interface as a tool for further interrogation and multivariate analysis of these data , which is accessible at https://vaccinebiomarkers . com . The mouse is typically the initial animal model used in early vaccine development and , rightly or wrongly , has traditionally functioned as an immunogenicity gatekeeper for the progression of vaccine candidates into higher animal models and ultimately to human clinical trials . Studies examining comparative transcriptomics in mice and men have variously described mouse models as having either a very poor or very high and significant correlation with human responses . Indeed a pair of articles famously utilized the same data sets to demonstrate both sides of this debate ( Seok et al . , 2013; Takao and Miyakawa , 2015 ) . The published datasets from these studies , though quite different from those in our current study in terms of the time points analysed , tissues examined and interventions used , did include an injection of LPS via the intraperitoneal route . An examination of the differential gene expression datasets after the LPS treatment in the Takao and Miyakawa ( 2015 ) work revealed that CXCL10 , CCL3 and CCL4 were among the top 50 positively regulated genes . SAA3 was not part of this dataset ( Seok et al . , 2013; Takao and Miyakawa , 2015 ) . Several studies have examined the transcriptomic signature of MF59 in blood , draining lymph nodes and injected muscle tissue ( Mosca et al . , 2008; Liang et al . , 2017; Francica et al . , 2017 ) . Transcriptomic analysis of mouse muscles injected with MF59 revealed a picture that is strikingly similar to that produced by our current study , in which significant fold-change elevations were noted for SAA3 , CCL2 , CCL4 , CCL5 and CXCL10 ( Mosca et al . , 2008 ) . A similar study that was also performed in mice , although using a different oil-in-water squalene emulsion ( AS03 ) , demonstrated changes in the gene expression levels that were similar to those observed in our study for the Fluad vaccination ( Morel et al . , 2011 ) . The CCL2 , CCL3 , CCL4 , CXCL1 , CXCL10 and IL-6 cyto/chemokines were significantly elevated at 4 and 24 hr post treatment in the injected muscle and also in the draining lymph nodes . Our identification of CCL2 and CXCL10 as potential markers of vaccine-elicited inflammation is therefore consistent with previously published studies . These chemokines are likely to play a significant role in the context of generating effective vaccine-elicited immunity as both are critical modulators of leukocyte trafficking and homing to the site of vaccine injection . CCL2 is a key chemokine regulating the movement of monocytes and macrophages , memory T cells and NK cells through vascular endothelium and the infiltration of the tissue into which the vaccine or inflammatory mediator has been applied ( Sozzani et al . , 1993 ) . CCL2 is readily produced by muscle cells in response to damage , inflammation or infection and potently recruits monocytes and macrophages upon engagement of the highly expressed CCR2 receptor , concomitantly setting up an autocrine amplification of the CCL2 chemokine in the monocytes and macrophages ( Cushing et al . , 1990; Yoshimura et al . , 1989; Chiu et al . , 2012 ) . This CCL2 amplification may further enhance inflammation at the site of injection and potentially the perceived immunological danger that a vaccine will lead to a subsequently higher level of immune responses . CCL2 has also been reported to influence the polarisation of developing antigen-specific T cells towards a Th2 phenotype , indicating that the presence of CCL2 in the cyto/chemokine milieu may influence the resulting vaccine-elicited T cell immunity ( Karpus et al . , 1997 ) . CXCL10 was first described as an interferon-induced chemokine that is produced by a wide range of cell types , including monocytes and dendritic cells ( Ciesielski et al . , 2002; Ge et al . , 2012; Ohmori and Hamilton , 1995; Holm et al . , 2012 ) . CXCL10 binds to CXCR3 and promotes immune cell trafficking and homing to inflamed tissues as well as the perpetuation of inflammation in a similar fashion to CCL2 . Likewise , CXCL10 has reported effects on T cell immunity and is critical for the generation of the protective CD8 T cell responses induced by activated dendritic cells . Both CCL2 and CXCL10 are potent recruitment signals for several of the key players in the generation of an immune response , monocytes or macrophages , neutrophils and dendritic cells . The presence of these cells at the site of injection has been shown to augment antigen-specific B and T cell immune responses significantly ( McKay et al . , 2004; Sumida et al . , 2004; Calabro et al . , 2011 ) . Our current study builds upon previous publications by comparing responses to four licensed vaccines with extensive safety and tolerability profiles to those produced by highly immunostimulatory TLR agonists and the IFA adjuvant . Furthermore , we performed a longitudinal transcriptomic analysis of the injected muscle site , the draining lymph nodes and the peripheral blood , as well as an analysis of proteins in the sera in order to begin to define the optimum time and the biomarkers that can be used to measure inflammatory responses after immunisation . While we have identified a set of inflammatory biomarkers that are applicable to mice , more work is needed to determine whether this set or a set of human biosignatures will act as potential signifiers of vaccine inflammation in humans . The animal studies were approved by the Ethical Review Board of Imperial College London , where the experiments were carried out and work was performed in strict compliance with project and personal animal experimentation licences granted by the UK government in accordance with the Animals in Scientific Procedures Act ( 1986 ) – PPL 70–7457 Protocol #1 . Animals received minimal handling and their physical condition was monitored at least twice daily . All procedures were performed under isoflurane anaesthesia when appropriate , and all efforts were made to minimize suffering . There was a detailed protocol in place , as required by the humane endpoints described in the animal licence , for early euthanasia in the event of onset of illness or significant deterioration in condition . At the end of the experiment , all animals were culled using a schedule one method and death confirmed before necropsy . Food and water were supplied ad libitum . Female CB6F1 mice of 6–8 weeks of age were purchased from Charles River . Animals received a single injection in their right hind leg quadricep muscle and were then culled at specific intervals after the immunisation , either 4 , 8 , 24 , 48 , 72 or 168 hr . A control group that did not receive an immunisation was culled and tissues harvested . A further control group received a single saline injection in the right hind leg quadriceps muscle and was then culled at the same intervals as the animals that received an active formulation . When each animal was culled , the injected muscle site and the iliac lymph nodes that drain the hind leg quadriceps muscles were harvested and flash frozen in liquid nitrogen . Peripheral blood , sampled from the mouse tail vein immediately before humane euthanasia , was collected ( 100 µL ) in RNAprotect animal blood tubes ( Qiagen , UK ) . Groups of mice ( n = 5 per group per time point ) received 1/10th of the human dose in 50 µL of one of the following licensed vaccines: Pentavac SD ( diphtheria , tetanus , pertussis ( whole cell ) , hepatitis B ( rDNA ) and Haemophilus influenzae type b conjugate vaccine ) ( Serum Institute India , Pune , India ) ; Agrippal ( trivalent flu subunits – H3N2 , H1N1 and influenza B ) ( Novartis Vaccines , now Seqirus , UK ) ; Fluad ( trivalent flu subunits – H3N2 , H1N1 and influenza B + MF59 ( oil-in-water emulsion ) ) ( Novartis Vaccines , now Sequirus , UK ) ; Engerix B ( recombinant hepatitis B surface antigen absorbed on aluminium ) ( GSK , Rixensart , Belgium ) , or either Poly I:C ( Sigma , UK – P0913: 50 µL of a 1 mg/mL solution ) , LPS ( Invivogen , UK – LPS-EB Ultrapure: 50 µL of a 0 . 5 mg/mL solution ) , IFA ( Seppic , France – Montanide ISA 51 VG: 50 µL of a 1:1 mixture of IFA and Saline ) , or saline alone ( Sigma , UK – 50 µL ) . The 1/10th of a human dose received by the mice was based on the ‘mouse equivalent dose’ . This estimation takes into account various measures and differences between animal species including the body surface area and metabolic rate and is an FDA accepted method for dose conversion ( Sharma and McNeill , 2009 ) . Gene expression data were generated from high-quality RNA samples on an Agilent microarray platform ( Agilent Technologies ) . RNA was labelled with a Low Input Quick Amp Labeling Kit ( Agilent Technologies ) according to the manufacturer’s instructions . Quantity and labelling efficiency were verified before hybridization to whole-genome 8 × 60 k mouse expression arrays ( Agilent design ID 028005 ) , and scanned at 5 μm using an Agilent scanner . Image analysis and data extraction were performed with Agilent's Feature Extraction software ( version 11 . 5 ) to generate the raw expression data . The complete set of microarray data was deposited in the NCBI’s Gene Expression Omnibus and is accessible through GEO accession number GSE120661 . Data analysis was performed in R version 3 . 3 . 2 ( 2016-10-31 ) . Microarray data were pre-processed , normalised and analysed for differential expression using R package limma v3 . 28 . 14 ( Ritchie et al . , 2015 ) . The raw data were first background corrected using the normexp method . Background corrected signals were quantile normalised between arrays . Linear models were fitted using the limma lmFit function . All treatments and time points were part of a single model and separate models were fitted for each tissue . There was an unvaccinated group for each treatment . Contrasts were designed to compare each of the different stimulus groups to unvaccinated animals and also to the saline control at each time point , using the interaction term ( treatment . time point – treatment . unvaccinated ) – ( saline . time point – saline . unvaccinated ) . Differential expression was evaluated using the moderated t-statistics and the p-values were adjusted using Benjamini and Hochberg’s ( BH ) method ( Benjamini and Hochberg , 1995 ) . The violin plots visualizing the strength of the transcriptomic responses were created using ggplot2 geom_violin , in which areas are scaled proportionally to the number of differentially expressed genes ( adj . p-value<0 . 01 ) . Genes that are orthologs in mice and humans were assigned using NCBI HomoloGene ( US National Library of Medicine , 2004 ) . Gene set enrichment analysis was performed with R package tmod ( version 0 . 34 ) using CERNO statistical test ( Weiner and Domaszewska , 2016; Yamaguchi et al . , 2008 ) . We calculated p-values corrected for multiple testing using the BH procedure and the effect size area under curve ( AUC ) of the gene set enrichment for blood transcriptional modules ( BTMs ) defined by Li et al . ( 2014 ) . BTMs were assigned to high-level annotation groups using the annotations defined by Kazmin et al . ( 2017 ) . Circular visualizations of the functional modules were performed using Circos version 0 . 69 . 4 ( Krzywinski et al . , 2009 ) . Highly concordantly as well as highly discordantly regulated genes between tissues were identified using the method described by Domaszewska et al . ( 2017 ) . Magnitude of gene expression change ( effect size ) , significance ( adj . p-value ) and direction of gene expression change are used to determine the discordance/concordance score ( disco . score ) . Weighted gene correlation network analysis ( R package WGCNA version 1 . 51 ) was used to find clusters of highly correlated genes among the stimulus groups ( Langfelder and Horvath , 2008 ) . Signed co-expression networks were constructed using the Pearson correlation as the similarity measure and the minimum module size was set to 20 . The first principal component of the expression matrix ( module eigengene ) of each constructed module is calculated using the moduleEigengenes function of the WGCNA package . Sample trait is provided as a binary indicator variable of the immunisation status . The Pearson correlations of the module eigengenes with traits were calculated to determine the association between several modules of co-expressed genes and the administered vaccines . We used BTMs to reveal the functional roles of the constructed WGCNA modules . A hypergeometric test was used to test the enrichment of co-expressed genes with genes specific to the BTMs defined by Li et al . ( 2014 ) . To find modules that are shared between muscle , lymph nodes , and blood networks , a consensus module analysis was carried out . Modules are connected on the basis of the number of genes they have in common and are visualized in a network diagram using the force-directed algorithm in R package igraph v1 . 0 . 1 . Groups of mice ( n = 5 per group per time point ) received 1/10th of the human dose in 50 µL of one of the licensed vaccines ( Table 1 ) . We analysed the sera of a subset of the conditions that were used for transcriptomics . These were selected on the basis of the differential gene expression analyses that had identified significant differences in the profiles and that also had the potential for these differences to be observed in the peripheral blood . Sera from immunised mice ( time points: 4 , 8 , 24 , 48 , 72 and 168 hr ) and baseline sera obtained from naïve mice ( n = 5 ) at time 0 hr were analysed using either a 9-plex Luminex or a Mouse Serum Amyloid A ELISA ( Bio-Techne , Abingdon , UK ) , according to the manufacturer’s protocols . Briefly , the Luminex filter-bottomed microplate was first pre-wet with 100 µL wash buffer then placed on the vacuum manifold to remove the buffer through the filter . 50 µL of the microparticle bead cocktail was added to each well followed by 50 µL of the diluted standard or sample and the mixture was incubated for 2 hr at RT on a horizontal orbital shaker set at 500 rpm . Wells were washed 3x by the addition of 100 µL wash buffer , with each wash being drawn through the filter membrane using the vacuum manifold . 50 µL of diluted biotin antibody cocktail ( specific for each analyte ) was added to each well and the microparticle beads were incubated with the biotinylated antibody mixture for 1 hr at RT on the orbital shaker at 500 rpm . The plates were washed again using 3 × 100 µL wash buffer per well , and then 50 µL of diluted streptavidin-PE was added to each well and the microparticle beads were incubated at RT for 30 min at 500 rpm on the orbital shaker . After a final 3 × 100 µL per well wash buffer on the vacuum manifold , the beads were resuspended with 100 µL wash buffer per well by incubation on the orbital shaker at 500 rpm for 2 min and immediately analysed on a Bio-Plex 200 System ( Bio-Rad Laboratories Ltd , UK ) . Serum samples were diluted 1:2 with the calibrator diluent provided within the Luminex kit ( LXSAMS-09: CCL2/MCP-1/JE; CCL3/MIP-1 alpha; CCL4/MIP-1 beta; CCL5/RANTES; CXCL1/GRO alpha/KC; CXCL10/IP-10; CXCL2/Gro beta/MIP-2/CINC-3; IL-6; TNF-alpha ) . Experimental samples were quantified against the standard for each cyto/chemokine . Briefly , for SAA3 , experimental samples were quantified by ELISA against the SAA3 standard on a pre-coated plate containing anti-mouse SAA3 capture antibody and a paired anti-mouse SAA3 detection antibody in a standard sandwich ELISA . Fold change of cyto/chemokine levels were calculated by dividing the mean values of proteins in immunisation groups by the mean values of proteins in the saline -alone group . The significance of the changes induced by different vaccines over those produced by saline alone were analysed using two-way ANOVA followed by Dunnett’s multiple comparisons test . Correlations between the fold changes of transcripts and proteins were calculated using the Pearson correlation coefficient .	Measles , whooping cough and other diseases can cause serious illness and death in humans , especially in young children and other vulnerable individuals . Giving people vaccines ‘trains’ their immune system to recognize and fight the microbes that cause the conditions . During an infection , the immune system triggers a set of responses that limit the spread of the infectious agent and eliminate it from the body . This can include swelling of tissues ( known as inflammation ) , which in rare cases , can be life threatening . Inoculations work by sparking a mild immune response in the body . Before a new vaccine is licensed for use , it is thoroughly tested in mice and rodents , and then in human volunteers , to ensure it will cause little or no inflammation . Finding a way to predict early on whether a vaccine candidate will trigger dangerous levels of inflammation would improve this process . To explore this , McKay , Cizmeci et al . injected the muscle tissue of different groups of mice with one of four licensed vaccines which , by definition , cause little or no inflammation . Other groups of animals were given one of three drugs known to trigger inflammation . Over the following seven days the team repeatedly collected blood as well as cells from the muscle tissue and the lymph nodes . These samples were then analysed to find out which genes were switched on or off at any given time . The experiments show that the responses of genes in the blood and lymph cells of the mice are connected to those in the muscle cells . Therefore , blood samples may provide a quick and convenient way to assess how an animal is responding to a potential new vaccine . By comparing the genes switched on or off in response to the different vaccines and drugs , McKay , Cizemeci et al . were able to identify a set of genes ( known as “biomarkers” ) that are associated with inflammation in animals . These biomarkers can be used to spot early on whether a new treatment is triggering inflammation . The next step would then be to identify a similar or identical set of biomarkers in other animals used in vaccine research , and in humans . Ultimately , this approach could make the assessment of the safety of a new vaccine candidate easier .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "tools", "and", "resources", "immunology", "and", "inflammation" ]	2019	Identification of potential biomarkers of vaccine inflammation in mice
Human neocortex expansion likely contributed to the remarkable cognitive abilities of humans . This expansion is thought to primarily reflect differences in proliferation versus differentiation of neural progenitors during cortical development . Here , we have searched for such differences by analysing cerebral organoids from human and chimpanzees using immunohistofluorescence , live imaging , and single-cell transcriptomics . We find that the cytoarchitecture , cell type composition , and neurogenic gene expression programs of humans and chimpanzees are remarkably similar . Notably , however , live imaging of apical progenitor mitosis uncovered a lengthening of prometaphase-metaphase in humans compared to chimpanzees that is specific to proliferating progenitors and not observed in non-neural cells . Consistent with this , the small set of genes more highly expressed in human apical progenitors points to increased proliferative capacity , and the proportion of neurogenic basal progenitors is lower in humans . These subtle differences in cortical progenitors between humans and chimpanzees may have consequences for human neocortex evolution . The expansion of the neocortex during primate evolution is thought to contribute to the higher cognitive capacity of humans compared to our closest living relatives , the great apes , and notably the chimpanzees ( Geschwind and Rakic , 2013; Rakic , 2009; Striedter , 2005 ) . Neocortex expansion in humans relative to chimpanzees involves an increase in the number of cortical neurons generated during fetal development ( Borrell and Reillo , 2012; Florio and Huttner , 2014; Herculano-Houzel , 2009; Lui et al . , 2011 ) . This reflects primarily a greater and prolonged proliferative capacity of human neural stem and progenitor cells ( NSPCs ) within the germinal zones of the developing neocortex ( Lewitus et al . , 2013 ) . Unravelling differences between human and chimpanzee NSPC behaviour is therefore a key issue , yet very little is known about such differences . The neocortex develops from two principal germinal zones , the ventricular zone ( VZ ) and the subventricular zone ( SVZ ) ( Angevine et al . , 1970 ) . In primates developing a folded ( gyrencephalic ) neocortex , and notably in humans , an inner SVZ ( iSVZ ) and an outer ( oSVZ ) can be distinguished ( Dehay et al . , 2015; Smart et al . , 2002 ) . Correspondingly , the VZ and SVZ harbour the cell bodies of two principal classes of NSPCs , called apical progenitors ( APs ) and basal progenitors ( BPs ) , respectively , each of which comprise several distinct NSPC types ( Borrell and Reillo , 2012; Götz and Huttner , 2005; Lui et al . , 2011; Taverna et al . , 2014 ) . APs ( neuroepithelial cells , apical radial glia , and apical intermediate progenitors ) divide at the ventricular surface , keep ventricular contact and exhibit apical cell polarity , whereas BPs ( basal ( or outer ) radial glia and basal intermediate progenitors ) lack this contact and type of cell polarity ( Taverna et al . , 2014 ) . Studies dissecting the switch between NSPC proliferation and differentiation have demonstrated that a central aspect of the cell division process , the orientation of the mitotic spindle , has a pivotal role , particularly in the case of APs ( Lancaster and Knoblich , 2012; Mora-Bermudez and Huttner , 2015; Mora-Bermudez et al . , 2014; Shitamukai and Matsuzaki , 2012 ) . The orientation of the spindle relative to the apical-basal axis of cell polarity in mitotic apical radial glia , the major cortical neural stem cells ( Götz and Huttner , 2005; Kriegstein and Alvarez-Buylla , 2009 ) , can determine whether their division is symmetric or asymmetric , and whether it is proliferative or neurogenic , with regard to their progeny ( Lancaster and Knoblich , 2012; Mora-Bermudez and Huttner , 2015; Mora-Bermudez et al . , 2014; Shitamukai and Matsuzaki , 2012 ) . Comparing spindle orientation in mitotic APs may therefore provide insight into the cell biological basis underlying the differences between humans and chimpanzees in NSPC proliferation versus differentiation during neocortex development . Protocols to generate structured cerebral tissue ( cerebral organoids ) from pluripotent stem cells in vitro constitute a major advance for studying neocortex development , in particular with regard to humans and non-human primates where fetal brain tissue is hard or impossible to obtain and manipulate ( Kadoshima et al . , 2013; Lancaster and Knoblich , 2014; Lancaster et al . , 2013; Mariani et al . , 2015; Qian et al . , 2016 ) . Human cerebral organoids form a variety of tissues that resemble specific brain regions , including the cerebral cortex , ventral forebrain , midbrain-hindbrain boundary , hippocampus , and retina . Moreover , their cerebral cortex-like regions exhibit distinct germinal zones , that is , a VZ containing APs and an SVZ containing BPs , as well as basal-most neuronal layers . Cerebral organoid APs include apical radial glia-like NSPCs that contact a ventricle-like lumen , express radial glia marker genes , undergo interkinetic nuclear migration , and divide at the apical surface , similar to their in vivo counterparts , and cerebral organoid BPs comprise both basal radial glia-like and basal intermediate progenitor-like NSPCs ( Lancaster et al . , 2013 ) . Finally , we have previously shown by single-cell RNA sequencing that the gene expression programs controlling neocortex development in human cerebral organoids are remarkably similar to those in the developing fetal tissue ( Camp et al . , 2015 ) . Together , these findings suggest that cerebral organoids constitute a valid system to explore potential differences in NSPC proliferation versus differentiation between humans and chimpanzees ( Otani et al . , 2016 ) , in particular with regard to spindle orientation in mitotic APs . Here , we have generated cerebral organoids from chimpanzee-derived induced pluripotent stem cells ( iPSCs ) , and used single-cell transcriptomics , immunohistofluorescence and live imaging to compare relevant features of chimpanzee NSPCs to human NSPCs in cerebral organoids and fetal neocortex . While most NSPC characteristics are found to be similar , we show that the prometaphase-metaphase in mitotic APs is longer in humans than in chimpanzees , indicating that a fundamental difference exists in the regulation of mitosis during neocortex development between the two species . Our data also provide a resource for further studies on human and chimpanzee differences in cortical development , and demonstrate the usability of cerebral organoids as a means to be able to perform such studies . We generated cerebral organoids from iPSCs derived from chimpanzee fibroblasts and lymphocytes ( Figure 1A left , Figure 1—figure supplement 1 ) . These chimpanzee cerebral organoids formed complex tissue structures that resembled the developing primate brain ( Figure 1A right ) , as reported previously for human cerebral organoids ( Lancaster et al . , 2013 ) . Similar to human iPSC-derived cerebral organoids ( [Camp et al . , 2015] , Figure 1B , C right ) , within the chimpanzee organoids grown for 52 days ( D52 ) , we observed cortex-like regions ( Figure 1A right ) with PAX6-positive APs ( such as radial glia ) residing predominantly in the apical-most zone facing a ventricular lumen ( Figure 1B left ) , similar to the ventricular zone ( VZ ) of developing primate neocortex at an early-mid stage of neurogenesis . Consistent with this , cells immunoreactive for the deep-layer neuron marker CTIP2 were observed in the basal region of the developing cortical wall ( Figure 1B left ) , corresponding to an early cortical plate . TBR2 ( also known as EOMES ) positive BPs ( presumably mostly basal intermediate progenitors ) were concentrated in a zone between the PAX6+ progenitors and the CTIP2+ neurons , corresponding to the subventricular zone ( SVZ ) . In the context of the time-lapse live imaging of apical mitoses described below , we observed apically directed nuclear migration prior to , and basally directed nuclear migration after , mitosis , consistent with the existence of interkinetic nuclear migration . Our results suggest that chimpanzee cerebral organoids recapitulate important aspects of fetal chimpanzee brain development and allow comparisons with cerebral cortex development in human cerebral organoids and fetal neocortex . 10 . 7554/eLife . 18683 . 003Figure 1 . Chimpanzee cerebral organoids recapitulate cortex development . ( A ) Bright-field image showing a representative chimpanzee organoid ( Sandra A , left ) and a cryosection from a chimpanzee organoid ( PR818-5 ) immunostained for PAX6 ( magenta ) and Ctip2 ( green ) combined with DAPI staining ( blue ) ( right ) at day 52 . Scale bars , 200 μm . ( B , C ) Cryosections of cortical regions from chimpanzee ( Sandra A ) and human ( SC102A-1 ) organoids at day 52 immunostained for PAX6 ( magenta ) , Ctip2 ( B , green ) and TBR2 ( C , green ) , without ( B ) and with ( C ) DAPI staining ( blue ) . Asterisks , ventricular lumen; scale bars , 50 μm . ( D ) Cartoon showing NSPC types ( APs , BPs ) and neurons enriched in zones within the neocortex at mid-neurogenesis . CP , cortical plate; N , neuron . ( E ) Heatmap showing normalized correlation ( Z-score ) of single-cell transcriptomes from chimpanzee cerebral organoid cortex with bulk RNA-seq data from laser-microdissected zones ( Fietz et al . , 2012 ) from 13 wpc human neocortex . CP , cortical plate . ( F ) Scatterplot showing NSPC and neuronal signature scores derived from analysis of fetal cerebral cortex single-cell transcriptomes ( Figure 1—figure supplement 1 ) calculated for each chimpanzee cerebral organoid cortical cell . ( G ) Heatmap showing expression of AP , BP , and neuron ( N ) marker genes . Each column represents a single cell , each row a gene . Cell type and maximum correlation to bulk RNA-seq data from cortical zones are shown in the top sidebar . APs and BPs were sub-classified based on G1 ( light grey ) or S-G2-M ( dark grey ) phases of the cell cycle . ( H ) Lineage network based on pairwise correlations between chimpanzee cerebral organoid cortical cells reveals a structured topology where VZ-APs connect to cortical plate ( CP ) neurons ( N ) through SVZ-BPs . Cells are coloured based on cortical zone ( top left ) or cell type assignment ( bottom left ) . APs , BPs , and neurons were classified based on maximum correlation with single-cell transcriptomes from the human fetal neocortex . Expression of markers PAX6 , TBR2 , and MYT1L are shown to the right . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 00310 . 7554/eLife . 18683 . 004Figure 1—source data 1 . Processed single-cell RNA-seq data for chimpanzee cells . * . txt file containing processed chimpanzee single-cell RNA-seq data ( 344 single cells ) in log2 ( FPKM ) with genes in columns and cells in rows . The first 7 columns contain metadata for each cell: cortex: assignment of cell to cortex ( 1 ) or to other regions within organoid ( 0 ) ; tSNE_1: tSNE1 loading for each cell; tSNE_2: tSNE2 loading for each cell; PC1: PC1 loading for each cell; PC2: PC2 loading for each cell; species: species of origin for each cell; cell_id: unique ID for each cell , with information about the experiment and the age of the organoid of origin for each cell . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 00410 . 7554/eLife . 18683 . 005Figure 1—source data 2 . Genes describing cell populations in the chimpanzee organoids . List of genes identified by PCA on all chimpanzee organoid single-cell transcriptomes as being most informative for defining cell populations . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 00510 . 7554/eLife . 18683 . 006Figure 1—figure supplement 1 . Characterization of chimpanzee iPSCs . ( A ) Chimpanzee iPSC line Sandra A stained for pluripotency markers SSEA5 ( red ) and NANOG ( Green ) . Nuclei are stained with DAPI . ( B ) PCA on bulk RNA-seq data from human iPSCs , chimpanzee and bonobo iPSCs , and human fibroblasts was used to describe the variation between cell types . RNA-seq data on chimp iPSC line Sandra A and human iPSC line 409b2 was generated in this study . Data from the other human , chimpanzee and bonobo , and fibroblast lines were previously published ( Ma et al . , 2014; Marchetto et al . , 2013 ) . ( C ) Dendrogram showing hierarchical clustering of human IPSC , chimpanzee and bonobo IPSC , and human fibroblast lines based on the Pearson correlation of the expression of 12 , 221 genes . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 00610 . 7554/eLife . 18683 . 007Figure 1—figure supplement 2 . Deconstructing cell type composition in chimpanzee cerebral organoids using single-cell RNA-seq . ( A ) scRNA-seq was performed on chimpanzee organoids dissociated at 45 , 50 , 51 , 55 , 62 , and 80 days ( d ) after embryoid body ( EB ) culture . PCA and unbiased clustering using tSNE reveals cell populations from hindbrain , midbrain , mesenchyme , and cerebral cortex ( shaded in grey ) within organoids . Different symbols indicate different experiments . ( B ) Marker genes are shown for each cluster with cells coloured based on gene expression level . Cerebral cortex cells have high expression of FOXG1 and NEUROD6 , and low expression of OTX2 and RSPO2 . Progenitors express marker SPAG5 . Cells are coloured based on expression level . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 00710 . 7554/eLife . 18683 . 008Figure 1—figure supplement 3 . Fetal human progenitor and neuronal neocortical signatures are recapitulated in chimpanzee cerebral organoids . ( A ) PCA of human fetal neocortex was used to identify genes describing cortical cell populations . Each dot represents a cell that is color-coded in shades of blue representing three different experiments . The genes correlating and anticorrelating with PC1 were used to define the NSPC and neuron signature , respectively . ( B ) Hierarchical clustering and heatmap visualization showing the expression of genes that have highest correlation ( NSPC signature ) and anti-correlation ( Neuron signature ) with PC1 . Cells are shown in rows , genes in columns . ( C , D ) Fetal cortical cells were classified as APs in G2-M ( AP1 ) , APs in G1-S ( AP2 ) , BPs in G2-M ( BP1 ) , BPs in G1-S ( BP2 ) , or migrating ( N1 , N2 ) and cortical plate ( N3 ) neurons . Each cell was scored for the NSPC ( top ) or neuron ( bottom ) signature and plotted in the order of pseudotemporal point on the neurogenic lineage ( C ) or plotted for each cell type ( D ) . ( E ) Scatterplot showing NSPC and neuronal signature scores for each human fetal , human organoid and chimpanzee organoid cortical cell . The signatures were derived from PCA of fetal cerebral cortex single-cell transcriptomes . ( F ) Heatmap showing gene expression of top NSPC and neuron signature genes across human fetal , human organoid ( hOrg ) , and chimpanzee organoid ( cOrg ) cells . ( G ) Monocle reveals a NSPC-to-neuron lineage in the chimpanzee organoid that correlates with the zones of the developing fetal primate neocortex . Cells ( circles , coloured by maximum correlation with cortical zones; CP , cortical plate ) are arranged in the 2-D independent component space based on genes identified using PCA . The minimal spanning tree ( grey lines ) connects cells , with the black line indicating the longest path . ( H ) Each chimpanzee cerebral organoid cortical cells scored for the NSPC ( top ) or neuron ( N , bottom ) signature and plotted in the order of pseudotemporal position on the neurogenic lineage . Cells are coloured by maximum correlation with cortical zones ( left ) or cell type ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 008 We next compared the proportion of various NSPC types , as revealed by expression of PAX6 and/or TBR2 , and neurons at a very early ( D28 ) and a mid-neurogenic ( D52-D54 ) stage between human and chimpanzee cerebral organoid cortices ( Figure 2 ) . In both species , we observed a decrease in PAX6+TBR2– apically located NSPCs ( presumably proliferating APs ) from D28 to D52 , concomitant with an increase in PAX6+TBR2+ and PAX6–TBR2+ basally located NSPCs ( presumably neurogenic BPs ) ( Figure 2A , B ) . Notably , whereas no significant differences were observed at D28 , at D52-D54 , the proportion of PAX6+TBR2+ NSPCs in the chimpanzee organoids was nearly twice that in the human organoids , and the proportion of PAX6+TBR2– NSPCs was correspondingly lower , whereas no significant difference between human and chimpanzee was observed for PAX6–TBR2+ NSPCs ( Figure 2B ) . In line with what would be expected with regard to neuron production , the proportion of PAX6–TBR2– cells , located in the basal-most zones of the developing cortical wall , was very low at D28 but increased by D52-D54 to about a third of the total cells for both , human and chimpanzee cerebral organoids ( Figure 2B ) . Immunostaining for CTIP2 corroborated the neuronal identity of these cells ( data not shown ) . 10 . 7554/eLife . 18683 . 009Figure 2 . Changes in the proportion of cortical NSPC subtypes and neurons during human and chimpanzee cerebral organoid development . ( A ) Cryosections of cortical regions from human and chimpanzee organoids at day 28 and day 52 immunostained for PAX6 ( magenta ) and TBR2 ( green ) combined with DAPI staining . Scale bars; D28 , 10 μm; D52 , 20 μm . Insets in the D52 merge images show selected areas with PAX6+TBR2+ double-positive nuclei ( arrowheads ) at higher magnification . ( B ) Quantification of the percentage of PAX6+TBR2– , PAX6+TBR2+ , PAX6–TBR2+ and PAX6–TBR2– cortical cells in human ( light grey ) and chimpanzee ( dark grey ) organoids at D28 ( n = 5 organoids , 50 μm wide field ) and D52-D54 ( n = 17 organoids , 100 μm wide field ) . Error bars , SEM; p<0 . 05 , p<0 . 01 . ( C ) Cryosections of cortical regions from human and chimpanzee organoids at D53 immunostained for KI67 ( yellow ) combined with DAPI staining ( blue ) . Scale bars , 20 μm . ( D ) Quantification of KI67+ cells in a 100 μm wide field in human and chimpanzee organoids at D52-D53 ( n = 7 ) . Error bars , SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 009 Consistent with the observation that the total proportion of NSPCs relative to neurons was virtually identical in human and chimpanzee organoids ( Figure 2B ) , the abundance of cycling cells , as revealed by KI67 immunostaining , was essentially similar ( Figure 2C , D ) . We conclude that at the two stages studied , there are – with the exception of the PAX6+TBR2+ NSPCs – no major differences between human and chimpanzee cerebral cortex developing in organoid culture with regard to the types of NSPCs and their abundance , or neuron output . To survey the cellular composition and cell type-specific transcriptomes of the chimpanzee organoids , we analysed 344 single cell transcriptomes from 7 organoids ranging in age from 45 to 80 days ( Figure 1D , Figure 1—source data 1 ) . We combined all transcriptomes and identified the genes most informative for defining cell populations by principal component analysis ( PCA ) ( Figure 1—source data 2 ) . Using these genes , we used tSNE analysis to cluster cells into transcriptionally distinct groups representing cerebral cortex , hindbrain , ventral midbrain and peripheral mesenchyme ( Figure 1—figure supplement 2 ) . These groups are similar to those identified in human cerebral organoids ( Camp et al . , 2015 ) . We identified 178 cortex-like cells based on strong expression of canonical NSPC and neuron marker genes ( i . e . , NSPCs and neurons: FOXG1 , NFIA , NFIB; NSPCs: PAX6 , SOX2 , GLI3; neurons: NEUROD6 ) and the lack of expression of the ventral and medial telencephalic markers OTX2 and RSPO2 ( Figure 1—figure supplement 2 ) . We sub-classified the 178 cerebral cortex-like cells based on the correlation between their transcriptomes and the bulk transcriptomes of laser-capture microdissected VZ , iSVZ , oSVZ , and cortical plate of fetal human neocortex ( GSE38805 , [Fietz et al . , 2012] ) . We found that groups of cells correlated best with one of the four zones , suggesting that the range of cell types present in the human fetal and organoid cerebral cortex are represented in our chimpanzee data ( Figure 1E ) . Consistent with this , each chimpanzee cell represents a cell state on a continuum from NPSCs to neurons based on gene expression signatures extracted from fetal human cerebral cortex transcriptomes ( Figure 1F , Figure 1—figure supplement 3 ) ( Camp et al . , 2015 ) . We next classified the chimpanzee cerebral cortex cells by determining the fetal cell type with which each cell most strongly correlates , resulting in 73 APs , 25 BPs , and 80 neurons . Analysis of known cell type markers revealed expression patterns consistent with what has been observed in human organoid and fetal cerebral cortex ( Figure 1G ) ( Camp et al . , 2015 ) . Though this classification is convenient to describe the cell types present in the chimpanzee organoid , we note that many of the cells can be described as intermediates between APs , BPs , and different stages of neuron maturation . We inferred lineage relationships among the chimpanzee cerebral cortex in an adjacency network based on pairwise correlations between cells ( Figure 1H ) , revealing a structured topology where VZ-APs connect to cortical plate neurons through SVZ-BPs . These lineage relationships were corroborated using a minimal spanning tree algorithm ( Figure 1—figure supplement 3G ) ( Trapnell et al . , 2014 ) . Together , these data allowed reconstruction of the chimpanzee organoid cerebral cortex from single-cell transcriptomes . To further explore transcriptome similarities and differences between chimpanzee and human cerebral cortex cells , we compared them to the single-cell transcriptomes of 220 fetal human cortex cells ( 12–13 weeks post-conception ( wpc ) , published in ( Camp et al . , 2015 ) , GSE75140 ) and 207 cortex-like cells from human cerebral organoids ( 40–80 days , 155 single-cell transcriptomes published in ( Camp et al . , 2015 ) , GSE75140; 52 single-cell transcriptomes acquired as part of this study ) ( Figure 3—source data 1 ) . In a PCA , the first principal component ( PC1 ) separated NSPCs and neurons , whereas PC2 separated species ( Figure 3A ) . Hierarchical clustering of organoid and fetal cells showed that human and chimpanzee organoid and human fetal cells were distributed together within the two main sub-clusters representing NSPCs and neurons ( not shown ) , and showed highly correlated expression of marker gene patterns ( Figure 3B ) . 10 . 7554/eLife . 18683 . 010Figure 3 . Comparing human and chimpanzee cerebral cortex gene expression . ( A ) PC1 and PC2 from PCA separated NSPCs and neurons , and human and chimpanzee , respectively . PCA was performed on all single-cell transcriptomes using genes expressed in more than two cells and with a non-zero variance . ( B ) Quasibinomial fit line of representative marker gene expression across cells ordered by correlation with PC1 . ( C ) Lineage network based on pairwise correlations between human fetal , human organoid , and chimpanzee organoid cells reveals a differentiation topology from VZ APs through BPs in iSVZ and oSVZ , to cortical plate ( CP ) neurons , with inter-species mixing in all stages . ( D ) Lineage network ( see ( C ) ) coloured by scaled expression level of marker genes . ( E ) Scatterplots showing z-scored significance estimates from single-cell differential expression ( SCDE ) analysis based on Bayesian probabilistic models . Reads from human and chimpanzee were mapped to a consensus genome , and human gene annotations were used for expression counting . The x-axis represents SCDE between human organoid APs vs . human organoid neurons . The y-axes on the left and right plots represent SCDE between human and chimpanzee APs and neurons ( N ) , respectively . Genes coloured as white triangles represent marker genes from Figure 1 and are generally not differentially expressed between human and chimpanzee , but do vary between APs and neurons , validating the SCDE analysis . Yellow and purple circles represent genes upregulated specifically in human APs and neurons , respectively . Circles are sized based on differential expression between human APs and neurons . Figure 3—figure supplement 1 shows a similar plot from the chimpanzee perspective . ( F ) Gene ontology enrichments ( -log10 P-value ) for differentially expressed gene groups shown in panel E . Left , human APs ( yellow ) and neurons ( N , purple ) that are not differential between human and chimpanzee . Center , upregulated in human APs ( top ) or neurons ( N , bottom ) compared to chimpanzee . Right , upregulated in chimpanzee APs ( top ) or neurons ( N , bottom ) from Figure 3—figure supplement 1 . ( G ) Left , expression profiles of ITGB8 and INSR are shown from human organoid , chimpanzee organoid , and human fetal cells ordered by correlation with PC1 . Right , bulk RNA-seq data from sorted aRG , bRG , and neurons ( N ) from human and mouse developing neocortex ( Florio et al . , 2015 ) confirms enriched expression of ITGB8 and INSR in human APs and neurons , respectively . ( H ) The same bulk RNA-seq data was used to confirm and estimate the origin of differential gene expression in APs versus neurons from single-cell organoid data . Pie chart shows the proportion of AP-enriched ( yellow ) or neuron-enriched ( N , purple ) genes that are observed in human , chimpanzee , and mouse . Pie charts also show the proportion of genes differential between APs and neurons that are observed only in human and chimpanzee , but not mouse ( human-chimp ancestor ) , or genes specific to human or chimpanzee . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 01010 . 7554/eLife . 18683 . 011Figure 3—source data 1 . Processed single-cell RNA-seq data for human cells . . txt file containing processed human single-cell RNA-seq data ( 207 single cells ) in log2 ( FPKM ) with metadata in first 4 columns for each cell: cell_id: unique ID for each cell; experiment: the experiment during which each cell was isolated; species: species of origin for each cell; cortex: assignment of cell to cortex ( 1 ) or to other regions within organoid ( 0 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 01110 . 7554/eLife . 18683 . 012Figure 3—source data 2 . Results of differential gene expression analyses . Excel file ( * . xlsx ) with multiple sheets containing results of all differential expression analyses presented in the manuscript as well as GO enrichment analysis for the differentially expressed ( DE ) genes: Sheet 1: Genes specific to APs , not DE between chimpanzee and human; Sheet 2: GO enrichment analysis for genes of sheet 1; Sheet 3: Genes specific to Neurons , not DE between chimpanzee and human; Sheet 4: GO enrichment analysis for genes of sheet 3; Sheet 5: Genes specific to APs and upregulated to human compared to chimpanzee; Sheet 6: GO enrichment analysis for genes of sheet 6; Sheet 7: Genes specific to Neurons and upregulated to human compared to chimpanzee; Sheet 8: GO enrichment analysis for genes of sheet 7; Sheet 9: Genes specific to APs and upregulated to chimpanzee compared to human; Sheet 10: GO enrichment analysis for genes of sheet 6; Sheet 11: Genes specific to Neurons and upregulated to chimpanzee compared to human; Sheet 12: GO enrichment analysis for genes of sheet 11; Sheet 13: GO enrichment data used to generate Figure 3F . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 01210 . 7554/eLife . 18683 . 013Figure 3—figure supplement 1 . Differential expression analysis between chimpanzee and human cerebral cortex cells from the chimpanzee perspective . ( A ) Scatterplots showing z-scored significance estimates from single-cell differential expression ( SCDE ) analysis based on Bayesian probabilistic models . Reads from human and chimpanzee were mapped to a consensus genome , and human gene annotations were used for expression counting . The x-axis represents SCDE between chimpanzee organoid APs vs . chimpanzee organoid neurons ( N ) . The y-axes on the left and right plots represents SCDE between human and chimpanzee APs and neurons , respectively . Genes coloured as white circles represent marker genes from Figure 1 and are generally not differentially expressed between human and chimpanzee , but do vary between chimpanzee APs and neurons , validating the SCDE analysis . Yellow and purple circles represent genes upregulated specifically in chimpanzee APs and neurons , respectively . Circles are sized based on differential expression between chimpanzee APs and neurons . ( B ) Plot showing the number of differentially expressed genes between human and chimpanzee cells as a function of standard deviations above the mean z-score from the Bayesian differential gene expression analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 013 We constructed an intercellular correlation network , which revealed a VZ sub-network of human and chimpanzees APs that link through BPs expressing iSVZ and oSVZ markers to cortical plate neurons . Generally , APs , BPs , and neurons from human and chimpanzee intermixed , confirming that cells in the chimpanzee organoid cortices have a zonal organization consistent with what is observed histologically ( Figure 3C , D ) . In conclusion , the major proportion of the variation in these data is not between in vitro and in vivo tissues or between species , but among cell states during neurogenesis , confirming that the major features of the genetic programs regulating the NSPC-to-neuron lineage are conserved between human and chimpanzees , and are recapitulated in cerebral organoids . To identify genes differentially expressed between chimpanzee and human cortex-like cells , we remapped all single-cell transcriptome reads to a consensus human-chimpanzee genome and used human annotations to identify 1-to-1 orthologous genes . We then used a Bayesian approach to identify differentially expressed genes by comparing cerebral organoid APs and neurons between species ( ignoring BPs due to the low number of BPs identified ) . We identified 297 and 279 genes that were more highly expressed in human APs and neurons , respectively , and 283 and 314 genes that were more highly expressed in chimpanzee APs and neurons , respectively ( Figure 3E , Figure 3—source data 2 ) . In addition to the between-species comparisons , we identified genes differentially expressed between human or chimpanzee APs and neurons to identify cell-type specific genes ( for human: 1328 AP-specific , 1132 neuron-specific; for chimpanzee: 1501 AP-specific , 1166 neuron-specific ) . The vast majority ( 94% ) of genes that are AP-specific and neuron-specific in humans are not differential between human and chimpanzee ( Figure 3E , Figure 3—figure supplement 1 ) . Of the differentially expressed genes between species , we identified 93 genes that are strongly upregulated in human organoid APs and 72 genes upregulated in human organoid neurons . Gene ontology enrichments suggest that the proteins encoded by some of these genes are integral to cell membranes and involved in intercellular signalling ( Figure 3F , Figure 3—source data 2 ) , for example integrin beta 8 ( ITGB8 ) in APs and insulin receptor ( INSR ) in human neurons . This upregulation of ITGB8 specific to human APs and INSR specific to human neurons is also observed in comparisons between human and mouse ( Florio et al . , 2015 ) ( Figure 3G ) . When comparing these results to bulk RNA-seq data from mouse APs and neurons ( Florio et al . , 2015 ) , we find that 75% of the genes with expression specific to APs or neurons in humans are also specific to each cell type in the mouse , suggesting that these gene expression programs were already established and likely present in the common ancestor of mouse , human and chimpanzee some 90 million years ago ( Figure 3F ) . Notably , a similar proportion of AP- and neuron-specific genes were gained on the chimpanzee and human branch subsequent to their separation , suggesting that our analysis did not have a strong human bias . About 12% of these genes specific to AP or neurons in both human and chimpanzee were not specific to these cell types in the mouse ( Florio et al . , 2015 ) , suggesting that they may be involved in developmental processes specific to the primate cerebral cortex . We used an established live imaging method ( Mora-Bermudez et al . , 2014 ) to compare dividing cortical APs , i . e . cells undergoing mitosis at the ventricular surface ( presumably mostly aRG ) , in slice cultures of both 11–13 wpc human fetal neocortex and human D30 cerebral organoids . We did not observe signs of strong perturbation during live image acquisition in either system , such as mitotic arrest ( Figure 4A , C , E; see also Figure 5A–C and Video 1 ) or lack of nuclear movements and cell death . Chromosome dynamics and spindle orientation of APs , as revealed by the orientation of the metaphase plate , were similar in human developing neocortex and human organoids , both before anaphase ( Figure 4A–D , G ) and during anaphase ( Figure 4A–D , H , I ) , when cell cleavage initiates . This strongly suggests that cerebral organoids are a suitable model to study live NSPC division and spindle orientation dynamics . 10 . 7554/eLife . 18683 . 014Figure 4 . Spindle orientation variability is similar between APs of human developing neocortex , human organoids and chimpanzee organoids . Live tissue imaging of spindle orientation , as reported by chromosome plate orientation , in organotypic slice culture of developing neocortex and cerebral organoids . Measurements were started after all chromosomes had formed a tight metaphase plate . 0 min is anaphase onset . Time-lapse is ∼1 . 1 min . ( A , C , E ) APs in a coronal slice of 13 wpc human frontal neocortex ( A ) , in a slice of a D30 human cerebral organoid from iPSC line SC102A-1 ( C ) , and in a slice of a D30 chimpanzee cerebral organoid from iPSC line Sandra A ( E ) . The time indicated on each image is when that image was taken , relative to anaphase onset ( 0 min ) . White dashed lines , ventricular surface . Yellow dashed lines indicate the two metaphase plate orientations with the greatest difference to each other . Scale bar , 5 μm . ( B , D , F ) Quantification of all orientations of the chromosome plates from the beginning of the metaphase plate stage to anaphase , for APs in the three respective tissues described in ( A , C , E ) . To facilitate tracing , individual tracks are colour-coded according to the initial range of the track , and the 90°−0° range is depicted twice ( green and yellow , 90°−75°; cyan and red , 75°−60°; blue and dark red , 60°−0°; 90° indicates perfectly vertical chromosome plates ) . ( G ) Maximal range of chromosome plate orientations for APs , from the beginning of the metaphase plate stage to anaphase onset , as determined in the measurements shown in ( B , D , F ) . Data are the mean ± SEM of ≥34 APs from 3 independent experiments each . ( H , I , J ) Orientation of chromosome plates at 2 . 2 min after anaphase onset , which indicates the predicted plane of cleavage , as determined in the measurements shown in ( B , D , F ) . 90° indicates a perfectly vertical cleavage plane . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 01410 . 7554/eLife . 18683 . 015Figure 5 . Differences in prometaphase-metaphase length between APs of human developing neocortex , human organoids , chimpanzee organoids and mouse developing neocortex . Live tissue imaging of mitotic phases , as reported by chromosomes , in organotypic slice culture of developing neocortex and cerebral organoids . 0 min is anaphase onset . Time-lapse is ∼1 . 1 min . ( A–D ) APs in a coronal slice of 13 wpc human frontal neocortex ( A ) , in a slice of a D30 human cerebral organoid from iPSC line SC102A-1 ( B ) , in a slice of a D30 chimpanzee cerebral organoid from iPSC line Sandra A ( C ) , and in a coronal slice of E14 . 5 mouse neocortex . The time indicated on each image is when that image was taken , relative to anaphase onset ( 0 min ) . White dashed lines , ventricular surface . Scale bar , 5 μm . ( E–G ) Time between the start of chromosome congression and anaphase onset ( referred to as 'prometaphase + metaphase' ) ( E ) , between the start of chromosome congression and the formation of a metaphase plate ( referred to as 'prometaphase' ) ( F ) , and between the formation of a metaphase plate and anaphase onset ( referred to as 'metaphase' ) ( G ) , for APs in the four tissues described in ( A–D ) . Data include APs from 11–13 wpc human neocortex , organoids from the human iPSC lines SC102A-1 and 409b2 , and chimpanzee iPSC lines Sandra A and PR818-5 , and are the mean ± SEM of ≥60 APs from ≥4 independent experiments each . Bracket with p<0 . 01; brackets with p<0 . 001; *p<0 . 001 ( mouse vs . all primate tissues ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 01510 . 7554/eLife . 18683 . 016Figure 5—source data 1 . Durations of all mitotic phases . Numerical values in minutes for the duration of all mitotic phases ± SEM used in the graphs in Figures 5 , 6 and 7 , in Figure 5—figure supplement 1 , 2 and 3 , and in Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 01610 . 7554/eLife . 18683 . 017Figure 5—figure supplement 1 . The length of the mitotic phases other than prometaphase-metaphase is similar between human and chimpanzee APs . Length of prophase , anaphase and telophase ( A ) , and of total mitosis ( B , sum of all mitotic phases described here and in Figure 5 ) between APs of human developing neocortex , human and chimpanzee cerebral organoids and mouse developing neocortex , determined from the experiments described in Figure 5 . Data are the mean ± SEM of ≥60 APs from ≥4 independent experiments each . p<0 . 05; **p <0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 01710 . 7554/eLife . 18683 . 018Figure 5—figure supplement 2 . Differences in prometaphase-metaphase length between APs of D30 and D52 human and chimpanzee cerebral organoids . Mitotic phase measurements similar to those in Figure 5E–G , but for APs in D52 organoids . Time between the start of chromosome congression and anaphase onset ( referred to as 'prometaphase + metaphase' ) ( A ) , between the start of chromosome congression and the formation of a metaphase plate ( referred to as 'prometaphase' ) ( B ) , and between the formation of a metaphase plate and anaphase onset ( referred to as 'metaphase' ) ( C ) . Data include APs from organoids from the human iPSC line SC102A-1 and chimpanzee iPSC line Sandra A , and are the mean ± SEM of 30 APs from 2 independent experiments each . For comparison , the relevant data for human and chimpanzee D30 cerebral organoid APs from Figure 5 are shown ( dashed lines ) . p<0 . 05; p<0 . 01; p<0 . 001; n . s . , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 01810 . 7554/eLife . 18683 . 019Figure 5—figure supplement 3 . Prometaphase-metaphase in orangutan organoid APs is similar to chimpanzee organoid APs . Live tissue imaging of mitotic phases , as reported by chromosomes , in organotypic slice culture of orangutan cerebral organoid . 0 min is anaphase onset . Time-lapse is ~3 . 5 min . ( A ) AP in a slice of orangutan D30 cerebral organoid ( Toba ) . The time indicated on each image is when that image was taken , relative to anaphase onset ( 0 min ) . White dashed line , ventricular surface . Scale bar , 5 μm . ( B–D ) Time between the start of chromosome congression and anaphase onset ( referred to as 'prometaphase + metaphase' ) ( B ) , between the start of chromosome congression and the formation of a metaphase plate ( referred to as 'prometaphase' ) ( C ) , and between the formation of a metaphase plate and anaphase onset ( referred to as 'metaphase' ) ( D ) . For comparison , the relevant mitotic phase lengths of human and chimpanzee cerebral organoid APs from Figure 5 are shown ( columns with dashed line ) . Data for orangutan are the mean ± SEM of 16 cells from 2 different cortex regions of an organoid . p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 01910 . 7554/eLife . 18683 . 020Figure 5—figure supplement 4 . Determination of cell cycle parameters of human and chimpanzee organoid APs using cumulative EdU labeling . ( A ) Schematic representation of the cumulative EdU labeling experiment . ( B ) Linear regression curves of human ( r2 = 0 . 984 ) and chimpanzee ( r2 = 0 . 998 ) PAX6+TBR2– cells after cumulative EdU labeling over 48 hr . The total cell cycle length minus the S-phase length ( Tc-Ts ) was obtained directly from the graph ( vertical dashed lines ) . The S-phase length ( Ts ) was determined from the x-axis and y-axis intercepts of the linear regression curves , and the total cell cycle length ( Tc ) was calculated therefrom . The growth fraction ( GF , solid horizontal line ) is indicated by orange arrowheads . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 02010 . 7554/eLife . 18683 . 021Video 1 . Differences in prometaphase-metaphase length between APs of human and chimpanzee cerebral organoids . Related to Figure 5B and C Live tissue imaging of mitotic phases , as reported by chromosomes , in organotypic slice culture of cerebral organoids . Time-lapse is ∼1 . 1 min . Datasets are the same as in Figure 5B and C . Left side: APs in a slice of a D30 human cerebral organoid from iPSC line SC102A-1 . Right side: APs in a slice of a D30 chimpanzee cerebral organoid from iPSC line Sandra A . Growing colour bars at the bottom indicate time progression of the respective dividing AP and are synchronized to the beginning of prometaphase ( in green ) . Metaphase plate time is in yellow and anaphase time is in red . Note the slower progression of the dividing human AP on the left . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 021 Spindle orientation can determine symmetric vs . asymmetric NSPC division ( Lancaster and Knoblich , 2012; Mora-Bermudez and Huttner , 2015; Mora-Bermudez et al . , 2014; Shitamukai and Matsuzaki , 2012 ) and is therefore a major candidate mechanism to explain the approximately 3-fold expansion of the neocortex in humans compared to great apes . We compared spindle orientation dynamics between human and chimpanzee APs in cerebral organoids . However , our data revealed no clear differences in spindle orientation , either during metaphase ( Figure 4C–G ) or shortly after anaphase onset ( Figure 4C–F , I–J ) . As deduced from the orientation of the chromosome plates , most APs in both human and chimpanzee divided with a cleavage orientation largely perpendicular to the apical , ventricular surface , showing deviations of fewer than 30° from a perfect orthogonal cleavage . Oblique and near-horizontal orientations were also observed , but at a much lower abundance and at similar frequencies in chimpanzee and human organoids ( Figure 4H–J ) . This shows that the frequency of asymmetric cell division caused by oblique spindle orientation is most likely not a major difference between human and chimpanzee APs . We noticed , however , unexpected differences between human and chimpanzee APs in their progression through mitosis . Specifically , measurement of the length of the various phases of mitosis ( for details , see Materials and methods ) revealed that APs in 11–13 wpc fetal human neocortex and D30 cerebral organoids remained approximately 5 min longer in prometaphase-metaphase than APs in chimpanzee organoids ( Figure 5A–C , E; Video 1 ) . By comparison , prometaphase-metaphase of APs in slice culture of mouse neocortex , a well-characterized model system for neurogenesis , lasted for only approximately half the amount of time than human APs ( Figure 5D , E; Figure 5—source data 1 ) . To trace the specific phase of mitosis when this difference arises , we used chromosome morphology and dynamics to determine the time chromosomes spent congressing toward the equatorial plane of the cell ( defined here as 'prometaphase' ) and the time they spent tightly aligned as a metaphase plate ( defined here as 'metaphase' ) . Remarkably , the longer prometaphase-metaphase of human than chimpanzee APs was specifically due to a ∼40-60% lengthening of metaphase ( Figure 5A–C , G ) , whereas prometaphase was not significantly different ( Figure 5A–C , F; Video 1 ) . By contrast , in mouse APs , both prometaphase and metaphase were found to be significantly shorter than the respective mitotic phases in human and chimpanzee APs ( Figure 5D , F , G; Figure 5—source data 1 ) . None of the other mitotic phases ( prophase , anaphase , telophase ) differed in length between APs in human fetal neocortex and human cerebral organoids vs . chimpanzee organoids . However , anaphase of mouse APs was found to be significantly shorter than that of human and chimpanzee APs ( Figure 5—figure supplement 1A; Figure 5—source data 1 ) . These differences between species in the individual mitotic phases were reflected in the cumulative length of total mitosis , which was significantly shorter in mouse APs than human and chimpanzee APs ( Figure 5—figure supplement 1B ) . To search for potential functional implications of these observations , we next quantified and compared the length of prometaphase-metaphase in human and chimpanzee APs of day 52 ( D52 ) cerebral organoids , and compared the results with those of D30 organoids . Prometaphase-metaphase ( Figure 5—figure supplement 2A ) and metaphase alone ( Figure 5—figure supplement 2C; Figure 5—source data 1 ) were shorter in D52 than in D30 human APs , and not anymore statistically significantly different in length from D52 chimpanzee APs . The longer metaphase of human than chimpanzee organoid APs may therefore characterise early phases of cortical development , when proliferative AP divisions are predominant . We also generated cerebral organoids from an orangutan iPSC line and determined the length of AP prometaphase-metaphase . This revealed that the length of prometaphase-metaphase in orangutan D30 organoid APs was similar to that of chimpanzee APs and significantly shorter than that of human organoid APs ( Figure 5—figure supplement 3A , B ) . As was the case for the human-chimpanzee AP comparison , the shorter prometaphase-metaphase of orangutan than human APs was due to a shorter metaphase ( Figure 5—figure supplement 3A , D ) rather than prometaphase ( Figure 5—figure supplement 3A , C; Figure 5—source data 1 ) . Together , these data indicate that human APs specifically lengthen prometaphase-metaphase as compared to great ape APs . In light of these differences in the duration of mitotic phases , it was of interest to compare the length of the total cell cycle of human and chimpanzee organoid APs . Using cumulative EdU labelling of D52-D54 cerebral organoids ( Figure 5—figure supplement 4A ) , we found a relatively minor ( ~6% ) difference in total cell cycle length , with human APs ( PAX6+TBR2– cells ) exhibiting a ~2 . 7 hr longer cell cycle ( 46 . 5 h ) than chimpanzee APs ( 43 . 8 h ) ( Figure 5—figure supplement 4B ) . However , a notable difference between the two species was the length of S-phase , which was nearly 5 hr longer in human ( 17 . 5 h ) than chimpanzee ( 12 . 8 h ) organoid APs ( Figure 5—figure supplement 4B ) . To investigate the origin of the longer metaphase in human than chimpanzee APs , we measured mitotic phase lengths in the original iPSCs used to grow the cerebral organoids . This revealed that both the human and chimpanzee organoid APs had a longer prometaphase-metaphase than their respective iPSCs of origin , showing that this general lengthening was due to the transition between iPSCs and the organoids of both species ( Figure 6A , B , E ) . In human APs , however , the lengthening was greater than in chimpanzee APs . In contrast to APs , human and chimpanzee iPSCs had similar prometaphase-metaphase lengths ( Figure 6A , B , E; Figure 5—source data 1 ) . 10 . 7554/eLife . 18683 . 022Figure 6 . Human and chimpanzee organoid APs exhibit longer prometaphase , and human organoid APs longer metaphase , than their iPSC lines of origin or B cells . Live imaging of mitotic phases , as reported by chromosomes , in human and chimpanzee iPSCs and B cells . 0 min is anaphase onset . Time-lapse is ∼1 . 1 min . ( A–D ) Human iPSC ( SC102A-1 ) ( A ) , chimpanzee iPSC ( Sandra A ) ( B ) , human B cell ( A158 ) ( C ) , and chimpanzee B cell ( Dorien ) ( D ) . The time indicated on each image is when that image was taken , relative to anaphase onset ( 0 min ) . Scale bar , 5 μm . ( E–G ) Time between the start of chromosome congression and anaphase onset ( referred to as 'prometaphase + metaphase' ) ( E ) , between the start of chromosome congression and the formation of a metaphase plate ( referred to as 'prometaphase' ) ( F ) , and between the formation of a metaphase plate and anaphase onset ( referred to as 'metaphase' ) ( G ) . Data include cells from each of the following iPSC lines: human , SC102A-1 and 409b2; chimpanzee , Sandra A and PR818-5; and from the following B cell lines: human , A144 , A156 and A158; chimpanzee , Jahaga , Ulla and Dorien . For comparison , the relevant mitotic phase lengths of human and chimpanzee cerebral organoid APs from Figure 5 are shown ( columns with dashed line ) . Data are the mean ± SEM of ≥30 cells from ≥3 independent experiments each . p<0 . 01; **p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 02210 . 7554/eLife . 18683 . 023Figure 6—figure supplement 1 . The length of prophase , anaphase and telophase is similar in human and chimpanzee iPSCs , B cells and organoid APs . Length of prophase , anaphase and telophase ( A ) , and of total mitosis ( B , sum of all mitotic phases described here and in Figure 6 ) in human and chimpanzee iPSCs and B cells , determined from the experiments described in Figure 6 . For comparison , the relevant mitotic phase lengths of human and chimpanzee cerebral organoid APs from Figure 5–figure supplement 1 are shown ( columns with dashed line ) . Data are the mean ± SEM of ≥30 cells from ≥3 independent experiments each . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 023 Further dissection into individual phases revealed that , whereas both human and chimpanzee APs had a longer prometaphase than their iPSCs of origin ( Figure 6A , B , F ) , only human APs had a longer metaphase when compared to the iPSCs of origin ( Figure 6A , B , G; Figure 5—source data 1 ) . This shows that prometaphase-metaphase lengthened in both species as APs were generated during cerebral organoid formation with the accompanying neural differentiation . However , the lengthening characteristics were species-specific . The lengthening was greater in humans than chimpanzees because the metaphase plate stage became longer only in human APs . To determine whether prometaphase-metaphase length may differ between chimpanzees and humans also in another cell type , we measured mitotic phases in human and chimpanzee B cells . In contrast to fetal tissue , these cells can be obtained not only from humans but also chimpanzees by collecting blood , that is , without major invasive procedures . The length of prometaphase-metaphase in B cells , as well as prometaphase and metaphase measured individually , were similar to that in iPSCs ( Figure 6 C–G ) , and hence significantly shorter than in human or chimpanzee APs . By contrast , the other mitotic phases were similar among organoid APs , iPSCs and B cells in both species ( Figure 6—figure supplement 1; Figure 5—source data 1 ) . This raises the intriguing possibility that lengthening of prometaphase-metaphase could be specific to ape and human NSPCs and , furthermore , that lengthening of the metaphase plate time could be specific to human NSPCs . To investigate potential functions of prometaphase-metaphase lengthening , we asked whether mitotic phases were different between proliferating and neurogenic APs . To this end , we measured mitotic phase lengths in a transgenic mouse line where EGFP is expressed under the promoter of the pan-neurogenic marker Tis21 in neurogenic but not proliferative NSPCs ( Haubensak et al . , 2004; Iacopetti et al . , 1999 ) . This revealed that prometaphase-metaphase was longer in proliferative AP divisions ( Tis21– ) than in neurogenic AP divisions ( Tis21+ ) , whereas the separate phases were not significantly different ( Figure 7; Figure 5—source data 1 ) . These results suggest that a lengthening of prometaphase-metaphase may be characteristic of proliferating NSPCs . 10 . 7554/eLife . 18683 . 024Figure 7 . Prometaphase-metaphase is longer in proliferative than neurogenic mouse APs . Live tissue imaging of mitotic phases , as reported by chromosomes , in organotypic slice culture of E14 . 5 mouse neocortex . 0 min is anaphase onset . Time-lapse is ~1 . 1 min . Data is from the same mouse dataset shown in Figure 5 , but distinguishes between Tis21::GFP– ( proliferative ) and Tis21::GFP ( neurogenic ) APs . ( A , B ) APs in a coronal slice of mouse E14 . 5 dorsolateral telencephalon , either negative ( A ) or positive ( B ) for expression of Tis21::GFP . The time indicated on each image is when that image was taken , relative to anaphase onset ( 0 min ) . White dashed lines , ventricular surface . Scale bar , 5 μm . Image panels in ( B ) are the same as in Figure 5D , but the Tis21::GFP fluorescence ( green ) is included in the prophase image ( merge ) . The GFP channel is also merged in the prophase image of ( A ) , and the other panels are DNA staining only . ( C , D ) Length of prometaphase and/or metaphase in proliferative vs . neurogenic APs . Data are the mean ± SEM of 41 Tis21::GFP– and 37 Tis21::GFP APs from 4 independent experiments . p<0 . 05 . ( C ) Time between the start of chromosome congression and anaphase onset ( referred to as 'prometaphase + metaphase' ) . ( D ) Time between the start of chromosome congression and the formation of a metaphase plate ( referred to as 'prometaphase' , left ) , and time between the formation of a metaphase plate and anaphase onset ( referred to as 'metaphase' , right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 024 We used the single-cell RNA-seq data to identify organoid APs in different phases of the cell cycle ( Figure 8A , Figure 8—figure supplement 1 ) and searched for genes that might be involved in human-specific lengthening of the metaphase . We compared human organoid APs in G1 with APs in G2-M and identified 395 genes with enriched expression in G2-M ( Figure 8B ) . We next compared human APs in G2-M with human iPSCs ( TkDA3-4 ) and an endothelial cell line ( HUVEC; both single-cell RNA-seq data sets in GSE81252 ) to understand the specificity of G2-M regulation in APs . We found that nearly all genes upregulated in human APs in G2-M compared with human APs in G1 were also upregulated during G2-M in iPSCs and endothelial cells ( Figure 8C ) . Therefore , the expression level of these genes is unlikely to contribute to the specificity of mitotic control of human APs in G2-M . However , we identified many genes that were highly expressed throughout the human AP cell cycle and were specific to APs . Genes with the highest specificity score encoded canonical cerebral cortex patterning transcription factors such as PAX6 , ID4 , and GLI3 , as well as proteins involved in cell adhesion and ECM signalling ( CDH4 , EFNB1/2 , COL4A2 ) . Notably , no genes associated with cell cycle , kinetochore , or spindle terms were specific to human APs ( Figure 8C , inset ) . Of genes specific to APs , a subset were also differentially expressed between human and chimpanzee cerebral organoids ( APOLD1 , BICC1 , EFNB1 , GSTM1 , IFI44L , ITGB8 , SDK2 , SEMA5A , SLC35F1 , ZNF516 ) , which makes them candidates for the unique regulation of AP proliferation in humans ( Figure 8D ) . 10 . 7554/eLife . 18683 . 025Figure 8 . Differential gene expression during AP mitotic phases . ( A ) G1 ( dark green ) and G2-M ( light green ) cell cycle phases were assigned to cells by performing PCA using genes involved in cell cycle regulation . PC1 and PC2 described cell cycle phases , and the top 50 correlating and anticorrelating genes were used to infer an intercellular correlation network for human and chimp APs , human iPSCs , and a human endothelial cell line . Networks are coloured based on the expression level of MKI67 . ( B ) Scatterplot shows z-scored significance estimates from single-cell differential expression ( SCDE ) analysis between human organoid APs vs . neurons ( N , x-axis ) and APs in G2-M vs . APs in G1 ( y-axis ) . Genes coloured as white circles represent marker genes and green circles represent genes upregulated specifically in APs in G2-M . Circles are sized based on differential expression between human APs and neurons . ( C ) iPSC and endothelial cell ( EC ) expression was used to assign a specificity score for genes enriched in human organoid APs compared to neurons ( higher in AP genes from panel B ) . The specificity score is plotted against the differential expression between APs in G2-M and APs in G1 . Cells with high AP specificity scores are in yellow in the main scatter plot . This shows that nearly all genes enriched in G2-M phase of the AP cell cycle are not specific to APs , but also enriched in G2-M of mitotic iPSCs and endothelial cells . ( D ) Heatmap shows the differential expression score between human and chimpanzee APs ( z-score ) and AP specificity score ( Log2 normalized ) of the same genes that are specific to APs relative to endothelial cells and iPSCs . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 02510 . 7554/eLife . 18683 . 026Figure 8—figure supplement 1 . Cell cycle assignment for differential gene expression analysis . ( A ) Hierarchical clustering was used to identify human organoid APs that most strongly expressed genes enriched in G2M phase of the cell cycle ( red ) . The genes were identified from PCA on fetal cortex progenitor cells ( top 100 correlating genes ) ( Camp et al . , 2015 ) . The cluster with weakest expression of these G2M associated genes was assigned as G1 phase ( blue ) . Intermediate cells ( grey ) were discarded from differential gene expression analysis . ( B ) A previously published method was used to computationally assign cell-cycle stage based on a machine-learning approach ( Scialdone et al . , 2015 ) . This method was generally consistent with our assignment based on the hierarchical clustering presented in panel A . ( C–F ) The same approach was used to identify the chimpanzee organoid APs , endothelial cells ( ECs ) , and iPSCs that most highly express G2M markers . Note that all iPSCs analyzed highly expressed most of the G2M marker genes . DOI: http://dx . doi . org/10 . 7554/eLife . 18683 . 026 We have characterized cerebral organoids generated from chimpanzee iPSCs , including a newly generated iPSC line , and shown that their cytoarchitecture , cell type composition , and neurogenic gene expression programs are remarkably similar to human cerebral organoids and to human fetal neocortex . This extends a very recent study ( Otani et al . , 2016 ) and establishes cerebral organoids as a valid system to compare human and chimpanzee NSPC behaviour . Using this system , we have shown that human and chimpanzee APs differ in that prometaphase-metaphase is longer in humans than in chimpanzees . This difference was also observed between human and orangutan and reflects a greater extent of prometaphase-metaphase lengthening that occurs as human APs are generated during cerebral organoid development from IPSCs . There are two intriguing implications as to the biological significance of this prometaphase-metaphase lengthening in human APs . One is related to the fate of the progeny arising from AP division . Mouse Tis21::GFP-negative APs , which are known to undergo proliferative divisions to generate more APs , have a longer prometaphase-metaphase than Tis21::GFP-positive APs , which are known to undergo neurogenic divisions to generate BPs ( Haubensak et al . , 2004 ) . The longer prometaphase-metaphase in human than chimpanzee APs would therefore be consistent with a greater tendency for proliferative than neurogenic divisions . In this respect , other changes in progeny fate have also been recently observed in a different context , upon an experimentally induced and considerable prolongation of AP mitosis in embryonic mouse neocortex ( Pilaz et al . , 2016 ) . Another set of observations are consistent with the notion that the longer prometaphase-metaphase in human than chimpanzee APs may indicate a greater tendency for proliferative than differentiative divisions . The human vs . chimpanzee prometaphase-metaphase difference decreased in the course of organoid cortical development from D30 to D52 , when one would expect proliferative AP divisions to decrease and differentiative AP divisions to increase . Further support for this notion was obtained by analysis of the interphase of the cell cycle , specifically S-phase . Mouse Tis21::GFP-negative ( proliferative ) APs have previously been shown to have a longer S-phase than Tis21::GFP-positive ( differentiative ) APs ( Arai et al . , 2011 ) . The substantially longer S-phase of human than chimpanzee APs observed here is therefore also in line with human APs having a greater tendency for proliferative divisions . Finally , the changes in the abundance of NSPC types in the course of cerebral organoid development yielded data supporting a greater AP proliferation in human than chimpanzee . Specifically , the proportion of PAX6+TBR2– NSPCs , located in the VZ and thus constituting proliferating APs , decreased in both human and chimpanzee cerebral organoids , but the value reached in human organoids was slightly higher than that in chimpanzee organoids ( Figure 2B ) . Conversely , the proportion of PAX6+TBR2+ NSPCs , located in the basal VZ and SVZ and constituting BPs with neurogenic potential , showed a greater increase in chimpanzee than human cerebral organoids . In sum , two independent lines of evidence , the detailed analysis of AP mitosis phase lengths and the determination of the proportions of the various NSPC types , support the concept that a longer neurogenic period ( Lewitus et al . , 2014 ) , which in turn implies a longer phase of NSPC proliferation ( Otani et al . , 2016 ) , contributes to the greater expansion of the neocortex in humans than the great apes . The second implication as to the biological significance of the longer prometaphase-metaphase in human than chimpanzee APs concerns the fact that these are the phases of mitosis when chromosomes prepare for segregation , to ensure that only one copy of each chromosome is distributed to each nascent daughter cell ( Musacchio and Salmon , 2007 ) . The longer duration of prometaphase-metaphase in human than chimpanzee APs , in particular of the metaphase plate stage ( Figure 5B ) , may therefore reflect some difference between the two species with regard to the preparation for chromosome segregation . If the longer prometaphase-metaphase in human than chimpanzee APs reflects a greater tendency for proliferative than neurogenic divisions in the human NSPCs , why did we not detect significant differences between human and chimpanzee APs in spindle orientation , a parameter previously shown to affect the mode of AP division ( Lancaster and Knoblich , 2012; Mora-Bermudez and Huttner , 2015; Mora-Bermudez et al . , 2014; Shitamukai and Matsuzaki , 2012 ) ? This may be due to spindle orientation variability between individual APs being greater than potential inter-species differences . This suggests that , in the cell types and stages analysed , spindle orientation may not play a key role in human vs . chimpanzee neurogenesis . Alternatively , this may reflect the fact that differences in proliferative versus neurogenic AP divisions can occur without a change in spindle orientation ( Konno et al . , 2008; Kosodo et al . , 2004; Mora-Bermudez and Huttner , 2015 ) . In this context , differences between human and chimpanzee NSPCs of relevance for neocortex expansion are likely to be small . Consistent with this view , our single-cell transcriptome analyses revealed only few differences between human and chimpanzee , and the differences in the proportions of organoid NSPC populations were in the range of a few percentage points . Furthermore , the ~5 min longer prometaphase-metaphase in human than chimpanzee APs constituted only a fraction of the total duration of their mitosis . These small differences nevertheless provide a set of clues as to which NSPC features may underlie the differential extent of neocortex expansion in humans versus apes , and are consistent with a scenario in which the accumulation of such small differences during evolution may have resulted in the distinct chimpanzee and human neocortices . Human fetal brain tissue ( 11–13 weeks post conception ( wpc ) ) was obtained with informed written maternal consent followed by elective pregnancy termination , and neocortex was dissected at room temperature , as described previously ( Florio et al . , 2015 ) . Research involving human fetal brain was approved by the Ethical Review Committee of the Universitätsklinikum Carl Gustav Carus of the Technische Universität Dresden ( reference number EK100052004 ) . In addition , research was approved by the Institutional Review Board of the Max Planck Institute of Molecular Cell Biology and Genetics . Mice were kept pathogen-free at the Biomedical Services Facility of the MPI-CBG . All mouse embryos were heterozygotes of the Tis21::GFP knock-in line ( Haubensak et al . , 2004 ) . Imaging was performed in the dorsolateral telencephalon of E14 . 5 embryos , at a medial position along the rostro-caudal axis . Embryonic day ( E ) 0 . 5 was defined as noon of the day of vaginal plug identification . All experiments using mice were performed according to the German Animal Welfare Legislation . Two human iPSC lines ( 409b2 , SC102A-1 ) , two chimpanzee iPSC lines ( PR818-5 , Sandra A ) , and one orangutan iPSC line ( Toba ) were used to generate cerebral organoids in this study . 409b2 was purchased from the RIKEN BRC cell bank and SC102A-1 was purchased from System Biosciences . PR818-5 ( 0818 ) was obtained as a kind gift from F . Gage ( Marchetto et al . , 2013 ) from the Salk Institute for Biological Studies ( La Jolla , CA ) . Sandra A and Toba were generated in collaboration with Shinya Yamanaka following a nonviral transfection method ( Okita et al . , 2013 ) . Briefly , blood was collected from a chimpanzee and an orangutan , both housed at the Leipzig Zoo , and leukocytes were isolated by Ficoll gradient centrifugation , which were then used for reprogramming to iPSCs . DNA sequencing revealed no chromosome aberrations , and RNA-seq and immunohistochemistry confirmed pluripotent gene and protein expression signatures . Primate blood samples used to generate iPSCs were obtained by certified veterinarians during annual medical examinations or other necessary medical interventions , meaning that no invasive procedures were performed on primates for the sole purpose of our research project . The Max Planck Institute for Evolutionary Anthropology has an institutional permit for the transport of biological material derived from endangered species ( DE216-08 , see http://cites . org/common/reg/si/e-si-beg . shtml ) . Human iPSC line TkDA3-4 ( Takebe et al . , 2013 ) was used to generate iPSC single-cell transcriptomes . iPSC lines were cultured under standard iPSC culturing methods on matrigel ( BD Biosciences ) using mTeSR1 ( Stemcell Technologies ) . Human endothelial cells . ( HUVECs , Lonza , Basel , Switzerland ) were maintained in endothelial growth medium ( EGM ) ( Lonza ) at 37°C in a humidified 5% CO2 incubator . Single cell transcriptome analysis confirmed the identity of human and chimpanzee iPSCs and human endothelial cells , and showed no contamination with other cell lines . B-cell lines were generated from blood obtained from three human ( A144 , A156 , A158 ) and three chimpanzee ( Dorien , Jahaga , Ulla ) individuals . Withdrawal and processing of blood samples was performed according to approved protocols , and was performed for chimpanzee during necessary veterinary interventions . Lymphocytes were isolated from blood using a Ficoll gradient centrifugation . Immortalization was performed by adding Epstein Barr virus ( EBV ) supernatant to the lymphocytes and further cultivation of the cells until colonies of immortalized B-lymphocytes were established ( Tosato and Cohen , 2007 ) . B-cells were maintained in RPMI with 10% FBS , 1% Glutamax and 2% penicillin/streptomycin . Cell lines were regularly tested for mycoplasma using a PCR-based test ( Minerva Biolabs ) and found to be negative . Human and chimpanzee cerebral organoids were generated from the above iPSCs and cultured for the indicated times as described previously for human cerebral organoids ( Lancaster and Knoblich , 2014; Lancaster et al . , 2013 ) , with minor modifications ( Camp et al . , 2015 ) . Cerebral organoids were fixed with 1% PFA in 120 mM phosphate buffer pH 7 . 4 for 20 min at room temperature and subjected to cryosectioning ( 14 µm ) and immunofluorescence as described ( Camp et al . , 2015 ) . The following primary antibodies were used: rabbit anti-PAX6 ( PRB-278P; Covance ) , sheep anti-TBR2 ( AF6166; R+D systems ) , rat anti-CTIP2 ( ab18465; Abcam ) , rabbit anti-KI67 ( ab15580; Abcam ) . The secondary antibodies , used in combination with DAPI staining , were all donkey-derived and conjugated with Alexa 488 , 555 or 647 ( Life Technologies ) . Images were acquired with a Zeiss LSM 880 Airy inverted microscope , using 10X ( 0 . 45 NA ) and 20X ( 0 . 8 NA ) Plan-Apochromat objectives , and analysed using Fiji . Quantifications were carried out in cortical regions of D28 and D52-54 cerebral organoids by counting , from the ventricular to the pial surface , either all PAX6 and TBR2 positive and negative nuclei stained by DAPI in 50 μm and 100 μm wide fields , respectively , or all KI67-positive cells in 100 μm wide fields . An average of 350 cells per sample were counted . Statistical significance was calculated using the Mann–Whitney U-test . EdU was added to 52 day old cerebral organoids at a final concentration of 1 μg/ml ( added from a 1 mg/ml EdU stock in PBS ) . The organoids were supplied with fresh medium containing EdU every six hours for up to 48 hr . Organoids were then collected in triplicates at the indicated time points ( 1 , 2 , 6 , 24 , 30/36 , 48 hr ) and processed as described above . For EdU detection , the Click-iT EdU Alexa Fluor 647 Imaging Kit ( Invitrogen C10340 ) was used according to the manufacturer’s instructions . Cell cycle parameters were determined using linear regression based on a model described previously ( Nowakowski et al . , 1989 ) . Live tissue imaging was performed as described previously ( Mora-Bermudez et al . , 2014 ) . In short , cerebral organoids or freshly dissected developing neocortex tissue were embedded in agarose ( Sigma , Germany ) , sectioned with a vibratome ( ~200 µm , Leica , Germany ) , embedded in type Ia collagen ( Cellmatrix , Japan ) , mounted in glass bottom microwell dishes ( MatTek , Germany ) , and incubated with Hoechst 33342 ( Sigma ) as vital DNA dye . Tissue slices in the dish were further cultured for observation in a microscope stage incubation chamber ( Pecon , Germany ) kept at 37°C . iPSCs and B cells were likewise mounted in glass bottom microwell dishes previously coated for 1h with matrigel ( BD Biosciecne ) and poly-D-lysine ( Sigma , Germany ) respectively , and imaged under their respective standard culturing conditions ( see above ) . Potential phototoxicity was stringently controlled as previously described ( Mora-Bermudez and Ellenberg , 2007 ) .	The human brain is about three times as big as the brain of our closest living relative , the chimpanzee . Moreover , a part of the brain called the cerebral cortex – which plays a key role in memory , attention , awareness and thought – contains twice as many cells in humans as the same region in chimpanzees . Networks of brain cells in the cerebral cortex also behave differently in the two species . How these species differences arise is not clear , but it likely occurs in the earliest phases of development when brain stem and progenitor cells divide and give rise to cerebral cortex cells in the growing brain . To study the earliest stages of brain development , researchers often use human brain cells grown in the laboratory . Under the right conditions , cells collected from adult humans and other animals can be reprogrammed to behave like brain stem cells . Recently , researchers have been able to use these reprogrammed cells to make tissue that resembles the brain in petri dishes , known as brain organoids . Mora-Bermúdez , Badsha , Kanton , Camp et al . have now analysed brain organoids grown from reprogrammed human , chimpanzee and orangutan cells . The experiments showed that the human and chimpanzee brain organoids were remarkably similar in many ways including in the mix of cell types and in how these cells were arranged . Mora-Bermúdez et al . then used live microscopy to show that progenitor cells that form the human cerebral cortex spend around 50% more time in a stage of the cell division process called metaphase compared to the same cells from chimpanzees or orangutans . Metaphase is the part of the division process when the cell makes sure that structures called chromosomes , which carry the cell’s DNA , can be separated and distributed equally between the two daughter cells . Mora-Bermúdez et al . also found that progenitor cells more likely to become neurons sooner had a shorter metaphase than progenitor cells more likely to remain proliferating as stem cells for longer . This suggests that a longer metaphase may be a feature of brain stem cells . Further studies are now needed to find out how the length of time these progenitor cells spend in metaphase affects how chimpanzee and human brains develop; and whether this can help explain why the human brain is so much larger .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2016	Differences and similarities between human and chimpanzee neural progenitors during cerebral cortex development
A 5 . 3 Å resolution , cryo-electron microscopy ( cryoEM ) map of Chikungunya virus-like particles ( VLPs ) has been interpreted using the previously published crystal structure of the Chikungunya E1-E2 glycoprotein heterodimer . The heterodimer structure was divided into domains to obtain a good fit to the cryoEM density . Differences in the T = 4 quasi-equivalent heterodimer components show their adaptation to different environments . The spikes on the icosahedral 3-fold axes and those in general positions are significantly different , possibly representing different phases during initial generation of fusogenic E1 trimers . CryoEM maps of neutralizing Fab fragments complexed with VLPs have been interpreted using the crystal structures of the Fab fragments and the VLP structure . Based on these analyses the CHK-152 antibody was shown to stabilize the viral surface , hindering the exposure of the fusion-loop , likely neutralizing infection by blocking fusion . The CHK-9 , m10 and m242 antibodies surround the receptor-attachment site , probably inhibiting infection by blocking cell attachment . Chikungunya virus ( CHIKV ) is a mosquito-transmitted viral pathogen that causes fever , myalgia , rash , and severe arthritis in humans ( Powers and Logue , 2007; Simon et al . , 2008 ) . The first reported human CHIKV infections occurred in East Africa in 1952 ( Robinson , 1955 ) . Prior to the 2005 epidemic on Réunion Island , CHIKV was not regarded as a highly prevalent virus . An adaptive mutation in the E1 protein ( E1-A226V ) that allowed CHIKV to replicate more efficiently in Aedes albopictus is considered to be the primary reason for its recent extensive spread , infecting millions of individuals in Africa and Asia ( Tsetsarkin et al . , 2007; Kumar et al . , 2008; Santhosh et al . , 2008 ) . In some CHIKV-infected patients , severe damage in joint tissues can cause debilitating chronic arthritis . In the recent outbreaks , a change in pathogenesis was observed , with some cases progressing to fatal encephalitis . An autochthonous CHIKV outbreak in Italy in 2007 and the presence of the Aedes albopictus vector in many areas of Europe and the Americas have raised concern of further spread of the virus . Currently , there is no vaccine or antiviral agent approved for use in humans . CHIKV belongs to the alphavirus genus of the Togaviridae family ( Kuhn , 2007 ) . Alphaviruses are a group of positive-sense , single-stranded RNA , enveloped viruses transmitted by arthropods ( Griffin , 2007 ) . The alphavirus genome encodes four non-structural and five structural proteins . The non-structural proteins are required for virus replication , protein modification , and immune antagonism . The structural proteins ( capsid-E3-E2-6K-E1 ) are synthesized as a polyprotein from a subgenomic promoter , and are cleaved post-translationally into separate proteins by an autoproteinase and signalase . The E1 glycoprotein participates in cell fusion ( Lescar et al . , 2001 ) , whereas the E2 glycoprotein binds to cellular receptors ( Smith et al . , 1995; Zhang et al . , 2005 ) and initiates clathrin-dependent endocytosis ( Solignat et al . , 2009 ) . Virus core assembly is initiated by interactions between the genomic RNA and the nuclear capsid protein ( NCP ) ( Tellinghuisen et al . , 1999; Tellinghuisen and Kuhn , 2000; Linger et al . , 2004 ) in the cytoplasm . The E3 protein is essential for the proper folding of p62 , the precursor to E2 , and the formation of the p62-E1 heterodimer ( Mulvey and Brown , 1995; Carleton et al . , 1997 ) . Although E3 remains part of mature Semliki Forest virus ( SFV ) and Venezuelan equine encephalitis virus ( VEEV ) , apparently , it is not a component of mature CHIKV ( Simizu et al . , 1984 ) . The small 6 kDa protein , 6K , associates with the p62-E1 heterodimer and is transported to the plasma membrane prior to assembly . The 6K protein facilitates particle morphogenesis but is not stoichiometrically incorporated into virions ( Gaedigk-Nitschko and Schlesinger , 1990 , 1991 ) . Alphaviruses are icosahedral particles that have T = 4 quasi-icosahedral symmetry ( Paredes et al . , 1993; Venien-Bryan and Fuller , 1994; Cheng et al . , 1995; Zhang et al . , 2002 , 2011; Kostyuchenko et al . , 2011 ) . The ectodomain forms 80 spikes , each consisting of three copies of E1-E2 heterodimers . There are 20 icosahedral “i3” spikes , situated on icosahedral 3-fold axes , and 60 quasi-3-fold “q3” spikes in general positions containing a quasi-3-fold axis consistent with T = 4 symmetry ( Figure 1A ) . Thus , each icosahedral asymmetric unit contains one complete q3 spike and one-third of an i3 spike . E1 of CHIKV is a 439 amino acid protein with an N-linked glycosylation site at residue Asn141 . The ectodomain is formed by the 404 N-terminal residues , followed by a 30 residue transmembrane ( TM ) helix and a five amino acid cytoplasmic domain . The E1 ectodomain consists of three β-barrel domains ( Lescar et al . , 2001; Gibbons et al . , 2004; Li et al . , 2010; Voss et al . , 2010 ) . Domain I is spatially located between domains II and III with the fusion peptide lying at the distal end of domain II . E2 of CHIKV is 423 amino acids in length and has two N-linked glycans at positions Asn263 and Asn345 . The E2 364 residue ectodomain is followed by a 26-residue TM helix and a 33-residue cytoplasmic domain . The E2 ectodomain consists of three distinct immunoglobulin-fold domains with domain A , the putative receptor-binding domain , lying between domains B and C ( Li et al . , 2010; Voss et al . , 2010 ) . In the mature virus , domain B covers the fusion loop in domain II of E1 . The virus becomes fusogenic in acid pH ( Wahlberg and Garoff , 1992 ) when sets of three E1 molecules combine to make a fusogenic trimer after expelling the three E2 proteins from the center of the spike . The three fusion loops are then exposed outwards for insertion into the host cell membrane ( Gibbons et al . , 2004 ) . Similar large conformational changes occur in other enveloped viruses ( Kielian and Rey , 2006 ) although little is known of the structural intermediates through which these changes take place . 10 . 7554/eLife . 00435 . 003Figure 1 . Structure of the CHIK VLPs . ( A ) Surface-shaded figure of ectodomain ( left ) and surface-shaded figure of nucleocapsid ( right ) , colored according to the radial distance from the center of the virus . White triangles indicate one icosahedral asymmetric unit . ( B ) Cross-section of the virus showing density above 1 . 5 σ also colored according to the radial distance from the center of the virus . ( C ) Resolution of β-strands in the E1 domain III . DOI: http://dx . doi . org/10 . 7554/eLife . 00435 . 003 The nucleocapsid core consists of 30 hexamers around each 2-fold axis and 20 pentamers at each 5-fold vertex of the nuclear capsid ( Figure 1A ) , and is separated from the ectodomains of E1 and E2 by a lipid membrane ( Figure 1B ) . The N-terminal 118 amino acids of Sindbis ( SINV ) ( Choi et al . , 1991 ) and SFV ( Choi et al . , 1997 ) nuclear capsid contain positively charged residues that associate with the genomic RNA ( Zhang et al . , 2002 ) . A similar positively charged N-terminal region exists in CHIKV NCP and functions to recognize the viral genome for packaging during NCP assembly ( Perera et al . , 2001; Hong et al . , 2006 ) . The carboxy-terminal region of the NCP has a chymotrypsin-like fold and acts as an auto-catalytic proteinase during processing of the structural polyprotein ( Hahn and Strauss , 1990; Choi et al . , 1991 , 1997 ) . In addition , there is a hydrophobic pocket in the C-terminal domain of the NCP that binds the 33 amino acid cytoplasmic tail of E2 , effectively linking the proteins of the inner core with those on the surface of the virion ( Lee et al . , 1996; Skoging et al . , 1996 ) . This association of E2 with the nuclear capsid is essential for virus assembly ( Owen and Kuhn , 1997 ) . Many studies have analyzed the interaction of antibodies with alphaviruses ( Meyer and Johnston , 1993; Griffin et al . , 2001; Hunt et al . , 2010 ) . Most neutralizing antibodies bind to E2 , as this glycoprotein is more exposed on the viral surface than E1 . The mode of neutralization differs considerably among antibodies , with some blocking attachment to cells and others inhibiting the conformational changes required for fusion ( Hernandez et al . , 2008 ) . The only published structural studies of alphavirus-antibody complexes were a cryoEM analysis of Ross River virus complexed with an antibody that blocked receptor attachment ( Smith et al . , 1995 ) and a cryoEM investigation of SINV complexed with an antibody that inhibited the conformational changes required to form fusogenic E1 trimers ( Hernandez et al . , 2008 ) . Expression of the CHIKV structural C-E3-E2-6K-E1 polyprotein gives rise to virus-like particles ( VLPs ) that have been used to immunize and protect non-human primates against CHIKV infection ( Akahata et al . , 2010 ) . Non-human primates immunized with these VLPs developed high-titer neutralizing antibodies that protected against viremia after high-dose challenges . Moreover , immunodeficient mice that were passively administered purified IgG from immunized non-human primates were protected against CHIKV challenge ( Couderc et al . , 2009 ) . Previously , we published the structure of the CHIK VLPs at a resolution of 18 Å using cryoEM reconstruction of single particles ( Akahata et al . , 2010 ) . Here , we report a 5 . 3 Å three-dimensional high-resolution reconstruction of the CHIK VLP . We also describe pseudo-atomic resolution structures of CHIK VLPs complexed with Fab fragments from four different neutralizing mouse monoclonal antibodies ( MAb CHK-9 , CHK-152 , m10 and m242 ) . These studies are the first structural investigations of CHIKV complexed with neutralizing MAbs , based on 15 Å resolution cryoEM reconstructions of the complexes , the 5 . 3 Å resolution structure of the VLPs and 3 Å resolution crystallographic studies of the Fab fragments . These structures suggest that MAb CHK-152 neutralizes infectivity by inhibiting fusion whereas the other MAbs probably neutralize by blocking attachment . The structure of CHIK VLPs was determined to a resolution of ∼5 . 3 Å . The trimeric appearance of individual spikes was recognizable immediately , along with the inclined positions of the three E1 molecules surrounding each spike . Further inspection of the E1 and E2 molecules showed β-barrel structures surrounding empty cavities , thus yielding a good indication of the positions of each domain . Although resolution of individual β-strands was possible only in some instances ( Figure 1C ) , identification of individual side chains was not possible . The E3 glycoprotein , which is observed in some alphavirus structures ( Zhang et al . , 2011 ) , was not apparent in the CHIK VLP cryoEM density map . Densities for the E2 B-domain were weak in all four E1-E2 heterodimers , consistent with the crystal structures of the CHIKV E1-E2 heterodimer ( Voss et al . , 2010 ) and the SINV trimeric spike ( Li et al . , 2010 ) . Within the CHIK VLP structure , the four quasi-equivalent chymotrypsin-like nucleocapsid proteins were organized as pentamers around the 5-fold vertices and hexamers around the 2-fold icosahedral symmetry axes ( Figure 1A ) . The cryoEM density height of the nucleocapsid proteins was roughly 5% less than that of the glycoproteins . The glycoproteins and the nuclear capsid were separated by a ∼45 Å-wide lipid membrane . The external and internal lipid leaflets had a density of about 5 σ , whereas the protein density of the glycoproteins and the NCPs had a density of about 7 σ . The two lipid leaflets were separated by a ∼15 Å wide region , where the density was less than 1 . 5 σ , representing the loosely packed aliphatic chains of the lipid molecules ( Figure 1B ) . The membrane was traversed by 60 × 4 pairs of α helices representing the E1 and E2 carboxy-terminal regions ( Figure 1B ) . The quality of the membrane density in the icosahedrally averaged map demonstrates that the symmetry of the glycoproteins and nuclear capsid impose icosahedral symmetry on the flexible membrane by confining it to a limited space . A number of procedures have been developed for the refinement of atomic parameters when the resolution of the structural data is insufficient to follow the polypeptide chains with certainty or to recognize the identity of amino acids ( Chapman , 1995; Volkmann and Hanein , 1999; Tama et al . , 2004; Fabiola and Chapman , 2005; Goddard et al . , 2007; Topf et al . , 2008; Trabuco et al . , 2008; Lasker et al . , 2010; Schröder et al . , 2010; Zhu et al . , 2010 ) . These procedures increase the number of refinable parameters beyond those necessary to position and orient a rigid known structure into a relatively low-resolution cryoEM density map , thus reducing the ratio of observed data points ( e . g . structure amplitudes ) to refinable parameters . However severe restraints are placed on the geometry of the structure ( thus providing additional observational data ) while improving agreement between observed and calculated data in real or reciprocal space . Verification in reciprocal space can be achieved by tests on data that had been excluded in the refinement process , for instance by calculating Rfree ( Brünger , 1992 ) . Verification in real space can be established by observing anticipated structural features that had not been considered in the restrained refinement process . In the analysis of the CHIKV VLPs described here , the initial interpretation was performed by fitting the structure of the CHIKV E1-E2 heterodimer ( Voss et al . , 2010 ) into the cryoEM map of the CHIKV VLPs assuming strict T = 4 symmetry between quasi-equivalent structures . Subsequently , the number of refinable parameters was increased by breaking the structure into domains , based on the assumption that the domains likely are more rigid than the hinges in the polypeptide chain that hold the domains together . In addition , constraints imposed by the assumption of strict T = 4 symmetry were abandoned , thus further multiplying the number of refinable parameters by four . Verification came from the improved correlation of the domain-fitted structure with the amino acid sequences , the improved correlation between density heights of quasi-equivalent parts of the structure with each other , and from the emergence of features such as the position of the N-linked carbohydrate moieties . In contrast to the interpretation of the cryoEM density of VEEV ( Zhang et al . , 2011 ) , which had a resolution of 4 . 8 Å , we were able to use the crystallographically determined structure of the glycoproteins to aid in the analysis . The crystal structure of CHIKV E1-E2 ( Voss et al . , 2010 ) was fitted into the 5 . 3 Å cryoEM map using the program EMfit to maximize the average value of the density ( sumf ) at all atomic positions ( Table 1A ) , to minimize the steric clash between symmetry related structures , and to minimize the percentage of atoms in low density ( Rossmann et al . , 2001 ) . Initially , the E1-E2 heterodimer crystal structure was fitted as a rigid body generating the four quasi-equivalent structures at positions #1 , #2 , #3 , and #4 ( Figure 2A ) while refining the best positions of the quasi-symmetry axes ( see Materials and Methods ) , and using the icosahedral symmetry to generate the structure in neighboring icosahedral asymmetric units . The resultant locations of the four quasi-equivalent CHIKV heterodimers were used as starting positions for refinement of individual domains . The average height of the density at all fitted atoms ( sumf ) improved by ∼12% for the independently fitted domains . The value of sumf was at least 38% lower for the B domains than for any of the other domains . This is consistent with the disorder of the B domain in the crystal structure of the SINV spike ( Li et al . , 2010 ) and the large temperature factors for the atoms in the B domain in the crystal structure of the CHIKV heterodimer ( Voss et al . , 2010 ) . The height of the average value of the cryoEM density taken over all atomic positions within a molecular domain was higher at the quasi-equivalent position #4 than at the other three quasi-equivalent positions ( Table 1A ) . This might be due to crowding around the 5-fold symmetry axes , which reduces the variability in the heterodimer position . 10 . 7554/eLife . 00435 . 004Table 1 . Quality of fitting the atomic structural fragments into the cryoEM densityDOI: http://dx . doi . org/10 . 7554/eLife . 00435 . 004A ) Average height of the densities at the atomic positions ( sumf ) ( Rossmann et al . , 2001 ) on fitting the E1E2 heterodimer CHIKV structure into the cryoEM map at four quasi equivalent positionsAll atomsT = 4 fittingIndependent domain fitting#1#2#3#4Average#1#2#3#4AverageE1I16 . 115 . 817 . 017 . 716 . 618 . 217 . 018 . 519 . 918 . 5II15 . 717 . 218 . 816 . 717 . 018 . 018 . 619 . 018 . 518 . 6III15 . 714 . 215 . 218 . 015 . 718 . 715 . 918 . 220 . 718 . 3E2A15 . 518 . 019 . 017 . 517 . 418 . 817 . 819 . 919 . 418 . 8B10 . 09 . 49 . 69 . 69 . 810 . 911 . 311 . 712 . 411 . 7C15 . 315 . 419 . 019 . 116 . 821 . 019 . 920 . 222 . 420 . 5D13 . 316 . 415 . 816 . 115 . 416 . 417 . 216 . 016 . 816 . 5Average14 . 915 . 817 . 016 . 816 . 117 . 817 . 318 . 219 . 018 . 1Main chain atoms onlyT = 4 fittingIndependent domain fitting#1#2#3#4Average#1#2#3#4AverageE1I20 . 518 . 421 . 522 . 920 . 823 . 221 . 622 . 626 . 323 . 4II17 . 020 . 923 . 020 . 220 . 321 . 723 . 323 . 522 . 822 . 8III18 . 915 . 317 . 522 . 318 . 522 . 317 . 922 . 626 . 622 . 4E2A16 . 521 . 122 . 219 . 419 . 822 . 020 . 323 . 422 . 622 . 1B10 . 79 . 49 . 910 . 110 . 010 . 811 . 213 . 013 . 412 . 1C18 . 719 . 527 . 426 . 923 . 127 . 727 . 127 . 430 . 828 . 3D13 . 418 . 718 . 518 . 517 . 319 . 419 . 819 . 119 . 019 . 3Average16 . 517 . 620 . 020 . 018 . 521 . 020 . 221 . 723 . 121 . 5B ) Average height of the densities at the atomic positions ( sumf ) ( Rossmann et al . , 2001 ) on fitting the CHIKV homology model of the capsid protein into the cryoEM map at the four quasi equivalent positionsAll atomsT = 4 fittingIndependent domain fitting#1#2#3#4Average#1#2#3#4AverageD1†14 . 913 . 313 . 916 . 114 . 616 . 114 . 615 . 417 . 315 . 8D2†17 . 017 . 418 . 118 . 117 . 618 . 319 . 117 . 419 . 018 . 6Average16 . 015 . 416 . 017 . 216 . 117 . 417 . 116 . 518 . 217 . 3Main chain atoms onlyT = 4 fittingIndependent domain fitting#1#2#3#4Average#1#2#3#4AverageD1†18 . 516 . 517 . 818 . 917 . 920 . 416 . 919 . 222 . 719 . 8D2†21 . 321 . 521 . 423 . 221 . 922 . 824 . 220 . 425 . 323 . 2Average19 . 919 . 019 . 621 . 119 . 921 . 821 . 019 . 924 . 121 . 6C ) Average height of the densities at the atomic positions ( sumf ) ( Rossmann et al . , 2001 ) on fitting the CHIKV model of the transmembrane protein into the cryoEM map at the four quasi equivalent positionsAll atomsT = 4 fittingIndependent domain fitting#1#2#3#4Average#1#2#3#4AverageE1‡13 . 714 . 613 . 218 . 615 . 013 . 914 . 414 . 717 . 515 . 1E2‡14 . 712 . 811 . 817 . 514 . 215 . 114 . 113 . 718 . 915 . 5Average14 . 413 . 512 . 317 . 914 . 614 . 514 . 214 . 218 . 215 . 3Main chain atoms onlyT = 4 fittingIndependent domain fitting#1#2#3#4Average#1#2#3#4AverageE1‡16 . 217 . 616 . 024 . 118 . 515 . 717 . 316 . 723 . 918 . 4E2‡16 . 715 . 013 . 620 . 716 . 517 . 716 . 715 . 823 . 718 . 5Average16 . 516 . 014 . 422 . 017 . 516 . 717 . 016 . 223 . 818 . 4Domain definition: E1 I ( residues 1-36 , 132-168 , 273-293 ) , II ( residues 37-131 , 169-272 ) , III ( residues 294-393 ) . E2 A ( residues 16-134 ) , B ( residues 173-231 ) , C ( residues 269-342 ) , D ( residues 7-15 , 135-172 , 232-268 ) . †Domain definition: D1 ( residues 119 to 183 ) , D2 ( residues 184 to 267 ) . ‡Domain definition: E1 ( residues 394-439 ) , E2 ( residues 343-423 ) . 10 . 7554/eLife . 00435 . 005Figure 2 . Diagrammatic representation of contacts between amino acid residues in adjacent subunits . ( A ) Diagrammatic organization of the E1 and E2 subunits according to T = 4 icosahedral symmetry . The white numbers show the sequence in which the four independent subunits were generated . The black capital letters indicate the sequence of generating seven of the icosahedral asymmetric units . Icosahedral symmetry elements are shown as filled triangles and ellipses . Quasi-symmetry elements are shown as red outlined triangles and ellipses . ( B ) Contacts between the E1 ( blue ) and E2 ( green ) molecules . ( C ) Contacts between NCPs . ( B ) and ( C ) show the number of contacts between the indicated molecules . The quasi T = 4 related positions #1 , #2 , #3 and #4 are indicated , prefixed by their icosahedral symmetry identification A , B , C and D . The center of the “i3” icosahedral spike is indicated by a filled red triangle and the center of the “q3” quasi-3-fold spike is indicated by an outlined red triangle . DOI: http://dx . doi . org/10 . 7554/eLife . 00435 . 005 Once the individual domains were fitted ( Figure 3A ) , the bond geometry between the carboxy end of one domain and the amino end of the next domain were regularized . The distance between these ends prior to regularization ( Table 2 ) did not exceed 6 . 9 Å ( excluding the flexible domain B ) , indicating that the quality of fit was reasonable . Comparison of the hinge angles between domains in the four quasi-equivalent positions ( Table 3 ) did not exceed 13° , except for defining the orientation of the B domain in E2 . Considering that the length of the longest domain ( DII of E1 ) is about 50 Å , this implies a relative difference of about 6 Å for atoms furthest from the hinge , corresponding to a maximum movement of about three standard deviations between superimposed structures . 10 . 7554/eLife . 00435 . 006Figure 3 . Fit of the atomic structure into the cryoEM density . ( A ) Fit of the E1E2 heterodimer , ( B ) Fit of the capsid protein . The quasi equivalent subunit closest to the icosahedral 5-fold axis was chosen for display . DOI: http://dx . doi . org/10 . 7554/eLife . 00435 . 00610 . 7554/eLife . 00435 . 007Table 2 . Distances in Å between N and C termini of fitted domains before regularizationDOI: http://dx . doi . org/10 . 7554/eLife . 00435 . 007Quasi equivalent positionsDomains#1#2#3#4E1I 36 to II 374 . 65 . 65 . 53 . 5II 131 to I 1324 . 84 . 14 . 44 . 1I 168 to II 1692 . 86 . 45 . 45 . 1II 272 to I 2735 . 84 . 73 . 92 . 1I 293 to III 2945 . 46 . 93 . 75 . 0E2D 15 to A 163 . 86 . 35 . 75 . 9A134 to D1353 . 15 . 04 . 42 . 3D172 to B1733 . 48 . 83 . 05 . 7B231 to D2326 . 27 . 75 . 97 . 2NCPD268 to C2695 . 15 . 24 . 14 . 0 ( D1 ) 183 to ( D2 ) 1841 . 22 . 21 . 61 . 610 . 7554/eLife . 00435 . 008Table 3 . Hinge angle change between domains in different heterodimersDOI: http://dx . doi . org/10 . 7554/eLife . 00435 . 008Hinge angle change in E1 between domain I and domain II ( in degrees ) Position #1Position #2Position #3Position #4Crystal structure-14 . 4-6 . 67 . 5-8 . 3Position #18 . 09 . 36 . 8Position #23 . 12 . 1Position #3-3 . 3Hinge angle change in E1 between domain I and domain III ( in degrees ) Position #1Position #2Position #3Position #4Crystal structure-7 . 3-8 . 0-9 . 42 . 1Position #1-3 . 510 . 37 . 7Position #213 . 09 . 2Position #37 . 6Hinge angle change in E2 between domain A and domain B ( in degrees ) Position #1Position #2Position #3Position #4Crystal structure-8 . 410 . 422 . 2-22 . 3Position #1-14 . 125 . 5-29 . 7Position #226 . 6-28 . 6Position #3-26 . 3Hinge angle change in E2 between domain A and domain C ( in degrees ) Position #1Position #2Position #3Position #4Crystal structure-11 . 5-10 . 4-7 . 6-17 . 1Position #1-4 . 8-8 . 6-7 . 3Position #27 . 28 . 3Position #3-14 . 8 After independent fitting of the domains , the root mean square ( rms ) distance between equivalent Cα atoms when making pairwise comparisons between E1-E2 heterodimers ranged from 0 . 5 Å to 2 . 4 Å ( Table 4 ) . This showed that the structures of the four quasi-equivalent parts of the model were more alike to each other than to the crystal structure of the CHIKV heterodimer ( Voss et al . , 2010 ) . As such similarity is unlikely to occur by accident , this observation helped to confirm the validity of the process used to interpret the cryoEM map . The systematic change of the E1-E2 heterodimers in the VLPs compared to the crystal structure is the result of accommodating the different quasi-equivalent environments . In contrast to the results for the related alphavirus , VEEV ( Zhang et al . , 2011 ) , the presence of differences larger than the standard deviation between equivalenced Cα atoms between the four quasi-equivalent structural components of the CHIKV VLPs explains why averaging of the quasi equivalent density did not improve the map . 10 . 7554/eLife . 00435 . 009Table 4 . RMS deviation in Å between quasi equivalent Cα atomsDOI: http://dx . doi . org/10 . 7554/eLife . 00435 . 009E1#1#2#3#4Xtal#10 . 001 . 211 . 251 . 081 . 69#20 . 001 . 281 . 271 . 52#30 . 001 . 101 . 68#40 . 001 . 17Xtal0 . 00E2#1#2#3#4Xtal#10 . 002 . 181 . 791 . 901 . 50#20 . 002 . 342 . 382 . 17#30 . 002 . 281 . 84#40 . 001 . 92Xtal0 . 00E1&E2#1#2#3#4Xtal#10 . 001 . 941 . 651 . 682 . 28#20 . 002 . 162 . 142 . 24#30 . 001 . 872 . 37#40 . 001 . 88Xtal0 . 00Capsid#1#2#3#4Xtal#10 . 000 . 490 . 730 . 710 . 48#20 . 001 . 090 . 940 . 71#30 . 000 . 630 . 56#40 . 000 . 52Xtal0 . 00 The quality of the side chain placements was determined by correlating their densities in the four quasi-equivalent positions . The density associated with each residue was taken as the average density of all the non-hydrogen atoms in that residue . A high correlation coefficient ( see Materials and Methods for definition ) indicates that the variation of cryoEM density along each polypeptide chain was similar . As the average density for each residue included all atoms in the side chain , a high correlation also implied similarity of the side chain densities . These correlation coefficients ( Table 5A ) varied from 0 . 44 to 0 . 30 , which was significantly greater than the values ( 0 . 40–0 . 08 ) for the initial T = 4 rigid body fitting . Furthermore , there was a reasonable correlation of the average crystallographic temperature factors for each amino acid residue of the CHIKV heterodimer crystal structure ( Voss et al . , 2010 ) with the height of the cryoEM density at each amino acid position ( Table 5A ) . 10 . 7554/eLife . 00435 . 010Table 5 . Correlation between amino acid sequence and cryoEM densityDOI: http://dx . doi . org/10 . 7554/eLife . 00435 . 010A ) Correlation between cryoEM densities in the four quasi equivalent positions of the E1 and E2 glycoproteins . All atomsT = 4 fittingIndependent domain fitting1234B1234B11 . 000 . 170 . 230 . 080 . 221 . 000 . 400 . 440 . 400 . 2121 . 000 . 280 . 380 . 301 . 000 . 410 . 300 . 2831 . 000 . 400 . 281 . 000 . 360 . 2841 . 000 . 191 . 000 . 15B1 . 001 . 00Main chain atoms onlyT = 4 fittingIndependent domain fitting12#34B1234B11 . 000 . 110 . 200 . 110 . 191 . 000 . 400 . 410 . 410 . 2221 . 000 . 310 . 300 . 301 . 000 . 360 . 320 . 2631 . 000 . 410 . 271 . 000 . 370 . 2641 . 000 . 161 . 000 . 12B1 . 001 . 00B ) Correlation between cryoEM densities of the four quasi equivalent capsid proteins . All atomsT = 4 fittingIndependent domain fitting#1#2#3#4#1#2#3#4#11 . 000 . 410 . 180 . 171 . 000 . 200 . 190 . 30#21 . 000 . 260 . 251 . 000 . 450 . 42#31 . 000 . 131 . 000 . 41#41 . 001 . 00Main chain atoms onlyT = 4 fittingIndependent domain fitting#1#2#3#4#1#2#3#4#11 . 000 . 370 . 160 . 261 . 000 . 220 . 170 . 33#21 . 000 . 310 . 271 . 000 . 340 . 43#31 . 000 . 151 . 000 . 31#41 . 001 . 00C ) Correlation between cryoEM densities of the four quasi equivalent E1&E2 TM and endodomain regions . All atomsT = 4 fittingIndependent domain fitting#1#2#3#4#1#2#3#4#11 . 000 . 430 . 170 . 411 . 000 . 370 . 340 . 35#21 . 000 . 240 . 381 . 000 . 440 . 34#31 . 000 . 201 . 000 . 46#41 . 001 . 00Main chain atoms onlyT = 4 fittingIndependent domain fitting#1#2#3#4#1#2#3#4#11 . 000 . 280 . 080 . 271 . 000 . 230 . 110 . 26#21 . 000 . 170 . 431 . 000 . 400 . 36#31 . 000 . 141 . 000 . 50#41 . 001 . 00 The fitting of the independent domains of the CHIKV heterodimer into the cryoEM map was further validated by finding the densities associated with N-linked glycans , after setting to zero all densities at grid points within 3 . 0 Å of every fitted atom . The remaining densities should correspond to the sugar moieties on E1 and E2 ( Pletnev et al . , 2001 ) . The shape and size of the carbohydrate moiety for each of the three N-linked glycosylation sites is consistent in each of the four quasi-equivalent positions . The average volume of the carbohydrates at the four E1 ( Asn141 ) sites in DI of E1 is 258 Å3 with an average height of 2 . 2 σ , with most of the noise density not being greater than 1 . 0 σ . This would correspond to about one sugar moiety . The densities corresponding to the four quasi-equivalent E2 ( Asn263 ) sites in the E2 connecting ribbon were somewhat fragmented , making it difficult to determine their volumes . However , at lower ( 15 and 10 Å ) resolution , the densities were well defined . The volume of the density and average height associated with the four quasi-equivalent sites at E2 ( Asn345 ) in domain C of E2 were as large as the sites at E1 ( Asn141 ) , but because of their proximity to the viral membrane , their boundaries were uncertain . The centers of the carbohydrate moieties were ∼6 Å from the N atom of the glycosylated Asn residues . Contacts between the E1 and E2 glycoproteins were determined ( Materials and Methods ) for the T = 4 fitted structure and compared with the improved , independently fitted , domain structure ( Table 6 ) . The E1 and E2 glycoproteins that form the four heterodimers in the VLPs have more contacts with each other in the independently fitted structure than in the crystallographic heterodimer structure . Particularly striking was the increase in hydrophobic contacts in the independently fitted results , which demonstrated how the heterodimer adapted itself to the quasi-equivalent symmetry of the VLPs compared to the constraints imposed by the crystal lattice . 10 . 7554/eLife . 00435 . 011Table 6 . Inter-molecular contacts ( See methods for descriptions ) DOI: http://dx . doi . org/10 . 7554/eLife . 00435 . 011Spike positionTotalHydrophobicPossible H bondsPossible salt bridgesA ) Number of atom-to-atom contacts between E1-E2 in heterodimeri3#123499 ( 42% ) 34 ( 13% ) 0q3#2254112 ( 44% ) 25 ( 10% ) 0#317959 ( 33% ) 33 ( 18% ) 0#4302118 ( 39% ) 39 ( 13% ) 4 ( 1% ) Crystal861 ( 1% ) 32 ( 37% ) 1 ( 1% ) B ) Number of atom-to-atom contacts of E2 with other E2s within a spikei3#122946 ( 20% ) 30 ( 13% ) 82 ( 36% ) q3#2708 ( 11% ) 10 ( 14% ) 33 ( 47% ) #34714 ( 30% ) 6 ( 13% ) 10 ( 21% ) #474118 ( 23% ) 24 ( 46% ) 7 ( 9% ) C ) Number of atom-to-atom contacts between glycoprotein spikesq3 to q3A#3 - D #412342 ( 34% ) 21 ( 17% ) 7 ( 5% ) A#4 - D#413052 ( 40% ) 18 ( 14% ) 1 ( 1% ) i3 to q3A#2 - B#15827 ( 47% ) 7 ( 12% ) 0A#1 - A#28829 ( 33% ) 19 ( 22% ) 0A#1 - A#3118 ( 73% ) 2 ( 8% ) 0D ) Number of atom-to-atom contacts between E1 and E2 within the spikes . q3A#2 - A#33016 ( 34% ) 5 ( 17% ) 0A#3 - A#4208 ( 40% ) 3 ( 14% ) 0i3A#4 - A#25625 ( 47% ) 11 ( 12% ) 0A#1 - B#1224112 ( 50% ) 21 ( 9% ) 22 ( 10% ) E ) Number of atom-to-atom contacts between capsid proteinsA#2 - B#17920 ( 25% ) 13 ( 16% ) 15 ( 19% ) A#1 - A#3113 ( 27% ) 1 ( 9% ) 3 ( 27% ) A#3 - D#218351 ( 28% ) 22 ( 12% ) 38 ( 21% ) A#4 - D#4195 ( 26% ) 2 ( 11% ) 8 ( 42% ) A#1 - B#1209 ( 45% ) 2 ( 10% ) 2 ( 10% ) F ) Number of atom-to-atom contacts between E1 & E2 in TM region#110060 ( 60% ) 3 ( 3% ) 0#28243 ( 52% ) 8 ( 10% ) 0#32012 ( 60% ) 1 ( 5% ) 0#42824 ( 86% ) 00See Figure 2 . Unlike the T = 4 fitted structure , the independently fitted structure lacked exact symmetry between the i3 and q3 spikes . Nevertheless , the q3 spike had apparent , although not exact , 3-fold symmetry within itself and with its surroundings . A comparison of the i3 and q3 structures showed that the E1 molecules wrap around the central E2 molecules more tightly in the i3 spike as indicated by the increased number of interactions between the E2 molecules in the i3 spike compared to the q3 spike . Thus , the i3 trimers in the VLPs were more compact with a greater number of interactions between its component molecules than the q3 trimer . The greater compactness of the i3 spike was confirmed by determining the radius of gyration of the i3 and the q3 spikes as a function of the radial distance from the viral center ( Materials and Methods ) . At the base of the spike , where the E1 molecules make contacts between spikes , the radius of gyration was up to 5 . 8 Å less for the i3 spike than for the q3 spike ( Table 7 ) . At the top of the spike , the i3 and q3 spikes showed less difference . In addition , the whole of the q3 spike was translated away from the center of the VLPs by ∼1 . 8 Å . Thus , in comparison with the 20 i3 spikes , the 60 q3 spikes may be poised to release the E2 molecules from the center of the spikes in preparation for the formation of the E1 fusogenic homotrimers early in the infection ( Li et al . , 2010 ) . 10 . 7554/eLife . 00435 . 012Table 7 . Spike radii of gyrationDOI: http://dx . doi . org/10 . 7554/eLife . 00435 . 012Distance from particle center ( Å ) Radii of gyration ( Å ) # of atomsi3q3SINV crystali3q3SINV* crystal25558 . 361 . 063 . 342346226553 . 557 . 354 . 913112611827541 . 146 . 942 . 612512210128530 . 133 . 130 . 184908029533 . 234 . 233 . 885809130533 . 331 . 931 . 390898231530 . 731 . 224 . 570809232526 . 725 . 9414733531 . 029 . 011Overall40 . 542 . 441 . 9669669626Sindbis virus ( SINV ) Prior studies ( Li et al . , 2010 ) have suggested that the trigger for movement of domain B to uncover the fusion loop on E1 at low pH might be the protonation of the highly conserved His169 and His256 residues in SINV on the β-ribbon that connects the ends of the B domain to the A and C domains of E2 ( Li et al . , 2010 ) . In the CHIK VLP structure , His170 ( equivalent to SINV His 169 ) does not interact closely with other negatively charged residues that would be required for a pH-sensing switch whereas CHIKV His 256 ( equivalent to SINV His256 ) contacts Glu166 . The crystal structures of SINV ( Choi et al . , 1991; Tong et al . , 1992 ) and SFV ( Choi et al . , 1997 ) NCPs had been determined previously . Of these , the amino acid sequence of the NCP of SFV has greater similarity to CHIKV . Therefore , a homology model for CHIKV NCP ( residues 119 to 267 ) was generated based on the crystal structure of the SFV NCP ( Protein Data Bank accession number 1VCP ) . This homology structure was fitted into the CHIK VLP cryoEM density map by assuming T = 4 symmetry , using the EMfit program ( Table 1B ) . The positions of the quasi-symmetry axes were refined to optimize the fit to the density and to minimize clashes between symmetry related NCP structures . The refinement found that the placement of the quasi-symmetry axes did not change significantly from those optimized while fitting the glycoproteins . The orientation and placement of the homology NCP in the cryoEM density was essentially the same as described for other alphaviruses ( Zhang et al . , 2002 ) . Furthermore , the slight asymmetric twist of the pentameric and hexameric rings about the 5 and 2-fold axes , respectively ( Figures 1A , 2B ) was the same as that observed for Ross River virus ( Cheng et al . , 1995 ) . The lack of a substantial rigid association between the two lobes of the NCP's chymotrypsin-like structure suggests that there could be conformational adjustments between the two lobes . Consequently , the NCP structure was split into two domains ( amino acid residues 119 to 183 and 184 to 267 ) representing the two lobes on either side of the substrate-binding cleft . These domains were then fitted independently at each of the four quasi-equivalent positions . The sumf values ( Table 1B ) for fitting the NCP into the pentameric capsomer density ( position #4 , Figure 3B ) were slightly greater than for the NCPs in the hexameric capsomer ( positions #1 , #2 , and #3 ) , as was the case for the E1 protein fitting near the 5-fold axes . The correlation of the average residue densities between the four quasi-equivalent NCP positions was roughly the same as for fitting of the glycoproteins ( Table 5 ) , indicating the same level of amino acid recognition for the glycoproteins and the NCPs . Contrary to the organization of the E1-E2 glycoproteins , where there are extensive inter-heterodimer contacts within the trimeric spikes , there were few contacts among the three NCP molecules around the icosahedral 3-fold axes and no contacts among the three NCP molecules around the quasi-3-fold axes . However , the six NCP molecules that form the hexameric capsomers around the icosahedral 2-fold axes made extensive contacts ( Figure 3B and Table 6 ) . These interactions were between the loop formed by residues 172 to 183 in one NCP molecule with residues in loops 190 to 206 , 239 to 243 and 261 to 262 in a neighboring NCP molecule . Moreover , salt bridges were formed between Lys178 and Glu240 . The NCP molecules around the icosahedral 5-fold axes made similar contacts and salt bridges , as expected by their quasi-similar environments . It was unexpected that the positions of the quasi-2- and 3-fold symmetry elements were matched almost perfectly between the external organization of the glycoproteins into 80 trimers and the internal organization of the NCP into 12 pentamers and 30 hexamers ( Figure 2 ) . The nucleoplasmic cores assemble in the cytoplasm ( Weiss et al . , 1989; Owen and Kuhn , 1996; Soonsawad et al . , 2010 ) and are transported to the plasma membrane where they interact with the endodomain of the E2 protein . This provides an opportunity for the C-terminal regions of E2 to bind to the hydrophobic pocket on the surface of the nucleocapsid ( Lee et al . , 1996 ) , thus nucleating the E1-E2 glycoproteins to form an icosahedral shell in synchrony with the icosahedral symmetry of the cores . The crystal structure of CHIKV E1-E2 ( Voss et al . , 2010 ) consists of residues 1-393 of E1 and 7-342 of E2 but does not include the TM regions and endodomains . Cα models were built of the missing part of the E1-E2 ectodomains , the TM helices , and the E2 endodomain , based on Rosetta structural predictions ( Lange and Baker , 2012 ) and on the cryoEM density . The alignment of the amino acid sequence with the cryoEM density was based on assigning the kink observed in the TM helix of E1 to the likely flexible Thr-Gly-Gly sequence . In addition , the alignment was based on placing the interaction of Tyr 400 into the hydrophobic pocket around Trp 251 in the NCP , corresponding to SINV NCP residue Trp 247 ( Lee et al . , 1996 ) . The TM regions of E1 and E2 were built independently for each of the four quasi-equivalent regions of the cryoEM map ( Figure 3C ) . The E1 and E2 helices were parallel and situated close together with some hydrophobic contacts between them ( Table 6 ) . The T = 4 quasi-symmetry operators that relate the pairs of E1-E2 TM helices to other E1-E2 TM helical pairs were in similar positions to the symmetry operators that relate E1-E2 ectodomains to each other . Although the lipid membrane is presumably flexible and likely to have different organization of lipid molecules in different particles , the cryoEM density showed a preference for roughly radial density features . Presumably , these densities represent an average of similarly placed lipid molecules , as was observed in the 4 Å resolution X-ray crystallographic map of the dsDNA PRD1 bacteriophage ( Cockburn et al . , 2004 ) . The only other high resolution structure of an alpha virus is that of VEEV ( Zhang et al . , 2011 ) . One way of comparing the structure of VEEV with the structure of CHIK VLPs is by superimposing their icosahedral symmetry elements . This shows that the overall radius of these two viruses is essentially the same . However , the positions and orientations of the molecular protein components have changed by up to 9 . 7 Å and 11 . 7° , respectively ( Table 8 ) , which is far greater than the RMS Cα atom differences between quasi equivalent positions of the component molecular components in the CHIK VLPs ( maximum RMS distance is 2 . 4 Å , Table 4 ) . It is also much larger than the RMS difference between Cα atoms on superimposing the equivalent molecular components of VEEV and CHIK VLPs without regard for the position and orientation of the icosahedral symmetry axes ( Table 8 ) . Thus the change in molecular positions and orientation with respect to the icosahedral framework is highly significant . Although the capsid protein amino acids are more conserved than any of the other structural proteins ( Table 8 ) , by far the largest differences in position and orientation of the molecular components occur for the capsid proteins ( Table 8 ) . The greater conservation of the amino acid sequences for the capsid protein is also reflected in a lower RMS differences between Cα atoms between superimposed capsid structures of VEEV and CHIK VLPs ( less than 1 . 2 Å ) as opposed to a comparison of the other molecular components ( greater than 2 . 0 Å , Table 8 ) . 10 . 7554/eLife . 00435 . 013Table 8 . Comparison between VEEV and CHIK VLPs . DOI: http://dx . doi . org/10 . 7554/eLife . 00435 . 013E1-E2TMCapsidA1A2A3A4A1A2A3A4A1A2A3A4d in Å2 . 63 . 53 . 52 . 24 . 82 . 32 . 13 . 09 . 67 . 79 . 09 . 7κ in degrees2 . 33 . 63 . 73 . 45 . 48 . 210 . 410 . 09 . 78 . 010 . 011 . 7dCα in Å2 . 72 . 52 . 42 . 12 . 62 . 22 . 52 . 01 . 11 . 21 . 21 . 2num69668070971411510099101149149149149% identityE1 = 56% , E2 = 34%TM = 24%Capsid = 62%Differences between the positions of the centers of mass ( d ) and orientation ( κ ) relative to the icosahedral symmetry axes of the ( E1-E2 ) heterodiners , the trans membrane ( TM ) helices and capsid proteins at positions A1 , A2 , A3 and A4 ( see Figure 2 ) . The RMS distances ( dCα ) are given between the number ( num ) of equivalent Cα positions of the superimposed molecules A1 to A4 . The percentage of identical amino acids in the aligned proteins is given on the last line of the table The above analysis shows that these two alpha-viruses maintain a fairly constant tertiary structure , notwithstanding moderate amino acid sequence variations . However , the quaternary organization has greater variability , although still retaining the same T = 4 quasi symmetry . Considering that the nucleocapsid core is assembled first and the subsequent association of the glycoproteins is based on the pre-existing cores , it is surprising that the difference of structure between the CHIKV and VEEV cores is less evident in the virus' ectodomains . Four cryoEM structures were determined to ∼15 Å resolution of CHIK VLPs complexed with the Fab fragments of the neutralizing mouse MAbs CHK-9 , CHK-152 , m242 and m10 ( Figure 4 ) . All Fab fragments bound the VLPs with one Fab fragment per E2 molecule . Except for CHK-152 , the concentration of the Fab molecules required for 50% neutralization was 100 or more times greater than the intact IgG ( Figure 5 ) . This implies a fundamental difference between how CHK-152 binds to the VLPs compared to the other Fabs . Because CHK-152 MAb and Fab fragment neutralize CHIKV at almost the same low concentration , it is unlikely that this MAb requires bivalent binding to inhibit infection . 10 . 7554/eLife . 00435 . 014Figure 4 . CHIKV VLP Fab complexes . Right: “Difference maps” showing the surface structure of the Fab molecules . The difference maps were produced by setting all density within 1 . 7 Å of an atom in the VLP structure to zero . Left: “Road maps” showing the projected surfaces of the VLP-Fab complexes for one ( triangular ) icosahedral asymmetric unit . The coloring is according to the radial distance of the surface from the center of the VLP . The footprints of the Fabs are shown in yellow . The radial projection of the whole Fab molecules onto the surface of the VLP is shown with white contours representing the height of the projected density . The mosaic background shows the amino acids that form the viral surface . DOI: http://dx . doi . org/10 . 7554/eLife . 00435 . 01410 . 7554/eLife . 00435 . 015Figure 5 . Inhibition of CHIKV by IgG and Fab fragments . Neutralizing activity of intact MAb and Fab fragments against CHIKV is shown for ( A ) CHK-9 , ( B ) CHK-152 , ( C ) m10 and ( D ) m242 , as determined by the reduction in the number of focus-forming units ( FFU ) on Vero cells . MAbs and Fab fragments were mixed with 100 FFU of infectious CHIKV ( strain La Reunion 2006 OPY-1 ) for one hour at 37°C before infecting Vero cells . Each data point is the average of three independent experiments performed in triplicate . Error bars represent standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 00435 . 015 The E1-E2 coordinates from the high-resolution VLP map were placed into the Fab bound cryoEM maps by aligning the icosahedral symmetry axes . The coordinates of the E2 B domain in the VLPs did not agree well with the B domain density in the maps of the VLP complexed with the CHK-9 or m242 Fab fragments , suggesting that the E2 B domain had a different orientation in these Fab-bound structures than in the non-complexed VLPs . To determine the modified position of the E2 B domain in these two complexes , the cryoEM density distributions were modified by setting to zero all density that was within 3 Å of all atoms in E1 and of all atoms in the A , C and D domains of E2 . The crystallographic coordinates of the E2-B domain ( Voss et al . , 2010 ) were then fitted into the modified cryoEM density distributions , assuming T4 quasi-icosahedral symmetry . The crystal structures of CHK-9 and m242 Fab fragments were determined to 3 . 0 Å and 3 . 1 Å resolution , respectively ( Table 9 ) . These Fab structures were fitted into the cryoEM maps of the corresponding VLP-Fab complexes . For the VLP-CHK-152 and VLP-m10 Fab complexes , for which no crystal structures of the corresponding Fab fragments had been determined , both the CHK-9 and m242 Fab crystal structures were tested to determine which Fab yielded the better fit to the cryoEM map of the complex . VLP-CHK-152 and VLP-m10 Fab cryoEM maps were best fitted with the Fab structures of CHK-9 and of m242 , respectively ( Figure 6 ) . The T = 4 symmetry parameters , determined by fitting the E1-E2 heterodimer into the higher resolution VLP density , were used for fitting the Fab fragments into the cryoEM densities of the appropriate VLP-Fab complexes . The height of the density for the variable domains of the Fab fragments was comparable to that of the CHIKV glycoprotein , suggesting a nearly 100% binding occupancy of the Fab fragments under saturating binding conditions . The density for the constant domains ( remote from the VLPs ) of the Fab fragments was about 17% lower , consistent with the flexibility of the elbow angle between the constant and variable domains . The footprints of the Fab fragments on the surface of the VLPs were determined with the RIVEM program ( Xiao and Rossmann , 2007 ) ( Figures 4 , 7 ) . 10 . 7554/eLife . 00435 . 016Table 9 . Data collection and refinement statistics for the m242 and CHK-9 Fab moleculesDOI: http://dx . doi . org/10 . 7554/eLife . 00435 . 016Fab m242Fab CHK9Data collectionX-ray source23-ID-B23-ID-BWavelength ( Å ) 1 . 031 . 03Resolution ( Å ) 3 . 13Space groupP21212C2Unit cell ( Å ) a = 137 . 0 , b = 89 . 1 , c = 94 . 4a = 87 . 9 , b = 57 . 7 , c = 118 . 1Unique reflections21 , 11535 , 876Average redundancy3 . 75 . 8I/σ17 . 5 ( 4 . 9 ) 25 . 0 ( 5 . 0 ) Completeness ( % ) 98 . 3 ( 96 . 4 ) 99 . 6 ( 99 . 9 ) Rmerge ( % ) †10 . 1 ( 23 . 4 ) 5 . 5 ( 39 . 2 ) RefinementResolution ( Å ) 3 . 11 . 8Rworking ( % ) ‡29 . 3 ( 34 . 5 ) 18 . 7 ( 23 . 3 ) Rfree ( % ) §33 . 1 ( 34 . 3 ) 22 . 0 ( 27 . 7 ) rmsd bonds ( Å ) 0 . 0050 . 01rmsd angels ( ° ) 0 . 850 . 91# of residues Ramachandran disallowed01*Values in parentheses throughout the table correspond to the outermost resolution shell†Rmerge = Σ\| I - <I> \| / Σ I , where I is the measured intensity of reflections‡Rworking = Σ\|\|Fobs\| -\|Fcalc\|\| / Σ\|Fobs\|§Rfree has the same formula as Rworking except that calculation was made with the structure factors from the test set10 . 7554/eLife . 00435 . 017Figure 6 . Fit of the Fab structures ( blue ) into the cryoEM density “difference” maps ( hatched grey surface ) calculated as described for Figure 4 . Shown also is how the Fab molecules bind to the E1 ( red ) -E2 ( green ) heterodimer . DOI: http://dx . doi . org/10 . 7554/eLife . 00435 . 01710 . 7554/eLife . 00435 . 018Figure 7 . “Road maps” showing footprint of four neutralizing Fabs on the VLP surface at position #3 as defined in Figure 2 . In order to differentiate between amino acids in different quasi 3-fold related subunits , their identity is defined as the amino acid sequence number in E1 + 2000 , 3000 , and 4000 , and in E2 + 2500 , 3500 , and 4500 for positions #2 , 3 , and 4 , respectively ( see Figure 2 ) . The surface is colored according to radial distance from the center of the VLP . The A , B , and D domains of E2 are bounded by a black , white and dashed white line , respectively . Residues in the putative receptor binding site on domain A of E2 are bounded by a yellow dashed line . The footprint of the Fabs onto the VLP surface is outlined in yellow . DOI: http://dx . doi . org/10 . 7554/eLife . 00435 . 018 The footprint of CHK-9 and m242 Fabs onto the VLP localized primarily onto the domain A and did not include any part of domain B . Indeed , binding of these fragments to the VLPs had nudged domain B ∼10 Å sideways , out of the way of the Fab fragment ( Figure 7 ) . In contrast , the footprint of the m10 Fab mapped exclusively on domain B and did not cause any observable movement ( Figure 7 ) . The footprint of CHK-152 , however , spanned both domains B and A as well as the linker peptides to domain B represented by domain D . Thus , unlike the other Fabs , CHK-152 has the capacity to cross-link domain B with domain A . Alphaviruses become fusogenic at low pH when domain B , held in place by the flexible β-ribbons of domain D , becomes mobile and exposes the fusion loop on domain II of E1 . As CHK-152 efficiently blocks viral fusion with host membranes ( P Pal and M Diamond , unpublished observations ) , it may do so by cross-linking the flexible domain B to the rest of the virus surface . An analogous mechanism of inhibition was described for neutralization of West Nile virus , in which the viral envelope proteins were cross-linked by the bound FAB fragments ( Kaufmann et al . , 2010 ) . All four MAbs examined here bind to the end of the trimeric spikes that includes the putative receptor attachment region ( Davis et al . , 1991; Dubuisson and Rice , 1993; Kinney et al . , 1993; Klimstra et al . , 1998; Bernard et al . , 2000; Lee et al . , 2002; Wang et al . , 2003; Pierro et al . , 2007 , 2008; Ryman et al . , 2007; Li et al . , 2010 ) , although only the CHK-9 Fab footprint actually overlaps the site ( Figure 7 ) . Nevertheless the proximity of these footprints to the possible receptor-binding site could probably allow neutralization by competitively limiting access of the virus to the cell surface receptor . To block formation of a fusogenic E1 trimer it is necessary to bind a Fab molecule to only one of the three E2 molecules in every spike . However , to block receptor attachment it is necessary to block all three receptor-binding sites on every spike . Such a stoichiometry would require substantially higher concentrations of Fabs to inhibit infectivity , as seen with CHK-9 , m10 and m242 Fabs ( Figure 5 ) . The difference between MAb and Fab neutralization efficiency is presumably the result of bivalent attachment and cross-linking on a single virion or aggregation between virions . In summary our structural studies suggest that the CHK-9 , m10 , and m242 Fab fragments neutralize CHIKV infectivity by blocking the cellular binding site on the A domain of E2 and that the CHK152 inhibits fusion by stabilizing domain B of E2 , preventing exposure of the fusion loop on E1 . CHIK VLPs have been reported as structural and immunologically indistinguishable from the mature infectious virus ( Akahata et al . , 2010 ) ; the use of CHIKV VLPs instead of live infectious virus has allowed us to avoid performing all structural studies under biosafety level 3 conditions . CHIK VLPs were produced and purified as described previously ( Akahata et al . , 2010 ) . Following purification , the buffer was exchanged to PBS and the VLPs were concentrated to 1 mg/ml . MAbs m242 , CHK-9 , CHK-152 , and m10 were purified from hybridoma superantants by protein A Sepahrose chromatography , and then digested with papain . The resultant Fab fragments were purified by sequential protein A Sepahrose and Superdex 75 16/60 size exclusion chromatography . Serial dilutions of MAbs or their Fab fragments were incubated with 100 focus-forming units ( FFU ) of CHIKV ( CHIKV La Reunion 2006 OPY-1 ) for one hour at 37°C . MAb or Fab-virus complexes were added to Vero cells in 96-well plates . After 90 min , cells were overlaid with 1% ( w/v ) methylcellulose in Modified Eagle Media supplemented with 4% FBS . Plates were harvested 18 hr later , and fixed with 1% PFA in PBS . The plates were incubated sequentially with 500 ng/ml of mouse-human chimeric version of CHK-9 ( containing a human γ1 constant region , P Pal and M Diamond , unpublished results ) and horseradish peroxidase conjugated goat anti-human IgG in PBS supplemented with 0 . 1% saponin and 0 . 1% BSA . CHIKV-infected foci were visualized using TrueBlue peroxidase substrate ( KPL ) , quantitated on an ImmunoSpot 5 . 0 . 37 macroanalyzer ( Cellular Technologies Ltd ) , and analyzed using GraphPad Prism software . Three microliter aliquots of purified CHIK VLP sample ( 1 mg/ml ) were applied to 400 mesh C-flat grids ( 1 . 2 μm hole size ) and double blotted inside a 100% humidified FEI vitrobot chamber , using a blotting time of 2 s . The grids were frozen by plunging into liquid ethane . A FEI Titan Krios electron microscope operated at 300 kV was used to collect cryoEM images on Kodak SO-163 films at a magnification of 59 , 000× and an electron dosage of ∼25 e−/Å2 at the boiling point of liquid nitrogen . The microscope beam was aligned parallel with the optical axis of the microscope using the coma-free alignment technique ( Kimoto et al . , 2003 ) . The quality of the alignment was checked by taking a CCD image of the carbon film under the same dosage as was used for data collection . The alignment was accepted when the Thon rings on the carbon support were visible beyond ( 1/4 . 5 ) Å−1 . A total of 1532 films were developed using full strength D19 ( Kodak ) solution . Micrographs were digitized with a Nikon Coolscan 9000ED scanner at a step size of 6 . 35 μm/pixel , which yielded a pixel size of 1 . 05 Å/pixel on the specimen . The absolute pixel size was determined by varying the pixel size from 1 . 05 to 1 . 14 in steps of 0 . 01 Å to find the “best” fit of the CHIKV E1E2 heterodimer crystal structure into the final cryoEM density . The fit was based on the average density of all the non-hydrogen fitted atoms ( sumf ) . The highest value of sumf was found when the pixel separation was 1 . 11 Å . Of the 1532 micrographs , 1012 did not have drift or charging problems and had Thon rings beyond ( 1/6 . 0 ) Å−1 observed through the sample . These were selected for further processing . A total of 52 , 183 particles were boxed with the EMAN2 program e2boxer ( Tang et al . , 2007 ) . The selected micrographs were found to be under-focused by between 1 . 1 μm and 2 . 9 μm using the EMAN program CTFIT ( Ludtke et al . , 1999 ) . Initially a “four-fold binned” map was used in which each set of 4 × 4 pixels was replaced by one pixel with the average height of the original sixteen pixels . Thus , the pixels in this map were separated by 4 . 44 Å , limiting the resolution to ∼13 Å . An initial CHIK VLP model was produced by selecting particles with 5- 3- and 2-fold symmetric projections from which an icosahedral reconstruction was made . This model was iteratively refined using a coarse 5° angular step size to improve the orientation and positioning of each particle until convergence was reached , as indicated by the lack of change of the Fourier shell correlation between successive cycles . At this point , the model had the characteristics of an alphavirus ( Cheng et al . , 1995 ) . In further iterations , the model was modified progressively by setting to zero both the inner density corresponding to the RNA genome and also the outer noise background density . Orientation and centers for each of the particles were determined by comparing the observed two-dimensional projections with projections of the model generated using a 1° angular step size . Each observed projection was allocated to the three best fits in the calculated projection classes . The images associated with each class were averaged and used to generate an improved model for the next cycle . Only those particles whose orientations and centers had changed their orientation and position by less than 2° and 2 pixels between successive cycles were retained for the next reconstruction . Unstable particles were discarded after each cycle and were not reconsidered . When the global searches had converged , local fine angular and positional refinements were conducted using 2-fold instead of 4-fold binned images . Particles were discarded when their three top solutions did not converge . After each cycle , the resolution of the map was estimated and used to modify the model by filtering out the Fourier coefficients beyond the current resolution . Further improvement was hindered because it was assumed that the magnification of each micrograph was the same and that the defocus distance of each particle on a given micrograph was the same . The program Frealign ( Grigorieff , 2007 ) was used to refine orientation , position , defocus distance and relative pixel size for each micrograph using the original scanned data without binning . Particles were rejected that were found after refinement to have changes in any of these parameters by more than 2σ . Overall convergence was established when the plot of the Fourier shell correlation vs resolution did not improve as estimated by visual inspection . The final map was reconstructed using 36 , 236 particles . The resolution of this map was estimated to be 4 . 6 Å , based on the spacing at which the Fourier shell correlation fell below 0 . 5 ( Van Heel , 1987 ) . To verify the effective resolution of the map ( Chang et al . , 2012 ) , the orientation of each image was determined with respect to the final map after having been low passed filtered to 7 Å . The resultant orientations of the images were used to calculate the Fourier shell correlation between two equal random groups of all the images . The map resolution was found to be 5 . 3 Å using the 0 . 5 Fourier shell correlation criterion . A series of maps were calculated using a B factor applied to the Fourier coefficients of the inverted map to estimate the noise level and backbone density continuity . The best value of B was -40 Å2 , which slightly increased the emphasis on the high order terms . Initially , the E1-E2 heterodimer crystal structure was fitted as a rigid body close to an icosahedral 3-fold axis ( Figure 2 ) defined by the polar angles psi = 69 . 1° , fi = -90° ( Rossmann and Blow , 1962 ) . This position was called “#1” ( Figure 2A ) . The icosahedral 3-fold axis was used to generate the complete icosahedral “i3” spike . Next , the i3 spike was rotated by the quasi-2-fold axis at psi = 74 . 6° , fi = -80 . 9° ( an orientation appropriate for T = 4 quasi-symmetry ) to produce the “q3” spike , generating positions #2 , #3 , #4 of the heterodimer ( Figure 2A ) . The position of this quasi-2-fold axis was refined to maximize the average height of the fitted density ( sumf ) while minimizing the number of clashes between symmetry related E1-E2 heterodimers ( Rossmann et al . , 2001 ) . The resultant positions of the four quasi-equivalent CHIKV heterodimers were used as starting positions for further position and orientation refinement of individual domains . E1 was divided into domain I ( residues 1-36 , 132-168 , 273-293 ) , II ( residues 37-131 , 169-272 ) and III ( residues 294-393 ) . E2 was divided into domain A ( residues 16-134 ) , B ( residues 173-231 ) , C ( residues 269-342 ) , and D the β-ribbon connector ( residues 7-15 , 135-172 , 232-268 ) . These domains were fitted sequentially ( I , II , III , A , C , B , and D ) . After each domain was fitted independently , the map was modified by zeroing out the density around each fitted atom within a radius of 3 . 0 Å . Thus , when fitting the next domain , atoms placed into zero density acted as a restraint by reducing the overall fitting criterion , “rcrit” ( Rossmann et al . , 2001 ) . This process substituted for avoiding clashes had all domains been fitted simultaneously . After the individual domains had been fitted , the bond geometry between the carboxy end of one domain and the amino end of the next domain were regularized . The average density height of the final atomic structure ( sumf ) was better for the individual domain fitting results than for the original T = 4 rigid body fitting results ( Table 1A ) , validating the fitting strategy . The quality of the independently fitted domain structure was also validated by comparing the fit of the atomic structures at each of the four quasi-equivalent densities . The density of each non-hydrogen atom in the CHIKV model was determined by interpolation using the densities at the eight surrounding grid points . The average density ( ρi ) of residue i was then set to the average densities of all the atoms associated with this residue . The correlation coefficient ( CC ) was calculated between the densities ρi ( A ) and ρi ( B ) where these are the cryoEM densities of residue i at the quasi-equivalent positions A and B . The sums are taken over i = 1 , N where N is the number of residues being matched . CC=∑[ ( ρi ( A ) -<ρi ( A ) > ) ( ρi ( B ) -<ρi ( B ) > ) ][∑ ( ρi ( A ) -<ρi ( A ) > ) 2∑ ( ρi ( B ) -<ρi ( B ) > ) 2]½ Non-covalent contacts between molecular entities were determined and classified by identifying the number of times a specific atom in one molecule was within a defined distance , Dlim = 3 . 5 Å , of any atom in a specific neighboring molecule . The contacts of each atom in a given molecule were added to yield a measure of how many close contacts the first molecule had with the second molecule . This count is roughly equivalent to measuring the surface contact area between the two molecules . In addition to measuring the total number of contacts , a count was made of that portion of contacts that represented hydrophobic interactions by counting only carbon-carbon atom distances less than Dlim . Also , the number of potential hydrogen bonds was estimated by counting the number of short ( Dlim < 3 . 5 Å ) distances between oxygen and nitrogen atoms . Lastly , a count was made of possible formation of salt bridges by counting the number of occasions that an Arg , Lys or His residue approached to within Dlim of an Asp or Glu residue . The radius of gyration of the spikes was computed as:rg=∑ ( ri2 ) N , where the Sum ( i = 1 to N ) is taken over the N Cα atoms in E1 and E2 of one spike , and ri is the radius of the ith atom measured from the spike axis . The latter is defined by the positions of the centers between the 3-fold ( in the i3 spike ) , or quasi-3-fold ( in the q3 spike ) related atoms . Alignment of the molecular components between VEEV and CHIK VLPs was performed with the HOMOlogy program ( Rossmann and Argos , 1975 ) . The position and orientation of each molecular component was calculated with respect to the equivalenced Cα atoms . CHIK VLP particles were incubated with Fab fragments at 4°C for two hours using a stoichiometric ratio of about four Fab fragments per E2 molecule . Samples were hand-blotted , flash-frozen on holey grids in liquid ethane . Images were recorded at 47 , 000× magnification with a CM300 field emission gun microscope using electron dose levels of approximately 20 electrons per Å2 . All micrographs were digitized at 6 . 35 μm per pixel using a Nikon scanner . Individual particle images were boxed using the program e2boxer in the EMAN2 package . Subsequently , the boxed images were two-fold binned , resulting in a sampling of the specimen images at 2 . 69 Å intervals . The CTFIT program from the EMAN package was used to determine the contrast transfer function parameters . A cryoEM reconstruction of CHIK VLP , low pass filtered to 18 Å , was used as an initial model for orientation determination and further refinement . The number of particles incorporated into the final reconstruction was 1728 , 1820 , and 1599 for Fabs m242 , CHK9 and m10 , giving final resolutions of 15 . 6 Å , 14 . 9 Å and 16 . 9 Å on the basis of a 0 . 5 Fourier shell correlation threshold , respectively . The m242 Fab was crystallized in 2M ammonium sulfate and 0 . 1 M sodium acetate pH 4 . 5 and the CHK-9 Fab was crystallized in 25% PEG 3350 , 0 . 1M Tris-Cl pH 8 . 5 and 0 . 2M lithium sulfate . Crystals were grown by vapor diffusion in hanging drops at 20°C . Crystals were flash-frozen and data were collected at 100K at the Advanced Photon Source ( APS ) GM/CA-CAT 23-ID-B beamline . Data were processed with the HKL2000 program ( Otwinowski and Minor , 1997 ) . The m242 Fab crystals diffracted to 3 . 1 Å resolution and had a P21212 space group with cell dimensions a = 137 . 0 , b = 89 . 1 and c = 94 . 4 Å . There were two Fab molecules per asymmetric unit . The CHK-9 crystals diffracted to 3 . 0 Å resolution and had a C2 space group with cell dimensions a = 87 . 9 , b= 57 . 7 and c = 118 . 1 Å . There was one Fab molecule per asymmetric unit . The structures of the m242 Fab and CHK-9 Fab crystals were determined by molecular replacement using a Fab crystal structure ( Protein Data Bank: 3DGG ) as a search model with the program Phaser ( Mccoy et al . , 2007 ) . The structures were refined using the program Refmac ( Murshudov et al . , 1997 ) ( Table 9 ) . The rms difference between the Cα atoms of the two molecules in the asymmetric unit of the m242 Fab crystals was 0 . 7 Å .	The Chikungunya virus is carried by mosquitos and can cause a number of diseases in humans including encephalitis , which can be fatal in some cases , and severe arthritis . A recent mutation in the E1 protein of the virus has allowed it to efficiently reproduce in a different species of mosquitos , leading to a Chikungunya epidemic in Réunion Island in 2005 and the subsequent infection of millions of individuals in Africa and Asia . The virus also has the potential to spread to many areas of Europe and the Americas . Chikungunya virus has a single-stranded RNA genome that codes for four non-structural proteins and five structural proteins . Based on this knowledge it has been possible to develop virus-like particles that can be used to immunise non-human primates against Chikungunya infection by inducing antibody production . However , the development of vaccines for Chikungunya in humans will require a deeper understanding of how these antibodies produced by the vaccine interact with the virus and more detailed information about the structures of the virus and antibodies . Sun et al . have used two techniques – X-ray crystallography and electron cryo-microscopy – to determine the structure of Chikungunya virus-like particles , and to obtain new insights into the interactions of these particles with four related antibodies . Electron cryo-microscopy was used to figure out the structure of the particles at near atomic resolution , and X-ray crystallography was used to determine the atomic resolution structures of two of the four Fab antibodies that neutralize the Chikungunya virus . Electron cryo-microscopy was also used to probe the complex formed by the interactions between the virus-like particles and the antibodies . Sun et al . were able to identify the likely viral receptor site that is blocked by three of the antibodies when they neutralize the virus; the fourth antibody is thought to act by immobilizing one of the domains of protein E2 , thereby hiding the “fusion loop” that allows the virus to enter and infect human tissue . It is hoped that these findings will contribute to efforts to combat the spread of the Chikungunya virus worldwide .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2013	Structural analyses at pseudo atomic resolution of Chikungunya virus and antibodies show mechanisms of neutralization
Transcranial magnetic stimulation ( TMS ) is a widely used non-invasive tool to study and modulate human brain functions . However , TMS-evoked activity of individual neurons has remained largely inaccessible due to the large TMS-induced electromagnetic fields . Here , we present a general method providing direct in vivo electrophysiological access to TMS-evoked neuronal activity 0 . 8–1 ms after TMS onset . We translated human single-pulse TMS to rodents and unveiled time-grained evoked activities of motor cortex layer V neurons that show high-frequency spiking within the first 6 ms depending on TMS-induced current orientation and a multiphasic spike-rhythm alternating between excitation and inhibition in the 6–300 ms epoch , all of which can be linked to various human TMS responses recorded at the level of spinal cord and muscles . The advance here facilitates a new level of insight into the TMS-brain interaction that is vital for developing this non-invasive tool to purposefully explore and effectively treat the human brain . In 1985 , when Barker et al . ( 1985 ) applied a pulsed magnetic field to a subject’s head to selectively evoke responses in a hand muscle , the scientific community was fascinated by this pain-free , electrodeless , and non-invasive tool for brain stimulation . Since then , transcranial magnetic stimulation ( TMS ) has attracted enormous interest in neurology , psychiatry , and applied neuroscience research for its unique capability of non-invasively activating neuronal populations and inducing plasticity . Despite its impressive array of applications and fast-growing popularity ( Cohen et al . , 1998; Walsh and Cowey , 2000; Reis et al . , 2008; Funke and Benali , 2011; Lefaucheur et al . , 2014 ) , TMS is poorly understood physiologically as we know very little about how TMS interacts with the brain at the level of neurons and circuitries . Although various sources of indirect evidence obtained from human studies support the use of TMS in multiple contexts ( Di Lazzaro et al . , 2008; Chung et al . , 2015; Suppa et al . , 2016 ) , our very limited insight into the neurophysiology of TMS remains a bottleneck that hampers the utilization and the development of TMS applications , blocking the exciting potential of this non-invasive brain stimulation tool . One critical reason behind this inadequacy is the absence of a research platform on which TMS-evoked neuronal activities can be investigated in vivo in real time . Extracellular electrophysiology ( EEP ) with microelectrodes is the gold standard for studying brain activities at the level of neurons ( Scanziani and Häusser , 2009 ) . However , a single TMS pulse , characterized by an alternating tesla-level magnetic field with a center frequency of approximately 4 kHz ( Ilmoniemi et al . , 1999; Wagner et al . , 2007 ) , generates an array of strong interferences disturbing the acquisition of EEP signal . As evidenced by the pioneering works on TMS-EEP ( Moliadze et al . , 2003; Pasley et al . , 2009 ) , artifacts from TMS resulted in a significant amount of data loss that precluded the investigation of TMS-evoked neuronal activities for up to 100 ms after each TMS pulse . Several groups used innovative imaging methods to circumvent this problem ( Allen et al . , 2007; Kozyrev et al . , 2014; Murphy et al . , 2016 ) ; however , their investigations were restricted to the upper cortical layers and they were unable to reach the single-neuron and millisecond precision needed to decipher the intricate interactions between TMS and neurons . Recently , two groups reported TMS-EEP methods for non-human primate research . One of the methods utilized custom-built TMS coil and offline correction to minimize the TMS-induced data loss to a median time of 2 . 5 ms ( Tischler et al . , 2011 ) , while the other used a combination of custom-built coil , amplifier modifications , field sensing , active compensation , and offline correction to minimize the data loss to 1 ms ( Mueller et al . , 2014 ) . Despite their success in artifact reduction , these methods face a major limitation that the technical expertise required for their implementations , especially custom-building TMS coils and field sensing , is not widely accessible to the neuroscience community . More importantly , these methods were developed solely for non-human primate research , which is used for investigating the neuronal underpinnings of high-level cognitive functions and therefore is not best suited for investigations concerning basic neurophysiology on the level of cells and detailed microcircuits . With the aim of establishing a widely applicable in vivo experimental platform to study the dynamics of TMS-evoked neuronal activities and to further develop the scientific and clinical applications of this powerful technique , we engineered a novel method for TMS-EEP that is suitable for , but not limited to , laboratory rodents , which are widely accessible and offer a rich repertoire of experimental techniques including transgenic and optogenetic tools . The method is compatible with existing standard TMS coils and allows the recording of neuronal activities 0 . 8–1 ms after the onset of various types of TMS stimuli by attenuating artifacts resulting from magnetic induction , electric field coupling , and vibrations . Furthermore , the method allows for the instantaneous determination of TMS-driven inadvertent charge injection into the neural tissue , a problem that has been overlooked by almost all prior TMS-EEP studies . In the following sections , we will present this methodological advance and demonstrate its potential by unveiling neuronal activities in the layer V of the primary motor cortex ( M1 ) that accompany cortically evoked unilateral muscle activations by monophasic single-pulse TMS ( mspTMS ) . Additionally , we will also demonstrate that mspTMS modulates different neuronal circuits depending on the orientation of the stimulating current or the time-window of investigation . The induction artifact ( Figure 1 , period indicated in blue ) is created as coil-generated rapid magnetic flux change induces voltages within loops formed along an EEP recording assembly . Owing to the large rate of flux change , the induction artifact , if transmitted to the high-gain and filter stages of an EEP amplifier , can easily lead to signal saturation and data loss ( Figure 1 inset ) . To address this , we developed a gated multi-stage TMS-EEP amplifier ( Figure 2 ) . It consists of a differential preamplifier ( Pamp ) stage of gain four and a filter-amplifier ( Famp ) stage of gain 500 , separated by an ultra-low capacitance/charge injection solid-state analog switch ( SSSW ) controlled by optically coupled digital signal synchronized to TMS . The components of the amplifier were chosen to provide the optimal balance between voltage and current noise with the source impedance of EEP microelectrodes . The Pamp stage must be able to maintain its high impedance character when being perturbed by TMS . Failing to do so will result in excessive induction current in the input wires that leads to electrical stimulation of the brain . In our design , the electronic components and supply voltage of the Pamp stage were chosen so it can tolerate ±7 . 9 V input during TMS . Due to its high-gain and filters , the Famp stage must be protected from TMS by SSSW that grounds the input to Famp for a user-defined time interval around TMS onset ( e . g . from 0 . 2 ms pre- to 0 . 8 ms post-TMS ) . The SSSW was strategically placed behind the input capacitor of a high-pass filter of the Famp so that the input capacitor is preconditioned to any DC bias in the microelectrodes before the end of the grounding period . To minimize ground bounce and to reduce ground loop , the external digital signal that controls SSSW is connected to the amplifier circuit through an optocoupler ( OC ) . In addition , to protect the amplifier circuit from TMS-induced fields , the circuit board of the amplifier is mounted inside a 1 . 5-mm-thick grounded aluminum enclosure . The Pamp , as well as DC-DC converters , is housed inside a metal shield for additional protection . Furthermore , to minimize artifacts from vibrations due to the loud click sound of TMS coils , solder contacts , instead of spring-loaded connectors , were used whenever possible , as well as polyphenylene sulfide film capacitors , since they do not generate piezoelectric voltages from vibration . The default frequency response of the amplifier was set from 300 Hz ( −3 dB ) to 5 kHz ( −6 dB ) , but the lower bound of the passband can be adjusted to 4 Hz as needed for different applications . A simplified circuit diagram of the amplifier , including the model number of its critical components , is shown in Figure 2 . The coil-emitted electric field gives rise to another type of artifacts . When a TMS pulse is triggered , a large current driven by a kV-level voltage pulse flows through the coil . Inadvertently , this process emits an electric field that injects a displacement current into the EEP recording assembly through capacitive coupling ( Figure 3A ) . In a short time-interval ( <0 . 4 ms; Figure 1 , period indicated in orange ) , as the coil current rapidly fluctuates , the displacement current generates a fast-changing artifact in EEP signal . In a long time-interval ( tens of ms; Figure 1 , period indicated in yellow ) , a decay-like artifact is observed as the displacement current polarizes the electrochemical double-layer of microelectrode tips and thereby generates a decaying waveform while the double-layer returns to its equilibrium potential . Depending on stimulation intensity , electrode impedance , and filter settings , the decay may persist with relatively high signal values for tens of milliseconds before re-entering the input range of a standard EEP amplifier ( green area in Figure 1 ) , contributing to an extended period of data loss . To address this problem , we developed an electrical shield for the TMS coil that substantially attenuates the coil-emitted electric field ( Figure 3B ) . One important consideration in shield construction is that the amount of eddy current in the shield should remain low; otherwise , strong vibration or even magnetic attenuation may occur . Therefore , we applied a layer of weakly conducting material in shield construction and the resulted shield possesses an electrical resistance of 10 kΩ ( see Materials and methods ) that does not result in vibration and magnetic attenuation . To verify the performance of our shield , we first used a magnetic pickup probe to confirm that at 10 kΩ , the shield does not attenuate the magnetic output of our TMS coil . As Figure 3C illustrates , induction voltage waveforms , with and without the shield , overlap perfectly , confirming the absence of any noticeable magnetic attenuation . Subsequently , using a high-impedance buffer , we measured the voltage between an EEP microelectrode and a ground electrode , both electrode tips immersed in a saline bath , under mspTMS at 100% maximum stimulator output ( MSO ) delivered with or without the shield . We expected that the shield would remove , to a large extent , voltage signal that is due to electric field coupling between the TMS coil and the EEP recording assembly . Figure 3D illustrates the results from these measurements . Without the shield , the captured waveform was drastically different from the induction waveforms resulted by mspTMS ( as seen in Figure 3C ) , and it ended with a strong voltage bias ( polarization ) . With the shield in place , the captured waveform appeared consistent with the induction waveforms and the voltage bias was no longer visible . These findings confirm that by interrupting electric field coupling , the shield is effective in preventing polarization and the decay artifact that follows . Upon elimination of artifacts resulted from induction and electric field coupling , vibration artifacts , which are normally masked by the other artifacts , become visible ( Figure 4B ) . Vibration can be generated by magnetic force , as well as by sound pressure perturbation . For the magnetically mediated vibration , an avoidance of ferromagnetic materials and large conductive surfaces in the close vicinity of the coil is adequate . However , the elimination of vibration artifacts driven by sound pressure is not straightforward . When a TMS pulse is triggered , a loud click sound is produced by coil wires due to the attractive forces between them . This sound is problematic as it generates micro-vibration in the amplifier input cables . The extremely weak signal ( μV-level ) these cables carry can be easily perturbed by micro-vibrations through the triboelectric effect ( Fowler , 1976 ) . Since the generation of such click sound is inevitable , we attenuated the vibration artifacts by using a special type of low-noise miniature coaxial cable with a semiconducting layer added between its dielectric and braided shield ( Figure 4A ) . The addition of this semiconducting layer provides a path that drains triboelectric charges , rendering the cables insensitive to vibration ( Figure 4C ) . Despite the impressive performance of the low-noise miniature coaxial cable in attenuating vibration artifacts , its length in an EEP recording assembly should be limited as the cable’s capacitance ( 100 pF/m ) , together with the amplifier input capacitance and electrode impedance , acts as a voltage divider that attenuates EEP signal . In our case , we kept the length of our cables at 16 cm to keep the signal attenuation less than 20% at 1 kHz . TMS-driven inadvertent charge injection is another major issue , which has been overlooked by most prior reports using EEP under TMS ( Moliadze et al . , 2003 , 2005; Pasley et al . , 2009 ) . By inserting electrodes into the brain and connecting them to the measurement electronics , multiple loops of electric circuit are formed ( Figure 5 ) . When being subjected to alternating electric and magnetic field , voltages can be readily developed along these loops that drive unwanted current injection into the brain through microelectrodes . If the amount and the temporal structure of the injected current are comparable to the threshold parameters reported in intracortical microstimulation ( ICMS ) literature ( bipolar charge transfer totaling from 150 to 800 pC , current waveform approximately similar to that of TMS; see Asanuma and Rosén , 1973 and Butovas and Schwarz , 2003 ) , such current will excite neuronal elements around the microelectrode tips and therefore severely confound the measurement of TMS effects . Therefore , it is crucial that the development of voltages along these loops be minimized . Since a large portion of the TMS-emitted electric field had already been filtered away by the coil shield , precautions were taken for magnetic induction . These included a compact arrangement of microelectrodes as well as cable twisting ( Figure 5—figure supplement 1 ) , both minimize the area of induction loops exposed to TMS . More importantly , we incorporated a low-gain monitoring channel ( LGM as seen in Figure 2 ) in our amplifier design that allowed us to conveniently determine the amount of inadvertent current flow , without any additional measurement devices , under each experimental setup . The conversion from voltage , which is measured by the LGM , to current is made possible since amplifier’s input capacitance and its voltage fluctuation are known , and the amount of current flow through the input capacitance is equal to the amount of current flow in the circuit . A detailed description of this conversion is presented in Materials and methods and Figure 5—figure supplement 2 . In six male Sprague-Dawley rats , we evaluated and optimized our method . Figure 6A offers an overview of our recording setup and the subsequent figures illustrate the performance of the method in vivo under a single monophasic ( Figure 6B ) and biphasic ( Figure 6C ) TMS pulse , as well as a triplet of 50 Hz biphasic pulses ( Figure 6D ) , which is the fundamental building block of theta burst stimulation ( Huang et al . , 2005 ) . The stimuli delivered here can be considered as the ‘worst-case scenarios’ as the stimulator-coil combinations used yield magnetic outputs ( peak strength up to four tesla ) that are one of the highest among commercially available TMS systems ( see Materials and methods ) . Nonetheless , the EEP signal recovered between 0 . 8 and 1 ms after the onset of each TMS pulse and was free from artifacts . Furthermore , the amount of inadvertent charge injection under each condition was far below ( by a factor of 200 or more; Figure 6—figure supplement 1 ) the modulation or activation thresholds reported in ICMS literature ( Asanuma and Rosén , 1973; Butovas and Schwarz , 2003 ) , confirming the validity of our measurements . With the newly developed method , we sought to address the question: what are the neuronal correlates of TMS cortically evoked muscle activation ? In another group of seven male Sprague-Dawley rats ( anesthetized by ketamine-xylazine ) , we recorded in vivo mspTMS-evoked neuronal activities in the layer V ( Figure 7—figure supplement 1 ) of the caudal forelimb area ( CFA ) , rodent’s equivalent to the forelimb area of primate M1 ( Rouiller et al . , 1993 ) . With the coil center positioned over the left CFA and the induced current pointing from the medial to the lateral part of the brain ( ML stimulation; Figure 6A inset ) , mspTMS evoked unilateral movement of the right forelimb . Simultaneous intramuscular electromyogram ( EMG ) recordings of the left and right biceps brachii muscle revealed motor unit action potentials ( MUAPs ) unilaterally in right biceps brachii ( contralateral to the stimulated hemisphere; Figure 7B and the insets of Figure 7C–F ) . The onset latency of the MUAPs was around 11 ms , similar to that found in our single-pulse ICMS experiment ( Figure 7—figure supplement 2 ) and in the rodent single-pulse ICMS literature ( Liang et al . , 1993; Deffeyes et al . , 2015 ) , confirming the cortical origin of TMS-evoked muscle activation . At the neuronal level , in layer V of the CFA , mspTMS evoked a rhythm of neuronal activities alternating between excitation and inhibition that lasted until approximately 300 ms post-stimulation . Figure 7A illustrates the multiunit spike raster and its corresponding peristimulus time histogram ( PSTH ) of multiunit firing rate ( FR ) from one animal . Figure 7C–F show the evoked normalized FR ( instantaneous FR subtracted by baseline average FR; see Materials and methods ) with increasing stimulation intensity ( 0% , 95% , 100% , and 120% motor threshold , MT ) averaged across all animals . Significance thresholds were drawn based on the 2 . 5 and 97 . 5 percentile of the empirical distribution of normalized FR during baseline ( 500 ms pre-TMS; see Materials and methods for details ) to control Type I error rate ( p<0 . 05 ) . We categorized the evoked significant excitatory and inhibitory events into three phases: intermediate excitation ( a period of increased FR that peaks around 20 ms ) , inhibition ( a long-lasting pause in FR after the intermediate excitation ) , and rebound excitation ( a period of increased FR following the inhibition ) . To investigate the effects of stimulation intensity on the normalized FR of each phase , we constructed hierarchical linear mixed-effects models . Stimulation intensity positively modulated the normalized FR for the intermediate excitation phase ( β = 2 . 75 ± 0 . 24 , F ( 1 ) =127 . 23 , p<0 . 001 ) and the rebound excitation phase ( β = 1 . 18 ± 0 . 19 , F ( 1 ) =38 . 65 , p<0 . 001 ) , while negatively modulated the normalized FR for the inhibition phase ( β = −0 . 23 ± 0 . 10 , F ( 1 ) =5 . 28 , p=0 . 02 ) . It is important to note here that despite the faithful EMG response in the contralateral biceps brachii muscle , the neuronal firing rate in the short-latency window ( 1–6 ms after TMS onset ) was low . This finding was rather surprising and will be further explored in the following section . Since we did not observe any significant modulation of neuronal FR in the short-latency window ( 1–6 ms ) after mspTMS despite faithful muscle activations in the contralateral forelimb , in another set of experiments ( N = 4 ) , we explored the possibility that neuronal response in this very early time window is dependent on the direction of mspTMS-induced current . In this set of experiments , we switched the TMS coil orientation so that the induced current flows from the posterior to the anterior part of the brain ( PA stimulation; Figure 6A inset ) . We could replicate most findings found in the previous set of experiments as the multiphasic response evoked by PA stimulation is qualitatively similar to that evoked by ML stimulation ( Figure 7—figure supplement 3 ) . However , in the short-latency window after TMS onset , the neuronal responses observed in ML and PA stimulation are drastically different . As the two examples in Figure 8A demonstrate , at 120% MT intensity , ML stimulation evoked scarcely any spike , whereas PA stimulation evoked robust neuronal firing generating a distinct temporal pattern with peaks at 1 . 2–1 . 6 ms and at 3 . 2–4 . 2 ms . To quantify the short-latency responses in the population , we constructed PSTHs of normalized FR across all animals under ML ( Figure 8B ) and PA ( Figure 8C ) stimulation . Significance thresholds were drawn using the 2 . 5 and 97 . 5 percentile of normalized FR distribution during baseline . ML stimulation evoked no significant excitation with the exception of the low albeit significant FR at 3 . 5–4 ms under stimulation intensity of 120% MT . On the contrary , under PA stimulation , multiple significant excitatory events were observed . At subthreshold level , significant excitatory events appeared at 2 . 5–3 . 5 ms and at 4–4 . 5 ms . As the stimulation intensity increased , FR was developed at particular time windows: 1–1 . 5 ms and 2 . 5–4 . 5 ms , reminiscent of the indirect wave ( I-wave ) phenomena observed in the corticospinal descending volleys in human and animal studies . Our understanding of the neuronal mechanism of TMS has been largely based on indirect evidence obtained at the level of cortical output reflected in spinal cord or muscle activities . Direct investigation of the dynamics of neuronal activities evoked by TMS was hindered by technical obstacles imposed by the strong electromagnetic pulse produced by TMS . We engineered a widely applicable experimental method for the in vivo study of TMS-evoked brain activities at the level of neurons using EEP . It allows the monitoring of neuronal activities as early as 0 . 8–1 ms after the strong electromagnetic perturbation of various TMS stimuli ranging from single pulse to the high-frequency theta burst stimulation . Our method encompasses solutions to all major challenges in concurrent TMS-EEP recording , including magnetic induction , electric field coupling , vibrations , and inadvertent charge injection . Despite the multidimensional approach of our method , it was developed with generalizability , simplicity , flexibility , and scalability in mind . It is compatible with , but not limited to , rodents , an animal model that is widely used for studying basic neurophysiology and offers a wide range of investigative tools . It does not require active compensation based on magnetic field sensing ( Logothetis et al . , 2001; Mueller et al . , 2014 ) or custom-made coils ( Tischler et al . , 2011; Mueller et al . , 2014 ) for artifact reduction . It can accommodate electrodes directly under a conventional TMS coil , a feature making it suitable even for awake animals with chronically implanted electrodes . In addition , the method can be scaled up for large-scale high-density EEP recording with silicon-based microelectrode arrays ( Buzsáki , 2004 ) as well as be accompanied by optogenetic tools for the in vivo control of neuronal circuits ( Scanziani and Häusser , 2009 ) . The amount of magnetic , electric , and vibrational interference TMS imposed on EEP depends on multiple factors . Some of the most critical factors include the waveform and magnitude of the pulsed magnetic and electric field emitted from a TMS coil , the size of circuit loops formed by an EEP recording assembly , and the coil position relative to these loops . Changes in these parameters will result in changes in the severity of different types of interference . For example , keeping the coil the same , by replacing a standard monophasic with a standard biphasic stimulator , coil-emitted fields will generate a longer period of magnetic and electric field interference due to the longer pulse waveform . However , the severity of interference might be lower if the coil and biphasic stimulator combination does not produce magnetic and electric outputs that are as high as those in the monophasic case . Similarly , miniaturization of TMS coils for small animals can also lead to a reduction in interference because of the reduced electromagnetic output of such devices . Furthermore , the integration of recording , reference , and ground electrode in one microfabricated electrode array can also reduce the severity of interference as such configuration significantly decreases the area of circuit loops exposed to TMS . One common criticism of TMS investigations in rodents is that the TMS coil is large compared to the size of a rodent brain . While we fully acknowledge this concern , we argue that it is not a problem of critical importance at this stage . With careful coil positioning , it is possible to achieve certain level of spatial selectivity as evident in the results of our study as well as those of several other reports ( Nielsen et al . , 2007; Rotenberg et al . , 2010; Muller et al . , 2014 ) . In addition , plasticity , assessed by motor output ( Muller et al . , 2014 ) , learning performance ( Mix et al . , 2010 ) , sensory-evoked neural activities ( Thimm and Funke , 2015; Murphy et al . , 2016 ) , or protein expressions ( Trippe et al . , 2009; Benali et al . , 2011 ) , can also be successfully induced in rodents using human TMS coils , making rodents a suitable experimental model for investigating the basic neuronal mechanisms underlying stimulation-induced plasticity . Furthermore , TMS can be used as a tool to deliver a strong transient stimulus to perturb neuronal populations of the neocortex ( Walsh and Cowey , 2000 ) . Being able to capture the neuronal response to such perturbation at spike resolution will undoubtedly open up another avenue to study the connectivity and the functional properties of neuronal networks . Nonetheless , the development of smaller and more compact coils specifically designed for small animals would be beneficial for their improved spatial resolution and smaller electromagnetic interference as the maximum magnetic output of these coils is much smaller ( at mT level; Makowiecki et al . , 2014; Tang et al . , 2016 ) than the 4T output tested in our development . To validate our method , we successfully translated mspTMS to rodents and unveiled the evoked neuronal activities underlying this classical TMS stimulus which has been widely used in humans since its introduction in 1985 ( Barker et al . , 1985 ) . At the behavioral level , mspTMS delivered in either ML or PA direction over left CFA evoked unilateral forelimb movement in the contralateral side . The similarity between the onset latency of mspTMS- and single-pulse ICMS-evoked MUAPs suggests the cortical origin of the TMS-evoked muscle activations . At the neuronal level , despite the similar evoked motor outputs , mspTMS delivered in the ML and PA orientation evoked different CFA layer V neuronal activities in the short-latency window ( 1–6 ms ) after TMS onset . While threshold or suprathreshold ML oriented stimuli evoked virtually no response in this time window , PA oriented stimuli evoked population spiking activities that occurred preferably around 1–1 . 5 ms and at 3–3 . 5 ms ( Figure 8 ) . Such discrepancy in neuronal response suggests that TMS-induced current of different orientations activates different microcircuits in the rodent forelimb M1: ML-oriented stimuli directly activated pyramidal cells of the descending motor pathways while PA-oriented stimuli evoked trans-synaptic high-frequency spiking activities in M1 ( Figure 9A ) . It is important to note here that neuronal activity within 1 ms after TMS onset is not visible . Therefore , any antidromic spike evoked by direct axonal activations could not be recorded . It might be argued that the observed discrepancy in short-latency response is a result of bias in neuronal sampling . We believe this is rather unlikely , as short-latency spikes evoked by ML stimulation were absent across multiple recording sites within CFA ( 0 out of 7 sites ) while the significant high-frequency spiking pattern was observed readily within CFA under PA stimulation ( 3 out of 4 sites ) . Additionally , in PA trials , the observed high-frequency spiking disappeared when we turned the stimulus orientation to ML . While we cannot rule out the possibility that mspTMS evoked early spike responses in areas other than the ones we monitored , our data supports the notion that in the layer V of CFA — the output layer of the rodent forelimb M1 — selectivity in stimulus orientation exists . Another confounding factor that might explain the discrepancy is the intensity difference of induced electric fields in the brain under ML and PA stimulation . Since the rodent skull is not spherical , with a given coil output , induced electric field in the ML direction ( along the short axis of the skull ) should be lower in intensity than that in the PA direction ( along the long axis of the skull ) , raising the possibility that the observed high-frequency spiking pattern under PA stimulation is a result of high intensity of the induced electric field . However , motor thresholds under ML stimulation , in which induced electric field intensity is lower , were significantly lower than their PA counterparts ( medianML = 61% MSO; medianPA = 74% MSO; Wilcoxon rank-sum test , p=0 . 03 ) . This is a strong indication that factors other than induced electric field intensity play a critical role in stimulus orientation selectivity . Therefore , we conclude that the observed response difference between ML and PA stimulation is unlikely to be caused solely by the difference in the intensity of induced electric fields . TMS works on human ( Kaneko et al . , 1996; Di Lazzaro et al . , 2001 ) and non-human primates ( Amassian et al . , 1990; Amassian and Stewart , 2003 ) also reported similar stimulus orientation selectivity but in the context of evoked motor outputs . However , we stress that the similarities between our results and those of humans and non-human primates rest solely at the level of a shared common principle: TMS-evoked direct activation is a product of the interaction between TMS-induced electric field and the anatomical and physiological properties of the neurons within . Despite different levels of complexity between primate and rodent brains , certain neuronal structures are preferably stimulated in one stimulus orientation rather than the others . But whether such similarity is based on shared anatomical and physiological properties warrants further investigation . Furthermore , the primate cortex is gyrencephalic while the rodent cortex is lissencephalic . As we could reliably stimulate a lissencephalic M1 and evoke muscle activation on the contralateral forelimb at the correct cortically evoked latency , the locus of direct TMS activation is most likely not dependent on the magnitude of induced electric field component normal to the cortical columns ( Fox et al . , 2004; Bungert et al . , 2017 ) . The evoked short-latency response in the PA orientation was characterized by population spikes at a very high frequency similar to that of the I-waves recorded in the corticospinal tracts of humans and animals in response to a transient shock delivered to the M1 by either transcranial electrical stimulation ( Patton and Amassian , 1954; Kernell and Chien-Ping , 1967 ) or TMS ( Kaneko et al . , 1996; Nakamura et al . , 1996; Di Lazzaro et al . , 2012 ) . What are the principles of anatomical and functional organization in M1 that drive such high-frequency neuronal response ? We recorded in layer V of the motor cortex ( Figure 7—figure supplement 1 ) where two types of excitatory projection neurons exist: the corticospinal tract ( PT ) neurons that project to midbrain , brainstem , and spinal cord , and the intratelencephalic ( IT ) neurons that project ipsi- or bilaterally within the cortex and striatum ( Harris and Shepherd , 2015 ) . It has been shown recently that PT neurons exhibit reciprocal connectivity characterized by short-term facilitation and that synaptic transmission time for a pair of reciprocally connected PT neurons is 1 . 6 ± 0 . 5 ms ( Morishima and Kawaguchi , 2006; Morishima et al . , 2011 ) . Therefore , it is plausible that the network formed by the interconnected PT neurons in layer V provides the physiological foundation for the high-frequency neuronal discharge and that a mspTMS pulse oriented in PA direction preferably delivers an input into this network that triggers the observed high-frequency spiking response ( Figure 9A ) . Furthermore , the interconnected PT network may also offer a neuronal explanation for the short-interval intracortical facilitation ( SICF ) described in the human literature ( Tokimura et al . , 1996; Ziemann et al . , 1998 ) . As we extend the window of investigation to 6–300 ms after TMS onset , a multiphasic response appears among the recorded CFA layer V neurons . The response is characterized by its excitation-inhibition-excitation pattern and is not qualitatively different between PA and ML stimulations ( Figure 7; Figure 7—figure supplement 3 ) . The strong excitation that peaks around 20 ms , given its latency , duration , presence in both layer V and II/III ( Figure 7—figure supplement 4 ) , and its apparent lack of motor output ( Figure 7B ) , reflects a high excitability state of the motor cortex . We hypothesize that this excitation is generated through the cortico-basal ganglia-thalamo-cortical loops ( Figure 9B ) . Evidence suggests that cortex projects monosynaptically to basal ganglia ( BG ) structures such as striatum and subthalamic nucleus ( STN ) ( Kita and Kita , 2012 ) , while the projection from striatum and STN back to cortex is polysynaptic ( Shepherd , 2013 ) . Deep brain stimulation ( DBS ) of the STN produces cortically evoked EEG potentials with a peak latency of 22 . 2 ± 1 . 2 ms , and TMS delivered at this latency after DBS showed facilitation of its cortically evoked motor outputs ( Kuriakose et al . , 2010 ) . It is likely that TMS activates IT and PT neurons that project to BG monosynaptically , and the response is then transmitted back to the cortex as the intermediate excitation observed here . But other cortico-cortical or cortico-subcortical loops could be involved as well . The neuronal mechanism of TMS protocols such as intracortical facilitation ( Ziemann et al . , 1996 ) and theta burst stimulation ( interpulse interval of 20 ms within each burst ) ( Benali et al . , 2011; Suppa et al . , 2016 ) remain unknown; however , it is conceivable that these protocols exploit this particular phase of excitation for their physiological effects . The long-lasting inhibition phase that follows the intermediate excitation is well-known , and evidence supports the notion that it is mediated by GABAB ( Butovas et al . , 2006; Murphy et al . , 2016 ) and underlies the long-interval intracortical inhibition as well as the cortical silent period in human TMS ( Valls-Solé et al . , 1992; McDonnell et al . , 2006 ) . However , the local or long-distance circuit mediating this phase of inhibition remains unknown . The rebound excitation phase , occurring after the inhibition , represents a period of excitation most likely resulting from the termination of GABAB inhibition , and corresponds to the late cortical disinhibition , which is being harnessed for augmenting plasticity induction in human TMS ( Cash et al . , 2016 ) . Similarly , the circuit mechanism behind this phase of rebound excitation remains to be elucidated as well . Would the same neuronal activity pattern be observed if a rodent-sized TMS coil is used to stimulate the forelimb M1 ? We believe that this is the case since we carefully calibrated coil position and stimulation strength according to MEP . Furthermore , the long-lasting inhibition and the rebound excitation are well-documented phenomena in ICMS ( Butovas and Schwarz , 2003 ) , which is a much more localized stimulation method than TMS . Additionally , as discussed above , data from human TMS is largely congruent with the pattern of neuronal activity reported here . However , we cannot rule out the possibility that the coil we used in this study directly activated structures outside of the forelimb M1 . Nonetheless , the role of stimulus spatial resolution in modulating neuronal networks is a highly interesting topic for future research . By combining the tool presented here with optogenetic , transgenic , anatomical , theoretical , and clinical methods , future work could take on two parallel directions concerning either the short- or long-latency evoked response of TMS . For the short-latency response , investigation could focus on discerning the circuit selectivity of different stimulus orientations by pinpointing the locus of direct activation in each case , and on elucidating the principles of anatomical and functional organization of the M1 microcircuitry . Additionally , utilizing TMS as a probe , other cortical areas can be investigated in a similar manner . For the long-latency response , the focus shall be on characterizing long-range circuits activated by TMS and examining their modulatory contributions in the treatment of various neurological and psychiatric conditions . We are convinced that studying the neuronal dynamics under TMS will undoubtedly advance our understanding of the functional organization of the brain , and drive the development of non-invasive brain stimulation therapies that are more specific , effective , durable and safe than hitherto possible . To determine the amount of TMS-induced charge injection in the electrode-electrode loop ( Figure 5A ) , we used the voltage signal from the low-gain monitoring channel to calculate the current flow via the amplifier’s input capacitance Cin . As shown in Figure 5—figure supplement 2A , since the input resistance Rin and the input capacitance Cin are parallel , voltage drop across Rin ( therefore , the recorded signal Vin ) is equal to the voltage drop across Cin . Because the value of Cin is known , its current ICincan be calculated using the equationICin=CindVindt Furthermore , since Rin is in the order of teraohm , the amount of current it draws can be neglected . Therefore , ICin is equal to the total amount of induction current present in the loop ( Iind ) . It is worth noting here that by adopting this method , the exact model of microelectrodes and its associated component values are not needed for the calculation . For determining the induced charge injection in the electrode-ground loop , we used a set of input cables in which both the recording and the reference electrode were connected to the amplifier’s positive input , and the ground electrode was connected to the amplifier’s negative input . Furthermore , the negative input was shorted to the amplifier ground . Under this configuration , Iind reflected the current in the electrode-ground loop ( Figure 5—figure supplement 2B ) . At the end of our validation , we conducted these measurements in vivo , under monophasic and biphasic TMS , at 100% MSO . By integrating Iind over time , the amount of charge transfer was determined . The results ( Figure 6—figure supplement 1 ) were then compared with the charge injection values reported in the ICMS literature . To construct the electrical shield ( Figure 3B ) , we first made a polyoxymethylene ( POM ) enclosure ( 1 mm thick at the bottom face ) according to the shape of our TMS coil . An even layer of conductive coating ( GRAPHIT 33 , Kontakt Chemie , Iffezheim , Germany ) was painted on the inner side of the enclosure until the desired electrical resistance ( 10 kΩ measured along the long axes of the shield body and cover ) was reached . A layer of non-conductive transparent coating was then applied to protect the conductive layer . As the body and the top covers of the enclosure are separate , protection coating was not applied along the contacting edges between the shield body and its top covers to allow good electrical contact . In addition , an electrical cable was connected directly to the conductive layer to provide a path for grounding . All experimental procedures involving animals were approved by the Tübingen Regional Council ( license number: N1/16 ) and performed in accordance with the Animal Welfare Act of Germany . Seventeen male Sprague-Dawley rats ( Charles River Laboratories , Sulzfeld , Germany; RRID:RGD_737891 ) 11–15 weeks of age were used ( six for method evaluation and optimization; seven for the ML experiments; four for the PA experiments ) . The animals were housed in environment-enriched transparent plastic cages under inverted 12 hr light/dark cycle with free access to water and food . Upon arrival , the animals were handled 10 min per day for 5 consecutive days for stress reduction . Animals were first sedated through a brief exposure to isoflurane ( 3% at 0 . 8 L/min ) . Upon sedation , a cocktail of ketamine ( 70 mg/kg ) and xylazine ( 1 mg/kg ) was injected intraperitoneally ( i . p . ) and ophthalmic ointment was applied to eyes . A 27-gauge catheter was implanted in the lower right quadrant of the abdomen to provide i . p . access throughout the experiment . Additional doses of ketamine ( 30 mg/kg ) were administered through the catheter to maintain a constant level of anesthesia , which was assessed by breathing rate , vibrissa whisking , and the toe-pinch reflex . During the incision phase of the surgery , xylocaine gel ( 2% ) was applied to the incision site . In addition , body temperature of the animals was maintained at 37°C using a feedback-controlled heating pad throughout the experiment . Animals were restrained in a stereotaxic frame with non-conductive ear bars . A 5 × 3 mm craniotomy was made over the left sensorimotor cortex . The resulted trepanation extended from −1 mm to +4 mm to bregma and from 1 mm to 4 mm lateral to the midline . Dura matter was carefully resected and the cranial window was covered with Ringer's solution . ICMS was used to map the spatial extent of the primary forelimb motor area ( caudal forelimb area , CFA ) . A platinum-tungsten microelectrode ( 1 MΩ at 1 k Hz ) was used for ICMS at depths around 1400 μm ( from the cortical surface ) , corresponding to layer V in rat neocortex , with a train of 13 biphasic square pulses ( 200 μs per phase ) delivered at 333 Hz . A stimulation site was considered non-responsive if it was not possible to elicit any visible movement with current intensity up to 100 µA . In one animal , we also used single-pulse ICMS ( one biphasic square pulse , 300 μs per phase , 300 µA ) to study the onset latency of MUAP of the biceps brachii in response to ICMS ( Figure 7—figure supplement 2 ) . 28-gauge monopolar EMG electrodes ( Ambu A/S , Ballerup , Denmark ) were implanted in both left and right biceps brachii muscle for recording , and in the finger pads bilaterally for reference . The electrodes were connected to a high-impedance amplifier through shielded cables . The signal was low-pass ( cutoff frequency 5 kHz ) filtered online and amplified 2000 times before digital conversion . During analysis , the signal was bandpass filtered ( 100–1000 Hz ) using digital Butterworth filters implemented anti-causally in MATLAB . TMS was delivered through a Magstim D25 figure-of-eight coil ( single circle radius 25 mm; Magstim Ltd . , Carmarthenshire , UK ) powered by either a Magstim 2002 stimulator for monophasic single-pulse stimulation ( mspTMS ) or a Magstim Super Rapid Plus stimulator ( with the inline inductor Magstim 3467 ) for biphasic single-pulse and repetitive stimulation . The Magstim 2002 and D25 combination is considered as the worst-case scenario since the resulting flux transient is as high as 4T ( based on data supplied by Magstim ) , which is two to three times higher than the output seen in combinations with larger coils that are routinely used in human stimulation . The TMS coil was held by a mechanical arm and positioned over the recording site in medial-lateral orientation , generating a current flowing from the medial to the lateral part of the brain ( under monophasic stimulation ) . In the PA orientation , the induced current flows from the posterior to the anterior part of the brain . The coil , controlled by a three-dimensional microdrive , was lowered as much as possible without touching the electrode assembly . The distance from the coil surface to the head of the animal was normally 8–10 mm ( including 1 mm due to the coil shield ) . TMS was triggered digitally by a controller PC , which also digitally controlled the behavior of our EEP amplifier ( Figure 6A ) . EEP was recorded through a pair ( signal-reference ) of microelectrodes ( ca . 1 . 5 MΩ impedance at 1 kHz ) fabricated in-house from glass-coated platinum-tungsten wires ( Thomas RECORDING , Giessen , Germany ) . A thin silver wire with silver-chloride coating was used as the ground electrode . The three electrodes were arranged in a three-pronged design ( Figure 5—figure supplement 1 ) that minimized the induction loop area between them . The assembly was held by a non-conductive non-magnetic L-shape holder that was mounted on a micropositioner ( David Kopf Instruments , Tujunga , USA ) . The recording electrode was lowered , through the cranial window , into CFA as determined by ICMS . The reference electrode was also lowered into the cortex but outside the boundary of CFA . The ground electrode was positioned to be in contact with unresected subcutaneous tissue by the border of the cranial window . Signals from the electrodes were transmitted through a set of 36-gauge low-noise miniature coaxial cables ( Axon’ Cable S . A . S . , Montmirail , France; Figure 4A ) to the amplifier . The operating mode of the amplifier was controlled by the controller PC as described in the main text . The signal from the amplifier output was digitized ( USB-ME64-System , MultiChannel Systems GmbH , Reutlingen , Germany ) at 40 kHz and subsequently visualized and stored on a PC . A schematic illustration of the entire recording setup is shown in Figure 6A . Upon completion of an experiment , the recording site was marked by an electrolytic lesion ( 1 cycle of cathode leading 0 . 1 Hz biphasic square pulse with 10 µA ) generated using a microelectrode powered by a waveform generator ( STG1002 , MultiChannel Systems , Reutlingen , Germany ) . Subsequently , the animal was deeply anesthetized with sodium pentobarbital ( 200 mg/kg ) and perfused using phosphate buffer ( 0 . 1 M ) followed by paraformaldehyde ( 4% ) . Afterward , the brain of the animal was processed using standard histological procedures . The recording layer was assessed by investigating lesions in hematoxylin and eosin stained coronal sections ( Figure 7—figure supplement 1 ) . Electrophysiological data was processed in MATLAB 2014b ( The Mathworks , Natick , USA; RRID:SCR_001622 ) . Spike detection was based on amplitude threshold that was set to 3 . 5 or 4 times of the median-based background noise standard deviation estimate in order to minimize the influence of high spike rates or amplitudes in biasing spike detection ( Quiroga et al . , 2004 ) . Spike isolation was performed using principal component analysis of the spike waveforms followed by a Gaussian mixture model with Kalman filters that track waveform drifts over time ( Ecker et al . , 2014 ) . A total of 51 single units were isolated ( L5ML = 19; L5PA = 14; L2/3ML=18 ) ; however , since at the present stage we are only interested in characterizing the response of M1 neuronal population to mspTMS , in the following analysis , we combine spikes from all single units as well as those that cannot be reliably isolated into a multiunit cluster . Each trial was defined by the time interval spanning from 500 ms pre-TMS to 1000 ms post-TMS . Normalized firing rate ( FR ) was calculated by subtracting the baseline ( 500 ms period prior to TMS onset ) average FR from the instantaneous FR of each time bin ( including baseline bins ) . This normalization procedure was performed on a trial-by-trial basis . For each animal under each stimulation condition , trains of normalized FR were averaged across trials . Thresholds for significant ( p<0 . 05 ) inhibitory and excitatory events were determined by the 2 . 5 and 97 . 5 percentile of the empirical distribution of normalized FR during baseline . To facilitate the detection of significant phasic response , each averaged train of normalized FR was filtered by a Gaussian kernel ( σ = 2 ms ) . An event is considered as a significant phasic response if the normalized FR exceeds either threshold for more than 10 ms and a gap up to 10 ms is tolerated to accommodate jittering . The onset and duration information of the detected phasic response was then used to extract FR for each phase in each individual trial . Statistical analysis was performed in R ( Core Team , 2016; RRID:SCR_001905 ) . Multiple hierarchical linear mixed-effects models were constructed using the lme4 package ( Bates et al . , 2015 ) to evaluate the effect of stimulation intensity on the normalized FR for each response phase . Stimulation intensity ( normalized to %MT ) was used as the fixed effect to model trials of normalized FR of each response phase . The animal’s identity was used as the random effect ( random intercept ) to control for intraclass correlation . We also explored the possibility of trial number being another fixed effect . However , it was dropped in the final models as it did not contribute significantly to model’s fit . Statistical significance of the fixed effect in each model was evaluated against the corresponding null model using the Kenward Roger-based F-test ( Halekoh and Højsgaard , 2014 )	Being able to tap into someone’s brain activity by holding loops of wires above their head sounds a little like the stuff of science fiction . And yet this technique , known as transcranial magnetic stimulation or TMS , is used in research and to treat many brain disorders . TMS emits a pulsed magnetic field that induces tiny electrical currents in the underlying brain tissue , activating that region of the brain . But exactly how these currents affect the individual neurons and networks within activated brain regions remains unclear . The main reason for this is that we cannot use conventional electrode-based techniques to study neuronal activity during TMS because its strong electromagnetic interferences mask the signals from the electrodes . Several groups have found ways to overcome this problem . However , their methods are technically demanding and specific to one single animal model –limitations that could present an obstacle for many laboratories . Li et al . therefore set out to develop a simple and widely accessible method to study neuronal activities under TMS . The resulting method makes it possible to measure the activity of individual neurons roughly 1/1 , 000th of a second after applying TMS . To show that the technique works , Li et al . induced small movements in the forelimbs of rats by applying TMS to the brain region that controls the forelimbs , while measuring the activity of neurons at the same time . This revealed , for the first time , how the neurons responsible for the forelimb movements responded to TMS . The observed TMS-triggered neuronal activity continued long after the TMS pulse had ended . The activity also varied depending on the direction of TMS-induced currents in the brain . This new method opens up the possibility to conveniently study – in rodents or other animals – how TMS procedures that are used in patients affect neuronal activity . Li et al . hope this will make it easier to develop , study and refine these procedures , and lead to advances in TMS therapies .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "tools", "and", "resources", "neuroscience" ]	2017	Lifting the veil on the dynamics of neuronal activities evoked by transcranial magnetic stimulation
Encoding of behavioral episodes as spike sequences during hippocampal theta oscillations provides a neural substrate for computations on events extended across time and space . However , the mechanisms underlying the numerous and diverse experimentally observed properties of theta sequences remain poorly understood . Here we account for theta sequences using a novel model constrained by the septo-hippocampal circuitry . We show that when spontaneously active interneurons integrate spatial signals and theta frequency pacemaker inputs , they generate phase precessing action potentials that can coordinate theta sequences in place cell populations . We reveal novel constraints on sequence generation , predict cellular properties and neural dynamics that characterize sequence compression , identify circuit organization principles for high capacity sequential representation , and show that theta sequences can be used as substrates for association of conditioned stimuli with recent and upcoming events . Our results suggest mechanisms for flexible sequence compression that are suited to associative learning across an animal’s lifespan . Whereas many behaviorally important events take place on timescales of seconds , neuronal membrane dynamics operate at a millisecond timescale . The discovery that during movement hippocampal place cells fire action potentials with timing that precesses relative to the hippocampal theta rhythm ( O'Keefe and Recce , 1993 ) , and that time-compressed representations of behavioral sequences occur as spike sequences within each theta cycle ( Skaggs et al . , 1996; Dragoi and Buzsáki , 2006; Foster and Wilson , 2007 ) , suggests that hippocampal activity is organized so that computations on a millisecond neural timescale can address events on behavioral timescales . Thus , spike sequences within theta cycles may form a neuronal substrate for episodic and spatial memory ( Pastalkova et al . , 2008; Lisman and Redish , 2009; Buzsáki and Moser , 2013; Wikenheiser and Redish , 2015 ) . Nevertheless , the circuit mechanisms through which theta sequences are generated are unclear and the mechanisms by which they may contribute to learning have received relatively little attention . Several features of theta sequences that may be important for their computational functions pose challenges to models attempting to explain their generation through biophysically constrained mechanisms . First , the rate at which action potentials precess relative to the theta rhythm depends on an animal’s speed of movement ( Geisler et al . , 2007 ) . Second , phase precession occurs along arbitrary two-dimensional trajectories ( Huxter et al . , 2008; Climer et al . , 2013; Jeewajee et al . , 2014 ) . Third , theta sequences emerge within theta waves that propagate along the dorsoventral axis of the hippocampus ( Lubenov and Siapas , 2009; Patel et al . , 2012 ) . Fourth , while place cells across the dorsoventral axis have field sizes that vary over an order of magnitude , spike phase nevertheless usually advances by a maximum of a single theta cycle across their place field indicating that the rate of phase precession varies dorsoventrally ( Kjelstrup et al . , 2008 ) . Finally , to successfully distinguish behavioral episodes , distinct theta sequences must be generated for experiences over an animal’s lifetime , implying that sequence generation must be both flexible and have a high capacity ( Chadwick et al . , 2015 ) . We previously introduced a phenomenological model which demonstrates that experimentally observed theta sequences can be accounted for by phase precession in independent place cells ( Chadwick et al . , 2015 ) . This is in contrast to suggestions that synaptic coordination within and between cell assemblies is required to explain theta sequences ( Tsodyks et al . , 1996; Harris et al . , 2003; Harris , 2005; Maurer and McNaughton , 2007; Geisler et al . , 2010; Lisman and Redish , 2009; Wikenheiser and Redish , 2015; Wang et al . , 2015 ) . Several cellular mechanisms for independent phase precession have been proposed ( O'Keefe and Recce , 1993; Mehta et al . , 2002; Harris et al . , 2002; Lengyel et al . , 2003; Burgess et al . , 2007; Leung , 2011; Chance , 2012 ) , but none appear able to account for the challenges above while maintaining consistency with the hippocampal circuitry ( see Figure 1—source data 1 and Discussion ) . Thus , the biophysical mechanisms through which an independent phase coding scheme could be implemented within the CA1 circuitry while accounting for known computationally important properties of theta sequences are not clear . The possible mechanisms underlying phase precession in CA1 are heavily constrained by the architecture of the CA1 circuit . CA1 pyramidal cells make few direct connections with one another ( Anderson et al . , 2007 ) ( but see Yang et al . , 2014 ) , suggesting that phase precession in CA1 arises through some combination of intrinsic cellular properties , external inputs to the circuit and local interactions between pyramidal cells and interneurons . Major sources of input to CA1 include spatially modulated signals from CA3 and from the entorhinal cortex , and temporally patterned GABAergic inputs from the medial septum , which target hippocampal interneurons and act as a pacemaker to entrain theta oscillations in the circuit ( Freund and Antal , 1988 ) . Previously proposed mechanisms for independent phase precession focus on integration of signals by place cells ( O'Keefe and Recce , 1993; Mehta et al . , 2002; Harris et al . , 2002; Lengyel et al . , 2003; Burgess et al . , 2007; Leung , 2011; Chance , 2012 ) . However , many interneurons also fire spikes that precess in phase against the theta rhythm , with interneuron phase precession exhibiting strong functional coupling to individual pyramidal cells ( Maurer et al . , 2006; Geisler et al . , 2007; Ego-Stengel and Wilson , 2007 ) . Thus , we asked whether phase precession underlying sequence generation could originate from interneuron dynamics . To address this possibility we introduce a minimal circuit model in which phase precession and theta sequences are generated through interactions between place cells and interneurons driven by pacemaker inputs . In contrast to the view that phase precession in interneurons is inherited synaptically from phase precessing place cell assemblies ( Maurer et al . , 2006; Geisler et al . , 2007 ) , interneuron phase precession in our model is crucial for the coordination of spike timing in place cells and for the generation of theta sequences . Due to the transient functional coupling between place cells and interneurons , phase precession occurs dynamically whenever a place cell is driven by external inputs . Consequently , phase precession and theta sequences are generated de novo within the network , and slow input sequences are automatically compressed into theta sequences in networks of interacting pyramidal cells and interneurons . Our model suggests that CA1 can function as a flexible compressor of its inputs to maintain a representation of temporal order occurring on a behavioral timescale within a faster timescale suitable for synaptic processing in downstream brain areas . The model enables predictions of pacemaker dynamics which account for velocity-dependence of network activity and dorsoventral organization of sequence generation , and predicts network configurations that may underlie the dissociation of phase precession and theta sequences ( Feng et al . , 2015; Middleton and McHugh , 2016 ) . The proposed mechanism not only generates sequences encoding spatial trajectories , but can also function as a general purpose circuit with a remarkably high capacity for encoding temporally extended sequences of events . We show how such a compression of ongoing experience into theta cycles enables flexible learning of behavioral associations through spike timing dependent plasticity ( STDP ) . Thus , CA1 may compress ongoing experiences during theta states into fast neural activity patterns suitable for online learning and decision making . Since theta phase precession in independent neurons is sufficient to account for experimentally observed theta sequences ( Chadwick et al . , 2015 ) , we first aimed to identify circuit mechanisms that account for experimentally observed features of phase precession in single neurons . Whereas in many previous models precession is assumed to arise from oscillatory drive targeting place cells , the frequency of theta is established by septal GABAergic projections to hippocampal interneurons ( Freund and Antal , 1988 ) , which in turn coordinate the spiking activity of local CA1 pyramidal cells ( Royer et al . , 2012 ) . We therefore reasoned that phase precession could emerge from the dynamics of interneurons driven by pacemaker inputs and interacting with pyramidal cells . To explore this possibility we constructed a minimal network model containing a single interneuron and pyramidal cell , with synaptic connectivity based on this architecture ( Figure 1A ) . The interneuron is driven to fire tonically by a constant depolarizing current , while pacemaker drive from the medial septum to the interneuron is simulated by an 8 Hz oscillatory current , which is sufficient to fully entrain spiking activity of the interneuron when the pyramidal cell is inactive ( Figure 1B–C ) . In this case , output from the interneuron drives rhythmic subthreshold theta frequency inhibitory synaptic potentials in the pyramidal cell ( Figure 1B ) . When spatial input to the pyramidal cell is simulated by a suprathreshold external drive , the resulting synaptic drive to the interneuron initiates phase precession in the coupled pair of cells , causing their firing frequency to elevate above that of the pacemaker theta and their firing phase to advance continuously over the place field ( Figure 1D–E ) . When the pyramidal cell is transiently driven by slow depolarizing current , the phase of spikes fired by the interneuron and by the pyramidal cell advances through a full 360 degrees relative to the 8 Hz pacemaker input . Hence , whenever pyramidal cells are activated by slow depolarizing drives , the basic architecture of the CA1 circuit , along with pacing inputs from the medial septum , is sufficient to generate phase precession in pyramidal cells and interneurons . 10 . 7554/eLife . 20349 . 003Figure 1 . A minimal CA1 circuit model for theta phase precession . ( A ) An interneuron ( red ) is driven by a pacemaker theta oscillation from the medial septum . The interneuron synapses reciprocally onto a pyramidal cell ( blue ) . The pyramidal cell is driven by slower external inputs occurring over behavioral timescales . ( B–E ) A simulation of this network as the animal crosses the place field of the pyramidal cell . ( B ) Interneuron spiking activity ( red lines ) and pyramidal cell spikes ( blue lines ) and membrane potential ( blue trace ) . ( C ) A sample of the interneuron spike train when the pyramidal cell is inactive ( i . e . , outside of the place field ) , with the pacemaker rhythm overlaid for reference . In this case , the interneuron locks to the pacemaker input . ( D ) A sample of the interneuron and pyramidal cell spike trains inside the place field . In this case , the interneuron precesses in phase against the pacemaker input and the pyramidal cell fires in bursts which also precess in phase . ( E ) The membrane frequency in the theta band and the spike phases of the interneuron ( red ) and pyramidal cell ( blue ) corresponding to the data shown in parts ( A ) - ( D ) . Phases are replicated over two cycles for clarity . Note that the pyramidal cell fires up to two spikes per theta cycle in this simulation . DOI: http://dx . doi . org/10 . 7554/eLife . 20349 . 00310 . 7554/eLife . 20349 . 004Figure 1—source data 1 . Table comparing the proposed model to previous models of phase precession . Our model is the first to successfully explain speed-modulation of precession frequency , two-dimensional phase precession and 360 degrees of phase precession without introducing unobserved circuit components , directionally modulated external inputs or inputs with speed-modulated oscillation frequencies . DOI: http://dx . doi . org/10 . 7554/eLife . 20349 . 004 To better understand the emergence of these phase precession dynamics , we developed a reduced model in which an interneuron is driven by weak pacemaker input and a slow depolarizing drive , and for which analytical solutions can be obtained ( Figure 2A , see Materials and methods for details of model ) . During injection of a constant input current the model generates either stable phase locking or phase precession at a constant rate against the pacemaker drive ( ‘frequency pulling’ ) depending on the strength of depolarizing drive relative to the strength of pacemaker input ( Figure 2B , C ) . Phase locking occurs for weak drives , where the interneuron becomes entrained to a fixed phase of the pacemaker input . Frequency pulling occurs for strong drives , in which case , for a given constant input current , the interneuron oscillates with a fixed frequency difference from the pacemaker , causing its phase to advance continuously relative to the pacemaker input . In the phase locking region , because the locking phase varies as a function of the input current , variation in the input current can be used to achieve variation through a maximum of 180 degrees of theta phases ( −90 to 90 degrees , see Figure 2B ) . In contrast , in the frequency pulling region , phase precession continues indefinitely at a fixed rate for a constant input current . 10 . 7554/eLife . 20349 . 005Figure 2 . Phase precession and phase locking in a reduced model of an interneuron driven by depolarizing current and weak pacemaker drive . ( A ) Schematic of the model . ( B–C ) Steady state dynamics for a constant depolarizing drive , assuming a linear f-I curve . ( B ) Phase locking as a function of input current . ( C ) Precession frequency as a function of input current . For sufficiently strong currents , the interneuron oscillates with a frequency above that of the pacemaker ( phase precession ) . For sufficiently weak currents , the interneuron oscillates more slowly than the pacemaker ( phase regression ) . Note that for more biophysical f-I curves the phase regression regime may be absent . ( D–E ) Evolution of interneuron phase during a transient , slowly varying current injection . ( D ) Input currents with Gaussian profiles and a range of amplitudes . ( E ) Interneuron phase as a function of time , for each current profile shown in ( D ) , showing a total of one cycle of phase precession for stronger drives and only transient phase precession before reversing in phase for weaker drives ( purple ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20349 . 00510 . 7554/eLife . 20349 . 006Figure 2—figure supplement 1 . Phase precession is robust to the temporal profile of excitatory drive . Square pulse excitatory currents ( A ) , upward-going asymmetric ramps ( C ) and downward-going asymmetric ramps ( E ) are each able to drive 360 degrees of phase precession ( B , D and E ) . Continuous phase precession through the firing field requires sufficiently strong excitatory drive , but is independent of the shape of the drive . This is in contrast to models that integrate an asymmetric input with a dendritic oscillation . DOI: http://dx . doi . org/10 . 7554/eLife . 20349 . 006 Previous models that focus on place cells also generate phase precession by integration of theta drives and slow depolarizing drives , achieving variable phase locking through a range of 180 degrees ( Mehta et al . , 2002; Harris et al . , 2002 ) . However , the underlying dynamics and mechanisms in our model are distinct from these schemes in several ways . First , the excitation-dependent locking of spike phase observed in our model for weak depolarizing drives ( the phase locking regime ) is the result of the active entrainment of an ongoing intrinsic cellular rhythm to a pacemaker drive , rather than a passive summation and thresholding of inputs to a silent pyramidal cell as in previous models . Moreover , the frequency pulling regime in our model , in which the external drive determines the rate of phase precession ( Figure 2C ) , is not generated by previous models . The dynamics inside the place field in our model take place almost entirely within this novel frequency pulling regime , with the phase locking regime instead governing the dynamics outside of the place field and therefore the alignment of spike phase at place field entry . Because our model relies on the frequency pulling rather than the phase locking regime to produce phase precession , continuous phase precession can be generated for arbitrary input profiles of sufficient strength , and does not require a monotonically increasing ramp input as in previous models ( Figure 2—figure supplement 1 ) . Second , for symmetrical place fields previous schemes predict a phase advance towards the center of a place field , but a phase reversal as the input current is reduced on leaving the place field ( Melamed et al . , 2004 ) . In contrast , when input currents are sufficient to drive the neuron into the frequency pulling domain in our model , then phase advances continuously throughout the input field ( Figure 2D , E ) . Provided that inputs are sufficiently strong and sustained , the phase of interneuron firing advances through a full 360 degrees , with the rate of phase precession determined by the strength of the injected current ( Figure 2D , E ) . Hence , this reduced model explains the dynamics observed in the network simulation of Figure 1 . Specifically , the interneuron remains in a stable phase locking regime while the pyramidal cell is inactive , but enters the frequency pulling regime whenever the pyramidal cell provides sufficient synaptic input , producing phase precession . Phase precessing synaptic inputs from the interneuron coordinate the spike timing of the pyramidal cell and confer phase precession , but phase precession in the interneuron is relatively insensitive to the timing of pyramidal cell inputs , instead requiring only a sufficient increase in excitatory drive . Experimentally the rate of phase precession in both place cells and interneurons increases with running speed , so that a constant relationship is maintained between spike phase and location ( Geisler et al . , 2007 ) . Because phase precession in our reduced model depends on pacemaker amplitude and excitatory drive , the precession frequency can be flexibly modulated by varying either parameter without needing to adjust the frequency of the pacemaker oscillation ( Figure 2C , Materials and methods ) . We therefore used the minimal circuit model of Figure 1 to test whether variation of these inputs to the interneuron can account for the experimentally observed speed-dependence of phase precession in pyramidal cells and interneurons . The reduced model predicts that either a decrease in pacemaker amplitude or an increase in depolarizing drive to interneurons with running speed would generate an increase in the rate of phase precession with running speed . However , for stability the pacemaker amplitude must be small for low running speeds ( see Materials and Methods ) . In this case the precession frequency can nevertheless be controlled independently through changes in the depolarizing drive with running speed . Indeed , we found that in the minimal circuit model a linear increase in pacemaker amplitude with running speed , combined with a linear increase in depolarizing current to interneurons with running speed , can generate an approximately linear increase in precession frequency while maintaining stable precession dynamics across running speeds ( Figure 3 ) . Hence , the dynamics required to maintain a fixed relationship between spike phase and place field position can be generated de novo in the local circuitry with inputs at a fixed theta frequency . Importantly , the predicted dependence on running speed of current input to the interneuron is consistent with findings of a velocity-dependent depolarizing current from glutamatergic circuits in the medial septum to interneurons in CA1 ( Fuhrmann et al . , 2015 ) . Similarly , the predicted dependence of the pacemaker amplitude on running speed is consistent with the dependence on running speed of both the LFP theta amplitude in CA1 ( McFarland et al . , 1975; Maurer et al . , 2005; Patel et al . , 2012 ) and the activity of inhibitory circuitry in the medial septum ( King et al . , 1998 ) . 10 . 7554/eLife . 20349 . 007Figure 3 . Running speed dependence of phase precession . ( A ) Illustration of the model circuit . To account for running speed dependence , pacemaker amplitude and depolarizing current amplitude are increased linearly with running speed . ( B ) Examples of phase precession at a slow and fast running speed , where the pacemaker amplitude and depolarizing current to interneurons are varied . ( C ) Phase precession frequency as a function of running speed . Individual dots illustrate the estimated precession frequency on a single lap . DOI: http://dx . doi . org/10 . 7554/eLife . 20349 . 007 The phase of theta activity varies systematically across the dorsoventral axis of the hippocampus ( Lubenov and Siapas , 2009 ) , spanning a range of 180 degrees ( Patel et al . , 2012 ) and creating the appearance of a dorsoventral traveling wave . This variation is difficult to account for by temporal delays in a common pacemaker drive , which has led to the suggestion that entorhinal-hippocampal or intrahippocampal interactions are required to account for dorsoventral phase offsets ( Patel et al . , 2012; Long et al . , 2015 ) . We asked if the present model can account for these observations without the necessity for additional circuit components . In the reduced interneuron model , the range of phases that stably lock to pacemaker input is precisely 180 degrees , with locking phase depending on the strength of the excitatory current and pacemaker amplitude ( Figure 2B ) . All other spike phases are unstable . This suggests that a gradient in excitatory inputs to interneurons ( or alternatively a gradient in input resistance or some intrinsic membrane current ) along the dorsoventral axis might be sufficient to generate the observed dorsoventral phase gradient , despite a coherent pacemaker input . To test this hypothesis using more biologically plausible neuronal dynamics we simulated integrate and fire interneurons driven by the same pacemaker inputs , but different levels of depolarizing input currents ( Figure 4A–C ) . This is equivalent to the full circuit model while the animal is outside of the place field and therefore the pyramidal cell is inactive . Figure 4B shows three examples of these simulations . In each case , the interneuron is attracted towards a stable locking phase of the pacemaker input , but the precise locking phase depends on the strength of depolarizing current . In Figure 4C we systematically analyzed how this locking phase depends on the strength of depolarizing current , finding a relationship remarkably similar to that predicted by the reduced model , including a range of 180 degrees of locking phases . Hence , in addition to explaining the change in precession frequency with running speed , the interplay between excitatory currents and pacemaker inputs can explain the phase gradient across the dorsoventral axis of the hippocampus , allowing the emergence of traveling theta waves based on variable locking to a single , common pacemaker input . 10 . 7554/eLife . 20349 . 008Figure 4 . Theta dynamics across the dorsoventral axis . ( A ) Inputs to interneurons across the dorsoventral axis hypothesized to produce a gradient in theta phase . ( B ) Interneuron spike phases for three simulations with different depolarizing currents . ( C ) Interneuron locking phase vs depolarizing current ( cf . Figure 2B ) . ( D ) A circuit model , and its dependence on dorsoventral location , which could produce simultaneous traveling theta waves and gradients in precession slope . ( E ) Phase precession in a ventral place cell/interneuron pair ( place field size 10 meters ) . ( F ) Phase precession in a dorsal place cell/interneuron pair ( place field size 0 . 3 meters ) . Note the change in both locking phase and precession slope from dorsal to ventral . DOI: http://dx . doi . org/10 . 7554/eLife . 20349 . 008 Place field size also varies along the dorsoventral axis of the hippocampus , ranging from less than one meter dorsally to approximately 10 meters ventrally ( Kjelstrup et al . , 2008 ) . This gradient is associated with a concomitant gradient in the slope of phase precession , such that phase precesses through approximately one cycle both dorsally and ventrally ( Kjelstrup et al . , 2008 ) . To test whether our minimal circuit model could account for these observations in addition to the traveling wave dynamics , we simulated place cell/interneuron pairs at the ventral and dorsal pole of CA1 , with place field sizes of approximately 10 meters and 0 . 3 meters respectively , and interneuron locking phases separated by approximately 180 degrees ( Figure 4D–F ) . We found that the gradient in both phase precession and theta phase along the dorsoventral axis could be accounted for simultaneously by a combination of a dorsoventral gradient in the amplitude of pacemaker drive , the depolarizing current to the interneuron and the strength of excitatory synaptic connections ( Figure 4E , F ) . Thus , our proposed mechanism predicts that depolarizing current input to interneurons ( or their excitability ) , the strength of excitatory synaptic connections from pyramidal cells to interneurons and the amplitude of the septal pacemaker drive all decrease from the dorsal pole to the ventral pole of the hippocampus ( Figure 4D ) . In line with these predictions , theta power is observed to decrease from dorsal to ventral hippocampus ( Royer et al . , 2010 ) . As a further test of the model we asked if in addition to accounting for phase precession on linear tracks , it can account for the properties of phase precession in open environments . In open environments , spikes always precess from late to early phases of theta , regardless of running direction ( Huxter et al . , 2008; Climer et al . , 2013; Jeewajee et al . , 2014 ) . These dynamics arise naturally from the depolarizing current envelope in the present model if the animal passes in a straight line through the center of a place field at a constant speed ( Figure 1 ) . No additional inputs such as from head direction cells are required . Experimentally , a more complex feature of phase precession in open environments is observed on passes through the edge of the place field , in which case the firing phase advances through around 180 degrees before reversing through 180 degrees over the second half of the field ( Supplementary Figure S2b in Huxter et al . , 2008 ) . In the present model , similar dynamics occur when the interneuron is not driven sufficiently strongly to pass through to the next cycle and is instead attracted back towards the initial phase ( Figure 2D–E and Figure 5A ) . Our model is also consistent with sequences observed during backwards travel , in which theta sequences reflect the ordering at which locations are visited rather than heading direction ( Cei et al . , 2014; Maurer et al . , 2014 ) . 10 . 7554/eLife . 20349 . 009Figure 5 . Robustness of phase precession to changes in the strength of place field drive . ( A ) Failure to precess through one full cycle . In this case , the external inputs were not strong enough to drive the interneuron past the threshold to be pulled into the next theta cycle , and instead it is pulled back towards the phase it started at . This is also seen in an initial increase followed by a decrease in frequency as the cell advances before reversing in phase against the pacemaker . ( B ) Precession through two full cycles . In this simulation , the amplitude of the slow envelope current was increased . This results in an increased firing rate of the pyramidal cell and hence an increased excitatory input to the interneuron . As a result , the interneuron received enough drive to pass through two cycles of pacemaker input . ( C ) The probability of an interneuron precessing through one , two , or three cycles of pacemaker theta phase as a function of the amplitude of the depolarizing envelope current onto the place cell . ( D ) The number of spikes fired by the place cell ( with standard deviation shown as error bars ) as a function of the amplitude of depolarizing envelope current . ( E ) The probability of the interneuron precessing through one , two , or three cycles of pacemaker theta phase replotted as a function of the number of spikes fired by the place cell . DOI: http://dx . doi . org/10 . 7554/eLife . 20349 . 00910 . 7554/eLife . 20349 . 010Figure 5—figure supplement 1 . Phase precession is robust to transient intrahippocampal perturbation . Zugaro et al . ( 2005 ) demonstrated that spike phase precession persists after transient inactivation of the hippocampus . To address whether our model can account for these observations , we simulated transient silencing of interneurons and pyramidal cells and simultaneous reset of the external pacemaker . Three representative example simulations are shown . In each case , phase precession resumes following the transient perturbation ( cf . Figure 3of Zugaro et al . , 2005 ) . Left column: The theta phase ( black trace ) , interneuron membrane potential and spikes ( red ) and pyramidal cell membrane potential and spikes ( blue ) . Right column: The interneuron and pyramidal cell spike phases relative to the pacemaker theta rhythm . Black bars show periods of silencing . DOI: http://dx . doi . org/10 . 7554/eLife . 20349 . 01010 . 7554/eLife . 20349 . 011Figure 5—figure supplement 2 . Perturbation of spike phase during interneuron silencing . ( Royer et al . , 2012 ) showed that phase precession by pyramidal cells is maintained following transient silencing of PV neurons , while the phase can appear to shift . To examine whether our model can account for these observations we simulated transient inhibition of interneurons . ( A–B ) When interneurons were silenced for 1 s , centered on the place cell’s firing field , phase precession was maintained and on average spike phase appears to advance . ( A ) Average spike phase ( ±circular standard error ) across five bins spanning the length of the place field , for the control simulation and for simulations in which interneurons are silenced ( cf . Figure 7c in Royer et al . , 2012 ) . ( B ) The mean shift in spike phase in each bin ( cf . Figure 7b in Royer et al . , 2012 ) . The relatively minor effects of interneuron silencing in our simulations is a result of the phase locking of cells outside of the place field ( before interneuron silencing begins ) , which ensures that pyramidal cells begin spiking at the correct phase upon place field entry . Despite the lack of any theta coordination via interneuron input inside the place field , their tonic spiking over the place field combined with the correct phase alignment at place field entry is sufficient to generate results similar to those of Royer et al . in the averaged data . ( C–D ) Simulations as for ( A–B ) , but with silencing of interneurons centered on a random location within 20 cm of the place field center , which may better approximate the conditions in Royer et al . ( 2012 ) . In these simulations phase precession is again maintained , while the phase change is reduced . Thus , when optogenetic silencing only covers part of the place field , interneuron inputs in the unsilenced portion of the place field further reduce the amount of disruption in the averaged data . DOI: http://dx . doi . org/10 . 7554/eLife . 20349 . 011 The phase advance and then reversal on passing through the edge of a place field results from failure of the weak synaptic depolarization to drive the model into the frequency pulling domain indicated in Figure 2 . What happens when the synaptic drive is instead very large ? We found that with strong and sustained inputs to place cells precession continues over multiple theta cycles ( Figure 5B ) . However , the pacemaker drive to the interneuron confers robustness against this effect , as the interneuron can only precess through a discrete number of theta cycles and requires considerable additional input to precess through two cycles of theta rather than one . Figure 5C shows how the number of theta cycles precessed by the interneuron varies with the amplitude of the slow envelope . Over a broad range of input currents ( Figure 5C ) , or more directly , a broad range of pyramidal cell spike counts ( Figure 5D , E ) , the interneuron will precess exactly one cycle over the place field . For the choice of parameters used here , robust phase precession through one cycle in the interneuron occurs provided the place cell fires between 10 and 25 spikes in its place field . Thus , phase precession is sufficiently robust to allow considerable rate remapping , but the mechanism nevertheless places constraints on the coexistence of phase precession and rate remapping within place cells ( Allen et al . , 2012 ) . Theta phase precession has also been shown to exhibit considerable robustness against experimentally induced circuit perturbations . For example , when CA1 is transiently silenced ( for ∼200 ms ) and the theta rhythm is simultaneously reset , phase precession resumes in CA1 unperturbed upon recovery ( Zugaro et al . , 2005 ) . We tested whether the model would exhibit similar robustness under such a perturbation by injecting a negative current into both place cells and interneurons to induce silencing while simultaneously resetting the phase of the external pacemaker drive . Indeed , we found that phase precession resumes upon recovery from this perturbation , as observed experimentally ( Figure 5—figure supplement 1 ) . Further robustness has been observed under optogenetic perturbations of the CA1 circuitry . Specifically , transient ( 1 s ) silencing of somatostatin-positive ( SOM ) interneurons has almost no effect on spike phase , altering mainly the burst firing of place cells , while silencing of parvalbumin-positive ( PV ) interneurons appears to introduce a small shift in average spike phase without compromising phase precession overall ( Royer et al . , 2012 ) . We replicated this experimental protocol by injecting a 1 s negative current pulse into the interneuron as the animal crossed the place field . When analyzing the resulting data using the methods of Royer and colleagues , we found a shift in spike phase of a similar magnitude and direction to that reported in experimental data ( Figure 5—figure supplement 2A , B ) . These findings can be explained as follows . The interneuron coordinates the pyramidal cell’s theta activity until place field entry so that pyramidal cell spike phase is correctly aligned at the start of the place field . Upon interneuron silencing , the pyramidal cell’s activity becomes independent of the theta rhythm , and depends only on the slow depolarizing drive . Nevertheless , because the pyramidal cell continues to spike tonically at a frequency higher than the theta rhythm during interneuron silencing , its spikes shift in phase continuously ( i . e . , precess ) against the theta rhythm over the place field . This precession within the place field , combined with the phase alignment at place field entry provided by the interneuron before the onset of optogenetic silencing , generates the apparent phase shift of Royer et al . ( 2012 ) in the trial-averaged data . In contrast to this transient manipulation , we expect that phase precession would be severely disrupted in experiments where phase precessing interneurons are silenced over an entire lap , so that the phase at place field entry is not correctly aligned . In summary , the model we outline here provides a robust mechanism for phase precession consistent with the circuitry in CA1 . The model accounts for the key features of phase precession observed in CA1 , including the dependence on running speed , place field size and dorsoventral location , phase precession along two-dimensional trajectories , the coupling of phase precession between place cells and interneurons , dorsoventral traveling theta waves and robustness to circuit perturbations . While the model that we propose in Figure 1 generates phase precession using only an isolated place cell and interneuron , CA1 place cells are embedded into much larger networks in which only 7–11% of neurons are interneurons ( Woodson et al . , 1989; Aika et al . , 1994; Bezaire and Soltesz , 2013 ) . The large disparity between the number of place cells and interneurons demands that a single interneuron in the model must couple to multiple pyramidal cells and generate phase precession in each one . To test if this is possible , we first simulated a single interneuron coupled synaptically to two pyramidal cells . We find that , when each pyramidal cell receives a depolarizing drive at a different time , the interneuron can be recruited for phase precession independently by each pyramidal cell ( Figure 6 ) . In this case , the interneuron shows two phase precession fields . As a direct consequence , and in contrast to the case in which there is just one active pyramidal cell per interneuron , the model predicts that outside of their suprathreshold firing fields place cells have subthreshold phase precession fields , characterized by transient increases in the frequency of their theta-modulated inhibitory input when the other place cell is active ( Figure 6B ) . However , if the pyramidal cells have overlapping place fields , these dynamics may be disrupted . In this case , the stronger synaptic input to the interneuron from two active place cells may increase its precession frequency ( Figure 2C ) , causing it to precess over multiple cycles ( Figure 5B ) . Moreover , as synaptic output from the same interneuron coordinates the theta activity of both place cells , their spiking may become synchronized . Thus , a single interneuron can support phase precession by more than one place cell , but overlap between the firing fields of place cells coupled to the same interneuron may disrupt precession-based codes by shifting the phase of coding relative to position and by causing multiple cycles of phase precession within a single firing field . 10 . 7554/eLife . 20349 . 012Figure 6 . Recruitment of an interneuron for phase precession by multiple pyramidal cells . ( A ) Circuit diagram showing two pyramidal cells connected to the same interneuron and receiving slow envelope currents at different points in time . ( B ) Simulation of this circuit showing the intrinsic theta frequency , spike phases and membrane potentials . When the blue cell recruits the interneuron for phase precession as the animal crosses its place field , this is also reflected in phase precession of the membrane potential oscillation of the green cell while the animal is outside of its firing field ( and vice versa ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20349 . 012 Given this potential sensitivity of the circuit to overlap between place fields of cells connected to the same interneuron , it is unclear whether our proposed mechanism can be extended to large networks with realistic numbers of interneurons and pyramidal cells . To address this we quantify the performance of larger networks while varying the density of active place cells per interneuron , and the spatial arrangement of the firing fields of place cells connected to the same interneuron ( Figure 7A–B ) . In these networks each pyramidal cell couples to only one interneuron , and these connections are bidirectional , so that each interneuron couples to multiple pyramidal cells ( see Materials and methods ) . This can be viewed as a simplified description of the interactions underlying phase precession , with other circuit interactions removed . These larger networks successfully compress slow input sequences into fast theta sequences when place field maps are sparse with low overlap ( Figure 7A top ) . In contrast , such sequences do not emerge when place field maps are dense and have high overlap ( Figure 7A bottom ) . 10 . 7554/eLife . 20349 . 013Figure 7 . Compression of slow input sequences in CA1 networks . ( A ) Network simulations at low and high mapping densities . For sparse , random place field maps , input sequences are compressed into robust theta sequences . For dense , random place field maps , no such sequence compression is observed . ( B ) Top: Examples of optimal maps given two different place field densities . A set of place cells attached to the same interneuron are mapped onto a linear track . In an optimal map , their place field centers are organized such that their overlap is minimized . For a certain number of place cells per interneuron ( here , four ) overlap occurs even for an optimal map . Bottom: Example of random maps . The location of each place field on the track is drawn from a uniform probability distribution . In this case , a larger number of place cells per interneuron causes an increase in the probability that place fields will overlap . ( C ) Network performance vs number of active place cells per interneuron . As more place cells become active ( or the number of interneurons is decreased ) , the compression of inputs into theta sequences is degraded ( red and blue traces ) . This is caused by a drop in the coherence of phase precession in the population , despite a relatively constant phase-position correlation in individual place cells on single laps ( blue trace vs gray trace ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20349 . 01310 . 7554/eLife . 20349 . 014Figure 7—figure supplement 1 . Three examples of random place field maps with a density of one active place cell per interneuron per meter , for which disruption of sequence compression occurs ( Figure 7C ) . A single interneuron is shown , as well as the active place cells which couple to that interneuron . Top panels: Place field inputs to the five pyramidal cells active on the track and coupled to the single interneuron . Second panels: Interneuron theta frequency over time for a single lap as the animal moves along the track . Third panels: Interneuron spike phases over time . Note that the interneuron precesses over multiple cycles when it receives sufficient pyramidal cell input , but that place field overlap does not always lead to multiple cycles of phase precession . Fourth panels: Interneuron spiking ( red ticks ) , pyramidal cell spiking ( colored ticks ) and pyramidal cell membrane potential ( colored traces ) over time . Fifth panels: Pyramidal cell membrane potential theta frequencies over time . Bottom Panels: Pyramidal cell spike phases over time . Note that pyramidal cell spiking is highly synchronous within a theta cycle when multiple pyramidal cells are coactive . This synchrony , as well as the multiple cycles of interneuron phase precession , underlies the disruption of sequence compression observed in Figure 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 20349 . 01410 . 7554/eLife . 20349 . 015Figure 7—figure supplement 2 . Robustness of phase precession under extraneous noise . ( A−B ) Examples of phase precession in simulations with low ( A ) and moderate ( B ) levels of noise injected into the pyramidal cell . ( C−D ) Dependence of phase precession and population sequence compression on the level of injected pyramidal cell ( C ) and interneuron ( D ) noise ( for an explanation of these metrics , see Figure 7 and its description in both the main text and the Materials and methods section ) . Note that , for a given amplitude in mV , the noise level also depends on the simulation timestep and membrane time constant ( see Equation 14 ) . Note also that , although interneurons and pyramidal cells appear to differ in their sensitivity to noise in these simulations , this is consistent with the amplitudes of other currents injected into the two cell types in the model , which are an order of magnitude smaller for interneurons than pyramidal cells ( e . g . , the pacemaker and velocity-dependent current to interneurons vs the place field drive and velocity-dependent current to pyramidal cells ) . As the overall scale of inputs depend strongly on the neuron model and biophysical parameters such as membrane resistance , they cannot be interpreted quantitatively in the present model due to a lack of biophysical detail . DOI: http://dx . doi . org/10 . 7554/eLife . 20349 . 01510 . 7554/eLife . 20349 . 016Figure 7—figure supplement 3 . Putative mechanisms for removing disruption from network theta sequences . ( A ) In one possible mechanism , synaptic weights between pyramidal cells and interneurons are altered so that pyramidal cell pairs with overlapping place fields no longer functionally couple to the same interneuron . ( B ) In a second mechanism , the place fields themselves undergo changes to remove overlap for pyramidal cells coupled to the same interneuron . DOI: http://dx . doi . org/10 . 7554/eLife . 20349 . 01610 . 7554/eLife . 20349 . 017Figure 7—figure supplement 4 . Distributions of single-cell phase precession strengths in random maps with varying degrees of disruptive place field overlap . Phase-position correlation is measured for each cell using its pooled spiking activity over 30 laps . Each map consisted of 200 pyramidal cells with place fields randomly organized under a uniform distribution over a five meter track . When there is just one active pyramidal cell per interneuron on the track ( top panel ) , the distribution of phase-position correlations simply reflects limited sample size effects , so that increasing the number of laps would yield a narrower distribution . As the number of active pyramidal cells per interneuron is increased , the mean phase-position correlation decreases due to disruptive place field overlap in the network . Nevertheless , strong phase precession continues to be generated in a fraction of pyramidal cells even for very dense maps ( bottom panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20349 . 017 To more systematically quantify factors affecting sequence compression within the network , we introduced two distinct metrics which measure the extent to which spiking within theta cycles faithfully recapitulates the slow sequence of place field inputs . We call these the single-cycle theta sequence metric and the population phase precession metric ( see Materials and methods for details ) . The single-cycle theta sequence metric measures the similarity between slow input sequences and individual theta sequences ( Figure 7C , solid red line ) . The population phase precession metric measures the robustness and coherence of phase precession in a population of cells , and therefore serves as an averaged measure of sequence compression over a dataset ( Figure 7C , solid blue line ) . For these metrics , correlations close to zero imply a lack of sequential organization and therefore poor performance , while strong ( positive or negative ) correlations signify the presence of sequential representations within theta cycles . Using these metrics , we tested how sequence compression varies depending on the properties of the place field map within the network . We observed decreases in the strength of both individual theta sequences and population phase precession with increasing place field density on the track ( Figure 7C ) . Strikingly , this quantitative analysis also revealed that with random place field mappings network performance degrades continuously with increasing place field density , whereas with optimal place field mappings designed to minimize overlap in the place fields of cells coupled to the same interneuron ( Figure 7B ) , high performance is maintained over a wider range of place field densities ( Figure 7C , dashed red and blue lines ) . In summary , both the number of active place cells in an environment and the spatial organization of their place fields influence the quality of sequence compression . In general , network performance is high when the spatial maps are sparse , but high levels of performance can also be maintained in denser spatial maps provided that the place fields of cells coupled to the same interneuron are well separated . Recent evidence suggests that theta sequences and phase precession on single laps may be dissociated under some circumstances , such that single-cell precession can occur without spatially ordered theta sequences . For example , on the first lap of a novel linear track place cells exhibit phase precession , in that their spikes advance continuously in phase against the theta rhythm , but the phase lags between cells are initially uncoordinated and do not generate population theta sequences until after further experience ( Feng et al . , 2015 ) . Similarly , when input from CA3 is permanently absent , robust phase precession is observed in each cell while spatially organized theta sequences fail to emerge ( Middleton and McHugh , 2016 ) . To test whether such a dissociation of phase precession and theta sequences is consistent with our model , we asked whether the changes in sequence compression observed in the simulations of Figure 7 are caused by changes in the robustness of phase precession in individual place cells , or whether they result from changes in the timing relationships between groups of place cells ( i . e . , a decoherence of phase precession in neuronal populations ) . When we quantified the fidelity of phase precession for individual cells on single laps ( see Materials and methods ) we found that robust single-unit phase precession persists as place field overlap is increased ( Figure 7C , gray line ) , despite the disruption of single-cycle theta sequences and population phase precession . Thus , sequence disruption is caused by a decoherence of phase precession within the population . This decoherence is caused by indirect interactions amongst place cells with overlapping place fields and shared interneurons ( e . g . see Figure 7—figure supplement 1 ) . Given this network configuration-dependent disruption of population activity in our model , we wondered if extraneous noise impacts phase precession and population sequences , and whether interneuron and pyramidal cell noise have similar or dissociable effects on circuit function . We found that with increasing amplitude noise injected into pyramidal cells , single-cell phase precession and population sequences were impaired in parallel ( Figure 7—figure supplement 2A–C ) . This is distinct from increasing place field overlap , which disrupts population sequences but not single-cell phase precession . In contrast , increasing noise injected into interneurons disrupts population sequences while leaving phase precession intact at the single trial level , revealing an additional mechanism for dissociating phase precession from population sequences ( Figure 7—figure supplement 2D ) . These results underscore the distinct roles of interneurons and pyramidal cells for generating phase precession and population sequences in the model . We next sought to establish how the experience dependent reorganization of network activity observed by Feng and colleagues might occur . Our model suggests two potential mechanisms . First , plasticity between place cells and interneurons could adjust synaptic weights such that place cells with overlapping place fields no longer couple strongly to the same interneurons ( Figure 7—figure supplement 3A ) . Second , the slow envelope inputs to place cells could rapidly reorganize in order to minimize the overlap of place fields of cells coupled to the same interneurons ( Figure 7—figure supplement 3B ) . If a plasticity mechanism were in place , synaptic changes which allow sequential activity in a new environment would cause disruption in previously stored maps . In contrast , place field reorganization could enable multiple stable maps to be formed without disruption or interference between different representations . Experimental evidence suggests that place field activity is indeed reorganized upon exposure to a novel environment , including a sparsification of the CA1 place code and a decrease in the number of active place cells ( Frank et al . , 2004; Karlsson and Frank , 2008 ) . Whether such a reorganization mediates removal of unwanted place field overlap as we predict here is yet to be determined . Further evidence suggests that such a mechanism may depend crucially on plasticity in CA3 to CA1 connections ( Dragoi and Tonegawa , 2013 ) . Hence , permanent silencing of CA3 would be expected to disrupt CA1 theta sequences without affecting phase precession in our model , as observed by Middleton and McHugh ( 2016 ) . What constraints does this sensitivity of sequence generation to connectivity impose on spatial mapping ? Intuitively , as more place cells are connected to a given interneuron , the fraction of place cells that can be active in a given environment without interfering with sequence generation becomes smaller . This intuition can be formalized by adopting a simplified model in which place cells can map to different locations on a linear track , under the constraint that place cells which functionally couple to the same interneuron cannot map to locations within a certain distance of each other , which we termed the exclusion zone ( see Materials and methods ) . With this model , we find the maximum fraction of pyramidal cells , F , which can express place fields in a given map is: ( 1 ) F < NINPLD where NI , NP are the number of interneurons and pyramidal cells respectively , L is the length of the track and D is the size of the exclusion zone ( approximately the size of a place field ) . The above inequality gives a bound on the density of the spatial representation . It implies that spatial maps generated by this network must be sparse and that the required sparsity depends on the ratio of pyramidal cells to interneurons , and on the size of the place fields . If the network is close to this upper bound , there will be a high density of subthreshold phase precession fields in place cells and interneurons will phase precess over most of the environment . If instead the network is operating well below this upper bound , so that the representation is sparser than the minimum requirement , there will be only occasional interneuron and subthreshold phase precession fields . While subthreshold phase precession fields have not yet been investigated , the density of reported interneuron phase precession fields can be high ( see Figure 2 of Maurer et al . , 2006 ) , suggesting that CA1 networks may operate close to this bound . Does the non-overlap constraint limit the capacity of the network for the representation of distinct environments and contexts ? When we quantify the capacity of the network under the non-overlap constraint ( see Experimental Procedures ) , we find that the number of distinct spatial maps , cell assemblies and sequences that can be generated by the network are each considerably larger than the number of environments , events or behavioral episodes that an animal could encounter within its lifetime . For example , assuming a population of 10 , 000 pyramidal cells of which 20% are active in each map , 1000 interneurons , an exclusion zone between place fields of 1 meter , a linear track of length five meters and that place field locations can be distinguished with a spatial resolution of 10 cm ( a conservative estimate ) , the number of spatial maps in which coherent theta sequences are generated is greater than 105000 . For the same population of cells , assuming each cell assembly consists of 100 pyramidal cells , there over 10500 possible cell assemblies , and assuming a phase sequence consists of 7 cell assemblies ( Lisman and Idiart , 1995 ) there are over 101500 possible sequences . Hence , despite the constraints imposed by the coupling between groups of pyramidal cells and interneurons , the capacity of the network to encode distinct environments , contexts and episodes can be considered to be unlimited from an ethological perspective . The above analysis quantifies the number of maps under which all pyramidal cells exhibit robust phase precession . However , experimental data show a distribution of phase precession strengths in simultaneously recorded pyramidal cell populations in CA1 ( e . g . , Skaggs et al . , 1996; Schmidt et al . , 2009 ) . We therefore asked whether randomly organized place field maps in our model might be sufficient to account for typical distributions of single-cell phase precession strengths in CA1 populations , despite disruptive place field overlap . In random maps , we found that pyramidal cells exhibit a broad range of phase-position correlations ( Figure 7—figure supplement 4 ) . Even for very dense random maps , in which phase precession is severely disrupted on average , a substantial proportion of the population continued to exhibit robust phase precession . Thus , even randomly organized place field maps may be sufficient to account for experimentally observed phase precession statistics in individual place cells . In summary , we find that overlap between the place fields of pyramidal cells which functionally couple to the same interneuron can disrupt sequence compression in the network . The level of disruption increases with the number of active place cells per interneuron . For random place field mappings , maintaining coherent sequence compression requires that place field maps are sparse . By introducing mechanisms to organize place field maps in order to avoid interference , coherent sequence compression can be maintained with much larger numbers of active place cells . While such mechanisms reduce the number of spatial maps available to the network , we find that even under these constraints , there is a practically unlimited capacity for generating distinct spatial maps , cell assemblies and theta sequences in the network . How might downstream neurons receiving synaptic input from place cells use compressed sequences for computation ? A longstanding hypothesis is that sequence compression enables the association of events through spike timing dependent plasticity ( STDP ) ( Skaggs et al . , 1996 ) . Because STDP acts on events correlated on a timescale of tens of milliseconds it is not well suited to directly associating behavioral events ( Levy and Steward , 1983; Markram et al . , 1997; Magee and Johnston , 1997; Bi and Poo , 1998 ) , but it may act on compressed theta sequences representing several seconds of recent and upcoming experiences ( Figure 8A , B ) . Theta sequence compression in conjunction with STDP has been suggested to lead to asymmetry in the firing fields of place cells receiving place cell input ( Mehta et al . , 2002 ) , but the use of compressed event sequences as conditioned stimuli in classical associative learning has not been evaluated . 10 . 7554/eLife . 20349 . 018Figure 8 . A proposed function of sequence compression for associative learning . ( A ) The animal explores an environment , activating different cells in CA1 in a particular temporal order on a behavioral timescale . ( B ) A population of CA1 place cells performs sequence compression on the slow Gaussian envelope inputs . These cells project onto a downstream neuron which signals some event of interest ( the unconditioned stimulus ) . When this event occurs , this cell fires tonically at the trough of the theta cycle . Synapses from CA1 place cells to the event cell are modifiable via STDP . ( C ) During each cycle of the theta rhythm , CA1 cell assemblies representing past , present and future events in behavioral time are activated sequentially . At the trough of the theta cycle , place cells representing the animal’s current location are active , whereas during the descending and ascending phases cells representing past and future locations respectively are active . If the downstream cell signaling the unconditioned stimulus fires an action potential at the trough of the theta cycle , STDP between pre- and post-synaptic spikes establishes an association between cells representing recently visited locations and the event . ( D ) If instead the downstream cell encoding spikes at the peak of the theta rhythm , an association between cells representing upcoming locations and this cell is formed , whereas cells representing recently visited locations and these cells have their synapses weakened ( i . e . , the temporal associations are reversed relative to those in C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20349 . 018 We consider a population of CA1 pyramidal cells performing sequence compression on its inputs and projecting to a downstream neuron which receives a second strong input encoding some particular outcome or event of behavioral relevance , termed the unconditioned stimulus ( US ) ( Figure 8B ) . When the US occurs , the downstream cell signals that event by firing action potentials . Importantly , because behavioral events extending up to several seconds into the past are represented in an orderly fashion along the descending phase of the theta cycle and events occurring up to several seconds into the future are ordered along the ascending phase , sequence compression using theta oscillations generates an absolute temporal reference frame in neural time for past , present and future events in real time on which STDP can act ( Figure 8C ) . The absolute temporal reference frame provided by the theta cycle enables the timing of the downstream US-driven action potential to determine the association made . If these downstream action potentials lock to the trough of the theta rhythm , a standard STDP rule will cause inputs from place cells centered on locations before the place where the US was experienced to undergo an increase in synaptic strength ( Figure 8C ) . This circuit therefore implements associative learning , forming an association between the conditioned and unconditioned stimuli . In contrast , if the downstream cell were to lock to a theta phase other than the trough , this would introduce a temporal shift to behavioral time lags at which potentiation and depression of synapses occurs . For example , a downstream neuron which fires at the peak of the theta oscillation will cause a decrease in synaptic strength from neurons representing past locations and an increase in synaptic strength from CA1 pyramidal cells representing the future locations ( Figure 8D ) . Thus , sequence compression with theta oscillations allows locations , or events , occurring in the past or future to be flexibly and selectively associated with a particular outcome by varying the spike phase of the downstream cell . The high capacity of the sequence compression mechanism that we propose here enables STDP to act on a practically unlimited set of potential behavioral experiences . In the model we propose here the dynamics of signal integration by interneurons are critical to phase precession and sequence generation . This is in contrast to previous models for phase precession which focus on place cells . A key difference is that interneurons generally exhibit ongoing rhythmic spiking activity throughout an environment , whereas place cells are typically silent across most of an environment , showing sustained firing activity only within spatially localized place fields . In our model phase precession requires entrainment of spontaneous spiking by a pacemaker input and acceleration of spiking due to further excitatory spatial input . The dynamics of our proposed model account for phase precession through a full 360 degrees , and suggest mechanisms for speed tuning and dorsoventral organization of phase that are consistent with experimental observations . We discuss below how the distinct dynamics of pyramidal cells lead to models of phase precession with different properties . Previous models for phase precession face challenges in fully accounting for experimentally observed features of theta phase precession and sequence generation ( see Figure 1—source data 1 ) . An initial model for phase precession was based on interference between oscillations with different frequencies ( O'Keefe and Recce , 1993 ) . This model can account for phase precession observed through the full 360 degrees of a theta cycle . However , because it generates repeated spatial firing fields its predictions map more closely onto the properties of entorhinal grid cells than place cells in the hippocampus ( O'Keefe and Burgess , 2005 ) . Moreover , to account for phase precession in two dimensions interference models require heading modulated speed-dependent oscillatory signals , but experimental evidence for signals with the required properties is so far quite limited ( Harvey et al . , 2009; but see Welday et al . , 2011 ) . Other models rely on interactions between slow depolarizing inputs to place cells and oscillatory inputs to their soma and / or dendrites . These models generate a unidirectional phase advance over the place field using either an asymmetric ramp drive ( Mehta et al . , 2002; Losonczy et al . , 2010; Magee , 2001 ) or using spike train adaptation so that firing ceases at the peak of a symmetric place field drive ( Harris et al . , 2002 ) . However , these models appear able to achieve phase advances over the place field of only 180 degrees . Moreover , to avoid phase reversal in later parts of the firing field these models rely on sustained depolarization and elevated firing of place cells to maintain an advanced phase , whereas for many place cells phase continues to advance after the center of their firing field while firing rate and membrane potential depolarization drop ( Huxter et al . , 2003; Harvey et al . , 2009 ) . In contrast to these previous models , the network architecture we propose here is able to account for phase precession through 360 degrees , does not require sustained depolarization following the place field center , is compatible with single rather than regularly repeating firing fields and does not rely on tuning of upstream velocity controlled oscillators . Observations following experimental manipulations of phase precession also constrain models of the underlying circuit and cellular mechanisms . Intrahippocampal administration of cannabinoids disrupts the temporal organization of CA1 activity during theta cycles without altering firing rates ( Robbe and Buzsáki , 2009 ) , consistent with the scenario in our model when place field maps are relatively dense and unorganized ( Figure 7 ) or when high levels of extraneous noise are injected into interneurons ( Figure 7—figure supplement 2 ) . Optogenetic inactivation of parvalbumin-positive interneurons ( PV ) , but not somatostatin-positive interneurons ( SOM ) , disrupts phase precession by shifting the firing phase of pyramidal cells towards the trough of the theta cycle ( Royer et al . , 2012 ) , which is consistent with our model if phase precessing interneurons are generally PV-positive ( Figure 5—figure supplement 2 ) . Following a transient silencing of CA1 activity and resetting of the hippocampal theta rhythm , phase precession is relatively unperturbed ( Zugaro et al . , 2005 ) , a finding which is also replicated by our model ( Figure 5—figure supplement 1 ) . Our proposed model for theta sequence compression makes a number of experimentally testable predictions . These can be grouped into core predictions of the model , and ancillary predictions that follow from constraining the model parameters to account for experimentally observed features of phase precession and to maximize the quality of sequence compression performed by the network . The core predictions are as follows . ( 1 ) Silent or inactive pyramidal cells should demonstrate subthreshold phase precession fields resulting from inhibitory input when their primary interneuron is activated by another place cell . This prediction should be testable through patch clamp recordings in awake animals ( Harvey et al . , 2009; Epsztein et al . , 2011 ) . ( 2 ) Groups of pyramidal cells which precess in tandem with a particular interneuron should have non-overlapping place fields , and if not will exhibit disrupted theta compression . This prediction may be testable with high density electrical recordings or advanced imaging methods . ( 3 ) Artificially depolarizing a place cell to generate a firing rate field should automatically produce phase precession . This requires that the interneuron driving phase precession is active , but otherwise should also be testable through awake patch-clamp recordings . ( 4 ) Phase precession in place cells should be accompanied by the presence of strong , phase precessing inhibitory synaptic inputs . ( 5 ) Entrainment of septal GABAergic inputs should set the basal theta frequency , but precession of CA1 interneurons and place cells against this basal theta should remain intact over a range of frequencies . ( 6 ) Phase precessing interneurons should show reciprocal synaptic connections onto pyramidal cells , and the interneuron to pyramidal cell synapse should be sufficiently strong to synchronize theta activity . ( 7 ) Sustained inactivation ( e . g . , over an entire lap ) of phase precessing interneurons should abolish pyramidal cell phase precession . Tuning of our model to account for experimentally observed features of phase precession leads to further predictions about expected properties of the network components . ( 1 ) Stable phase precession across different running speeds emerges when phase precessing interneurons receive a velocity-modulated excitatory drive and a pacemaker drive with velocity-dependent amplitude . ( 2 ) A dorsoventral gradient in excitation to phase precessing interneurons , alongside a gradient in pacemaker amplitude and excitatory synaptic strength , simultaneously generates dorsoventral traveling theta waves and changes in precession slope across the dorsoventral axis . We note that these ancillary predictions pertain only to the specific implementations that we have considered , and that alternative mechanisms are possible within the model circuitry . For example , dorsoventral traveling waves could equally emerge from a gradient in the phase of pacemaker input within our model , and alternative mechanisms for generating velocity-dependence of phase precession may also be possible within the circuit . Because there is a greater number of pyramidal cells than interneurons in CA1 , our model requires that each phase precessing interneuron couples to several pyramidal cells . For successful sequence generation pyramidal cells which couple to the same interneuron must have largely non-overlapping place fields . This constraint leads to predictions for network topographies that support and determine the quality of sequence compression . ( 1 ) For randomly organized place field maps , reducing the density of firing rate fields in the pyramidal cell population increases the quality of the sequence-compressed representation of behavioral events . ( 2 ) Sequence compression can be maintained with maximal performance at far greater place field densities when these place fields are organized so as to minimize coactivity of pyramidal cells which precess with the same interneuron . Interestingly , this implies that CA1 does not exhibit topographically organized place field maps . This is consistent with an apparent lack of anatomical organization of place cells ( Redish et al . , 2001; Dombeck et al . , 2010 ) . ( 3 ) Because the maximum place field density is determined by the ratio of phase precessing interneurons to pyramidal cells , and because experimentally observed coding densities appear to increase along the dorsoventral axis , either: there are more phase precessing interneurons per pyramidal cell in the ventral hippocampus , the fraction of silent pyramidal cells in the ventral hippocampus is higher , or sequences are disrupted in the ventral hippocampus . When the constraint underlying these predictions is violated population theta sequences can be disrupted despite the presence of single-cell phase precession on individual laps . This may explain the dissociation of phase precession and theta sequences during exploration of novel environments ( Feng et al . , 2015 ) and when CA3 is silenced ( Middleton and McHugh , 2016 ) . In order to avoid such disruption of sequential activity , our model requires that CA1 networks can learn to decorrelate spatial maps following global remapping , most likely via a reorganization of firing rate fields in order to remove disruptive place field overlap . Understanding the learning rules and circuit mechanisms underlying such a decorrelation poses an interesting challenge , and points towards the importance of investigating the initial reorganization and stabilization of place field maps in novel environments ( Frank et al . , 2004; Karlsson and Frank , 2008 ) , with a particular focus on the emergence of spatiotemporally structured representations within theta cycles ( Dragoi and Tonegawa , 2013; Feng et al . , 2015 ) . The model that we propose automatically compresses slow sequences of inputs occurring on timescales of seconds into fast sequences of spiking activity within each cycle of the network theta rhythm . This mechanism could in principle be implemented in parallel with some previously proposed mechanisms for phase precession . For example , in addition to inputs from local interneurons considered here , dendritic and somatic interference in pyramidal cells might also contribute to the phase advance over the first half of the place field ( Magee , 2001; Losonczy et al . , 2010 ) , and oscillatory or phase precessing inputs from CA3 and entorhinal cortex might contribute to membrane phase in CA1 place cells ( Chance , 2012; Jaramillo et al . , 2014 ) . Moreover , additional architectures could extend our proposed model . For example , in brain areas such as CA3 and entorhinal cortex , attractor mechanisms may generate the firing rate fields through local circuit interactions ( Samsonovich and McNaughton , 1997 ) , such that the circuit mechanism proposed here generates theta phase precession when excitatory neurons are driven by slow depolarizing inputs arising from within the local circuitry . How might theta sequences generated by pyramidal-interneuron interactions contribute to hippocampal-dependent learning and memory ? Our analysis suggests a scenario in which , during theta oscillations , CA1 provides a time-compressed ongoing narrative of behavioral episodes ( Figures 1–7 ) . This enables downstream STDP mechanisms to form associations between ongoing behavioral events and specific outcomes such as reward or punishment ( Figure 8 ) . During sharp wave ripple events the CA1 network can then explore its state space and thereby test outcomes of different behavioral choices based on associations stored during theta activity ( e . g . , Singer et al . , 2013; Gomperts et al . , 2015 ) . This allows a form of mental exploration in which possible behavioral sequences can be simulated and the likely outcomes determined based on associations learned during theta states ( Hopfield , 2010 ) . The model that we suggest here provides a mechanism for real-time generation of theta sequences with capacity for storing novel associations experienced across an animal’s lifetime . This proposed framework makes several additional experimentally testable predictions for neurons downstream from CA1 . First , we predict that , during theta states , the spiking of downstream neurons encoding unconditioned stimuli is locked to the theta rhythm , but is not strongly influenced by the activation of specific cell assemblies in CA1 . During sharp wave ripple events that take place following learning , we predict that activation of these same downstream neurons can be driven by cell assemblies in CA1 whose outputs were previously associated with the conditioned stimulus . In line with these predictions , reward responsive neurons in the VTA lock more strongly to the hippocampal theta rhythm than non-reward responsive neurons , and VTA neurons that lock more strongly to the hippocampal theta rhythm exhibit greater coordination with CA1 cell assemblies representing reward locations during awake sharp wave ripple events ( Gomperts et al . , 2015 ) . Second , to form associations between conditioned stimuli and rewards occurring in the future , present or past on behavioral timescales , the proposed learning mechanism predicts that the timing of downstream reward-encoding neurons relative to the theta rhythm should shift , firing near the peak for future rewards and near the trough when rewards have been obtained . This behavior has been observed in reward-encoding neurons in the ventral striatum , which precess in phase relative to the hippocampal theta rhythm as the animal approaches a reward site ( van der Meer and Redish , 2011 ) . Hence , during theta states a primary function of interneurons in CA1 and other hippocampal structures may be to support compression of ongoing events into neuronal sequences in order to store associations in synaptic projections to downstream brain areas , which may then be utilized during sharp wave ripple events for mental exploration , planning and decision making . To understand how phase precession emerges in the circuit of Figure 1 , we developed a reduced model of an isolated interneuron driven by a constant excitatory current and a pacemaker current ( Figure 2A ) . In this simplified description , we treat the interneuron as an oscillator whose baseline frequency ω⁢ ( I ) is determined by the amplitude of the depolarizing current I through its f-I curve ( note that we do not make any explicit assumptions about the form of this f-I curve in the reduced model , although in Figure 2 we assumed a linear f-I curve ) . We consider the pacemaker input as a weak perturbation to this oscillator , which allows a reduced description of the interneuron in which only its phase is considered ( e . g . , Ermentrout , 1986 ) . We show below that under some general assumptions the following equation is obtained ( Adler , 1946 ) : ( 2 ) d⁢Δ⁢ϕ⁢ ( t ) d⁢t=Δ⁢ω-A⁢sin⁡ ( Δ⁢ϕ⁢ ( t ) ) where: Δ⁢ϕ=ϕ-θ is the instantaneous phase difference between the interneuron and the pacemaker input; the detuning Δ⁢ω⁢ ( I ) =ω⁢ ( I ) -ωθ is the frequency difference between the pacemaker input and the intrinsic frequency of the interneuron in the absence of pacemaker input; A is the synchronization factor , which depends on the amplitude of pacemaker input and on the intrinsic properties of the interneuron ( we describe this dependence in the derivation below ) . This equation approximates the phase relationship of the interneuron to the pacemaker input for different strengths of pacemaker drive and excitatory drive . For a constant input current , Equation ( 2 ) generates two distinct dynamical states depending on the relative values of A and Δ⁢ω ( i . e . , depending on the amplitude of pacemaker input and excitatory input to the interneuron ) . The first is stable phase locking and the second is phase precession . Phase locking occurs when \|Δ⁢ω/A\|<1 , with a stable locking phase of Δ⁢ϕlock=arcsin⁡ ( Δ⁢ωA ) ( Figure 2B ) . Phase precession occurs when \|Δ⁢ω/A\|>1 , where there are no stable phases and the interneuron precesses continuously , but nonlinearly , in phase against the pacemaker input ( Figure 2C ) . To derive Equation ( 2 ) and the dynamics described above , we assume that the pacemaker input is weak so that we may introduce an approximation based on its infinitesimal phase response curve z⁢ ( ϕ ) . Specifically , the dynamics of an oscillator with frequency ω driven weakly by an external perturbation Q⁢ ( t ) can be approximated by the reduced phase model: ( 3 ) d⁢ϕd⁢t=ω+z⁢ ( ϕ ) ⁢Q⁢ ( t ) where amplitude variations have been neglected . To model the case of an oscillator driven by a weak pacemaker ( i . e . , an interneuron driven by the septal theta rhythm ) we consider a perturbation of the form Q⁢ ( t ) =Q0⁢cos⁡ ( θ⁢ ( t ) ) . Equation ( 3 ) can then be expressed as: ( 4 ) d⁢Δ⁢ϕd⁢t=Δ⁢ω+z⁢ ( θ+Δ⁢ϕ ) ⁢Q0⁢cos⁡ ( θ⁢ ( t ) ) where Δ⁢ϕ=ϕ-θ and Δ⁢ω=ω-ωθ . If z is also sinusoidal , the above equation can be further approximated by averaging out fast fluctuations to obtain Equation ( 2 ) . To see this , we define the theta-average of a variable X as: ( 5 ) ⟨X⟩θ=12π∫02πXdθ Averaging out fluctuations on a sub-theta cycle timescale then gives: ( 6 ) ⟨dΔϕdt⟩θ=Δω+Q0⟨z ( θ+Δϕ ) cos⁡ ( θ ) ⟩θ which for sinusoidal phase response curves of the form z⁢ ( ϕ ) =z0-z1⁢sin⁡ ( ϕ ) is: ( 7 ) ⟨dΔϕdt⟩θ=Δω+Q0⟨cos⁡ ( θ ) ( z0−z1sin⁡ ( θ+Δϕ ) ) ⟩θ ( 8 ) =Δω−Q0z1⟨cos⁡ ( θ ) sin⁡ ( θ+Δϕ ) ⟩θ ( 9 ) =Δω−12Q0z1⟨sin⁡ ( 2θ+Δϕ ) +sin⁡ ( Δϕ ) ⟩θ ( 10 ) ≈Δω−12Q0z1sin⁡ ( Δϕ ) where in the last line it was assumed that Δ⁢ϕ does not change over a single theta cycle . This recovers Equation ( 2 ) and provides an explicit formula for the synchronization factor in terms of the phase response curve and pacemaker drive A=Q0⁢z1/2 . In other words , the synchronization factor depends on the amplitude of pacemaker drive and the sinusoidal component of the phase response curve of the interneuron . The general solution to Equation ( 2 ) is given by: ( 11 ) Δ⁢ϕ⁢ ( t ) =2⁢arctan⁡[A- ( Δ⁢ω ) 2-A2⁢tan⁡ ( 12⁢ ( Δ⁢ω ) 2-A2⁢ ( c-t ) ) Δ⁢ω] where c is a constant determined by the initial conditions . This equation is valid for both the phase locking and phase precession regimes . In the case of phase precession , where \|Δ⁢ω/A\|>1 , this gives the following precession frequency: ( 12 ) f= ( Δ⁢ω ) 2-A2/2⁢π which is obtained by noting that tan is a π-periodic function , while arctan is a monotonic function . Assuming that the phase precession frequency scales with running speed v and field size as f=v/ ( 2⁢R ) , where R is the radius of the place field ( Chadwick et al . , 2015 ) , we obtain a constraint on the detuning and synchronization factor: ( 13 ) ( Δ⁢ω ) 2=A2+ ( π⁢v/R ) 2 which shows how the phase precession frequency can be controlled across running speeds and dorsoventral locations by changing the strength of pacemaker input and excitatory drive to interneurons . For the stable phase locking case , where \|Δ⁢ω/A\|<1 , the expression for the precession frequency yields complex values . To recover the steady state locking dynamics shown in Figure 2B , note that for complex arguments the tan function in Equation ( 11 ) becomes a tanh , and in the limit t→∞ this tanh term tends to 1 so that Δ⁢ϕ⁢ ( t ) becomes independent of t and the initial condition c . The rate at which the decay to steady state occurs will therefore vary with \|Δω/A\| . Simulations were performed using the Brian simulator ( Goodman and Brette , 2009 ) . To model to experimental protocol of Zugaro et al . ( 2005 ) , we repeated the simulation with parameters as described above ( with running speed v=40 cm/s ) and in addition injected a negative current of amplitude 50 pA into the pyramidal cell and 20 pA to the interneuron for a duration of 200 ms , centered on the peak of the place field input , while simultaneously resetting the phase of the pacemaker drive . These parameters were sufficient to generate silencing for around 200–250 ms in both cells . To model the experimental protocol of Royer et al . ( 2012 ) , we again repeated the simulation with parameters as described above ( v=40 cm/s ) , but in this case delivered a negative current of amplitude 10 pA to the interneuron for a duration of 1 s . This was sufficient to silence the interneuron for the duration of the current injection . In the first simulation , silencing was centered on the peak of the place field input . 1050 laps were simulated , and the data were then pooled and averaged to generate Figure 5—figure supplement 2A , B ( see Royer et al . ( 2012 ) for details of data analysis ) . In the second simulation , silencing was centered on a random location within 20 cm of the peak of the place field input . We performed this additional simulation to account for the fact that the silencing was centered on a fixed portion of the track in the protocol of Royer and colleagues , so that the place cells analyzed typically were only silenced over part of their place field . We again simulated 1050 laps , with silencing centered on a different location in each lap , and then pooled and averaged the resulting data ( Figure 5—figure supplement 2C , D ) . To model changes in theta dynamics along the dorsoventral axis , we simultaneously varied the place field width σ , noise σnE , excitatory synaptic weight wE , depolarizing current to interneurons I0I and the pacemaker drive Iθ . For Figure 4A , B , only I0I was varied and all other parameters were as above . For Figure 4C , D , we chose two parameter sets representing the dorsal and ventral poles . For the dorsal pole , the parameters were: σ=45 cm; σnE=0 . 7 mV; I0I=80 . 455 pA; Iθ=1 . 95 pA; wE=0 . 53 nS . For the ventral pole , the parameters were: σ=600 cm; σnE=3 mV; I0I=79 . 525 pA; Iθ=0 . 12 pA; wE=0 . 081 nS . In both cases , the running speed was set to v=30 cm/s . To estimate the theta frequency of the simulated neurons , the membrane potential was bandpass filtered at 6 . 25-10 Hz and the instantaneous phase was calculated via a Hilbert transform . The phase was unwrapped and then smoothed using a moving average of width 250 ms . The gradient was calculated at each time point to obtain the instantaneous frequency . To determine the precession frequency at different running speeds , we calculated the average membrane frequency within a radius of 15 cm around the place field center on each pass through the place field . To remove artefactual frequency estimates arising due to the bursting dynamics within theta cycles , we excluded individual runs based on the variability of the instantaneous place cell frequency within this 15 cm radius . Specifically , we excluded runs on which the standard deviation was greater than 1 . 75 times the mean standard deviation over all runs at that speed . This excludes cases in which the estimated frequency fluctuated rapidly on a short timescale . To estimate the strength of phase precession in each pyramidal cell ( the single-cell phase precession metric ) , we calculated the Pearson correlation between the vector of spike phases Φ and the vector of the animal’s location X at the time of each spike on a single lap . The phase offset was chosen in order to minimize this correlation , i . e . to obtain the most negative possible ( strongest ) correlation between spike phase and the animal’s location ( Foster and Wilson , 2007; Feng et al . , 2015 ) . Specifically , given the vectors X and Φ , we calculated the correlation ρ ( X , Φ+ϕ~ ) , where ϕ~=argminϕ ( ρ ( X , Φ+ϕ ) ) . This metric was also used to measure single-cell phase precession pooled over multiple laps ( Figure 7—figure supplement 4 ) . To obtain the measure of population phase precession ( the population phase precession metric ) , we pooled the spikes of all pyramidal cells on a single lap . We again calculated the correlation between the vector of pooled spike phases Φpop and the vector whose entries are given by the distance of the animal from the place field center of the corresponding cell in the pooled spike phase vector at the time of that spike Xpop . As for the single cell case , the phase offset was chosen in order to minimize this correlation by calculating ρ ( Xpop , Φpop+ϕ~pop ) , where ϕ~pop=argminϕpop⁡ ( ρ ( Xpop , Φpop+ϕpop ) ) . To measure the strength of sequential activity in the population ( the single-cycle theta sequence metric ) , we analyzed the data on a cycle-by-cycle basis . For each cycle , we calculated the Pearson correlation between the vector of spike times in the population and the vector whose entries are given by the place field center corresponding to each spike in this first vector . Theta windows for this method had a temporal width equal to the period of the pacemaker input to the network . The offset of theta windows was given by the phase offset ϕ~pop which maximized the population phase precession measure for that lap . This allows for the possibility of an offset between the simulated CA1 network theta activity and the septal input oscillation . For network simulations , the number of simulated pyramidal cells was held constant ( Np=1000 ) and the number of interneurons was varied . This choice was made to avoid changes in correlation values introduced by changes in sample size . The number of interneurons was always chosen to be a divisor of the number of pyramidal cells so that there was an equal number of place cells for each interneuron . Each simulated place cell was given exactly one place field . For random place field mapping , place field locations were generated by a uniform distribution over a linear track . For optimal place field mapping , place field locations were defined so that the place cells associated with a single interneuron were equally spaced along the track and so that the entire population of place cells uniformly covered the track . Here we quantify the capacity of the network under the assumption that pyramidal cells which couple to the same interneuron cannot have overlapping place fields . We use three distinct measures of the network capacity: the number of spatial maps at a given spatial acuity; the number of cell assemblies; the number of phase sequences . These derivations were used to provide the capacity estimates stated in the main text .	Nerve cells in the brain exchange information via electrical impulses . In a given brain area , the electrical impulses at any particular moment can be thought of as forming a code that represents an aspect of the outside world . For example , groups of nerve cells in the hippocampus generate a type of code called a theta sequence , which represents a series of recent and upcoming events . The specific timing of electrical impulses within a theta sequence is crucial in creating certain types of memory . There are two major classes of nerve cell in the brain: excitatory cells activate impulses in neighbouring cells , while inhibitory cells act to temporarily block impulses from other nerve cells . Many groups , or “circuits” , of nerve cells contain combinations of both cell types to control how and when they communicate . Previous studies show that both types of cell are active within theta sequences , but it is not known precisely how they contribute to creating the sequences . Chadwick et al . developed a new mathematical model that simulates how theta sequences can emerge from circuits of both excitatory and inhibitory nerve cells . The connections between these simulated cells are based on experimental data from real nerve cells in the hippocampus . The model predicts that inhibitory cells play an important role in generating theta sequences by interacting with groups of excitatory cells to coordinate the timing of electrical impulses . Furthermore , the model shows how memory capacity depends on these connections . The next step following on from this work is to carry out experiments to test the model’s predictions . This will include monitoring the same group of nerve cells in multiple different situations to find out how their theta sequences change , and recording electrical events in individual nerve cells during theta sequences . If the theory’s predictions are confirmed this would lead to a deeper understanding of how our brains remember sequences of events .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "neuroscience" ]	2016	Flexible theta sequence compression mediated via phase precessing interneurons
The rodent hippocampus represents different spatial environments distinctly via changes in the pattern of “place cell” firing . It remains unclear , though , how spatial remapping in rodents relates more generally to human memory . Here participants retrieved four virtual reality environments with repeating or novel landmarks and configurations during high-resolution functional magnetic resonance imaging ( fMRI ) . Both neural decoding performance and neural pattern similarity measures revealed environment-specific hippocampal neural codes . Conversely , an interfering spatial environment did not elicit neural codes specific to that environment , with neural activity patterns instead resembling those of competing environments , an effect linked to lower retrieval performance . We find that orthogonalized neural patterns accompany successful disambiguation of spatial environments while erroneous reinstatement of competing patterns characterized interference errors . These results provide the first evidence for environment-specific neural codes in the human hippocampus , suggesting that pattern separation/completion mechanisms play an important role in how we successfully retrieve memories . Place neurons ( e . g . “place cells” ) in the rodent hippocampus preferentially fire in a particular spatial location ( O'Keefe and Dostrovsky , 1971 ) , the combination of which provide a neural code for that spatial environment ( O'Keefe and Nadel , 1978; Wilson and McNaughton , 1993; Muller and Kubie , 1987 ) . The collection of active place cells in an environment is thought to serve as a “cognitive map , ” providing a spatial framework for both navigation and memory more generally ( O'Keefe and Nadel , 1978; Redish , 1999; Buzsáki and Moser , 2013 ) . Two fundamental properties of place cells are their stability ( Thompson and Best , 1990; Hill , 1978 ) and their environmental specificity , also known as “remapping” ( Muller and Kubie , 1987 ) . Without reliable recapitulation of the ensemble of place cells representing a specific “map , ” spatial memory is impaired ( e . g . , Kentros , 1998; Morris et al . , 1986; McHugh et al . , 2007 ) . Remapping , a form of reorganization of hippocampal “maps” for different environments , is theorized to be a fundamental mechanism to navigation and memory more generally . However , the exact link between memory performance and remapping has yet to be fully established ( Jeffery et al . , 2003; Colgin et al . , 2008 ) . In humans , invasive recordings from the hippocampus have demonstrated place-coding neurons in single environments ( Ekstrom et al . , 2003; Jacobs et al . , 2013; Miller et al . , 2013 ) . Additionally , the human hippocampal formation is important to episodic memory more broadly ( Spiers et al . , 2001 ) , with place cells activating during item recall ( Miller et al . , 2013 ) and several studies demonstrating the ability to decode both location and episode-specific details from hippocampal fMRI blood-oxygen-level-dependent ( BOLD ) patterns ( Guterstam et al . , 2015; Hassabis et al . , 2009; Chadwick et al . , 2010 ) . Whether the human hippocampus represents one spatial environment as either the same or different from another , however , – a cornerstone of the idea that the hippocampus may compute spatial “maps” as part of a larger role in processing memories—remains unknown and untested . In addition to serving as a basic marker of memory , the environmental specificity of the hippocampus is thought to elucidate critical theoretical mechanisms of hippocampal function known as pattern separation and completion . These processes were predicted by early computational models and are thought to account for the memory interference errors commonly encountered in memory research and our everyday lives ( Yassa and Stark , 2011; Marr , 1971; Kohonen , 1977; Hunsaker and Kesner , 2013 ) . This theory states that pattern separation is a process that makes memories neurally distinct during memory storage and pattern completion a process by which memories are retrieved from a neural cue . Pattern separation and completion are thus thought to be important complements to each other ( Yassa and Stark , 2011 ) . Theoretical models and several empirical findings additionally suggest that CA3/DG and CA1 subfields mediate pattern separation and completion in the hippocampus ( Guzowski et al . , 2004; Leutgeb et al . , 2007; Leutgeb , 2004; Bakker et al . , 2008 ) . Yet exactly how these findings relate to human spatial memory remains unclear . Pattern completion is thought to rely on neural “attraction” between the cues that precede recall and stored representations , therefore allowing the cue to trigger re-instantiation of the full memory ( Rolls , 2010 ) . This property of attraction has the important implication that memories that are neurally similar will compete , producing interference in the case that the incorrect memory wins this competition ( Colgin et al . , 2008; Shapiro and Olton , 1994 ) . Theoretical models , therefore , postulate the central importance of pattern separation as critical to making memories less similar and thus avoiding interference due to neural attraction . Alternative accounts of memory interference , however , instead argue against a pure pattern separation/completion based account in favor of a model which posits inhibition of interfering memories from executive control regions during memory recall ( Anderson , 2003 ) . This account instead predicts that similar representations can co-exist , but can be selected , maintained , and strengthened by executive control centers during memory retrieval . Therefore , a definitive neural link between behavioral interference , neural pattern separation , and spatial remapping is necessary to resolve this debate and clarify the function of the hippocampus in memory . The aims of this study , thus , were three fold . The first was to examine whether humans also recapitulate neural codes for the same environment as well as bifurcate codes for different environments using fMRI and a multivariate pattern analyses . A second and critical test of whether remapping occurs in humans , however , is whether situations involving highly interfering spatial contexts can produce remapping failures ( e . g . , Skaggs and McNaughton , 1998; Spiers et al . , 2015 ) , and if so , what neural mechanisms characterize these errors . A final goal was to provide a link between behavioral measures of environment knowledge and neural measures of spatial remapping in humans . Our first and most basic prediction was that our human participants , analogous to remapping the rodent , would exhibit hippocampal voxel patterns that could uniquely identify each spatial environment . To address this prediction , we performed a searchlight classifier throughout the MTL ( see Materials and methods for details and Figure 2c ) . This approach allowed us to naively identify MTL regions where voxel patterns carried city specific information . Our inclusion of the interference city into this analysis allowed us to address additional questions of pattern separation/completion . For instance , if the reduced retrieval performance of the interfering city could be attributed to insufficiently separated neural patterns where models predict neural competition at retrieval , the classifier should disproportionally misclassify the interference city as one of the similar cities ( Cities 1&2 ) but not the distinct city ( City4 ) . 10 . 7554/eLife . 10499 . 006Figure 2 . Analysis methods . ( a ) Single trial parameter estimates were generated by building a single model with a separate regressor for each trial . ( b ) Subfields were demarcated manually to create separate ROIs for CA3/DG , CA1 , Subiculum , and PHG . ( c ) The searchlight classifier was trained using single trial estimates from half of the retrieval blocks and tested on the remaining retrieval data . Training/testing was repeated for all searchlight spheres in each subjects MTLs , creating subject specific statistical maps . ( d ) Within-city similarity was assessed for each ROI by extracting the trial parameter estimates from the subfields and correlating between matched trials of a city’s “A” and “B” retrieval blocks . ( e ) Between-city similarity was calculated consistent with within-city similarity . DOI: http://dx . doi . org/10 . 7554/eLife . 10499 . 00610 . 7554/eLife . 10499 . 007Figure 2—figure supplement 1 . Snapshot of virtual environment . DOI: http://dx . doi . org/10 . 7554/eLife . 10499 . 007 The pattern classifier correctly identified three of the four cities at levels above chance , revealing a cluster in hippocampal regions left CA3/DG and CA1 that significantly classified city identity ( Figure 3a–c ) . Analyzing this cluster in a one-way repeated-measures ANOVA , with classifier performance of each city as a separate factor , revealed significant differences between cities ( Figure 3C , F ( 3 , 54 ) = 12 . 9 , p<0 . 001 ) . Testing each city’s classifier performance against chance revealed that the classifier performed above chance on all cities except the interference city ( Cities 1 , 2 , & 4 above chance: t ( 18 ) > 3 . 2 , p<0 . 006 , two-tailed ) . Conversely , interference city classification performance was consistently below chance levels ( t ( 18 ) = −3 . 2 , p = 0 . 006 , two-tailed ) , despite overall classification ( across all cities ) being well above chance ( Figure 3C , t ( 18 ) = 5 . 6 , p = 2 × 10−5 , two-tailed ) . We note that within our search space , which included the hippocampus , parahippocampal , fusiform , lingual , and inferior gyrus , only this cluster in CA3/DG and CA1 exceeded the family-wise error rate cluster size correction ( see Materials and methods ) . Also , a control analysis that altered the classifier training protocol produced similar results suggesting these results were robust to differing analysis approaches ( Figure 3—figure supplement 1 , see Materials and methods for more details ) . This finding confirmed our prediction that , like the rodent , the human hippocampus contains environment specific representations in the CA1 and CA3/DG subfields for the environments that were most easily retrieved ( Cities 1 , 2 , & 4 ) . Additionally , the below chance classifier performance for City 3 and close resemblance of the retrieval accuracy and classifier accuracy seemed to indicate that low behavioral performance on City 3 may have related to lower levels of voxel pattern remapping for this environment . 10 . 7554/eLife . 10499 . 008Figure 3 . Environment classification . ( a ) City classification searchlight revealed a cluster of above chance classification performance throughout much of left CA3/DG and CA1 . ( b ) Pie chart of distribution of voxels in the searchlight showing their predominance in CA3/DG and CA1 . ( c ) Classifier performance of each city revealed above chance performance on cities 1 , 2 , and 4 and below chance performance on city 3 . Further analysis of city 3 classification performance revealed above-chance misclassification of city 3 trials as cities 1 & 2 . ( d ) City 3 ( interference city ) retrieval performance and city 3 classifier performance were positively correlated . p<0 . 05 , p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 10499 . 00810 . 7554/eLife . 10499 . 009Figure 3—figure supplement 1 . Classifier trained with matched number of trials from each city . DOI: http://dx . doi . org/10 . 7554/eLife . 10499 . 00910 . 7554/eLife . 10499 . 010Figure 3—figure supplement 2 . City 1 & 2 classification results broken down by correctly classified and incorrectly classified as each city . DOI: http://dx . doi . org/10 . 7554/eLife . 10499 . 010 To further explore the idea that low classification levels of the interference city ( City 3 ) were due to the competing representations of the similar cities , we inspected interfering city classification results . Here , we predicted that the classifier would misclassify the interference city trials as either City 1 or City 2 on more than 50% of trials ( chance level ) . This would be consistent with our behavioral results , which indicated that the interfering city was most often confused with Cities 1 and 2 during learning ( see Materials and methods ) . This would potentially support a pattern separation/completion account of hippocampal remapping of spatial memory errors . As predicted , interference city trials were incorrectly labeled as one of the two similar cities at levels well above 50% ( t ( 18 ) = 2 . 7 , p = 0 . 01 , two-tailed , Figure 3c ) . This suggested that a disproportionate amount of interference city trials resembled the similar cities . In contrast , trials for Cities 1 & 2 , on which retrieval performance was well above chance , were incorrectly classified as City 3 significantly less than chance ( t ( 18 ) = -8 . 6 , p<0 . 001 corrected; see Figure 3—figure supplement 2 ) . Overall , these findings are consistent with the idea that retrieval errors on City 3 could be attributed to , at least in part , insufficient differentiation of neural patterns from Cities 1&2 . A second critical prediction from the pattern separation/completion account would suggest that better individual performance on the interference city should be attributable to more distinct and therefore more readily classified representations of this city . Therefore , we also examined the link between interference city classifier performance and participant retrieval performance by seeing if the two measures were correlated . We found that interference city retrieval and classifier performance were significantly correlated ( r ( 17 ) = 0 . 61 , p = 0 . 006 , Figure 3d ) , a result which persisted even when matching the number of classifier training trials for each city ( Figure 3—figure supplement 1 ) . The link between participants who performed better on the interference city and those that had more readily classified neural representations of the interference city thus suggested that our best performing participants had neural representations that were more differentiated from each other than those of poorly performing participants . Together , the findings from the searchlight classification analysis support the idea that the human hippocampus exhibits environmental specificity . Further , the low classification performance of the interference city , and correlation between behavioral and neural classifier performance , favor a pattern separation/completion based account of memory interference . Although searchlight classifier analyses have advantages , in our case , the ability to naively identify regions within the human hippocampus exhibiting environment specific coding , they are not as well suited to hypotheses involving functional dissociations between subfields . For instance , searchlight clusters do not necessarily carry unique signals from different brain regions ( Woo et al . , 2014; Etzel et al . , 2013 ) . Pattern classification also cannot indicate whether neural codes are more similar within the same vs . between different environments , a cornerstone of spatial remapping findings in the rodent ( Wilson and McNaughton , 1993; Muller and Kubie , 1987 ) . We therefore employed a region-of-interest ( ROI ) based , multivariate pattern similarity ( MPS ) approach to 1 ) determine whether different human hippocampal subfields played different roles in spatial remapping and 2 ) provide more specific alignment with findings from rodents indicating higher neural similarity for the same vs . a different spatial environment . As outlined in the prior section , our operational definition of remapping was voxel similarity within the same city ( context reinstatement ) and dissimilarity between different cities ( remapping ) . To quantify this using multivariate pattern similarity ( MPS ) , we created a voxel “remapping index” defined as within-environment similarity minus the average between-environment similarity ( Figure 2d–e ) . Because our searchlight classifier analysis implicated CA1 and CA3/DG subfields in exhibiting voxel pattern remapping and remapping in the rodent is predominantly studied in CA1 and CA3/DG , here , we only include data from only CA1 and CA3/DG ( for other subfield results , please see Figure 4—figure supplement 1 ) . The results of this analysis are presented in Figure 4a–c . A 2x4 subfield by city repeated measures ANOVA revealed a main effect of subfield ( F ( 1 , 18 ) = 6 . 93 , p<0 . 02 ) , and a marginal subfield by city interaction ( F ( 3 , 16 ) = 3 . 14 , p<0 . 06 ) . This suggested that CA3/DG tended to exhibit more remapping but also that the pattern of results across cities differed for CA3/DG vs . CA1 . Also of interest was whether each city remapping score was significantly different from chance . In left CA3/DG , remapping scores were significantly above zero for Cities 1 & 2 ( all t-tests one tailed , t ( 18 ) >1 . 8 , ps<0 . 05 corrected; see Materials and methods ) and marginally above zero City 4 ( t ( 18 ) = 1 . 6 , p = 0 . 06 corrected ) . Thus , left CA3/DG MPS patterns indicated higher pattern similarity when participants retrieved spatial distances within Cities 1 , 2 , and 4 compared to the correlations of the patterns between different cities . 10 . 7554/eLife . 10499 . 011Figure 4 . Multivariate pattern similarity analysis ( MPS ) ofenvironment similarity during retrieval . ( a ) Similarity matrix of all pairwise city MPS conditions in CA3/DG . Diagonal depicts within-city and off-diagonal depicts between city MPS conditions . ( b ) Same as ( a ) for CA1 . ( c ) Voxel remapping index for CA3/DG ( green ) and CA1 ( blue ) . Remapping index for each city was the z-transformed contrast between within city and average between cities MPS ( see legend below ) . Left CA3/DG showed overall more remapping than CA1 , with significant remapping for Cities 1 & 2 and marginally significant remapping for City 4 . Left CA1 showed significant remapping only for City 4 . p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 10499 . 01110 . 7554/eLife . 10499 . 012Figure 4—figure supplement 1 . Cortical region MPS analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 10499 . 012 Left CA1 , in contrast , had remapping scores that were significantly above zero for City 4 ( t ( 18 ) = 2 . 29 , p<0 . 05 , corrected ) , but not for Cities 1-3 ( ts<−0 . 25 , p>0 . 6 ) . Consistent with the searchlight classifier results , City 3 did not exhibit significant remapping in either CA3/DG or CA1 ( ts<0 . 98 , ps>0 . 17 ) . This analysis further clarified the results of the searchlight classifier by suggesting that when characterizing entire subfields , more cities showed significant remapping effects in CA3/DG than CA1 , but that CA1 did exhibit remapping for the most distinct city ( City 4 ) . Together , these findings suggested that CA1 and CA3/DG showed remapping for distinct city retrieval , while only CA3/DG showed remapping between similar cities . Overall , these findings also confirm the importance of pattern separation mechanisms to remapping in both human CA3/DG and CA1 , an issue we consider with greater depth in the Discussion . Our searchlight classifier results suggested that correct trials tended to disproportionally resemble the similar cities but this effect was reduced for our best performing participants . This begged the question , if even correct interference city trials resembled the similar cities , did incorrect trials show even greater resemblance to the similar cities than the correct trials ? If the representation of the interference city was unstable and easily attracted to the similar cities , then voxel patterns for incorrect interfering city trials should be highly correlated with voxel patterns of correct trials from the similar cities . Conversely , we would not expect to see high pattern similarity between similar cities and interference city trials if they were correctly answered . An important control comparison was included to make sure that this effect could be attributed to interference rather than a general property of incorrect retrieval . We would not expect the incorrect interference city trials to show similarity to the distinct city ( because the distinct city was substantially different from Cities 1–3 ) , and thus neither incorrect nor correct interfering city trials should have correlated voxel patterns with the distinct city . To address these issues , we calculated MPS to compare correct and incorrect interference city trials with other city trials using matched visual stimuli ( triads ) ( see Materials and methods ) , the results of which are presented in Figure 5 . Voxel patterns in left CA3/DG ( and right CA3/DG ) on incorrect high interference city retrieval trials were significantly correlated with correctly retrieved voxel patterns in the similar cities ( Figure 5A ) . Importantly , incorrect interference city trials were significantly more correlated with correct similar city ( City 1 & 2 ) trials than were correct interference city trials with correct similar city or distinct city trials in CA3/DG ( left bar greater than others , two-tailed t-test , t ( 18 ) >2 . 2 , p<0 . 04 ) . This effect was present in left CA3/DG ( it was also present in right CA3/DG , two-tailed t-test , t ( 18 ) >3 . 7 , p<0 . 001 , Figure 5B , Figure 5—figure supplement 1A ) but not present in CA1 nor any other subfield ( t ( 18 ) <1 . 8 , p>0 . 09 , Figure 4—figure supplement 1C , Figure 5—figure supplement 1B ) . These results augment our searchlight classification results by demonstrating resemblance of incorrect interference city trials to the similar cities . Specifically , our findings support the idea that the unstable , weakly differentiated neural patterns of City 3 allowed the stable representations of Cities 1 & 2 to occasionally “outcompete” City 3 during retrieval . This in turn led to pattern completion of the wrong representation and selection of an incorrect response . 10 . 7554/eLife . 10499 . 013Figure 5 . Analysis of incorrect and correct interference city trials . ( a ) Analysis of interference city trials reveals higher similarity between incorrect city 3 ( interfering city ) and correct city 1 or 2 trials than between correct city 3 and correct cities 1 and 2 trials in CA3/DG . Control comparisons suggest that this effect could be attributed to interference from cities 1 & 2 . Left bar greater than all other bars t ( 18 ) >2 . 2 , p<0 . 04 . ( b ) CA1 did not exhibit similar behavior for incorrect vs correct between-city 3 comparisons . p<0 . 05 , *p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 10499 . 01310 . 7554/eLife . 10499 . 014Figure 5—figure supplement 1 . Right hemisphere hippocampal interference city MPS analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 10499 . 01410 . 7554/eLife . 10499 . 015Figure 5—figure supplement 2 . Empirical HRF plotted beside Canonical HRF convolved with 4 s boxcar function ( average response time was 3 . 8 s ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10499 . 015 One potential issue with multivariate pattern analysis techniques such as classification and MPS is that they could be driven by simple effects related to increases or decreases in the BOLD signal at specific voxels and subfields rather than changes in distributed neural patterns ( Etzel et al . , 2013 ) . It is also important to demonstrate that regions that carried multivariate information were recruited during the task by showing activation above baseline . To address these issues , we employed a simple univariate model comparing correct responses on each retrieval block against the baseline task ( see Materials and methods ) . We found significant levels of activation across hippocampal subfields ( average parameter estimates of CA1 , CA3/DG , and subiculum , left hippocampus t ( 18 ) = 3 . 6 , p = 0 . 002 , right hippocampus: t ( 18 ) = 5 . 4 , p = 3 × 10−5; all t-tests two-tailed ) , confirming that the hippocampus was broadly activated by our task , consistent with our past work ( Kyle et al . , 2015; Copara et al . , 2014; Stokes et al . , 2015 ) . We then tested whether MPS differences could be explained based on differences in univariate activation , which would challenge our findings of subfield specific changes in BOLD activation patterns ( Etzel et al . , 2013; Davis et al . , 2014 ) . To test this idea , we performed an 8x4 subfield ( left and right CA1 , CA3/DG , Subiculum , and PHC ) by city repeated measures ANOVA on mean activation . This analysis revealed a main effect of subfield ( F ( 7 , 126 ) = 29 , p<0 . 001 ) driven by larger parameter estimates in PHC than hippocampus proper ( t ( 18 ) >4 . 3 , p<4 × 10−4 ) . Neither of the remaining effects , however , ( main effect of city and subfield by city interaction ) were significant ( F<1 , p>0 . 5 ) . We also specifically tested regions of interest CA3/DG and CA1 with a 2 × 4 subfield ( left CA1 vs left CA3/DG ) by city repeated measures ANOVA which revealed no significant effects ( Fs<2 , ps>0 . 15 ) . These findings suggest that city-specific differences in univariate activation levels ( i . e . , greater activation to the distinct city than other cities ) could not account for our overall pattern of results . We believe that four novel components of our findings aid in understanding of human hippocampal function and its relation to memory processing . First , we extend environment specific coding to the human hippocampus using voxel-pattern based analyses . Using a searchlight classifier approach , which naively identified medial temporal lobe regions carrying city specific information , we found a cluster of voxels in CA3/DG and CA1 whose patterns decoded specific cities during retrieval . Next , using an MPS ROI approach which utilized all voxels from a subfield to characterize similarity within and between cities using simple correlations , both CA3/DG and CA1 showed higher similarity within city than between cities . Although past studies in humans have confirmed the presence of location specific coding in the hippocampus ( Ekstrom et al . , 2003; Jacobs et al . , 2013 ) , measuring remapping typically requires a large number ( >40 ) of simultaneously recorded cells ( Wilson and McNaughton , 1993 ) , which are difficult to obtain in most human studies . Additionally , although past fMRI studies in humans have suggested location-specific ( Guterstam et al . , 2015; Hassabis et al . , 2009 ) , distance-specific ( Morgan et al . , 2011 ) , and episode-specific spatial coding within a single environment ( Brown et al . , 2014 ) , demonstrating remapping between different spatial environments in particular has been elusive because altering the environment changes the visual scenes and trajectories experienced by the subject . Here , we dealt with this issue by minimizing visual confounds inherent in navigation by instead having subjects retrieve spatial distances from specific environments during retrieval . Thus , our findings suggest that indeed the human hippocampus contains neural codes that differentiate specific spatial environments . Second , our results provide support for a pattern separation/completion based account of memory disambiguation . Here , we probed the neural underpinnings of both successful and unsuccessful disambiguation during memory retrieval . Retrieval of Cities 1 , 2 , & 4 , which were more easily learned and retrieved , were shown to involve orthogonal voxel patterns , as demonstrated by a searchlight classifier and voxel similarity analyses . City 3 , however , which contained repeated landmarks but in a novel arrangement from Cities 1 & 2 , did not exhibit neural characteristics consistent with remapping or pattern separation , i . e . higher within than between city similarity or above chance classification . Rather , when attempting to classify City 3’s correct retrieval trials , most trials were classified as City 1 or 2 , although this effect was reduced for higher performing participants , suggesting that high performers exhibited more stable hippocampal patterns than low performers . Thus , one possible explanation for the poor performance on City 3 is that its neural patterns were insufficiently separated from those of Cities 1 & 2 , resulting in a tendency to incorrectly pattern complete to stable representations of City 1 & 2 . An alternative interpretation could be to attribute such errors to inhibition failure , for example , insufficient inhibition of City 1 & 2 representations by prefrontal cortex could lead to those being erroneously retrieved when attempting to retrieve City 3 ( Anderson , 2004 ) . The inhibition model , though , would not appear to predict low classification of correct City 3 trials and misclassification of these trials as Cities 1 & 2 trials since correct City 3 trials should involve trials in which traces from Cities 1 & 2 were successfully suppressed ( Aron et al . , 2014 ) and thus show no correlation with Cities 1 & 2 ( see Figure 5 ) . Furthermore , it is not clear how inhibition from higher cortical areas alone could lead to different patterns of suppression across the hippocampal subfields as prefrontal cortex projects primarily to subiculum and entorhinal cortex and not differentially to the CA fields , at least in non-human primates ( Goldman-Rakic et al . , 1984 ) . Thus , our findings overall support the importance of pattern completion and separation , particularly in CA3/DG , to spatial remapping and appear less easily reconciled with an inhibition-based account . A third important insight provided by our findings is a potential link between remapping-like mechanisms in humans , spatial learning , and rodent hippocampal remapping . The relationship between hippocampal remapping and behavior , however , remains unclear from the few studies to address both ( Colgin et al . , 2008 ) . Part of the issue , as acknowledged in past such studies , is that it is difficult to assay whether a rat “knows” it is in a different environment or not , although dwell time and reversing direction may be important behavioral assays ( Spiers et al . , 2015 ) . Jeffery et al . , ( 2003 ) show that in a hippocampally dependent place-reward discrimination task , rodents perform only slightly worse after small environment modifications that induce global ( ∼85% of cells ) remapping ( Jeffery et al . , 2003 ) , suggesting that remapping can occur quickly and have little negative effect on performance . McHugh et al . , 2007 , in contrast , demonstrated that dentate gyrus NMDA knockout mice experienced less hippocampal remapping between contexts and less behavioral discrimination ( McHugh et al . , 2007 ) , suggesting that remapping is important to behavior . In the current study , we assessed map drawing performance after each round of spatial exploration , which provided a more direct link to the formation of a cognitive map during navigation . We found that maps of Cities 1 & 2 were easily learned because information could be readily transferred between the two cities . Later , when assessing neural patterns during retrieval , Cities 1 & 2 were shown to have mutually orthogonal hippocampal voxel patterns . In contrast , City 3’s maps were less accurate and took more trials to acquire due to interference from Cities 1 & 2 . During retrieval , performance was lower and hippocampal patterns were not orthogonal to those of Cities 1 & 2 . However , we found that participants who performed better on City 3 did show voxel patterns that were more readily differentiated from the other cities . Thus , our findings from Cities 1 & 2 appear consistent with the results of Jeffrey et al . 2003 as we show remapping between Cities 1 & 2 despite the map acquisition data arguing for shared information between the two distinct representations . Our results for City 3 , though , appear consistent with McHugh et al . , 2007 , with less remapping negatively impacting memory performance . Thus , we think our data provide a potentially important link between behavioral memory performance in humans and measures of remapping and pattern separation/completion . Finally , our findings also provide important extensions and challenges regarding the function of the human hippocampal subfields in spatial context disambiguation . Specifically , when correctly retrieving spatial distances from two overlapping cities differing only in terms of two swapped stores , neural patterns were uncorrelated to each other and all other cities , an effect primarily shown in left CA3/DG . These findings support a role for CA3/DG in differentiation of competing spatial inputs , suggesting that this subfield in particular may be important for fine-grained discriminations amongst overlapping contexts as a part of a larger role in pattern separation/completion ( Yassa and Stark , 2011 ) . We also found that CA3/DG pattern remapping was ( marginally ) significant for City 4 , suggesting that it differentiated the distinct city as well . These findings are consistent with the idea of CA3/DG as a universal pattern separator/completer , with failures linked to low performance on City 3 ( Yassa and Stark , 2011 ) . In contrast , pattern similarity in CA1 was higher when participants correctly judged spatial distances from a distinct city featuring novel landmarks and geometry compared to retrieval-induced patterns from the other cities . These findings appear somewhat inconsistent with models that suggest CA1 serves as a complement to CA3/DG in pattern separation/completion ( Guzowski et al . , 2004 ) . Instead , our data appear to be more consistent with the emerging idea that CA1 may play a specific role in detection or representation of novelty ( Duncan et al . , 2012; Larkin et al . , 2014 ) , possibly acting as an important hub for integrating cortical input ( Stokes et al . , 2015; Remondes and Schuman , 2004 ) . Although our study cannot provide specific insight beyond this speculation regarding the functional role of CA1 , our findings suggest that its role in processing spatial contexts goes beyond a pattern separation/completion function defined by its position between CA3/DG and entorhinal cortex . One potentially perplexing aspect of our findings is that we found similar degrees of remapping , and therefore putative neural pattern separation processes , for both the similar cities ( Cities 1 & 2 ) and City 4 , despite the fact that information content was significantly different for City 4 than Cities 1&2 . Past studies , for example , have found that neural pattern differences may scale as a function of environment dissimilarity , with geometrically more distinct environments showing lower neural pattern similarity than more similar-shaped environments ( Stokes et al . , 2015; Leutgeb et al . , 2005 ) . In contrast to these two studies , though , we did not employ a continuous measure of environmental similarity and retrieval success hinged on being able to successfully maintain separate representations for the different environments but not necessarily tracking differences in the details of the different environments themselves . Thus , it is possible that our paradigm involved a more discrete form of pattern separation ( Yassa and Stark , 2011 ) than would be needed in experiments involving a continuous change between environments . Consistent with the discrete pattern separation interpretation , studies of human episodic memory suggest that even overlapping episodes can be successfully decoded from multivariate patterns in the hippocampus , suggesting that retrieval of even very similar episodes can involve pattern separation-like processes ( Chadwick et al . , 2010 ) . Thus , one possible interpretation of our results is that pattern separation in CA3/DG may not always scale with behavioral details and thus may be a more discrete process that depends on the exact demands required by encoding and retrieval . In conclusion , while past studies in rodents suggest that hippocampal remapping might be one possible mechanism whereby memories are differentiated , no studies have demonstrated a comparable phenomenon in humans . Our findings thus provide several important insights about the human hippocampus: 1 ) the hippocampus contains unique codes about specific spatial environments , with these codes showing significant neural pattern similarity within the same spatial environment and low similarity between different spatial environments 2 ) when interference between spatial environments is high , pattern completion/separation processes may fail , resulting in difficulties discriminating competing cities 3 ) remapping , in humans at least , appears tightly linked with behavioral performance at discriminating different environments during retrieval 4 ) human CA3/DG appears to play a fairly ubiquitous role in pattern separation/completion during retrieval of specific spatial environments while CA1 plays a more specific role in representing features of novel environments . Together , our findings provide new insight into how the human hippocampal circuit processes competing spatial information . Nineteen healthy individuals participated in the experiment ( 9 female ) from the community surrounding University of California at Davis . All participants had normal or corrected-to-normal vision and were screened for neurological or psychiatric illness . This study was approved by the Institutional Review Board at the University of California at Davis . Written informed consent was obtained from each participant before the experiment . The study consisted of a learning session ( not scanned ) and retrieval session ( scanned ) . During the spatial learning session , subjects played a video game where they learned four virtual environments on the computer in a randomized order . Each virtual environment consisted of six “stores” ( for a snapshot , see Figure 2—figure supplement 1 ) . Participants learned store locations by traversing the environment and then drawing a map of the store locations after visiting each store . This process repeated four times in the imaging study and six times in the behavioral study ( see Materials and methods ) before participants moved on to the next city . Traversals involved traveling to each store in the environment in a randomized order . The layout of each environment is shown in Figure 1a . Similar cities ( Cities 1 and 2 ) were the same except the locations of two stores were swapped in these environments . City 3 ( called the interference city based on the behavioral results ) contained the same stores as Cities 1 and 2 , but in a novel layout and with novel ground and wall textures . Finally , City 4 ( distinct city ) contained a novel store set , a novel layout , and novel ground and wall textures . Novel ground and wall textures were used in Cities 3 and 4 to reduce interference in these cities based on the findings of Newman , et al . , 2007 . We instructed participants that there would be four cities , some involving repeated stores , and that they would need to distinguish each city from one another to successfully perform the retrieval portion of the task . The retrieval portion took place in the scanner , where participants completed eight retrieval blocks ( two per city ) . Each retrieval block consisted of ~4½ min of memory judgments pertaining to a single city . Before the start of each retrieval block , text and verbal confirmation informed participants of which city they would be retrieving next , followed by a 40 s refresher video . After completing a retrieval block , participants were permitted a brief break before moving on to the next block and thus a new city . The order of retrieval blocks was pseudo randomized with rules dictating that no city could be tested twice in a row and that each city must be tested once before a city could be repeated . Each block consisted of 20 trials of judgments of the relative distances of stores in that city . Each trial consisted of an image of 3 stores , one store on top and two below . Participants judged which of the two bottom stores was closer to the top store , and indicated their choice by pressing the corresponding key on an MR-compatible button box . A “one” response indicated that the lower left store was closer to the top store and a “two” indicated the lower right . Because Cities 1–3 shared the same stores , they also shared the same stimulus set , while city 4 contained a novel stimulus set . To control for effects of motor response , the position of the two bottom probe stores on each stimulus was swapped during “A” vs “B” retrieval blocks of the same city , effectively switching the correct button responses for the different sessions of the same city ( see Figure 2c ) . To determine behaviorally whether participants employed similar , competing , or novel representations for spatial context , we tested each participant on how they encoded four different virtual cities ( Figure 1a ) . This involved navigating one of the four cities and then drawing a map immediately afterward . The first two cities ( Cities 1 & 2 involved the same stores and geometry with two locations swapped [similar cities] ) ; the third city involved the same stores as cities 1 &2 but a novel geometry and locations ( interference city ) . City four involved a completely novel set of stores and geometrical arrangement ( distinct city; see Figure 1 , Materials and methods , and Zhang et al . 2014 ) . We tested a total of 31 participants . Map drawing accuracy improved for all four cities as a function of navigation ( Figure 1—figure supplement 1: City 1 & 2: Beta = 0 . 065 , F ( 1 , 170 ) = 34 . 95 , p = 1 . 8 × 10−8; second city 1 & 2: Beta = 0 . 025 , F ( 1 , 175 ) = 7 . 36 , p = 0 . 007; City 3: Beta = 0 . 058 , F ( 1 , 179 ) = 38 . 41 , p = 4 × 10−9; City 4: Beta = 0 . 033 , F ( 1 , 179 ) = 14 . 52 , p = 0 . 0002 ) , suggesting that participants were able to form stable representations of each one . We next wished to address the extent to which acquiring a representation for one environment might enhance or impede acquiring a representation for another environment . This , in turn , would provide insight into the extent to which forming a representation of the different environments involved overlapping , different , or competing representations , as hypothesized . To address this issue , we computed the difference in map drawing scores during city transitions , in other words , the extent to which drawing a map on the last trial of learning interfered with drawing the map of a new city on the first trial ( i . e . , first drawn map of the new city – last drawn map of the previous city . Comparing changes in map scores when transitioning from one city to another , we found a main effect of city transition ( Figure 1—figure supplement 2 , 1- Way ANOVA , F ( 3 , 74 ) = 3 . 83 , p<0 . 02 ) . Post hoc two-tailed t-tests demonstrated that transitioning to City 3 from City 1 or 2 resulted in significantly greater learning costs than transitioning from City 1 to City 2 or from City 2 to City 1 ( t ( 41 ) = 2 . 4 , p<0 . 02; t ( 38 ) = 2 , 7 , p<0 . 01 ) . Similarly , transitioning from city 3 to cities 1 or 2 resulted in significantly greater learning costs than transitioning to City 4 ( t ( 33 ) = −2 . 2 , p <0 . 04 ) . This supports the assertion that City 3 representations interfered with cities 1 and 2 . The only city to city transition that did not result in significant learning costs was the City 1 to City 2 transition ( two-tailed t-test: t ( 27 ) = -1 . 9 , p = 0 . 07; all other city to city transitions involved significant costs in map drawing performance ( one tailed t-tests against zero , ts>−3 . 3 , ps<0 . 007 ) . This supports the idea that Cities 1 and 2 facilitated learning of each other . Transitioning to City 4 from any other city compared to transitioning from Cities 1 and 2 , however , did not differ ( t ( 49 ) = 1 . 0 , p = 0 . 3 ) . This suggests that City 4 was likely represented by a new representation entirely . Together , our findings support the idea that Cities 1 and 2 likely involved similar , largely overlapping representations ( similar cities ) , City 3 involved a representation interfering with City 3 ( interference city ) , and City 4 likely involved a novel non-overlapping , novel representation ( distinct city ) . We designed our stimuli such that trials involving stores that swapped locations between Cities 1 and 2 were over-represented . For instance , 12 of 20 trials involved at least one swapped store between Cities 1 & 2 and 9 of these 12 had a different correct response in City 1 vs City 2 . Therefore , a subject could score a maximum of 55% accuracy in City 2 based on knowledge of City 1 alone , and vice versa and all subjects were well above this threshold . Additionally , even if we assume that subjects correctly answer all trials involving the same response in Cities 1 & 2 and guess on trials involving different correct responses , we would expect a chance level accuracy of 77 . 5% for cities 1 & 2 . All but 2 subjects had accuracy above 77 . 5% for both Cities 1 and 2 ( one scored 77 . 5% on City 1 and 92 . 5 percent on City 2; the other scored 72 . 5 and 57 . 5 on Cities 1 & 2 , respectively ) . Furthermore , taking the lower performance for Cities 1 and 2 for each subject and testing the result against 77 . 5% , subjects still performed significantly above chance ( t ( 18 ) = 5 . 4 , p<0 . 0001 ) . Thus , performance on the swapped cities ( Cities 1 & 2 ) could not be explained by a strategy involving using the same response on both cities . We employed the same imaging sequences and preprocessing steps described in Kyle et al . , 2015 and Stokes et al . , 2015 . Imaging took place in a Siemens 64-Channel 3T “Skyra” scanner . High-resolution structural images were acquired employing T2-weighted turbo-spin echo ( TSE ) anatomical sequences ( TR = 4200 . 0 ms , TE = 93 . 0 ms , FOV = 1 . 9 mm , flip angle = 139° , bandwidth = 199 Hz/pixel ) , involving a voxel resolution of 0 . 4 × 0 . 4 × 2 mm . High-resolution functional echo-planar imaging ( EPI: TR = 3000 ms , TE = 29 ms , slices = 36 , field of view ( FOV ) = 192 mm , flip angle = 90° , bandwidth = 1462 Hz/pixel ) involved a resolution of 1 . 6 × 1 . 6 × 2 mm . Sequences were acquired perpendicular to the long axis of the hippocampus . An additional matched-bandwidth sequence was acquired to aid in registration of the EPI sequence to the high-resolution scan ( TR = 3000 ms , TE = 38 ms , slices = 36 , FOV = 245 mm , flip angle = 90° , bandwidth = 1446 Hz/pixel ) . Each EPI sequence underwent band pass filtering , slice-timing , and motion correction in SPM8 before parameter estimation . Parameter estimation for univariate analyses used a canonical hemodynamic response function ( HRF ) , and modeled all correct responses above baseline for each EPI sequence ( Friston et al . , 1995 ) . Analysis of multivariate pattern similarity requires maximally orthogonalized hemodynamic response functions ( HRFs ) as collinearity can inflate MPS-related correlations ( Mumford et al . , 2012 ) . Consistent with past work , we modeled each trial as a separate regressor ( Copara et al . , 2014; Mumford et al . , 2012; Rissman et al . , 2004 ) using finite impulse response ( FIR ) functions to model the average HRF to retrieval stimuli . This produced 10 parameter estimates for the first through the tenth TR after stimulus onset , corresponding to a 30 s long time course estimate of the HRF for each subject , block , and voxel ( Mumford et al . , 2012; Mourão-Miranda et al . , 2006 ) . This ensured the greatest ability to detect when spatial contextual retrieval might occur for the different cities but without selecting specific HRFs for different subjects or conditions . To select the HRF that explained the most variance for all subjects , sessions , and voxels , we employed independent component analysis decomposition using logistic infomax ICA ( Bell and Sejnowski , 1995 ) and identified a single HRF component that explained 38% variance ( shown in Figure 5—figure supplement 2 ) . This HRF was then resampled using a cubic spline interpolation to match the 16 time-bin per scan default that SPM8 uses to build regressors . Separate left- and right- hemisphere anatomical ROIs were manually traced ( using FSLview ) based on each participant’s high resolution T2 as described previously ( Copara et al . , 2014; Ekstrom et al . , 2009 ) . Demarcated subregions included hippocampal subregions CA1 , CA3/DG , Subiculum , and the extrahippocampal region parahippocampal cortex . We combined the CA3/DG subfield as finer distinctions cannot be made at the acquired resolution . MPS analyses were based on all voxels identified within ROIs . We performed classification using the Princeton mvpa toolbox ( Detre , 2006 ) , with alterations to the code to allow three hidden layers and a searchlight across MTL subfields . The searchlight was performed as in our previous manuscripts ( Copara et al . , 2014; Stokes et al . , 2015; Kyle 2015 ) . Briefly , for each 31 voxel ellipsoid throughout each subject’s MTL , we trained the classifier on one half of retrieval blocks ( one block per city ) and used the second half to test classification accuracy then swapped training and testing data . Two classifier training protocols were used . The first trained the classifier using all correct trials from one half of retrieval blocks . This method maximized the amount of training data but did not balance the number of trials used to train each city ( Figure 2c and Figure 3 ) . A second classifier training protocol used a random subset of correct trials from each city so that the classifier would be trained using the same number of trials from each city ( Figure 3—figure supplement 1 ) . Next , the average classifier performance for each searchlight position created a subject specific statistical map . Maps were warped to common space of a template subject using Advanced Normalization Tools ( AVANTS et al . , 2008 ) . Finally , group-space maps were contrasted and clustered by t-value corresponding to alpha = 0 . 05 . Permutation tests corrected for false positives by providing a corrected p<0 . 05 cluster size from the distribution of max cluster size of 1000 label-shuffled permutations . We note that a control analysis expanded search volume outside of the MTL to include fusiform gyrus and inferior temporal cortex , no clusters from these regions passed threshold . Pattern similarity analysis involves measuring the similarity of voxel patterns by calculating the correlation between parameter estimates of different trials within a common collection of voxels ( Mumford et al . , 2012; Kriegeskorte et al . , 2006 ) . To measure pattern similarity , we identified trials that were correctly retrieved during two separate retrieval blocks . For within city MPS correlations were made between blocks of the same city and for between city MPS between blocks of different cities . MPS values measured the average r value between matching , correctly answered trial pairs for each participant and each subfield ( Figure 2c ) . The bottom stores of each stimuli were swapped between different testing sessions of the same city , eliminating contributions to within-city MPS from the same motor response ( Figure 2d ) . Because Cities 1–3 shared a common stimulus set , within and between city MPS could be calculated identically for these cities . The ability to match stimuli identity for within and between city MPS for Cities 1–3 allowed excellent control for visual aspects of the task as any differences in patterns could be attributed solely to retrieval environment . Because City 4 contained novel stimuli , necessitated by our behavioral testing to ensure that this city involved a new representation , between-city comparisons involving City 4 had no logical matching stimuli . Thus , instead of matching trials based on stimuli identity , all correctly retrieved stimuli from City 4 were correlated with all correctly retrieved stimuli from the other city for all possible pairwise combinations . A control analysis calculated between-city MPS forCities 1-3 using the same method as City 4 , with all pairwise combinations of non-visually matched triads . This control analysis did not reveal any significant deviations to our effects and thus visual matching was maintained when possible . The significance of the remapping index was tested with a t-test against 0 . The family-wise error rate was corrected for using a bootstrapping approach . Ten thousand iterations of t-tests on each subfield and city were performed on randomly permuted data . The distribution of t-values was then used to determine the corrected t-value at a given percentile . Although we present only results from Left CA1 and Left CA3/DG in the body of the manuscript , CA1 , CA3/DG , subiculum , and parahippocampal cortex ( PHC ) were demarcated from both hemispheres . Use of all hippocampal subfields in a region-of-interest ( ROI ) multivariate pattern similarity ( MPS ) analysis provided a complementary approach to the searchlight classifier . Searchlight techniques are limited in that they can demonstrate the location of information content , but generally do not support functional dissociations between regions . ROI approaches allow better access to investigate functional dissociations but suffer from the multiple comparison problem . To control for multiple comparisons here , we performed ANOVAs using all demarcated subfields to attempt to control for multiple comparisons . First , we tested whether within city vs . between city neural similarity varied as a function of subfield , we performed a 2 ( Within/Between ) × 8 ( Left and right , CA1 , CA3/DG , Subiculum , and PHC ) repeated-measures ANOVA . We found a main effect of subfield ( F ( 7 , 126 ) = 8 . 4 , p<0 . 001 ) and an interaction effect of Within/Between city retrieval and subfield F ( 7 , 126 ) = 2 . 14 , p = 0 . 04 ) . This suggested that the relationship of within/between city MPS varied by subfield . We then broke down within vs between for each city using the remapping index ( Figure 4 ) . Here an 8 × 4 subfield by city remapping repeated measures ANOVA revealed a main effect of subfield ( F ( 7 , 126 ) = 2 . 835 , p = 0 . 009 ) suggesting that MTL subfields varied significantly in their tendency to “remap . ” These analyses suggested further investigation was warranted . This analysis is provided in the manuscript under “Classification of city-specific retrieval patterns in the hippocampus demonstrates successful decoding of all spatial contexts except the interfering environment” in the results section . In the next section , we address incorrect vs correct retrieval of the interfering city . Here , we calculated MPS to compare correct and incorrect interference city trials with other city trials using matched visual stimuli ( triads ) during retrieval ( see Experimental Procedures ) . An 8 × 4 subfield by condition repeated measures ANOVA revealed that the correlations between correct and incorrect trials of the different city comparisons ( Figure 5a , b and Figure 4—figure supplement 1 and Figure 5—figure supplement 1 ) varied as a function of subfield ( significant interaction effect: F ( 21 , 378 ) = 2 . 5 , p <0 . 001 ) . One possibility is that our analysis approach for Cities 1–3 or our presentation of novel stores for City 4 allowed for a visual confound . Thus , could our results be due to differences in visual features during retrieval rather than city-specific neural representations ? Several lines of evidences argue against this possibility . Cities 1–3 were perfectly visually matched in terms of what was presented to the participant and how we correlated these triads during our analyses . To provide more detail on the visual matching in our incorrect trial analysis using MPS ( Figure 5 ) was perfectly visually matched ( matched incorrect City 3 trials with corresponding City 1&2 trials vs . matched correct City 3 trials with corresponding City 1 & 2 trials ) , eliminating a visual confound as a possible counter interpretation . Thus , these findings for Cities 1–3 cannot be explained based on a trivial visual stimulus confound . One unavoidable aspect is that our use of a truly novel city , City 4 , did not involve visually matched triads as these stores were necessarily different than those in Cities 1-3 . The pattern of findings based on a visual confound from City 4 alone , however , would have predicted a qualitatively and quantitatively different pattern of results . Specifically , a visual confound would predict higher classifier performance for City 4 and similar performance for Cities 1 , 2 , & 3 . Our pattern of results , however , was that City 3 had lower classifier performance than Cities 1 , 2 , 4 . Thus , the prediction provided solely by a visual confound was not supported by our data . Second , the visual confound would not predict the presence of a performance correlation with classification accuracy on City 3 ( Figure 3d ) . Specifically , as we report in Figure 3d , classification performance improves with better individual subject retrieval performance on City 3 . Thus , together , our paradigm and pattern of findings argue against a visual confound based explanation alone .	How do we remember the different places that we have visited during our day ? Studies of brain activity in rats suggest that each place a rat visits is represented by a different pattern of activity in a region of the brain called hippocampus . However , it is not clear what role this “spatial remapping” plays in the formation of memories in humans . Kyle et al . used a technique called functional magnetic resonance imaging ( fMRI ) to investigate how our brain represents the different places we’ve visited , and how this is linked to how well we can remember these locations . For the experiments , human volunteers played a video game where they visited four virtual environments in a different order . Later , the volunteers were asked to remember details about the virtual environments they had visited while their brain activity was monitored using fMRI . The experiments show that in order to remember distinct locations – even if they have some features in common – the hippocampus produces patterns of activity that have very little overlap with each other . This process is termed “pattern separation” . Sometimes our memory confuses different locations , which could be due to a failure of the brain to distinguish between the patterns of brain activity that represent these locations . Kyle et al . ’s findings provide the first evidence for spatial remapping in the human hippocampus and its importance in forming memories of locations . The next steps are to find out how many different environments our brain might be able to store at the same time , and to identify factors that could aid in our memory for spatial locations .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Successful retrieval of competing spatial environments in humans involves hippocampal pattern separation mechanisms
Regeneration responses in animals are widespread across phyla . To identify molecular players that confer regenerative capacities to non-regenerative species is of key relevance for basic research and translational approaches . Here , we report a differential response in retinal regeneration between medaka ( Oryzias latipes ) and zebrafish ( Danio rerio ) . In contrast to zebrafish , medaka Müller glia ( olMG ) cells behave like progenitors and exhibit a restricted capacity to regenerate the retina . After injury , olMG cells proliferate but fail to self-renew and ultimately only restore photoreceptors . In our injury paradigm , we observed that in contrast to zebrafish , proliferating olMG cells do not maintain sox2 expression . Sustained sox2 expression in olMG cells confers regenerative responses similar to those of zebrafish MG ( drMG ) cells . We show that a single , cell-autonomous factor reprograms olMG cells and establishes a regeneration-like mode . Our results position medaka as an attractive model to delineate key regeneration factors with translational potential . The ability to regenerate individual cells , lost organs or even the structure of the entire body is widespread in the animal kingdom . The means by which certain species achieve remarkable feats of regeneration , whereas others have restricted or no capacity to do so is poorly understood . Teleost fishes are widely used models to study development , growth and regeneration of the visual system ( Centanin et al . , 2011; Raymond et al . , 1988 , 2006; Rembold et al . , 2006 ) . The retina of these fish undergoes lifelong neurogenesis , and the range of retinal cell types is generated from two sources . The first are the cells of the ciliary marginal zone ( CMZ ) , which include retinal stem cells that give rise to progenitor cells and ultimately differentiated cell types of the growing neural retina ( Centanin et al . , 2011 , 2014; Raymond et al . , 2006 ) . A second source for new retinal cells are Müller glia ( MG ) cells , which generate new cell types during homeostasis and regeneration ( Bernardos et al . , 2007 ) . Some teleost species , including goldfish ( Carassius auratus ) and zebrafish ( Danio rerio ) , have been analyzed with respect to their ability to regenerate the retina and recover visual function after injuries ( Bernardos et al . , 2007; Braisted and Raymond , 1992; Raymond et al . , 1988; Sherpa et al . , 2008 ) . Among these , zebrafish is the best-studied and has been shown to contain multipotent MG cells which can self-renew and regenerate all retinal neuronal and glial cell types after injuries . It is currently assumed that other teleost species possess the same regenerative capacities; however , detailed analyses have been lacking . To investigate MG cell-mediated retina regeneration in a distantly related teleost , we chose the Japanese ricefish medaka ( Oryzias latipes ) , which is a well-established model organism that shared its last common ancestor with zebrafish between 200 and 300 million years ago ( Schartl et al . , 2013 ) . Few regeneration studies have been carried out in medaka , but the literature reveals some interesting differences to zebrafish . Whereas fins can be fully regenerated in adult medaka ( Nakatani et al . , 2007 ) , the heart has no regenerative capacity ( Ito et al . , 2014; Lai et al . , 2017 ) . The development and growth of the neural retina of medaka has been studied ( Centanin et al . , 2011 , 2014; Martinez-Morales et al . , 2009 ) , but regeneration studies are missing . After injuries , multipotent MG cells of the zebrafish retina have been shown to upregulate the expression of pluripotency factors including lin-28 , oct-4 , c-myc and sox2 ( Ramachandran et al . , 2010 ) . Sox2 is well known for its role in maintaining the pluripotency of embryonic stem cells ( Masui et al . , 2007 ) and is one of the four original Yamanaka factors required for the generation of induced pluripotent stem cells ( Takahashi et al . , 2007 ) . Sox2 has been frequently used in reprogramming studies , such as the conversion of mouse and human fibroblasts directly into induced neural stem cells ( Ring et al . , 2012 ) , or the transformation of NG2 glia into functional neurons following stab lesions in the adult mouse cerebral cortex ( Heinrich et al . , 2014 ) . In the regenerating zebrafish retina , sox2 expression is upregulated 2 days post injury ( dpi ) and is necessary and sufficient for the MG proliferation associated with regeneration ( Ramachandran et al . , 2010; Gorsuch et al . , 2017 ) . In the present study , we find that medaka MG ( olMG ) cells display a restricted regenerative potential after injury and only generate photoreceptors ( PRCs ) . We observed that olMG cells can re-enter the cell cycle after injures but fail to divide asymmetrically or generate neurogenic clusters , two steps which are essential to full regeneration . Using in vivo imaging , two-photon mediated specific cell ablations and lineage tracing , we find that olMG cells react preferentially to injuries of PRCs and are only able to regenerate this cell type . We demonstrate that sox2 is expressed in olMG cells in the absence of injury but , in contrast to zebrafish , is not maintained in proliferating olMG cells after injury . We show that inducing targeted expression of sox2 in olMG cells is sufficient to shift olMG cells into a regenerative mode reminiscent of zebrafish , where they self-renew and regenerate multiple retinal cell types . In contrast to zebrafish and goldfish , where MG cells are described as the source of rod PRCs that gradually accumulate during the early larval period ( Bernardos et al . , 2007; Nelson et al . , 2008 ) , it has been shown previously that olMG cells are quiescent at a comparable developmental stage in the hatchling ( 8 dpf ) retina ( Lust et al . , 2016 ) . While the zebrafish retina massively increases its rod PRC number during post-embryonic growth ( Figure 1—figure supplement 1A–B''' ) via the proliferation of MG cells ( Bernardos et al . , 2007 ) , the medaka retina maintains its rod PRC layer from embryonic to adult stages ( Figure 1—figure supplement 1C–D''' ) and rod PRCs are born from the CMZ ( Figure 1—figure supplement 2 ) . In order to address the regenerative abilities of olMG cells we used the rx2::H2B-eGFP transgenic line that labels the CMZ , olMG cells and cone PRCs but no rods in hatchling ( 8dpf ) and adult medaka ( Martinez-Morales et al . , 2009; Reinhardt et al . , 2015 ) ( Figure 1—figure supplement 3 ) . To investigate the reaction of olMG cells and the retina upon injury , we performed needle injuries on rx2::H2B-eGFP transgenic fish . To label cells re-entering the cell cycle we subsequently analyzed the fish either by immunohistochemistry for the mitotic marker phospho-histone H3 ( PH3 ) at 3 dpi or incubated them in BrdU for 3 days to label cells in S-phase . We detected proliferating cells in the central retina , on the basis of both labels PH3 ( Figure 1A–1A’’ ) and BrdU ( Figure 1B–1B’’ ) 3 days after a needle injury . These proliferating cells were also positive for rx2-driven H2B-eGFP , showing that the olMG cells had re-entered the cell cycle . These results demonstrate that olMG cells in hatchling medaka are quiescent in an uninjured background ( Lust et al . , 2016 ) , but begin to proliferate upon injury . The onset of MG proliferation in zebrafish has been observed between 1 and 2 dpi ( Fausett and Goldman , 2006 ) . To understand if olMG cells show a similar mode of activation , we performed BrdU incorporation experiments and analyzed time-points after injury ranging from 1 dpi until 3 dpi . At 1 dpi , no BrdU-positive cells were detected in the retina ( data not shown ) . At 2 dpi , the first BrdU-positive cells were detected in the inner nuclear layer ( INL ) and the outer nuclear layer ( ONL ) of the central retina ( Figure 1—figure supplement 4A–B''' ) . Co-localization with GFP showed that these cells are olMG cells or olMG-derived cells ( Figure 1—figure supplement 4A–B''' ) . In response to injury olMG cells initiate DNA synthesis and divide maximally once as indicated by the appearance of single or a maximum of two BrdU-positive cells next to each other in the INL at both 2 dpi and 3 dpi ( Figure 1C and C' ) . In contrast , the injury response of zebrafish MG ( drMG ) cells at comparable stages ( 4dpf ) is characterized by the formation of large nuclear , neurogenic clusters in the INL ( Figure 1D and D' ) . This is consistent with the response of adult drMG cells to injury in which a single asymmetric division produces a MG cell and a progenitor cell that divides rapidly to generate neurogenic clusters ( Nagashima et al . , 2013 ) . These results show that olMG cells start re-entering the cell cycle between 1 and 2 dpi but do not generate neurogenic clusters . For proper regeneration to occur , the appropriate cell types must be produced . This requires not only the regulation of the proliferation of stem or progenitor cells , but also the proper control of lineage decisions in the progenitors . If and when fate decisions are made by the MG cells or proliferating progenitors during regeneration is largely unknown . To study whether different injury sites ( PRC or retinal ganglion cell ( RGC ) injury ) result in a differential response of olMG cells , we used two-photon mediated ablations and consecutive imaging ( Figure 2—figure supplement 1A–D ) and addressed their behavior in immediate ( up to 30 hours post injury ( hpi ) ) and late ( until 6 dpi ) response to injury . We induced PRC injuries in medaka and observed that olMG nuclei below the wound site started migrating apically at 17 hpi ( Figure 2A–2A''' , see also Figure 2—video 1 ) . These migrations were not coordinated between individual cells . Some nuclei migrated into the ONL , whereas others stayed at the apical part of the INL . Nuclei farther from the wound site did not migrate in response to the injuries . In contrast , after RGC injuries , there was no migration of olMG nuclei , either apically or basally toward the wound , within the first 30 hpi ( Figure 2B–2B''' , see also Figure 2—video 1 ) . To investigate whether olMG nuclei migrate back at later time-points after PRC injuries or show any migratory behavior after RGC injuries , we re-imaged the injury site at 2-day intervals to follow an injured retina up to 6 dpi . At 2 dpi , retinae with PRC injuries showed a gap in the INL below the injury site , at a position where olMG nuclei are normally found , reflecting the migration of olMG nuclei towards the ONL from this location ( Figure 2C–2C'' ) . The gap in the INL persisted until 6 dpi ( Figure 2C'' ) . The reaction of olMG cells in retinae with RGC injuries differed . Here , we neither observed an apical nor basal migration of olMG nuclei ( Figure 2D–2D'' ) and in fact no migration of olMG nuclei was observed at all until 6 dpi . To rule out that this is due to too little damage in the RGC layer we increased the injury size . This led to swelling and secondary cell death of PRCs and activated olMG nuclei to migrate apically ( Figure 2—figure supplement 2A–B ) , indicating further that their preferential reaction is toward PRC injuries . Taken together , these results show that olMG nuclei migrate toward PRC injury sites within 24 hpi and remain in this location up until 6 days , whereas they display no discernible reaction toward RGC injuries . This indicates a clear preferential reaction of olMG nuclei to refill the injured PRC layer . Long-term in vivo imaging of fish that were injured in the ONL made it apparent that olMG nuclei migrate apically into the wound site but remain there which might indicate a complete remodeling of the soma of these neuroepithelial cells . To understand whether cell bodies of the olMG cells remain intact during this nuclear migration , we analyzed nuclear movements ( transgenic line rx2::H2B-eGFP ) in the context of the olMG cell body ( transgenic line rx2::lifeact-eGFP ) . We imaged the double transgenic animals at 2-day intervals following ONL injuries . As previously observed , olMG nuclei migrated out of the INL into the wound site ( Figure 3A–3A'' ) . The rx2::lifeact-eGFP labeled cell bodies of the olMG cells spanning the entire apico-basal distance remained intact until 6 dpi in the absence of an apparent ( labeled ) nucleus in the INL ( Figure 3A'' ) . The earlier position of the nucleus was still recognizable by a slight enlargement of the soma . Additionally , to extend the range of analysis , we performed immunohistochemistry on injured fish at 3 and 10 dpi . After injury , incubation in BrdU for 3 days and direct fixation at 3 dpi we found that at the site of injury the GFP-positive cell bodies , labeled by rx2::lifeact-eGFP , did not contain a nucleus anymore while the neighboring , more distant GFP-positive cell bodies contained elongated olMG nuclei ( Figure 3B–B'''' , see also Figure 3—video 1 ) . After incubation in BrdU for 3 days and fixation at 10 dpi , we observed similar results ( Figure 3C–3C'' ) . Here , we used immunohistochemistry to detect the olMG cell bodies via a GS-staining ( Figure 3C ) . BrdU-positive cells in the ONL mark the site of the injury ( Figure 3C'' ) . In the region directly underneath the site of injury , the majority of olMG nuclei , which had been labeled by rx2::H2B-eGFP , were absent from the INL ( Figure 3C'' ) . GS-positive cell bodies remained spanning the entire apico-basal height , but without the apparent presence of nuclei . In contrast , unaffected GS-positive olMG cells located on either side of the wound site still contained their nuclei , as easily detected by the large size of the soma . This data shows that the cell bodies of injury-activated olMG cells are still intact despite the migration of their nuclei into the ONL . Since the injury response of olMG cells apparently does not involve self-renewal of olMG cells we wondered about the position and orientation of the cell division plane , a factor which has been associated with cell fate in various systems . We first addressed the apico-basal position of dividing olMG cells by PH3 immunohistochemistry after injury . We detected PH3-positive cells only in the INL ( Figure 4A–4A'' ) . Some dividing cells were located more apically ( Figure 4A–4A'' ) , while others were located more basally ( Figure 4—figure supplement 1A–B ) . This is in contrast to findings in zebrafish where , in a light injury paradigm , PH3-positive drMG cells can be found in the ONL 2 days after injury ( Nagashima et al . , 2013 ) . To address the cleavage plane of dividing olMG cells , we employed in vivo imaging of rx2::H2B-eGFP fish , which permits visualizing the separation of chromatin and thus gives a measurement of the orientation of division . The first injury-triggered olMG divisions were observed at 44 hpi ( Figure 4B–4B''' , see also Figure 4—video 1 ) . They occurred in the INL , both in the center and close to the ONL ( Figure 4—figure supplement 1C–C''' ) . The mode of division was preferentially apico-basal ( 5 out of 6 divisions in 5 out of 6 animals ) , while only a single horizontal division was observed ( 1 out of 6 divisions in 1 out of 6 animals ) . In contrast , drMG cells are reported to predominantly divide with a horizontal division plane ( Lahne et al . , 2015 ) . These results show that injury-induced olMG cell divisions occur in different positions in the INL and have a strong preference to occur apico-basally . In zebrafish , drMG cells are able to regenerate all neuronal cell types and self-renew after injury ( Nagashima et al . , 2013; Powell et al . , 2016 ) . We followed a BrdU-based lineage-tracing approach successfully applied in zebrafish ( Fausett and Goldman , 2006; Powell et al . , 2016 ) to address the potency of olMG cells . Transgenic rx2::H2B-eGFP fish retinae were injured either by two-photon laser ablation of PRCs or RGCs specifically or using a needle ablating all cell types . The injured fish were incubated in BrdU for 3 days to label proliferating cells . This allows to efficiently detect all injury-triggered S-phase entry of olMG cells ( Figure 5—figure supplement 1A–D ) . For lineaging , fish were grown until 14 dpi to allow a regeneration response and subsequently analyzed for BrdU-positive cells in the different retinal layers ( Figure 5A ) . PRC injuries led to the detection of 97% of all BrdU-positive cells in the ONL , mostly in the rod nuclear layer , indicative for PRC fate ( Figure 5B and E ) . No BrdU-positive cells could be detected in the INL . Additionally , we found that the INL below the injury site was devoid of olMG cell nuclei , both consistently arguing for the absence of injury-triggered olMG self-renewal . Strikingly , RGC injuries did not trigger BrdU-uptake in olMG cells or any other differentiated cell type ( data not shown ) . Needle injuries affecting all retinal cell types triggered the same response as the specific lesions in the PRC layer . 97% of all BrdU-positive cells were present in the ONL , and only a single BrdU-positive olMG cell was found in 1 of 10 fish ( Figure 5C and E ) . Also later application of BrdU after injury ( 4 to 7 dpi ) did not result in BrdU-positive olMG cells ( Figure 5—figure supplement 2A–C ) . Importantly , BrdU-positive nuclei were not positive for GS , indicating that they were not olMG cells anymore ( Figure 5D ) , but were positive for Recoverin , a PRC marker ( Figure 5E ) . These results demonstrate that olMG cells do not self-renew and rather function as mono-potent repair system restricted to the generation of PRCs , most of which belong to the rod lineage . The previous results show that olMG cells re-enter the cell cycle after injuries introduced by needle to the complete retina or by two-photon ablation to the PRC layer . They regenerate PRCs but do not undergo self-renewal . This suggests that olMG cells lack intrinsic factors that trigger self-renewal and multipotency upon injury . One transcription factor which is well known for its involvement in the self-renewal of stem cells – particularly neural stem cells – is Sox2 ( Sarkar and Hochedlinger , 2013 ) . It has been shown that cells expressing sox2 are capable of both self-renewal and the production of a range of differentiated neuronal cell types ( Sarkar and Hochedlinger , 2013 ) . Data from zebrafish have shown that a ubiquitous gain of sox2 expression triggers a proliferative response of drMGs in the absence of injury ( Gorsuch et al . , 2017 ) . To investigate the expression of sox2 in MG cells , we performed immunohistochemistry on uninjured retinae in medaka and zebrafish . In the medaka retina , Sox2 protein is detected in amacrine cells ( ACs ) and olMG cells in the central retina ( Figure 6A–6A''' ) . In zebrafish , the pattern is similar: Sox2 protein is present in ACs and drMG cells in the central retina ( Figure 6B–6B''' ) . This data is consistent with data from other vertebrates including human , whose MG cells also maintain sox2 expression ( Gallina et al . , 2014 ) . To investigate the expression of sox2 in the olMG and drMG cells responding to injury by proliferation , we performed needle injuries , incubated the fish in BrdU and fixed them at 3 dpi . We could detect BrdU-positive MG cells both in medaka and zebrafish . The vast majority of proliferating olMG cells did not express sox2 anymore at 3 dpi ( Figure 6C–6D , 6% of all BrdU-positive cells were Sox2-positive ) . We saw a similar scenario in response to either PRC or RGC injury , where 9% and 10% respectively of all BrdU-positive cells were Sox2-positive . Conversely , in zebrafish , sox2 expression was still detected after 3 days in drMG cells that proliferated in response to needle injury ( Figure 6E–6F , 84% of all BrdU-positive cells were Sox2-positive ) . These findings strongly argue that the downregulation of sox2 expression in proliferating olMG cells restricts their regenerative properties . The results presented above indicate that after injury , olMG cells and olMG-derived progenitors do not maintain the expression of sox2 , in contrast to the situation in zebrafish . We hypothesize that the prolonged sox2 expression facilitates drMG cells to undergo self-renewal and to generate neurogenic clusters and ultimately all cell types necessary to regenerate a functional retina . To test this hypothesis , we chose the inducible LexPR transactivation system ( Emelyanov and Parinov , 2008 ) targeted to olMG cells ( rx2::LexPR OP::sox2 , OP::H2B-eGFP ) to sustain sox2 expression . In mifepristone-treated retinae , we detected increased levels of Sox2 protein in induced olMG cells ( Figure 7A–B'' ) . To address the proliferative behavior of Sox2-sustaining olMG cells in response to injury , we treated fish with mifepristone and BrdU for 2 days , performed a needle injury , maintained the fish in mifepristone and BrdU until 3 dpi and analyzed immediately ( Figure 7C ) . We observed increased formation of proliferating clusters as well as the distribution of BrdU-positive cells in all layers of the retina after needle injury ( 4 out of 6 fish ) ( Figure 7D–E' ) . To address the long-term potential of Sox2-induced olMG cells , we ablated all retinal cell types by needle injury and performed BrdU-mediated lineage tracing as described above . We induced sox2 expression for 2 days and provided BrdU in parallel , performed a needle injury and maintained the expression of sox2 until 3 dpi . After a chase until 14 dpi , the retinae and regenerated cell types were analyzed ( Figure 8A and B ) . In needle-injured wild-type fish which were also treated with mifepristone as well as in transgenic fish ( rx2::LexPR OP::sox2 , OP::H2B-eGFP ) which were not treated with mifepristone , olMG cells did not self-renew and gave predominantly rise to PRCs , mostly rod PRCs ( Figure 8F ) . In contrast , olMG cells experiencing persistent expression of sox2 showed self-renewal and differentiation into different cell types in the ONL and INL as indicated by BrdU lineage tracing . In particular , the olMG cells maintaining sox2 expression after injury regenerated olMG cells ( Figure 8C–C'' and F ) and exhibited a significant increase in regenerated ACs and RGCs , which were positive for HuC/D ( Figure 7D–E'' and 8F ) . Furthermore , a slight increase in cone PRCs and a decrease in rod PRCs was observed after sox2 induction ( Figure 8C–F ) . These data indicate that a targeted maintenance of sox2 expression after injury is sufficient to induce self-renewal and increase potency in olMG cells in the medaka retina turning a mono-potent repair system into a regeneration system with increased potency . Here , we have characterized a differential regenerative response between two teleost fish and used it as a framework to address the molecular determinants of regeneration during evolution . By using a combination of in vivo imaging , targeted cell type ablation and lineage tracing , we investigated the dynamics of the injury response in the medaka retina . We focused on MG cells , which play a prominent role in zebrafish retinal regeneration . While upon injury olMG cells re-enter the cell cycle , they fail to undergo self-renewal . Furthermore , olMG cells do not generate the neurogenic clusters which arise in zebrafish , nor do they produce all neuronal cell types in the retina . We traced this effect prominently to Sox2 , the expression of which is maintained in proliferating drMG cells after injury , but not in olMG cells . We demonstrated that the sustained expression of sox2 is sufficient to convert an olMG into a dr-like MG cell . The fact that this response is acquired cell-autonomously and in the context of a non-regenerative retina can be relevant for putative translational approaches . Since olMG cells do not self-renew after injuries and only have the capacity to regenerate PRCs , olMG cells are not true multipotent retinal stem cells . Instead , olMG cells should be considered lineage-restricted progenitors . They re-enter the cell cycle between 1 and 2 dpi , similar to the re-entry observed in zebrafish . This indicates that the signals that are essential for cell cycle re-entry are present in medaka and are activated in a window of time similar to that in zebrafish . After retinal injures olMG nuclei migrate into the PRC layer; the cell bodies of nucleus-depleted olMG cells are maintained in the retina . This could be important since MG cell bodies play a crucial role in mechanical stability of the retina ( MacDonald et al . , 2015 ) as well as light guiding through the retina ( Franze et al . , 2007 ) . After retinal injures , olMG cell bodies were maintained in the absence of a nucleus in the INL reflecting the necessity to preserve this structural and optical element . In the uninjured retina , olMG cells express sox2 , as is the case for many other vertebrates , including humans . However , sox2 expression in olMG cells is downregulated in response to injury , in contrast to the injury response of drMG cells , which upregulate sox2 ( Gorsuch et al . , 2017 ) . We speculate that this upregulation is due to epigenetic modifications of the sox2 locus . A recent study in the mouse retina showed that the expression of oct4 is upregulated shortly after injury and then downregulated at 24 hpi ( Reyes-Aguirre and Lamas , 2016 ) . This correlates with a decrease in the expression of DNA methyltransferase 3b and its subsequent upregulation at 24 hpi , triggering a decrease in methylation and subsequent re-methylation of oct4 . Furthermore , a recent study on zebrafish regeneration discovered the existence of so-called tissue regeneration enhancer elements ( TREEs ) ( Kang et al . , 2016 ) . One TREE was associated with leptin b , which is expressed in response to injuries of the fin and heart . This TREE acquires open chromatin marks after injury , can be divided into tissue-specific modules and can drive injury-dependent expression in mouse tissue . This raises the possibility that the sox2 locus in olMG cells experiences epigenetic modifications after injury which differ from modifications in zebrafish . The fact that sox2 expression is detected in all vertebrate MG cells analyzed to date in the absence of injury raises the question whether a decrease in sox2 expression after injury might be a common feature of non-regenerative species like chicken , mouse and even humans . Data from a conditional sox2 knockout in mouse shows that Sox2 is necessary for maintenance of MG morphology and quiescence ( Surzenko et al . , 2013 ) . While its expression is maintained in response to the injection of growth factors after retinal damage ( Karl et al . , 2008 ) its regulation in response to injury alone has not been described . Data obtained in cultures of human MG cells ( Bhatia et al . , 2011 ) provide additional important insights . Strikingly similar to medaka , silencing the expression of sox2 caused MG cells to lose stem and progenitor cell markers and adopt a neural phenotype ( Bhatia et al . , 2011 ) . These findings align well with the results from medaka presented here und suggest that olMG cells and their behavior as progenitor cells can serve as a model for mammalian and in particular human MG cells . The results shown here may provoke an evolutionary question: is retinal regeneration an ancestral or derived feature within the infraclass of teleosts ? The question might be resolved by investigations of this capacity in other fish species more closely related to medaka , such as Xiphophorus maculatus , whose last common ancestor with medaka lived around 120 million years ago ( Schartl et al . , 2013 ) . Additionally , species like the spotted gar , whose lineage diverged from teleosts before teleost genome duplication ( Braasch et al . , 2016 ) , might provide insights about the ancestral mode of retinal regeneration . Recently , the retinal architecture of the spotted gar has been analyzed ( Sukeena et al . , 2016 ) . There , proliferative cells have been detected in the central retina likely representing proliferating MG cells , which generate rod PRCs during homeostasis as seen in zebrafish , suggesting that regeneration is indeed an ancestral feature in the ray-finned fish lineage . Additionally , one wonders whether the ability of MG cells to regenerate injured retinal cells is directly related to their involvement in rod genesis during post-embryonic growth and conversely whether the lack of regenerative abilities of MG cells is a result of the lack of rod genesis ? The differences in rod layer increase in zebrafish and medaka as well as the differences in rod genesis by MG cells might be due to the natural habitats and photic environment of the fish . While larval zebrafish live near the water surface , adults live in deeper waters where rods become more important for visual function ( Lenkowski and Raymond , 2014 ) . Medaka on the other hand are surface fish their entire life , since they live in shallow waters like rice paddies ( Kirchmaier et al . , 2015 ) which decreases the need for a massive increase in rods . With a potential translational perspective , regenerating and non-regenerating systems can now be systematically compared to delineate the underlying factors and mechanisms . To date , our cumulative results show that the regenerative potential of olMG cells in the context of homeostasis and injury in medaka resemble that of mammals and birds more than zebrafish . We propose that this provides an added value to medaka as a model species for regeneration studies that bridge the differences between zebrafish and mammals . Studies of heart regeneration that have compared zebrafish and medaka lend additional support to this statement ( Ito et al . , 2014; Lai et al . , 2017 ) . As reprogrammable multipotent retinal stem cells , MG cells harbor a great potential for treating degenerative retinal diseases . Our work indicates that the addition of a single reprogramming factor facilitates a regeneration-like response mediated by olMG cells . Their multiple resemblances of features of mammalian and human MG cells position them as an ideal model for the development of new treatments preventing the degeneration and initiating the regeneration of the retina . Medaka ( Oryzias latipes ) and zebrafish ( Danio rerio ) used in this study were kept as closed stocks in accordance to Tierschutzgesetz 111 , Abs . 1 , Nr . 1 and with European Union animal welfare guidelines . Fish were maintained in a constant recirculating system at 28°C on a 14 hr light/10 hr dark cycle ( Tierschutzgesetz 111 , Abs . 1 , Nr . 1 , Haltungserlaubnis AZ35–9185 . 64 and AZ35–9185 . 64/BH KIT ) . The following stocks and transgenic lines were used: wild-type Cabs , rx2::H2B-eGFP , rx2::lifeact-eGFP , rx2::H2B-eGFP QuiH , rx2::LexPR OP::sox2 OP::H2B-eGFP cmlc2::CFP , rx2::CreERT2 , GaudíRSG ( Reinhardt et al . , 2015 ) , AB zebrafish and Albino zebrafish . All transgenic lines were created by microinjection with Meganuclease ( I-SceI ) in medaka embryos at the one-cell stage , as previously described ( Thermes et al . , 2002 ) . For BrdU incorporation , fish were incubated in 2 . 5 mM BrdU diluted in 1x Embryo Rearing Medium ( ERM ) or 1x Zebrafish Medium for respective amounts of time . For induction of the LexPR system , fish were induced by bathing them in a 5 µM to 10 µM mifepristone solution in 1x ERM for respective times . For induction of the Cre/lox system , fish were treated with a 5 µM tamoxifen solution in 1x ERM over night . For in vivo imaging fish in a Cab background were kept in 5x 1-phenyl-2-thiourea ( PTU ) in 1x ERM from 1 dpf until imaging to block pigmentation . Fish in a QuiH background could be imaged without any treatment . Fish were anesthetized in 1x Tricaine diluted in 1x ERM and mounted in glass-bottomed Petri dishes ( MatTek Corporation , Ashland , MA ) in 1% Low Melting Agarose . The specimens were oriented lateral , facing down , so that the right eye was touching the cover-slip at the bottom of the dish . Imaging and laser ablations were performed on a Leica ( Germany ) SP5 equipped with a Spectra Physics ( Santa Clara , California , USA ) Mai Tai HP DeepSee Ti:Sapphire laser , tunable from 690 to 1040 nm and Leica Hybrid Detectors . A wound was introduced using the bleach point function or the region of interest function , together with the high-energy two-photon laser tuned to 880 nm . The wound size was defined between 40 and 60 µm diameter for medium-sized wounds . Wounds bigger than 60 µm diameter were defined as large wounds . Rx2::H2B-eGFP or rx2::lifeact-eGFP fish were used for the ablations . Since rx2 is expressed during retinal development residual GFP could be visualized in rod PRCs as well as in RGCs when increasing the gain of the Hybrid Detectors . Follow-up imaging was performed using same laser at 880 nm and a 40x objective . Larvae ( zebrafish 5 dpf , medaka 8 dpf ) were anesthetized in 1x Tricaine in 1x ERM and placed on a wet tissue . Under microscopic visualization , the right retina was stabbed multiple times in the dorsal part with a glass needle ( 0 . 05 mm diameter ) . Left retinae were used as controls . Fish were euthanized using Tricaine and fixed over night in 4% PFA , 1x PTW at 4°C . After fixation samples were washed with 1x PTW and cryoprotected in 30% sucrose in 1x PTW . To improve section quality , the samples were incubated in a half/half mixture of 30% sucrose and Tissue Freezing Medium ( Leica ) for at least 3 days . 16-µM-thick serial sections were obtained on a Leica cryostat . Sections were rehydrated in 1x PTW for 30 min at room temperature . Blocking was performed for 1–2 hr with 10% NGS ( normal goat serum ) in 1x PTW at room temperature . The respective primary antibodies were applied diluted in 1% NGS o/n at 4˚C . The secondary antibody was applied in 1% NGS together with DAPI ( 1:500 dilution in 1xPTW of 5 mg/ml stock ) for 2–3 hr at 37˚C . Slides were mounted with 60% glycerol and kept at 4°C until imaging . Primary antibodySpeciesConcentrationCompanyAnti-BrdUrat1:200AbD Serotec , BU1/75Anti-eGFPchicken1:500Thermo Fisher , A10262Anti-HuC/Dmouse1:200Thermo Fisher , A21271Anti-GSmouse1:500Chemicon , MAB302Anti-pH3 ( Ser10 ) rabbit1:500Millipore , 06–570Anti-Recoverinrabbit1:200Millipore , AB5585Anti-Sox2rabbit1:100Genetex , GTX101506Anti-Zpr-1mouse1:200ZIRC Secondary antibodySpeciesConcentrationCompanyAnti-chicken Alexa Fluor 488donkey1:750Jackson , 703-485-155Anti-mouse Alexa 546goat1:750Thermo Fisher , A-11030Anti-mouse Cy5donkey1:750Jackson , 715-175-151Anti-rabbit DyLight549goat1:750Jackson , 112-505-144Anti-rabbit 647goat1:750Thermo Fisher , A-21245Anti-rat DyLight549goat1:750Jackson , 112-505-143Anti-rat Alexa633goat1:750Thermo Fisher , A-21094 BrdU antibody staining was performed with an antigen retrieval step . After all antibody stainings and DAPI staining , except for BrdU , were complete , a fixation for 30 min was performed with 4% PFA . Slides were incubated for 1 hr at 37°C in 2 N HCl solution , and pH was recovered by washing with a 40% Borax solution before incubation with the primary BrdU antibody . TUNEL stainings on cryosections were performed after all other antibody stainings were completed using the In Situ Cell Death Detection Kit TMR Red by Roche . Stainings were performed according to the manufacturers protocol with the following modifications . Washes were performed with 1x PTW instead of PBS . All immunohistochemistry images were acquired by confocal microscopy at a Leica TCS SPE with either a 20x water objective or a 40x oil objective . Images were processed via Fiji image processing software . Statistical analysis and graphical representation of the data were performed using the Prism software package ( GraphPad ) . Box plots show the median , 25th and 75th percentiles; whiskers show maximum and minimum data points . Unpaired t-tests were performed to determine the statistical significances . The p-value p<0 . 05 was considered significant and p-values are given in the figure legends . Sample size ( n ) and number of independent experiments are mentioned in every figure legend . No statistical methods were used to predetermine sample sizes , but our sample sizes are similar to those generally used in the field . The experimental groups were allocated randomly , and no blinding was done during allocation .	All animals have at least some ability to repair their bodies after injury . But certain species can regenerate entire body parts and even internal organs . Salamanders , for example , can regrow their tail and limbs , as well as their eyes and heart . Many species of fish can also regenerate organs and tissues . In comparison , mammals have only limited regenerative capacity . Why does regeneration vary between species , and is it possible to convert a non-regenerating system into a regenerating one ? Laboratory studies of regeneration often use the model organism , zebrafish . Zebrafish can restore their sight after an eye injury by regenerating the retina , the light-sensitive tissue at the back of the eye . They are able to do this thanks to cells in the retina called Müller glial cells . These behave like stem cells . They divide to produce identical copies of themselves , which then transform into all of the different cell types necessary to produce a new retina . Lust and Wittbrodt now show that a distant relative of the zebrafish , the Japanese ricefish ‘medaka’ , lacks these regenerative skills . Although Müller glial cells in medaka also divide after injury , they give rise to only a single type of retinal cell . This means that these fish cannot regenerate an entire retina . Lust and Wittbrodt demonstrate that in medaka , but not zebrafish , levels of a protein called Sox2 fall after eye injury . As Sox2 has been shown to be important for regeneration in zebrafish Müller glial cells , the loss of Sox2 may be preventing regeneration in medaka . Consistent with this , restoring Sox2 levels in medaka Müller glial cells enabled them to turn into several different types of retinal cell . Sox2 is also present in the Müller glial cells of other species with backbones , including chickens , mice , and humans . Future experiments should test whether loss of Sox2 after injury contributes to the lack of regeneration in these species . If it does , the next question will be whether restoring Sox2 can drive a regenerative response .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine" ]	2018	Activating the regenerative potential of Müller glia cells in a regeneration-deficient retina
Joubert syndrome ( JBTS ) is a severe recessive neurodevelopmental ciliopathy which can affect several organ systems . Mutations in known JBTS genes account for approximately half of the cases . By homozygosity mapping and whole-exome sequencing , we identified a novel locus , JBTS23 , with a homozygous splice site mutation in KIAA0586 ( alias TALPID3 ) , a known lethal ciliopathy locus in model organisms . Truncating KIAA0586 mutations were identified in two additional patients with JBTS . One mutation , c . 428delG ( p . Arg143Lysfs4 ) , is unexpectedly common in the general population and may be a major contributor to JBTS . We demonstrate KIAA0586 protein localization at the basal body in human and mouse photoreceptors , as is common for JBTS proteins , and also in pericentriolar locations . We show that loss of TALPID3 ( KIAA0586 ) function in animal models causes abnormal tissue polarity , centrosome length and orientation , and centriolar satellites . We propose that JBTS and other ciliopathies may in part result from cell polarity defects . Joubert syndrome ( JBTS ) is a rare ciliopathy characterized by a specific midhindbrain malformation presenting as ‘molar tooth sign’ on axial MRI . Patients typically have a perturbed respiratory pattern in the neonatal period and pronounced psychomotor delay . Depending on the genetic subtype , there may be additional retinal degeneration , nephronophthisis , liver fibrosis , and skeletal abnormalities ( such as polydactyly ) . JBTS is genetically heterogeneous , with recessive mutations reported in more than 20 genes encoding proteins related to the function of cilia and associated structures ( Romani et al . , 2013; Bachmann-Gagescu et al . , 2015 ) . Cilia are axoneme-based organelles which protrude into the extracellular milieu , anchored to the cell by a modified centriole ( basal body ) . They are present in virtually every cell type ( Christensen et al . , 2007 ) . Non-motile ‘primary’ cilia play essential roles in mechanotransduction , chemosensation , and intracellular signal transduction , including Hedgehog ( Hh ) , PDGFα , and WNT pathways , in embryonic development and adult tissue homeostasis ( Goetz and Anderson , 2010 ) . In addition , highly modified and specialized cilia constitute the light-sensitive outer segments of retinal photoreceptor cells . Dysfunction of cilia , centrioles of basal bodies , and centrosomes can lead to a spectrum of developmental single- or multi-organ disorders termed ‘ciliopathies’ ( Bettencourt-Dias et al . , 2011 ) . KIAA0586 ( TALPID3; MIM #610178 , MIM #000979-9031 ) is essential for vertebrate development and ciliogenesis . The KIAA0586 ( TALPID3 ) protein is localized at the centrosome in human , chicken , mouse , and zebrafish cells ( Yin et al . , 2009; Ben et al . , 2011; Wu et al . , 2014 ) , and in particular , at the distal end of the mother centriole—the basal body of cilia ( Kobayashi et al . , 2014a ) . In model organisms , KIAA0586 null mutations cause failure of basal body docking and loss of cilia , leading to early embryonic lethal phenotypes ( Davey et al . , 2006; Bangs et al . , 2011; Ben et al . , 2011; Stephen et al . , 2013 ) . KIAA0586 ( TALPID3 ) binding partners include PCM1 , Cep120 , and CP110 , which interact with a known JBTS protein , CEP290 ( Tsang and Dynlacht , 2013 ) . Here , we report three JBTS families with loss-of-function mutations in KIAA0586 . Using animal models , we demonstrate that TALPID3 ( KIAA0586 ) is not only essential for transduction of Hedgehog signaling but plays an important role in centrosomal localization , orientation , and length . Finally , and beyond its established requirement for ciliogenesis , TALPID3 ( KIAA0586 ) plays a key role in cell and tissue polarity . The diagnosis of JBTS was based on the presence of a molar tooth sign in all three families . Family 1 ( Figure 1A ) is a consanguineous Kurdish family from northeast Syria . The two affected siblings were examined at the age of 6 years and 10 months ( MR026-01 ) and 2 years and 2 months ( MR026-04 ) , respectively . Pregnancy , delivery , and birth parameters of both children were unremarkable . In the neonatal period , both were hypotonic and weepy . Motor and speech development in MR026-01 were delayed , and his IQ was estimated to be between 50 and 70 . Further symptoms were severe myopia , scoliosis , brachydactyly , distinct facial characteristics , and recurrent febrile seizures . Height was reduced ( 108 cm , −2 . 6 SD ) , weight was normal ( 22 kg , −0 . 27 SD ) , and head circumference was increased ( 57 cm , +2 . 3 SD ) . MR026-04 had not reached any milestones , and at the age of 7 years , she was wheelchair-bound . Cognitive abilities were weaker than in her brother , with an IQ estimated to be below 35 . MR026-04 had similar physical characteristics as her brother , severe muscular hypotonia , prolonged and therapy-resistant seizures since the age of 14 months , and hypothyroidism . At the time of examination , her height was 91 cm ( 1 SD ) , weight was 11 . 5 kg ( −0 . 7 SD ) , and there was macrocephaly ( head circumference of 59 cm , +8 SD ) . 10 . 7554/eLife . 08077 . 003Figure 1 . Patients with Joubert syndrome ( JBTS ) and KIAA0586 mutations ( A–C ) . ( WT , wildtype; M , mutation ) . The ‘molar tooth sign’ in cranial axial MRI is indicated by arrows . ( A ) Family 1: Homozygosity mapping yielded eight homozygous chromosomal candidate regions ( not shown ) , including the JBTS23 locus comprising KIAA0586 . Patients MR026-01 and MR026-04 carry a homozygous splice site mutation , c . 2414-1G>C . ( B ) Patient MD1 of Family 2 is compound heterozygous for two truncating mutations , including the prevalent c . 428delG ( p . Arg143Lysfs4 ) allele . ( C ) Family 3: Patient G2 is double heterozygous for c . 428delG in KIAA0586 , and a frameshift mutation in KIF7 ( JBTS12; c . 811delG , p . Glu271Argfs51 ) . He also carries three potentially pathogenic variants in the ciliopathy genes CEP41 , KIF14 , and WDPCP ( blue ) . ( D ) Genomic structure of KIAA0586 with mutations in exons 5 and in/adjacent to exon 18 indicated . The gel electrophoresis shows the aberrant transcripts due to c . 2414-1G>C . ( E ) Scheme of human KIAA0586 protein and predicted consequences of JBTS-associated mutations . Orange color: unrelated residues included due to frameshift mutations . The third coiled-coil domain is the counterpart of the functionally essential fourth coiled-coil domain in chicken ( framed in red ) . ( F ) Chicken TALPID3 ( KIAA0586 ) is highly similar to the human protein . The talpid3 mutation results in an early frameshift and loss of three coiled-coil domains , including the fourth one . The in-frame deletion of exons 11 and 12 of mouse KIAA0586 ( 2700049A03Rik ) is depicted above the scheme of the chicken ortholog . DOI: http://dx . doi . org/10 . 7554/eLife . 08077 . 00310 . 7554/eLife . 08077 . 004Figure 1—figure supplement 1 . Analysis of potential interactions between Talpid3/TALPID3 , Kif7/KIF7 and IFT57 in the mouse and in chicken . Biallelic KIF7 mutations cause JBTS type 12 in human ( Dafinger et al . , 2011 ) . Although both the KIAA0586 mutation c . 428delG and the KIF7 mutation c . 811delG were paternally inherited in patient G2 , we sought to test for subtle abnormalities resulting from this double heterozygosity . In addition , we had previously found through a microarray analysis of talpid3 limb buds that IFT57 , a protein associated with ciliopathy phenotypes in mice and zebrafish , is downregulated in talpid3 embryos ( Bangs et al . , 2011 ) . Using in ovo complementation of the talpid3 neural tube , we could not detect induction of ISLET1 expression or ventralized PAX7 in the wildtype or talpid3 neural tube by overexpression of KIF7 or IFT57 ( Figure 1—figure supplement 1A , B ) . We then used siRNA constructs against KIF7 to model a heterozygous loss of KIF7 in the TALPID3+/− neural tube . Knock-down with two siRNA constructs had a weak effect on neural tube patterning compared to the mouse KIF7−/− knockout ( Cheung et al . , 2009 , Liem et al . , 2009 ) . Although the NKX2 . 2 expression domain could be marginally expanded in wildtype embryos ( not shown ) , there was no expansion of ISLET1-positive motorneuron progenitors in wildtype or TALPID3+/− embryos . PAX7 , however , was weakly dorsalized in both wildtype and talpid3+/− embryos ( Figure 1FC ) . These results suggested that some KIF7 function may be cilia-independent as has been suggested ( Liem et al . , 2009 ) . To more precisely investigate for a possible epistatic relationship between Kif7 and Talpid3 , particularly in the organs primarily affected in JBTS , such as the cerebellum , we undertook a Talpid3+/− × Kif7+/− mouse cross in order to determine if double Talpid3+/−/Kif7+/− heterozygous animals had brain patterning malformations . We first dissected embryos at E15 . 5 , 16 . 5 , and 17 . 5 and found that Talpid3+/−/Kif7+/− embryos were morphologically normal , including size , situs and limb patterning . MEFs derived from E12 . 5 embryos were normally ciliated , with the percentage of ciliated cells and cilia length comparable to those seen in wildtype , Talpid3+/+/Kif7+/− and Talpid3+/−/Kif7+/+cells ( Figure 1—figure supplement 1F , G ) . MRI and sectioning of the brain also showed no brain patterning abnormalities ( Figure 1—figure supplement 1D , E ) . Subsequently , Talpid3+/−/Kif7+/− animals were born and grew normally compared to their litter mates and showed no abnormal brain morphology ( Figure 1—figure supplement 1 ) . We conclude that KIAA0586 ( TALPID3 ) and KIF7 do not act epistatically and hypothesize that additional genetic alterations in ciliopathy genes of patient G2 , eventually including those identified in CEP41 ( JBTS15 ) , KIF14 , and WDPCP , may contribute to a mutational load that is sufficient to elicit a JBTS phenotype on a KIAA0586+/−; KIF7+/− background . ( A ) Overexpression of IFT57 does not have an effect on patterning of the neural tube in the talpid3 chicken . ( B ) Overexpression of KIF7 does not rescue or alter neural tube patterning in the talpid3 chicken . ( C ) siRNA knockdown of KIF7 resulted in a weak dorsalization of PAX7 but no expansion of ISLET1 . ( D ) Talpid3+/− Kif7+/− mice showed no gross anatomical abnormalities , neither were developmental brain defects identified through MRI ( D ) or histology ( E ) . ( F , G ) No abnormalities were identified in either the percentage of ciliated cells ( F ) , nor the length of cilia ( G ) in MEFs derived from wildtype , Talpid3+/+Kif7+/− , Talpid3+/− Kif7+/+ or Talpid3+/− Kif7+/− mice . DOI: http://dx . doi . org/10 . 7554/eLife . 08077 . 004 Family 2 ( Figure 1B ) is of North American origin . Patient MD1 was born at 34 3/7 weeks gestation following preterm premature rupture of membranes at 26 weeks . At birth , patient MD1 was found to have cardiac defects including a patent ductus arteriosus ( PDA ) , patent foramen ovale ( PFO ) and a 3/6 ventricular septal defect ( VSD ) causing persistent pulmonary hypertension 24 hr after birth . The PDA and PFO resolved , and VSD was at 2/6 within 22 days . At 7 months , MD1 was found to have a superior vena cava duplication . At 2 years of age , MD1 had hypotonia which inhibited motor actions , although she crawled proficiently , used sign language and single words , and self-fed by hand and with utensils . In addition , she had type I bilateral Duane syndrome with no abduction in either eye , narrowing of the palprebal fissure of the inturned eye , was farsighted , had thin tooth enamel , held her jaw sideways in a cross-bite pattern , and had long fingers with a slight clinodactyly of the fifth finger . She had a broad forehead , arched eyebrows , ptosis of the right eye , and a triangle-shaped mouth . Her receptive language was good . There was intermittent hyperpnea/apnea during awake periods . Patient MD1 had no liver , kidney , or eye abnormalities at 2 years of age . Family 3 ( Figure 1C ) is of German origin: patient G2 displayed a relatively mild JBTS phenotype with developmental delay and behavioral abnormalities , but no dysmorphic signs and no renal , retinal , skeletal , or liver abnormalities . His symptoms were described previously ( Figure 1C , Dafinger et al . , 2011 ) . We have identified KIAA0586 mutations in three JBTS families ( Figure 1A–D ) . Genome-wide SNP genotyping in Family 1 identified eight homozygous chromosomal candidate regions with a total range of 67 . 1 Mb . By WES , the homozygous mutation c . 2414-1G>C in intron 17 , affecting the invariant consensus of the exon 18 acceptor splice site , was found in the index patient , MR026-01 , and his affected sister , MR026-04 . Segregation analysis in the family was compatible with causality ( Figure 1A ) . The mutation was absent from 372 healthy Syrian controls , including 92 of Kurdish origin , and not listed in the ExAC database . In patient MD1 from Family 2 , WES identified compound heterozygosity for the KIAA0586 mutations c . 428delG ( p . Arg143Lysfs4; rs534542684; MAF of 0 . 39% in ExAC db ) and c . 2512C>T ( p . Arg838* ) , each inherited from a healthy parent ( Figure 1B ) , and both resulting in premature stop codons . Because the coiled-coil domain , which is essential for KIAA0586 function in mouse , chicken , and zebrafish ( residues 531–571 and residues 497–530 in human and chicken KIAA0586 ( TALPID3 ) , respectively; Figure 1F ) , would be lost in a truncated protein derived from the c . 428delG mutation , we consider it a loss-of-function mutation ( as is the case for the talpid3 chicken mutation which introduces a frameshift 3′ to c . 428 in the chicken ortholog , Figure 1F ) . Like the Talpid3/TALPID3 null mutations in mouse and chicken , c . 428delG is clearly recessive because the father of the patient is a healthy carrier . The c . 2512C>T ( p . Arg838* ) mutation is predicted to lead to nonsense-mediated decay ( NMD ) or a truncated protein , but with preservation of the essential coiled-coil domain . The simplex patient of Family 3 , G2 , was a known carrier of a heterozygous N-terminal frameshift mutation in exon 3 of the JBTS12 gene KIF7 , c . 811delG ( p . Glu271Argfs51 ) ( Dafinger et al . , 2011 ) . WES of the family trio ( patient G2 and his parents ) additionally identified the c . 428delG ( p . Arg143Lysfs4 ) mutation in KIAA0586 , in the patient ( Figure 1C ) . We hypothesized that disease in patient G2 could be due to biallelic mutations either in KIF7 ( JBTS12 ) or in KIAA0586 ( JBTS23 ) , assuming that the ‘missing mutation’ has escaped detection by sequencing due to an extra-exonic localization . Genome-wide CGH ( Affymetrix 6 . 0 SNP array ) did not reveal structural alterations adjacent to or within KIF7 ( Dafinger et al . , 2011 ) or KIAA0586 , thereby largely excluding a large deletion or duplication . PCR amplification and subsequent sequencing of KIAA0586 exons from cDNA did not reveal aberrant splicing as a potential hint for a deep intronic splice site mutation . Because KIF7 and KIAA0586 both encode modulators of GLI processing and c . 428delGKIAA0586 and c . 811delGKIF7 likely represent recessive loss-of-function mutations , we investigated the possibility of a potential epistatic effect predisposing to JBTS . No such interactions were identified in mouse and chicken experiments ( details are fully described in Figure 1—figure supplement 1 ) . Therefore , unidentified mutations are likely to be involved , either mutations in KIF7 , KIAA0586 ( e . g . , deep intronic mutations or alterations in non-coding regulatory regions which would both not be covered by WES ) or biallelic mutations in another ( yet unknown ) JBTS gene . WES revealed further heterozygous missense variants in three known recessive ciliopathy genes in patient G2 ( Figure 1C ) , all affecting evolutionarily conserved residues of the respective proteins: ( 1 ) c . 536 G>A ( p . Arg179His , rs140259402; MAF of 0 . 001647% in ExAC db ) in CEP41 , the gene associated with JBTS15 ( Lee et al . , 2012 ) . ( 2 ) c . 3181A>G ( p . Ile1061Val; MAF of 0 . 01155% in ExAC db ) in KIF14 , a gene associated with a lethal fetal ciliopathy phenotype ( Filges et al . , 2014 ) . ( 3 ) c . 1333G>C ( p . Ala445Pro , rs61734466; MAF of 0 . 6609% in ExAC db ) in WDPCP , the gene associated with Bardet–Biedl syndrome type 15 ( BBS15 ) , and a putative contributor to Meckel Gruber syndrome ( Kim et al . , 2010 ) . All variants were of paternal origin and rare in the general population except the WDPCP allele , which had been maternally inherited and which has been annotated homozygously in five healthy individuals ( ExAC db ) , indicating that this is a benign variant . Genome-wide CGH ( Affymetrix 6 . 0 SNP array ) did not show structural alterations adjacent to or within CEP41 , KIF14 , or WDPCP . In addition , we searched the WES data of patient G2 for heterozygous putative loss-of-function ( that is , truncating ) variants in genes with documented ciliary function . This revealed a paternally inherited frameshift variant , c . 206_207insA ( p . Ser70Valfs3 ) , in PLA2G3 , the gene encoding phospholipase A2 . In a functional genomic screen , PLA2G3 was found to be a negative regulator of ciliogenesis and ciliary membrane protein targeting ( Kim et al . , 2010 ) . The p . Ser70Valfs3PLA2G3 variant is relatively common , but has not been documented in homozygous state in healthy individuals ( MAF of 0 . 4060 in ExAC db ) . We also filtered for known JBTS genes carrying at least two rare variants in patient G2 , but we did not identify such a constellation . When applying this to all genes captured in the WES approach , there was also no potentially causative double heterozygosity in a gene of known or probable ciliary function . Filtering for homozygous rare and likely pathogenic variants was negative , compatible with lack of consanguinity in the parents of patient G2 . The c . 2414-1G>C mutation affects the invariant consensus of the acceptor splice site of exon 18 . RT-PCR and Sanger sequencing of the fragments amplified from cDNA revealed three aberrant splicing products due to usage of alternative exonic acceptor splice sites at AG motifs within exon 18 and due to skipping of exon 18 ( Figure 1D , E ) : a 13-bp deletion that results in a premature termination codon ( alternative acceptor splice site at c . 2425/2426AG ) , a 108-bp in-frame deletion ( alternative acceptor splice site at c . 2520/2521AG ) , and a 188-bp deletion due to skipping of exon 18 that results in a premature stop codon . These aberrant transcripts were present in the cDNA from both patients , but not in the cDNA of a healthy control individual ( Figure 1D ) . The mutant mRNA molecules are likely to be degraded by NMD . If the mutant transcripts were stable , the essential coiled-coil domain ( Figure 1E ) , which mediates centrosomal localization and function of KIAA0586 protein ( Yin et al . , 2009; Wu et al . , 2014 ) , would be preserved . KIAA0586 ( Talpid3 ) is a centrosomal protein and localizes to the basal body and the adjacent centriole of primary cilia in human RPE1 , IMCD3 cells ( Figure 2A ) , and other cell types ( Kobayashi et al . , 2014a; Wu et al . , 2014 ) . Immunofluorescence analysis of the retina of wildtype C57BL/6 mice allowed us to allocate Talpid3 expression to different retinal layers , namely the photoreceptor layer , the outer and inner plexiform layer , and the ganglion cell layer ( Figure 2B ) . Co-staining with the ciliary marker centrin-3 ( Trojan et al . , 2008 ) demonstrated Talpid3/KIAA0586 localization in the ciliary region at the joint between the inner and outer segment of photoreceptor cells in cryosections through the mouse retina and the retina of a human donor eye ( Figure 2B , C , E ) . Higher magnification revealed that Talpid3/KIAA0586 specifically localized at the basal body ( mother centriole ) and the adjacent centriole as well as between the two centrioles , but not in the connecting cilium of mouse and human photoreceptor cells ( Figure 2D , F ) . These findings were confirmed by immunoelectron microscopy of Talpid3 labeling on sections through mouse photoreceptor cilia ( Figure 2G , H ) . Immunostaining was found at centrioles and in the pericentriolar region in the apical inner segment of photoreceptor cells . The spatial distribution of Talpid3/KIAA0586 labeling at the ciliary base of photoreceptor cells is summarized in the scheme of Figure 2J . 10 . 7554/eLife . 08077 . 005Figure 2 . Localization of KIAA0586/Talpid3 in primary cilia and in photoreceptor cilia of mammalian retinas . ( A ) Triple labeling of a ciliated IMCD3 cell demonstrates localization of Talpid3 ( green ) in the basal body ( BB ) and the adjacent centriole ( Ce ) at the base of the primary cilium co-stained by antibodies against Pericentrin-2 ( PCNT2 , red ) and anti-acetylated tubulin ( acTub , cyan ) , a biomarker of the axoneme ( Ax ) . ( B ) Longitudinal cryosections through a mouse retina stained for Talpid3 ( green ) and counterstained for the ciliary marker Centrin-3 ( Cen3 , red ) and for the nuclear DNA marker DAPI reveal Talpid3 localization in the ciliary region ( CR ) at the joint between the inner ( IS ) and the outer segment ( OS ) of the photoreceptor layer , the outer ( OPL ) and inner plexiform layer ( IPL ) . Overlay of DIC ( differential interference contrast ) image with DAPI ( blue ) nuclear stain in the outer ( ONL ) and the inner nuclear layer ( INL ) and in the ganglion cell layer ( GC ) . ( C–F ) Immunostaining of cryosections through the photoreceptor layer of a mouse ( C ) and a human retina ( E ) demonstrates co-localization of KIAA0586/Talpid3 and Cen3 in the CR of photoreceptor cells . Higher magnification of double-labeled mouse ( D ) and human ( F ) photoreceptor cilium reveals substantial localization of Talpid3/KIAA0586 at the centriole ( Ce ) , the BB and between the Ce and BB of the photoreceptor cilium , but not in the connecting cilium ( CC ) . ( G , H ) Immunoelectron microscopy analysis of longitudinal section through the cilium of a mouse rod photoreceptor cell and ( G ) higher magnification of the ciliary base ( H ) labeled for Talpid3 reveals Talpid3 in the periciliary region namely in the Ce and BB . ( J ) Schematic representation of Talpid3/KIAA0586 localization in the photoreceptor cilium . Scale bars: A , 1 μm; B , 10 μm; C , E , 5 μm; D , F , 0 . 5 µm; G , H , 200 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 08077 . 005 The talpid3/Talpid3−/− phenotype in model animals has thus far been attributed to the role of TALPID3 in ciliogenesis and the subsequent loss of Hh-dependent patterning . However , the patients in this study did not display any overt defects typical for impaired Hh signaling such as polydactyly or hypotelorism , which have been described in other patients with JBTS ( Bachmann-Gagescu et al . , 2015 ) . Talpid3 chicken embryos also have polycystic kidneys ( Yin et al . , 2009 ) , a phenotype that is frequently ascribed to a loss of oriented cell division ( Happe et al . , 2011; Carroll and Yu , 2012 ) , as well as cell migration defects ( Bangs et al . , 2011 ) , which may also occur due to loss of cell polarity ( Happe et al . , 2011; Carroll and Yu , 2012 ) . To investigate if tissue and cell polarity is impaired by a loss of TALPID3 function , we first examined the patterning of the skin and the inner ear , two highly polarized tissues independent of Hh signaling . At E10 , embryonic chicken feather buds express β-catenin in an oriented manner , with a larger domain in the anterior part of the bud ( Figure 3A , C ) . While 88% of wildtype feather buds at E10 are oriented in this manner ( n = 117/133 ) , only 21% of stage-matched talpid3 feather buds were ( n = 38/179 ) . 22% of talpid3 buds were oriented in the wrong direction ( n = 39/179 ) and 57% had failed to show any orientation of β-catenin expression ( n = 102/179; Figure 3C′ ) . Talpid3 feather buds also frequently merged ( 29% of buds; asterisk , Figure 3B′′ ) . Thus , the skin of talpid3 embryos did not show the characteristic rostral-caudal polarization of wildtype skin . The hair cells ( HCs ) of the inner ear ( known as the basilar papilla ( BP ) in chicken ) have a highly polarized structure determined by the non-canonical Wnt-PCP signaling pathway . In the wildtype chicken , as in mouse , individual HCs exhibit an orientated actin-based stereocilia bundle , the apex of which lies at the abneural side of the cell , where within an actin-free ‘bare zone’ , a single kinocilium ( a microtubule-based true cilium ) forms ( arrow , Figure 3D ) . HCs are frequently used to assess how cell polarity and ciliogenesis are perturbed in mouse mutants ( Goetz and Anderson , 2010 ) . The HCs of talpid3 embryos formed actin filament bundles ( curved line , Figure 3E ) , but no kinocilium , demonstrating that , as with other tissues studied , loss of TALPID3 impairs ciliogenesis . Furthermore , although stereocilia were present in talpid3 HCs , stereocilia bundles frequently lacked polarity compared to wildtype HCs as indicated by either centrally located stereocilia bundles in SEM or actin filaments throughout the cell ( talpid3 n = 1086/1195; wt n = 258/502; Figure 3D–G , L ) . Orientation of the polarized stereocilia bundles that did form in talpid3 HCs was also abnormal ( Figure 3E , G , N ) . The orientation of stereocilia was determined in relation to their position to the abneural side of the BP ( Figure 3F , G , M ) . 73% of stereocilia of wildtype cells ( n = 244 ) were oriented within 40° of the expected angle ( 90° , compared to 38% of talpid3 cells ( n = 237; Figure 3M′ ) ) . Thus , talpid3 HCs showed disrupted polarity . 10 . 7554/eLife . 08077 . 006Figure 3 . Loss of TALPID3 ( KIAA0586 ) causes abnormal tissue and cell polarity and abnormal intracellular organization . ( A , A′ , A′′ ) β-catenin expression is localized anteriorly within feather buds of the wildtype chicken at day 9 . 5 and ( B , B′ , B′′ ) in the talpid3 chicken at day 9 . 5 . Black circles indicate feather buds with correct polarity; dashed black circles represent no polarity; blue circles represent abnormal polarity ( Schematic C ) . The talpid3 chicken ( B′′ ) demonstrates feather buds which lack polarity ( blue circles ) which is not seen in the wildtype chicken ( A′′ ) . Asterisks represent merged feather buds . ( C′ ) Quantification of the percentage of feather buds with correct , abnormal or no polarity in wildtype and talpid3 ( D , E ) SEM of the basilar papilla in wildtype ( D ) and talpid3 ( E ) chickens . Arrows indicate cilia . Curved lines represent the base of stereocilia hair bundles . ( F–K ) Actin bundles identified by phalloidin ( green ) and centriolar localization identified by γ tubulin ( red ) . ( F′ , G′ ) overview of wildtype and talpid3 basilar papilla , higher magnification in ( F , G ) , red circles with line represent orientation of polarized actin bundles in basilar papilla; dashed red circles represent unpolarized actin bundles ( Schematic L , M ) . ( L′ ) Quantification of polarized haircells . ( M′ ) Quantification of the angle of polarised hair cells . ( H–K ) Dashed white circles represent magnified images ( Ji–Kii′ ) . ( Ji–Kii ) White arrows indicate aligned centrosomes; blue arrows indicate unaligned centrosomes ( Schematic N ) . ( N′ ) Quantification of cells with aligned centrosomes . ( O , P ) Orientation based on placement of Golgi ( TGN46 , red ) in comparison to actin indicating the leading edge ( phalloidin , green ) and nucleus ( Dapi , blue , schematic in Q ) in MEFs . Asterisks represent areas of higher magnification ( not all represented at lower magnification ) . ( Q′ ) Quantification of the angle of orientation of MEF cells in scratch assay . Scale Bars: A , B 5 mm; A′ , A′′ , B′ , B′′ 1 mm; D , E 1 μm; F , G , H , I , J , K 20 μm; F′ , G′ 100 nm; Ji , Ji′ , Jii , Jii′ , Ki , Ki′ , Kii , Kii′ 10 μm; O , P 100 μm; Oi , Oii , Oiii , Oiv , Ov , Ovi , Pi , Pii , Piii , Piv , Pv , Pvi 25 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08077 . 006 Loss of TALPID3 function prevents basal body docking ( Yin et al . , 2009 ) , which we have previously suggested to be due to failure of centrosome migration ( Stephen et al . , 2013 ) . The migration and subsequent localization and docking of the centriole is crucial to establish polarity and placement of the actin bundle in the HC ( Tarchini et al . , 2013 ) , and we therefore hypothesized that disturbed cell polarity may result from defective centrosome migration in talpid3 HCs . Using antibodies against γ tubulin to determine the localization of the centriole within the actin-negative abneural bare zone in the HCs , 95% of wildtype HCs exhibited a basal body ( centrosome ) within the abneural bare zone ( n = 632 from 7 samples , Figure 3H , J , N ) . In contrast , only 49% of talpid3 cells exhibited a centriole within the bare zone ( either abneural or abnormally polarized; n = 219 from 6 samples; Figure 3I , K , N ) , thus demonstrating that the intracellular organization of talpid3 cells was frequently abnormal . Furthermore , and in agreement with the failure of correct polarization of the stereocilia , centrioles were frequently observed on the neural side of talpid3 HCs ( Figure 3Kii ) . We conclude that failure of centriolar migration in talpid3 cells results in abnormal cell polarization and stereocilia formation in HCs . Because 49% of talpid3 cells did exhibit a centriole correctly localized yet ciliogenesis was completely disrupted , the failure of ciliogenesis may not only be due to impaired centriolar migration . Directional cell migration is also intimately linked to the localization of the centrosome between the leading edge of the migrating cell and the Golgi apparatus . Talpid3−/− MEFs show abnormal cell migration ( Bangs et al . , 2011 ) , and we therefore examined if the orientation of the Golgi apparatus to the leading edge of migrating cells was also disrupted by a loss of Talpid3 function in mouse , in an in vitro scratch assay ( Figure 3O , P ) . The angle between the leading edge and Golgi was taken as the angle of orientation , with an angle of 0° suggesting perfect alignment of the Golgi to the leading edge of the migrating cell ( Figure 3Q ) . The angle of orientation was within 40° in 69% of wildtype cells and 55% of Talpid3−/− cells , whilst 20% of Talpid3−/− cells exhibited orientation angles greater than 60° compared to 11% of wildtype cells ( Figure 3Q′; wildtype cells = 132 , Talpid3−/− cells = 117 from two experiments; Figure 3O–Q ) , suggesting a reduction in intracellular polarization of the Golgi apparatus to the leading edge in the Talpid3−/− MEFs ( Figure 3O , P ) . Thus , KIAA0586 ( TALPID3;Talpid3 ) plays an essential role in the internal organization and polarization of cells , likely through its action on the centrosome . JBTS primarily affects the brain of the patients . The choroid plexus is a highly polarized multiciliated neuroepithelium in which we have previously shown , as now in HCs , a failure of centrosome migration in talpid3 mutant chickens ( Stephen et al . , 2013 ) . To determine if talpid3 mutant neuroepithelia exhibit cell polarity defects , we examined the intracellular organization of choroid plexus cells in E8 talpid3 mutant chickens . Wildtype choroid plexus cells exhibited a distinctive polarization with an apical , centriolar zone ( CZ , Figure 4A ) above a separate zone of mitochondria ( MZ , Figure 4A ) ; the most apical mitochondria were found an average of 7 µm from the apical surface ( Figure 4C ) . In contrast , the mitochondria in talpid3 choroid plexus are found in the most apical zone , an average of 3 µm from the apical surface ( m , Figure 4B ) , and centrioles are present throughout the cell ( asterisk in Figure 4B ) . We conclude that the neuroepithelium has an abnormal intracellular organization of centrosomes and mitochondria and therefore , like the HCs and migratory fibroblasts , is not correctly polarized . Although we have previously suggested that a failure of centrosome migration to the apical surface is the primary reason that cilia fail to form ( Stephen et al . , 2013 ) , our analysis of the HCs suggest an additional requirement for TALPID3 during ciliogenesis , independent of the centriole migration . We therefore investigated the maturation of the mother centriole , crucial for the basal body to dock to the membrane and initiate ciliogenesis . Subdistal appendages were identified in approximately 40% of wildtype and talpid3 centrioles ( wt n = 35 , talpid3 n = 48 , Figure 4D , E , G ) , whereas distal appendages were noted in 28% of wildtype centrioles and only 6% of talpid3 centrioles ( Figure 4D–G ) . To determine if there was a loss of distal appendages , we examined localization of CEP164 , a protein known to localize to the distal appendages of the mature mother centriole , the basal body . CEP164 localized correctly at the mother centriole and not at the sister centriole , in both wildtype and talpid3 cells of the neuroepithelium and fibroblasts ( Figure 4H–M ) . However , CEP164 puncta were smaller , disorganized and frequently orientated away from the apical cell surface in talpid3 cells ( Figure 4K , L ) . This confirmed our previous EM analysis ( Yin et al . , 2009 ) and data in this study , which demonstrated that centrioles frequently failed to migrate or orientate correctly in talpid3 cells . Smaller sized CEP164 puncta also suggested that distal appendages were not formed normally in talpid3 cells . As abnormal or absent distal appendages can result in elongation of the centriole due to improper capping , centriolar length was studied in wildtype and talpid3 choroid plexus cells ( Figure 4N–Q ) . Centrioles in wildtype tissue were on average 0 . 7 µm in length compared to 0 . 9 µm in the talpid3 chicken , suggesting that talpid3 centrioles may indeed fail to undergo complete maturation and are subsequently elongated ( Figure 4R ) . 10 . 7554/eLife . 08077 . 007Figure 4 . Loss of TALPID3 causes abnormal intracellular organization and centriolar orientation ( A , B ) The chicken choroid plexus at E8 is a highly polarized structure with docked centrioles ( asterisk , A ) identified within a clear centriolar zone apically ( CZ , A ) and a mitochondrial zone ( MZ; m indicates mitochondria ) . The talpid3 choroid plexus ( B ) lacks these defined zones , with mitochondria identified in the most apical zone ( m , B ) centrioles identified throughout the cell , failing to dock ( asterisk , B ) . Quantification of distance of mitochondria to cell surface ( C ) . ( D–G ) talpid3 tissue is capable of producing mature centrioles . Wildtype centrioles ( D ) and talpid3 centrioles ( E , F ) exhibited subdistal appendages ( SD ) , and distal appendages ( DA ) , although DA were less frequently observed on talpid3 centrioles , quantified in ( G ) . CEP164 localizes to the distal mother centriole in wildtype and talpid3 choroid plexus neuroepithelium ( purple arrow indicated distal mother centriole , green arrow proximal centriole; H , I , I′ , I′′ , K , L , L′ , L′′ ) and fibroblasts ( J , J′ , J′′ , M , M′ , M′′ ) , but CEP164 puncta are smaller and disorganized in talpid3 choroid plexus and fail to orientate to the apical surface of the cell ( arrows L ) . Centrioles in wildtype tissue were on average 0 . 7 µm ( red line indicating centriole/basal body; N , R ) compared to 0 . 9 µm in the talpid3 choroid plexus ( O , P , Q , R ) . Scale bars: A , B = 1 μm , D , E , F = 100 nm; H , K = 10 μm I , J , L , M = 5 μm , N , O , P , Q = 200 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 08077 . 007 In human cells , KIAA0586 is also required for centriolar satellite dispersal ( Kobayashi et al . , 2014a ) . Compatible with this , we observed electron-dense condensations around the centrioles in the neuroepithelium of talpid3 chicken , which were absent from wildtype centrioles ( basal body; 80% of wildtype cell exhibited electron-dense clear area around the centriole , whereas only 21% of talpid3 cells did; wt n = 35 , talpid3 n = 48; Figure 5A , D , G ) . To determine if these were centriolar satellites , we examined the localization of PCM1 , a marker for centriolar satellites . Compared to wildtype centrioles ( Figure 5B , C ) , PCM1 puncta were larger around talpid3 centrioles ( Figure 5E , F ) , possibly reflecting an increase in centriolar satellites . Because we observed KIAA0586 immunostaining around the pericentriolar region ( Figure 2G , H ) , we used the centriolar satellite marker AZI1 in human RPE1 cells to determine if KIAA0586 localized to centriolar satellites ( Figure 5H–J ) but found that KIAA0586 and AZI1 did not colocalize . Thus , as observed in human cell lines , TALPID3 is essential for centriolar satellite dispersal . As TALPID3 protein does not localize to the centriolar satellites , we assume that this is an indirect consequence of TALPID3 deficiency . 10 . 7554/eLife . 08077 . 008Figure 5 . Analysis of centriolar satellites in the talpid3 choroid plexus . An area clear of electron-dense condensations was observed around the basal body in wildtype cells ( area outlined by dots; A ) , electron-dense condensations were observed adjacent to talpid3 centrioles ( indicated by arrows , D ) . Quantified in ( G ) . Immunostaining for a centriolar satellite marker in the choroid plexus , PCM1 ( magnified area outlined by dashed line; PCM1 = red , γ tubulin , green B , B′ , C , C′ , C′′ , E , E′ , F , F′ , F′′ ) . KIAA0586 protein does not colocalize with AZI1 , a satellite protein in human RPE1 cells ( KIAA0586 = red , AZI1 = green H , I , J ) . Scale bars: A , D = 500 nm; B , E 10 = μm; C , F = 2 μm H , I , J 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08077 . 008 We conclude that KIAA0586 ( TALPID3 ) is essential for several distinct roles in centriole function , including centriole migration and orientation which can subsequently affect cell and tissue polarity and ciliogenesis , centriole maturation which affects docking of the basal body and ciliogenesis and through an indirect mechanism , centriolar satellite dispersal , which may also affect ciliogenesis . JBTS is a genetically heterogeneous condition , caused by mutations in several genes related to the structure and function of cilia ( Romani et al . , 2013 ) . Through homozygosity mapping and WES , we identified a novel disease locus ( JBTS23 ) , defined by mutations in the KIAA0586 gene , encoding a centrosomal protein ( Andersen et al . , 2003 ) ( Figure 1 ) , which is supported by simultaneous concurrent studies ( Bachmann-Gagescu et al . , 2015; Roosing et al . , 2015 ) . We used Talpid3/TALPID3−/− mouse and chicken models to understand the corresponding pathomechanisms causing the phenotypes of these patients and discovered centrosome abnormalities and loss of cell polarity . We confirm localization of KIAA0586 at centrosomal structures at the basal bodies and the adjacent daughter centrioles of primary cilia of mouse and human photoreceptor cells as well as in pericentriolar regions ( Figure 2 ) . KIAA0586 has previously been associated with recessive ciliopathy phenotypes in mouse ( Bangs et al . , 2011; Davey et al . , 2014 ) , chicken ( Davey et al . , 2006; Davey et al . , 2007 ) and zebrafish ( Ben et al . , 2011 ) . These animal models have either naturally occurring or induced 5′ mutations which disrupt an essential coiled-coil domain , resulting in loss of protein function , consecutive loss of Hh signaling and early embryonic lethality . The talpid3 chicken is a thoroughly examined animal model with polydactyly , holoprosencephaly , abnormal neural tube patterning , polycystic kidneys , liver fibrosis , short ribs , and endochondral bones with defective ossification ( Lewis et al . , 1999; Buxton et al . , 2004; Davey et al . , 2006; Davey et al . , 2014 ) . The c . 428delG ( p . Arg143Lysfs4 ) mutation was identified in heterozygous state in patient MD1 , in trans to a nonsense mutation ( Figure 1B ) , and in a patient G2 who is also heterozygous for a KIF7 ( JBTS12 ) frameshift mutation and variants in three other known ciliopathy genes ( Figure 1C ) ( Dafinger et al . , 2011 ) . Our experiments did not indicate epistatic interaction between KIAA0586 and KIF7 , and a secondary occult mutation in either gene cannot be excluded . The c . 428delG mutation results in a premature termination codon in five human KIAA0586 isoforms , causing either a major protein truncation 5′ to the essential coiled-coil domain or NMD . It is comparable to the talpid3 chicken loss-of-function mutation which introduces a frameshift in the orthologous region ( Figure 1F ) . The c . 428delG mutation is annotated in dbSNP ( rs534542684 ) , and its MAF in the general population is surprisingly high ( 0 . 39% , 378 out of 96 , 534 alleles in the ExAC db ) , reminiscent of the most common deafness ( c . 35delG in GJB2; 0 . 60% in the ExAC database ) or cystic fibrosis ( p . Phe508del in CFTR; 0 . 67% in the ExAC database ) mutation . In two concurrent studies reporting KIAA0586 mutations in patients with JBTS , c . 428delG represented the most prevalent mutation ( Bachmann-Gagescu et al . , 2015; Roosing et al . , 2015 ) . While c . 428delG was clearly enriched in patients with biallelic KIAA0586 mutations in both studies ( present in 20 of 24 such patients ) , only two were homozygous . Despite its commonness , c . 428delG was neither observed in homozygous state in healthy individuals in the TGP , ESP , or ExAC databases . Such rarity of homozygosity could indicate that it causes embryonic lethality , early death or severe illness leading to underrepresentation of the respective samples . Embryonic lethality in talpid3 chicken and Talpid3−/− knockout mice would support such an interpretation . On the other hand , c . 428delG was not found in a simultaneous study that reports biallelic KIAA0586 mutations in early lethal ciliopathies ( Alby et al . , 2015 ) . Of note , a very recent study on rare human knockouts identified in the genomes of 2636 healthy Icelanders lists one individual of 57 years with homozygosity for c . 428delG ( Sulem et al . , 2015 ) . This could either be due to protective modifiers or a low mutational load in the ciliome of the respective person . Assuming the latter , c . 428delGKIAA0586 could represent a hypomorphic allele that increases susceptibility to develop JBTS , with more severe mutations required either in trans ( in heterozygous carriers , as in most patients reported by Bachmann-Gagescu et al . ( 2015 ) ; Roosing et al . ( 2015 ) ) , or in other genes ( in homozygous carriers ) for disease manifestation . The presence of a heterozygous potentially deleterious C5orf42 ( JBTS17 ) variant in the only c . 428delGKIAA0586-homozygous patient reported by Bachmann-Gagescu et al . ( 2015 ) , and the co-occurence of such variants in four ciliopathy genes in patient G2 ( including a truncation in the JBTS gene KIF7 ) support the categorization of c . 428delGKIAA0586 as a hypomorphic mutation of incomplete penetrance . Of note , no secondary KIAA0586 mutation was identified in c . 428delG-heterozygous JBTS patients in the two other studies ( Bachmann-Gagescu et al . , 2015; Roosing et al . , 2015 ) , which could be due to the contribution of other genes . The homozygous mutations c . 2414-1G>C ( Family 1 ) and c . 2512C>T ( p . Arg838 , Family 2 ) would not disrupt the 5′ functionally essential coiled-coil domain in the consecutive KIAA0586 protein , and partial function may be maintained ( possibly due to preserved , albeit truncated , KIAA0586 protein ) . We have shown that KIAA0586 has several functions in the centriole , and this may be mediated by different protein residues . The occurrence of retinal degeneration in JBTS depends on the genetic subtype , but is variable even within a family . The localization of KIAA0586 at the ciliary base of retinal photoreceptor cells corresponds to other JBTS proteins . Proteins of the periciliary compartment at the base of the photoreceptor cilium are thought to be critical for the handover of cargo from the dynein-mediated transport through the inner segment to the kinesin-powered anterograde intraflagellar transport in the ciliary compartment ( Roepman and Wolfrum , 2007; Sedmak and Wolfrum , 2010 ) . KIAA0586 may be part of the protein networks implicated in these processes . The lack of retinal disease in the patients described herein may be due to the less strongly developed structure of the distal appendages and/or the possible functional redundancy in the cilia of retinal photoreceptor cells . Nevertheless , patients with KIAA0586-related JBTS should be investigated for signs of retinal degeneration , and given that mutations in the JBTS gene CEP290 may cause non-syndromic Leber congenital amaurosis ( den Hollander et al . , 2006 ) , KIAA0586 represents a candidate gene for isolated retinopathies . Loss of KIAA0586 ( TALPID3 ) function in animal models results in a failure to produce both primary and motile cilia . Previously , it has been suggested that this is due to a failure of the centrosome to migrate apically or dock at the plasma membrane ( Yin et al . , 2009; Stephen et al . , 2013 ) . The subsequent failure of cilia formation results in abnormal Hh signaling and disrupted GLI processing ( Davey et al . , 2006 ) . Most patients with KIAA0586-related JBTS exhibit few classical Hh phenotypes such as polydactyly ( this study , Bachmann-Gagescu et al . , 2015; Roosing et al . , 2015 ) , unlike the corresponding mouse , chicken , and zebrafish models ( Davey et al . , 2006; Bangs et al . , 2011; Ben et al . , 2011 ) . We show that , independent of Hh signaling , cell and tissue polarity are disrupted upon loss of TALPID3 . JBTS is characterized by cerebellar hypoplasia and loss of decussation of neuronal projections from the cerebellum ( Romani et al . , 2013 ) . While Hh signaling is required for controlling the growth of the embryonic cerebellar primordia ( Lewis et al . , 2004 ) , the failure of decussation has been proposed to result from defective axonal guidance ( Romani et al . , 2013 ) , a process depending on centrosome-guided cell polarity ( Solecki et al . , 2006 ) . Furthermore , we have shown that in inner ear HCs , cell polarity and ciliogenesis , albeit closely linked , are differentially affected in talpid3 cells . Thus , loss of decussation may reflect loss of polarity . We propose that KIAA0586 exerts a role in intracellular trafficking and cell polarity distinct from its role in docking of the centriole . Talpid3 cells have abnormal microtubule dynamics ( Yin et al . , 2009 ) . Microtubules are required for the recruitment of satellites and proteins in the distal centriole ( Kim et al . , 2008; Schmidt et al . , 2009 ) , a process known to be impaired by loss of KIAA0586 . Abnormal cell polarity in talpid3 cells may be due to the effect of TALPID3 on microtubule dynamics and a direct role in centrosome organization: microtubules are essential for intracellular trafficking , cellular structure , and polarity . We have shown that localization of the centrosome , mitochondria , and Golgi is disrupted in talpid3 cells . Moreover , Rab8 , a GTPase which binds to the Golgi and is required for vesicular trafficking and ciliogenesis ( Nachury et al . , 2007; Henry and Sheff , 2008; Feng et al . , 2012 ) , is mislocalized in KIAA0586-depleted cells ( Kobayashi et al . , 2014a ) . In JBTS patients with mutations in AHI1 ( JBTS3 ) , encoding an interactor of RAB8 ( Hsiao et al . , 2009 ) , non-ciliary trafficking from the Golgi and ciliogenesis are impaired . Of note , Golgi mislocalization in the talpid3 choroid plexus is similar to what has been observed in Ahi1−/− mice ( Hsiao et al . , 2009 ) . This suggests a similar pathogenesis of JBTS23 and JBTS3 , with defective cell polarity , intracellular trafficking , and Hh signaling . KIAA0586 interacts with CP110 and Cep120 ( Kobayashi et al . , 2014a ) , distal centriolar proteins implicated in centriole duplication and maturation , and ciliogenesis . The predominant expression of Cep120 on the daughter centriole throughout most of the cell cycle depends on Kiaa0586 , as indicated by high expression of Cep120 on both centrioles and absence of CP110 from the mother centriole prior to ciliogenesis in Talpid3−/− cells ( Spektor et al . , 2007 ) . Although equal expression of KIAA0586 on the mother and daughter centrioles has been reported ( this study , Kobayashi et al . , 2014a ) ; there is evidence that KIAA0586 predominantly localizes at the mother centriole ( Wu et al . , 2014 ) . In addition , loss of chicken KIAA0586 ( TALPID3 ) causes centriole elongation whereas overexpression of Cep120 causes elongation of the mother centriole , suggesting that KIAA0586 ( TALPID3 ) may control centriole length through depletion or suppression of Cep120 on the mother centriole . Similarly , depletion of CP110 also increases centriole length ( Schmidt et al . , 2009 ) , suggesting that KIAA0586 regulates centriolar length through controlling CP110 localization and centriolar capping of the distal mother centriole . Loss of other centriolar proteins , such as OFD1 , likewise results in elongated centrioles and loss of distal appendages ( Singla et al . , 2010 ) . Based on the colocalization of KIAA0586 , CP110 , and Cep164 , it has been proposed that KIAA0586 regulates ciliary vesicle docking adjacent to Cep164 localization ( Kobayashi et al . , 2014a ) , but not distal appendage formation itself , and this is supported by evidence from human patient KIAA0586−/− cells which show Cep164 within the distal centriole ( Alby et al . , 2015 ) . We also find evidence for a vesicle docking defect , demonstrated by an increase in centriolar satellites . However , we propose that KIAA0586 loss primarily causes abnormal distal appendages and impaired Cep164 localization , similar to what can be observed in OFD1 mutants ( Singla et al . , 2010 ) . In addition , determination of Cep164 expression in cells of highly polarized tissue demonstrates a further centriolar defect not easily distinguished in in vitro assays—the loss of centriole orientation to the apical membrane of the cell . Whether this defect is due to the depletion of KIAA0586 from the centriole or impairment of another function of KIAA0586 in pericentriolar regions or cytoskeleton remains to be elucidated . We have identified KIAA0586 as a novel gene for JBTS , and we propose that it is not only required for ciliogenesis , but also to establish cell and thus tissue polarity . JBTS23 , and possibly other JBTS subtypes , may result from impairment of both functions . Blood samples for DNA extraction were obtained with written informed consent . All investigations were conducted according to the Declaration of Helsinki , and the study was approved by the institutional review board of the Ethics Committees of the University of Erlangen-Nürnberg , the University of Bonn , and the University Hospital of Cologne . In accordance with the Human Gene Nomenclature Committee ( HGNC ) , we have used KIAA0586/KIAA0586 for designation of the human gene and protein , respectively . In accordance with the Chicken Gene Nomenclature Committee ( CGNC ) , we use TALPID3/TALPID3 for designation of the chicken gene and protein , respectively . Although the current gene symbol for the mouse gene is 2700049A03Rik ( protein: 2700049A03RIK ) , we use Talpid3/Talpid3 as the gene and protein names , respectively . Where we refer to a generic conclusion on the function of the orthologs of KIAA0586 , we use KIAA0586 . As in previous publications , the chicken model is referred to as talpid3 , and the mouse model is referred to as Talpid3−/− . The nomenclature of human KIAA0586 mutations refers to reference sequence NM_001244189 . 1 ( corresponding protein: NP_001231118 . 1 ) . The Exome Aggregation Consortium ( ExAC ) database ( Cambridge , MA , United States; http://exac . broadinstitute . org ) , which aggregates numerous databases including the current versions of the Exome sequencing project ( ESP , Fu et al . , 2013 ) and the Thousand Genomes Project ( TGP , Via et al . , 2010 ) was last accessed on 11 July 2015 for the presence and frequency of identified variants in healthy individuals . Family 1: genotyping and homozygosity mapping were performed in Family 1 ( MR026 ) as previously reported ( Abou Jamra et al . , 2011 ) . DNA from patient MR026-01 underwent exome capture and whole-exome sequencing ( WES ) using the SureSelect Human All Exon 50 Mb Kit ( Agilent Technologies , Santa Clara , United States ) and a SOLiD4 instrument ( Life Technologies , Carlsbad , United States ) as described previously ( Abou Jamra et al . , 2011 ) . Of the targeted regions , 73 . 2% were covered at least 20× , and 83 . 4% were covered at least 5× . To validate the results , we also conducted WES in the likewise affected sibling , MR026-04 , analogous to previously described disease gene identification approaches ( Ahmed et al . , 2015; Riecken et al . , 2015 ) . 96% of the target sequence were covered at least 20× . Family 2: samples from the index patient , MD1 , and her parents underwent WES at GeneDX ( Gaithersburg , MD , United States ) . Family 3: WES and mapping of reads for the index patient ( G2 ) and both parents were carried out as previously described ( Basmanav et al . , 2014; Beck et al . , 2014 ) . In brief , filtering and variant prioritization was performed using the varbank database and analysis tool ( https://varbank . ccg . uni-koeln . de ) of the Cologne Center for Genomics . In particular , we filtered for high-quality ( coverage >15-fold; phred-scaled quality >25 ) , rare ( MAF [minor allele frequency] ≤0 . 01 ) variants ( dbSNP build 135 , the 1000 Genomes database build 20110521 , and the public Exome Variant Server , NHLBI Exome Sequencing Project , Seattle , build ESP6500 ) . To exclude pipeline-related artifacts ( MAF ≤ 0 . 01 ) , we filtered against variants from in-house WES data sets from 511 patients with epilepsy . The Affymetrix genome-wide Human SNP Array 6 . 0 utilizing more than 906 , 600 SNPs and more than 946 , 000 copy number probes was used for genome-wide detection of copy number variations in patient G2 . Quantitative data analyses were performed with GTC 3 . 0 . 1 ( Affymetrix Genotyping Console ) using HapMap270 ( Affymetrix ) as reference file . In the index patient ( G2 ) , all coding KIAA0586 and KIF7 exons were Sanger-sequenced in search of a second mutation . In addition , we amplified and sequenced all KIAA0586 exons from cDNA ( derived from whole blood mRNA , PAXgene Blood RNA Tube , PreAnalytiX , Hombrechtikon , Switzerland ) in search of potential hints of aberrant splicing due to extra-exonic variants . Continuous PCR amplification of KIF7 exons from whole blood mRNA was not successful . The sample of patient G2 was analyzed by genome-wide CGH ( Affymetrix 6 . 0 SNP array ) to exclude structural alterations adjacent to or within KIAA0586 , KIF7 , CEP41 , KIF14 , or WDPCP . Confirmation of the identified mutations and segregation analyses were carried out by Sanger sequencing . In Family 1 , we isolated mRNA using the RNeasy kit ( Qiagen , Hilden , Germany ) from lymphoblastoid cell lines that have been established based on standard protocols from patients MR026-01 and MR026-04 . We transcribed mRNA to cDNA using SuperScriptII reverse transcriptase and random primers ( Invitrogen; Van Allen Way Carlsbad , California , United States ) . To test if the KIAA0586 mutation c . 2414-1G>C impairs splicing , we used two pairs of primers ( KIAA0586_exprF1 , 5′-TCCATCTCCTAAGTCCAGACCAC-3′ and KIAA0586_expR1 , 5′-TCCAAGTTTGCACAGGAGG-3′ , located in exons 16 and 19 , and KIAA0586_exprF2 , 5′-TCAGGTACATTGGAAGGTCATC-3′ and KIAA0586_expR2 , 5′-AACTGGCGGAAATGGAGG-3′ , located in exons 17 and 21; NM_001244189 . 1 ) and standard PCR methods . Electrophoresis on standard agarose gel followed by cutting out the DNA bands , purifying the DNA using QIAquick gel extraction kit ( QIAgene; Hilden , Germany ) , and Sanger sequencing were performed . Eggs were obtained from talpid3 flock ( MG Davey; talpid3 chicken lines are maintained at the Roslin Institute under UK Home Office license 60/4506 [Dr Paul Hocking] , after ethical review ) . Mice were maintained at the Human Genetics Unit , Western General , Edinburgh , under UK Home Office license PPL 60/4424 ( Ian Jackson ) . The Talpid3+/−/Kif7+/− line was produced by crossing of the previously described Talpid3+/− knockout mouse line ( Bangs et al . , 2011 ) and the reported Kif7+/− mouse line ( Cheung et al . , 2009 ) . Animal experiments carried out at the JGU Mainz corresponded to the statement of the Association for Research in Vision and Ophthalmology ( ARVO ) as to care and use of animals in research . Adult mice were maintained under a 12-hr light–dark cycle , with food and water ad libitum . Chicken eggs from talpid3 flock were incubated at 38°C until 12 days at the latest , staged as per Hamburger and Hamilton ( 1951 ) , dissected into cold PBS , and fixed in 4% PFA/PBS . Mouse timed matings were established between Talpid3+/− mice ( Bangs et al . , 2011 ) and Kif7+/− mice ( Cheung et al . , 2009 ) and confirmed by vaginal plug . Pregnant females were sacrificed at day 10 of pregnancy for production of mouse embryonic fibroblasts ( MEFs ) . Otherwise between day 12 and 16 of pregnancy and embryos were dissected in cold PBS , decapitated , and fixed immediately in 4% PFA for histological analysis . Pups were sacrificed between 7 and 21 days after birth by lethal injection and brains were dissected into 4% PFA/PBS . Embryos used in comparisons were dissected as family groups and genotyped after analysis . Tissues were collected on dissection , lysed in 10 mM Tris ( pH8 ) , 10 mM EDTA ( pH 8 ) , 1% SDS , 100 mM NaCl , and 20 mg/ml proteinase K at 55°C overnight before DNA extraction using Manual Phase Lock Gel Tubes ( 5 Prime ) for phenol/chloroform extraction . For chicken TALPID3 , sequencing primers used were 5′-TCATTTCATTAGCTCTGCCG-3′ ( forward ) and 5′-CCATCAAACCAACAGCTCAG-3′ ( reverse ) . For mouse Talpid3 , PCR primers were 5′-TGCCATGCAGGGATCATAGC ( forward ) , 5′-GAGCACACTGGAGGAAAGC-3′ ( reverse ) and 5′-GAGACTCTGGCTACTCATCC-3′ , 5′-CCTTCAGCAAGAGCTGGGGAC-3′ , respectively . For mouse Kif7 , PCR primers were 5′-CACCACCATGCCTGATAAAAC-3′ ( P1 forward ) , 5′-CTATCCCCAATTCAAAGTAGAC-3′ ( P1 reverse ) , 5′-CCAAATGTGTCAGTTTCATAGC-3′ ( P2 forward ) , 5′-TTCTCACCCAAGCTCTTATCC-3′ ( P2 reverse ) . Fixed samples from mouse brain and chicken legs were embedded in paraffin , sectioned , and stained in haematoxylin and eosin as described previously ( Davey et al . , 2014 ) . Mouse and chicken embryos were rehydrated through a methanol gradient and in situ hybridization carried out for chicken β-catenin ( codons 1–127 ) as previously described ( Nieto et al . , 1996 ) . The following Kif7 sequences were targeted for knockdown: Target 1: TTATCGACGAGAACGACCTCAt , Target 2: cATCCAGAACAAAGCGGTGGTG , Target 3: gTCCTCTAACACTAAGAACATT , Target 4: gACAGATGACATAGTCCGTGTG to which 22mer sequences were designed in Genscript and cloned into pRFPRNAiC ( Das et al . , 2006 ) ( Dundee Cell Products , Dundee , United Kingdom ) . Embryos were electroporated at stage 12HH ( as described Yin et al . , 2009 ) , observed for RFP expression at stage 24HH , fixed , and prepared for sectioning and immunohistochemistry at stage 22HH as below . Tissue from embryos was collected and genotyped . MEFs were prepared from E10 . 5 eviscerated and decapitated embryos . Cells were dissociated in trypsin/versin and maintained to passage 2 as per Hall et al . ( 2013 ) and serum removed from media for 48 hr to induce ciliogenesis . RPE1 cells ( ATCC ) were grown in DMEM-F12 , 10% FCS Gold , 50 μl hygromycin , 5 ml L-glut . IMCD3 ( mouse inner medullary collecting duct cells ) cells were grown in DMEM-F12 10% FCS . To induce ciliogenesis , RPE1 and IMCD3 cells were starved in DMEM:F12 or Opti-MEM I ( Life Technologies , Carlsbad , California , United States ) for 72 hr . Cells were fixed with methanol at −20°C for 2–5 min . After washing in PBS , cells were immunolabeled with polyclonal antibodies against acetylated tubulin ( T7451 , Sigma-Aldrich , St . Louis , Missouri , United States ) , pericentrin-2 ( sc-28145 , Santa Cruz Biotechnology , Dallas , United States ) , and KIAA0586 ( HPA000846 , Atlas Antibodies , Stockholm , Sweden ) before incubation with appropriate secondary antibodies conjugated to Alexa 488 ( A21206 , Molecular Probes , Invitrogen ) , CF 568 ( 20106-1 , Biotrend , Köln , Germany ) , and CF 640 ( 20177 , Biotrend ) , and with DAPI ( 6335 . 1 , Roth , Karlsruhe , Germany ) . Eyes from a healthy human donor ( #199-10; 56 years of age , dissection 29 hr post mortem ) were obtained from the Department of Ophthalmology , University Hospital of Mainz , Germany , according to the guidelines of the declaration of Helsinki . After sacrifice , eyeballs from adult C57BL/6J mice were dissected , cryofixed in melting isopentane , cryosectioned and immunostained as previously described ( Overlack et al . , 2011 ) . Cryosections were incubated with monoclonal antibodies to centrin-3 as a molecular marker for the ciliary apparatus of photoreceptor cells as previously characterized ( Trojan et al . , 2008 ) , and polyclonal antibodies against KIAA0586 ( Atlas HPA000846 ) . Washed cryosections were incubated with appropriate antibodies conjugated to Alexa 488 ( Molecular Probes A21206 ) and Alexa 568 ( Molecular Probes A11031 ) in PBS with DAPI ( Roth 6335 . 1 ) to stain the nuclear DNA and mounted in Mowiol 4 . 88 ( Hoechst , Germany ) . Specimens were analyzed in a Leica DM6000B deconvolution microscope ( Leica , Germany ) . Image contrast was adjusted with Adobe Photoshop CS using different tools including color correction . For section immunocytochemistry on chicken tissue , chicken embryos were dissected into PBS , fixed , sectioned , and stained as described ( Davey et al . , 2006 ) , except for CEP164 , in which an antigen retrieval step was undertaken ( incubation in 0 . 1% BME/PBS for 5 min , incubation in 55°C PBS for 4 hr ) . For bone sections , legs were dissected at E12 . For immunocytochemistry , cells were fixed as above . Antibodies were used against: acetylated α-tubulin ( Sigma–Aldrich T7451 ) , γ-tubulin ( Sigma–Aldrich T5192; T5326 ) , TGN46 ( Abcam , Cambridge , United Kingdom , ab16059 ) , PCM1 ( Abcam ab72443 ) , AZI1 ( kind gift of Jeremy Reiter , UCSF ) , centrin-3 ( Trojan et al . , 2008 ) , KIAA0586 ( Atlas Antibodies , HPA000846 , ProteinTech 24421-1-AP ) , CEP164 ( ProteinTech , Manchester , United Kingdom , 22227-1 ) RFP ( Life Technologies R10367 ) , GFP ( Life Technologies A-21311 ) , Pax7 ( Developmental Studies Hybridoma Bank , Iowa City , United States ( DSHB ) ) , ISLET1 ( DSHB ) , NKX2 . 2 ( DSHB ) , Phalloidin ( Life Technologies A12379 ) , Anti-mouse ( Life Technologies A11017 ) , anti-rabbit ( Life Technologies A21207 ) . Imagining was undertaken on a Zeiss LSM 710 or a Nikon Air confocal microscope or Leica DMLB . Chicken embryos were dissected into PBS at 8 days of incubation , avoiding contamination with yolk , heads were removed and placed into 4% PFA , 2 . 5% glutaldehdye in 0 . 1 M cacodylate buffer . The choroid plexus was immediately removed and placed into fresh fixative ( as previous ) for 24 hr . Tissue was prepared and visualized for transmission electron microscopy as described previously ( Davey et al . , 2007 ) , and axoneme/basal body structure was compared to what was observed and reported previously ( Paintrand et al . , 1992 ) . Anti-KIAA0586 antibody ( Atlas Antibodies , HPA000846 ) was used for pre-embedding labeling in mouse retinas as previously described ( Maerker et al . , 2008; Sedmak and Wolfrum , 2010 ) . Ultrathin sections were cut on a Leica Ultracut S microtome and analyzed with a Tecnai 12 BioTwin transmission electron microscope ( FEI , The Netherlands ) . Images were obtained with a charge-coupled device SIS Megaview3 SCCD camera ( Surface Imaging Systems , Herzogenrath , Germany ) and processed with Adobe Photoshop CS . Angles of proliferation , migration , orientation , and localization were calculated using Axiovision Angle3 software , and cilium length was measured using Zen software ( Zeiss , Oberkochen , Germany ) . Scratch assays were carried out in wildtype and Talpid3−/− MEFs grown to confluence and serum starved ( DMEM + 0 . 5% FCS ) for 48 hr with a p10 pipette tip . Medium was then renewed and MEFs incubated for four hours before fixation in ice-cold methanol prior to immunofluorescence . Angles of orientation were then taken as a measurement of the angle from the center of the nucleus , through the center of the leading edge ( towards the wound , identified by phalloidin staining for F-actin ) and again through the center of the Golgi apparatus ( identified by TGN46 antibody staining ) . Tiled Z stacks of the scratch/wound were analyzed for greater accuracy . The expected orientation of the stereocilia of the basilar papilla hair cells were taken as being at 90° to the abneural edge of the basilar papilla . The angle of orientation was taken by drawing a line through the cell perpendicular to the abneural edge and a second from the center of the cell , intersecting with both the perpendicular line and center of the actin bundle . The internal angle was taken to be the angle by which cell orientation deviated from the expected . Cilium length was measured using Zen software ( Zeiss , Oberkochen , Germany ) .	Joubert syndrome is a rare and severe neurodevelopmental disease in which two parts of the brain called the cerebellar vermis and brainstem do not develop properly . The disease is caused by defects in the formation of small projections from the surface of cells , called cilia , which are essential for signalling processes inside cells . Mutations in at least 25 genes are known to cause Joubert syndrome , and all encode proteins that create or maintain cilia . However , these mutations account for only half of the cases that have been studied , which indicates that mutations in other genes may also cause Joubert syndrome . Here , Stephen et al . used genetic techniques called ‘homozygosity mapping’ and ‘whole-exome sequencing’ to search for other mutations that might cause the disease . They found that mutations in a gene encoding a protein called KIAA0586 also cause Joubert syndrome in humans . One of these mutations ( c . 428delG ) is unexpectedly common in the healthy human population . It might be a major contributor to Joubert syndrome , and the manifestation of Joubert syndrome in individuals with this mutation might depend on the presence and nature of other mutations in KIAA0586 and in other genes . The TALPID3 protein in chickens and other ‘model’ animals is the equivalent of human KIAA0586 . A loss of TALPID3 protein in animals has been shown to stop cilia from forming . This protein is found in a structure called the basal body , which is part of a larger structure called the centrosome that anchors cilia to the cell . Here , Stephen et al . show that this is also true in mouse and human eye cells . Further experiments using chicken embryos show that a loss of the TALPID3 protein alters the location of centrosomes inside cells . TALPID3 is also required for cells and organs to develop the correct polarity , that is , directional differences in their structure and shape . The centrosomes of chicken brain cells that lacked TALPID3 were poorly positioned at the cell surface and abnormally long , which is likely responsible for the cilia failing to form . Stephen et al . 's findings suggest that KIAA0586 is also important for human development through its ability to control the centrosome . Defects in TALPID3 have a more severe effect on animal models than many of the identified KIAA0586 mutations have on humans . Therefore , the next step in this research is to find a more suitable animal in which to study the role of this protein , which may inform efforts to develop treatments for Joubert syndrome .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2015	TALPID3 controls centrosome and cell polarity and the human ortholog KIAA0586 is mutated in Joubert syndrome (JBTS23)
Non-additive interaction between genetic variants , or epistasis , is a possible explanation for the gap between heritability of complex traits and the variation explained by identified genetic loci . Interactions give rise to genotype dependent variance , and therefore the identification of variance quantitative trait loci can be an intermediate step to discover both epistasis and gene by environment effects ( GxE ) . Using RNA-sequence data from lymphoblastoid cell lines ( LCLs ) from the TwinsUK cohort , we identify a candidate set of 508 variance associated SNPs . Exploiting the twin design we show that GxE plays a role in ∼70% of these associations . Further investigation of these loci reveals 57 epistatic interactions that replicated in a smaller dataset , explaining on average 4 . 3% of phenotypic variance . In 24 cases , more variance is explained by the interaction than their additive contributions . Using molecular phenotypes in this way may provide a route to uncovering genetic interactions underlying more complex traits . The discrepancy between the contribution of known genetic factors to variation of a trait and the estimated total contribution of all genetic variants has become known as ‘missing heritability’ ( Manolio et al . , 2009 ) . Some of the explanations for this discrepancy are: many common variants with small effects; many rare variants with larger effects; and interactions between genetic variants ( epistasis ) or between variants and environment ( GxE ) . Here , we focus on the discovery and characterisation of epistasis , by which we mean that the effect of a genetic variant on a trait depends on the genotype at one or more other locations in the genome . Statistically we define this as a joint effect of two loci on a trait , significant beyond the sum of additive effects . On long time frames , epistasis plays an important role in evolution ( Breen et al . , 2012 ) , and has been used to explain the persistence of deleterious mutations under selection ( Hemani et al . , 2013 ) . Epistasis has frequently been seen in crosses between model organism strains . Huang et al . ( 2012 ) looked at mapping variants associated with three traits in two distinct Drosophila populations and found very little concordance between the results . They postulated that this could be because the effect of genetic variants was dependent on the genetic background , and found frequent evidence of genetic interactions between one or more variants and the originally associated SNPs . Annotating these interacting SNPs to genes revealed common networks of highly connected genes across both populations . In a study of sources of variation in yeast crosses , Bloom et al . ( 2013 ) carried out a scan for epistasis which discovered 78 pairs of loci where the effect of one was dependent on the genotype of the other , affecting 24 traits . In most cases these interactions explained little of the genetic variation in trait , the median was 3% , but in one case 14% of this variance was explained . Significant interactions between variants have also been seen to affect rice yields ( Huang et al . , 2014 ) and metabolic traits in yeast ( Wentzell et al . , 2007 ) . An extended recent review of study designs appropriate to detect epistasis in model organisms , and the evidence thus far collected , can be found in Mackay ( 2014 ) . However , epistasis has proved harder to identify in human genome-wide association studies . In particular , with classical complex traits there has not been evidence of epistasis on the scale seen in model organisms . This may be in part because of the large number of possible interactions to test in the human genome , and possibly because the genetic architecture is different in a homogeneous outbred population from that of a cross between inbred lines . Paré et al . ( 2010 ) have described how an interaction , either genetic or environmental , can induce genotype dependent variance in phenotypes . This effect can be observed without directly modeling the interacting factor . They suggested that SNPs which showed such effects on variance could be prioritized in the search for interactions . We see an example of why this could be true in Figure 1A: carriers of C allele of SNP rs230273 show reduced expression when also carriers of the G allele of SNP rs3131691 . For carriers of this G allele , this induces a bimodality in expression which appears as a large variance in expression . For those with AA genotype at rs3131691 , expression appears independent of rs230273 genotype; in the absence of the induced bimodality , the variance within this group is much reduced . The interactions causing genotype dependent variance could be with another genetic variant ( epistasis , as in our example and the focus of this paper ) or an environmental factor . 10 . 7554/eLife . 01381 . 003Figure 1 . Genotype dependent variance analysis identifies candidate SNPs for interactions . These SNPs cluster close to the transcription start site . ( A ) The plot shows expression of the gene TRIT1 , broken down by v-eQTL genotype ( rs3131691 ) , to illustrate how an interaction can be observed as an increase in variance . The genotype at rs3131691 interacts with the genotype of rs230273 . Orange individuals are carriers of the C allele at rs230273 , which decreases expression only in the AG and GG genotype groups of rs3131691 . Observing only expression conditioned on rs3131691 , this induced bimodality increases the variance of the observations within these groups . Jitter has been introduced in the x axis to reduce overplotting . ( B ) Histogram of distance from transcription start site in kilobases for the 508 peak v-eQTL hits . Figure shows the clustering of the 508 v-eQTL discovered in the TwinsUK cohort around the transcription start site , with downstream of the TSS counted as positive . The orange triangles below mark the positions of the 26 v-eQTL which replicated in the GEUVADIS cohort . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 00310 . 7554/eLife . 01381 . 004Figure 1—figure supplement 1 . Peak v-eQTL signals for 13 , 660 genes . p-values for SNPs associated with variance in gene expression ( v-eQTL ) are plotted against their genomic position . Horizontal line indicates FDR = 0 . 05 cut off . Only the most significant v-eQTL for each gene is plotted , explaining isolated signals and there being few signals with p-value >0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 00410 . 7554/eLife . 01381 . 005Figure 1—figure supplement 2 . −log10 p value for v-eQTL against–log10 p value for eQTL for 508 v-eQTL hits estimated in the TwinsUK cohort . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 00510 . 7554/eLife . 01381 . 006Figure 1—figure supplement 3 . Variance of expression of ENSG00000164978 ( NUDT2 ) is dependent on genotype dosage of rs10972055 . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 00610 . 7554/eLife . 01381 . 007Figure 1—figure supplement 4 . Variance of expression of ENSG00000105499 ( PLA2GC4 ) is dependent on genotype dosage of rs8109684 . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 00710 . 7554/eLife . 01381 . 008Figure 1—figure supplement 5 . Variance of expression of ENSG00000043514 ( TRIT1 ) is dependent on genotype dosage of rs3131691 . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 00810 . 7554/eLife . 01381 . 009Figure 1—figure supplement 6 . Variance of expression of ENSG00000075234 ( TTC38 ) is dependent on genotype dosage of rs6008743 . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 00910 . 7554/eLife . 01381 . 010Figure 1—figure supplement 7 . Variance of expression of ENSG00000164111 ( ANXA5 ) is dependent on genotype dosage of rs6857766 . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 01010 . 7554/eLife . 01381 . 011Figure 1—figure supplement 8 . Variance of expression of ENSG00000137054 ( POLR1E ) is dependent on genotype dosage of rs7033474 . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 01110 . 7554/eLife . 01381 . 012Figure 1—figure supplement 9 . Variance of expression of ENSG00000168765 ( GSTM4 ) is dependent on genotype dosage of rs542338 . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 01210 . 7554/eLife . 01381 . 013Figure 1—figure supplement 10 . Variance of expression of ENSG00000232629 ( HLA-DQB2 ) is dependent on genotype dosage of rs114183935 . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 01310 . 7554/eLife . 01381 . 014Figure 1—figure supplement 11 . Variance of expression of ENSG00000196735 ( HLA-DQA1 ) is dependent on genotype dosage of rs9276807 . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 01410 . 7554/eLife . 01381 . 015Figure 1—figure supplement 12 . Variance of expression of ENSG00000160284 ( C21orf56 ) is dependent on genotype dosage of rs16978976 . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 015 We therefore adopt the following two step strategy for uncovering epistasis affecting gene expression . We search for: ( 1 ) SNPs affecting the variance of expression ( v-eQTL ) within the 2 Mbp region around the transcription start site ( TSS ) of the gene , and then ( 2 ) SNPs in epistasis with these v-eQTL . Previous work that looked for variance QTL for height and BMI in ∼150 , 000 samples identified one replicated locus ( Yang et al . , 2012 ) . Wang et al . ( 2014 ) also looked at v-eQTL in gene expression in the same cohort as presented here , where expression was quantified using microarrays rather than sequence based technology ( Grundberg et al . , 2012 ) . They concluded that v-eQTL can often be induced by partial linkage disequilibrium with eQTL . They also discovered differences in expression between monozygotic twins which were dependent on genotype of the twin pair , such differences cannot be induced by these partial linkages and thus point to a gene–environment interaction . The haplotype effect explanation for v-eQTL , combined with a literature which has concluded in many cases epistasis does not contribute to variation in complex traits ( Hill et al . , 2008 ) , led them to conclude epistasis is not a cause of v-eQTL . However , they do not search for examples of epistasis; we do so in this paper , explicitly ruling out haplotype effects . We note that microarray data are also less suitable than RNA-seq for the purpose of detecting v-eQTL , because saturation of signal limits discrimination at extremes ( Wang et al . , 2009 ) . In neither Yang et al . ( 2012 ) nor Wang et al . ( 2014 ) were variance QTL directly used to identify epistatic or GxE interactions . Two papers have also looked at producing a phenotype related to variance , in both cases using the coefficient of variance ( CV ) within inbred lines to map variants which control the stochastic influence in phenotypic variation ( Ansel et al . , 2008; Jimenez-Gomez et al . , 2011 ) . In single cell work , and animal models where the environment can be strictly controlled , variance within inbred lines could be seen as stochastic . But we focus our work on where genotype dependent variance is the consequence of a hidden factor , in our case the presence of an interaction between genetic variants , rather than examples where the observations are due to differences in random processes . There are two other mechanisms by which genotype dependent variance can be induced . Firstly , as Sun et al . ( 2013 ) have described , standard eQTL working on mean gene expression levels can be mistaken for having variance effects in the presence of a mean–variance relationship . With RNA-seq data , the relationship between mean and variance is clear; as RNA-seq reads are sampled from a Poisson distribution , a square root transformation breaks this link . Secondly , as discussed by the Wang et al . ( 2014 ) paper described above , haplotype effects can appear as v-eQTL . For example , the situation where a recent strong eQTL co-segregates with a more common SNP ( i . e . , the SNP is in low R2 with the eQTL , but high D′ ) could be observed as variance effects of a single SNP . This could also by mistaken for epistasis between two variants which jointly tag the eQTL . We control for this possibility by explicitly considering all possible explanatory eQTL in the full sequence data available for our replication sample . We searched for v-eQTL in a dataset of 765 LCL samples from female Caucasian adult twins in the TwinsUK cohort , including 134 monozygotic ( MZ ) twin pairs and 192 dizygotic ( DZ ) pairs . The same samples from this cohort have previously been used for eQTL analysis , with expression quantified using microarrays ( Grundberg et al . , 2012 ) . The level of expression of 13 , 660 genes was determined using whole transcriptome sequencing ( RNA-seq ) . Using a non-parametric association test between SNPs within a cis window of ±1 Mbp around the TSS and the square of the residuals ( ‘Materials and methods’ ) , we identified 497 SNPs as peak v-eQTL for 508 genes ( false discovery rate ( FDR ) <0 . 05 , Figure 1—figure supplement 1; Supplementary file 1A ) , 23 reaching Bonferroni significance ( nominal p-value <8 . 9 × 10−10 ) . Many of the FDR defined v-eQTL cluster close to the TSS ( 9 . 3% are within 10 kb ) but they are found at all positions in the window ( Figure 1B ) . Of the 508 v-eQTL , 181 are also significant eQTL at a false discovery rate ( FDR ) of 0 . 05 ( Figure 1—figure supplement 2 ) . To search for epistasis , we scanned the cis windows for a second variant statistically interacting with each of the peak v-eQTL . A forward stepwise analysis identified independent examples of epistasis , not induced by linkage disequilibrium; a statistical test was applied to remove signals related to dominance ( ‘Materials and methods’ ) . This identified 256 independent SNPs in apparent epistasis with the peak v-eQTL for 173 genes ( Bonferroni , p-value <1 . 98 × 10−8; Supplementary file 1B ) . To call these signals as genuine genetic interactions we required two further criteria: ( i ) significant replication in an independent dataset , and ( ii ) that the interaction could not be explained by the effect of a third , possibly rare , variant effecting expression as discussed above . We replicated our scan for v-eQTL and epistatic interactions in 462 samples with LCL RNA-seq data from 1000 Genomes samples collected by the GEUVADIS consortium ( Lappalainen et al . , 2013 ) . Table 1 reports the results of replication for v-eQTL and epistasis using both FDR and Bonferroni correction for threshold determination . For the 23 v-eQTL that are significant using the Bonferroni threshold , 16 are significant in the GEUVADIS cohort ( FDR <0 . 05 ) , 15 with same direction of effect . Of the 508 v-eQTL , 28 replicated with an FDR <0 . 05 , 26 with same direction of effect . The ten most significant v-eQTL in the GEUVADIS cohort , with matching direction of effect across the two cohorts , are shown in Figure 1—figure supplements 3–12 . 10 . 7554/eLife . 01381 . 016Table 1 . Replication analysisDOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 016TestThresholdAssociations ( available for testing in GEUVADIS ) Replicate , FDR <0 . 05 ( % success ) Same direction of effect ( % success ) π1v-eQTLFDR <0 . 05508 ( 485 ) 28 ( 5 . 8% ) 26 ( 93% ) 0 . 30v-eQTLBonf <0 . 0523 ( 23 ) 16 ( 70% ) 15 ( 94% ) 0 . 72EpistasisBonf <0 . 05256 ( 246 ) 137 ( 56% ) 131 ( 96% ) 0 . 71Significant associations ( at FDR and Bonferroni thresholds ) from the TwinsUK sample were replicated in GEUVADIS samples . The number of overlapping SNPs and genes in both datasets per analysis is shown , as well as the percentage of replicated associations . π1 is an estimate of the proportion of replicating loci in the GEUVADIS cohort ( Storey , 2002 ) . Of the 256 epistasis associations , information on both the SNP and the gene was available for 246 in the GEUVADIS data . We found that 137 replicated with FDR <0 . 05 , 131 of which had the same direction of effect ( Supplementary file 1B ) . p-value enrichment analysis ( Storey , 2002 ) indicated that there was replication evidence for 71% of the 246 . Moreover , we observed a correlation of 0 . 58 between the effect sizes of the interactions in both datasets ( p-value = 5 . 9 × 10−24 ) , with 202 of the 246 interactions sharing the same direction of effect ( p-value = 2 . 2 × 10−25 ) ( Figure 2—figure supplements 1 , 2 ) . As discussed in the introduction , it is possible that an observed statistical interaction between two SNPs can be caused by a single true eQTL in linkage disequilibrium with them . For example , a particular combination of alleles across the pair of SNPs could tag a rare causative eQTL . To rule out this possibility , we took advantage of the full sequence for the GEUVADIS replication samples obtained by the 1000 Genomes Project ( The 1000 Genomes Project Consortium , 2012 ) . For the 131 replicated examples of epistasis we identified all eQTL for the relevant genes amongst all sequenced cis SNPs or indels ( a forward stepwise scan identified all eQTL significant with p<10−5 , ‘Materials and methods’ ) . The aim was for good characterisation of eQTL down to low frequency variants , though this is complicated by power and poorer imputation accuracy at such frequencies . We then tested whether the epistatic interaction was still significant in models incorporating each eQTL individually at the same threshold as previously applied . Fifty seven epistasis signals remain significant . Figure 2A shows the effect of the epistasis SNP broken down by genotype group on expression of TRIT1 , Table 2 and Figure 2—figure supplements 3–12 report the 10 most significant examples of epistasis in the GEUVADIS cohort , a full list is in Supplementary file 1B . For all plotted interactions , the direction of effect was consistent within v-eQTL genotype groups across cohorts . In at least two instances we see sign epistasis , the effect of one SNP reverses direction conditional on the other SNP ( Figure 2—figure supplements 7 , 9 ) . 10 . 7554/eLife . 01381 . 017Figure 2 . TRIT1 expression is affected by an interaction between two SNPs , lying on the boundaries of two separate enhancer regions , in both TwinsUK and GEUVADIS cohorts . ( A ) Expression of TRIT1 is shown , with a separate panel for each v-eQTL ( rs3131691 ) genotype group . Relationship between expression and imputed genotype dosage of the epistasis SNP ( rs230273 ) is shown to be conditional on v-eQTL genotype . Expression from TwinsUK individuals is shown in the upper panels , GEUVADIS individuals in the lower panels . Best fit lines show different SNP effects for the epistatic SNPs in different v-eQTL genotype groups , these lines are constructed ignoring twin structure in the case of the TwinsUK sample and population in the GEUVADIS cohort . ( B ) SNPs affecting TRIT1 expression are near regulatory elements . Position of v-eQTL ( rs3131691 ) , interacting epistasis SNP ( rs230273 ) and a nearby eQTL ( rs34387655 ) affecting TRIT1 expression are shown . ENCODE segmentation analysis shows regulatory elements around TRIT1 ( reverse strand gene ) . Colours indicating regions are: yellow = weak enhancer , orange = strong enhancer , red = strong promoter , light red = weak promoter , purple = poised promoter , dark green = transcriptional transition/elongation , light green = weakly transcribed , blue = insulator , and light grey = heterochromatin or repetitive/copy number variation . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 01710 . 7554/eLife . 01381 . 018Figure 2—figure supplement 1 . Evidence for epistasis in twins against evidence for epistasis in 1000 Genomes for the 246 significant hits . The 57 replicated associations after removing possible haplotype effects are shown in blue . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 01810 . 7554/eLife . 01381 . 019Figure 2—figure supplement 2 . Estimate of interaction effect size in 1000 Genomes and twins cohorts . Effect size is reported as proportion of variance explained by the interaction , where sign is positive if when both variants have the alternate allele the combined effect is a greater increase in expression than predicted by the separate additive effects , negative if expression is decreased comparatively . The 57 replicated associations are shown in blue . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 01910 . 7554/eLife . 01381 . 020Figure 2—figure supplement 3 . ENSG00000164978 ( NUDT2 ) expression is affected by an interaction between two SNPs in both TwinsUK and GEUVADIS cohorts . Expression of NUDT2 is shown , with a separate panel for each v-eQTL ( rs10972055 ) genotype group . Relationship between expression and imputed genotype dosage of the epistasis SNP ( rs10814083 ) is shown to be conditional on v-eQTL genotype . Expression from TwinsUK individuals is shown in the upper panels , GEUVADIS individuals in the lower panels . Best fit lines indicate the different epistatic SNP effects in the different v-eQTL genotype groups and are illustrative only . These lines are constructed ignoring twin structure in the case of the TwinsUK sample and population in the GEUVADIS cohort and do not represent model fit for the analysis performed . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 02010 . 7554/eLife . 01381 . 021Figure 2—figure supplement 4 . ENSG00000232629 ( HLA-DQB2 ) expression is affected by an interaction between two SNPs in both TwinsUK and GEUVADIS cohorts . Expression of HLA-DQB2 is shown , with a separate panel for each v-eQTL ( rs114183935 ) genotype group . Relationship between expression and imputed genotype dosage of the epistasis SNP ( rs1049130 ) is shown to be conditional on v-eQTL genotype . Expression from TwinsUK individuals is shown in the upper panels , GEUVADIS individuals in the lower panels . Best fit lines indicate the different epistatic SNP effects in the different v-eQTL genotype groups and are illustrative only . These lines are constructed ignoring twin structure in the case of the TwinsUK sample and population in the GEUVADIS cohort and do not represent model fit for the analysis performed . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 02110 . 7554/eLife . 01381 . 022Figure 2—figure supplement 5 . ENSG00000232629 ( HLA-DQB2 ) expression is affected by an interaction between two SNPs in both TwinsUK and GEUVADIS cohorts . Expression of HLA-DQB2 is shown , with a separate panel for each v-eQTL ( rs114183935 ) genotype group . Relationship between expression and imputed genotype dosage of the epistasis SNP ( rs9274666 ) is shown to be conditional on v-eQTL genotype . Expression from TwinsUK individuals is shown in the upper panels , GEUVADIS individuals in the lower panels . Best fit lines indicate the different epistatic SNP effects in the different v-eQTL genotype groups and are illustrative only . These lines are constructed ignoring twin structure in the case of the TwinsUK sample and population in the GEUVADIS cohort and do not represent model fit for the analysis performed . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 02210 . 7554/eLife . 01381 . 023Figure 2—figure supplement 6 . ENSG00000006282 ( SPATA20 ) expression is affected by an interaction between two SNPs in both TwinsUK and GEUVADIS cohorts . Expression of SPATA20 is shown , with a separate panel for each v-eQTL ( rs12943759 ) genotype group . Relationship between expression and imputed genotype dosage of the epistasis SNP ( rs1122634 ) is shown to be conditional on v-eQTL genotype . Expression from TwinsUK individuals is shown in the upper panels , GEUVADIS individuals in the lower panels . Best fit lines indicate the different epistatic SNP effects in the different v-eQTL genotype groups and are illustrative only . These lines are constructed ignoring twin structure in the case of the TwinsUK sample and population in the GEUVADIS cohort and do not represent model fit for the analysis performed . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 02310 . 7554/eLife . 01381 . 024Figure 2—figure supplement 7 . ENSG00000204531 ( POU5F1 ) expression is affected by an interaction between two SNPs in both TwinsUK and GEUVADIS cohorts . Expression of POU5F1 is shown , with a separate panel for each v-eQTL ( rs116627368 ) genotype group . Relationship between expression and imputed genotype dosage of the epistasis SNP ( rs115631087 ) is shown to be conditional on v-eQTL genotype . Expression from TwinsUK individuals is shown in the upper panels , GEUVADIS individuals in the lower panels . Best fit lines indicate the different epistatic SNP effects in the different v-eQTL genotype groups and are illustrative only . These lines are constructed ignoring twin structure in the case of the TwinsUK sample and population in the GEUVADIS cohort and do not represent model fit for the analysis performed . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 02410 . 7554/eLife . 01381 . 025Figure 2—figure supplement 8 . ENSG00000021355 ( SERPINB1 ) expression is affected by an interaction between two SNPs in both TwinsUK and GEUVADIS cohorts . Expression of SERPINB1 is shown , with a separate panel for each v-eQTL ( rs318452 ) genotype group . Relationship between expression and imputed genotype dosage of the epistasis SNP ( rs6940344 ) is shown to be conditional on v-eQTL genotype . Expression from TwinsUK individuals is shown in the upper panels , GEUVADIS individuals in the lower panels . Best fit lines indicate the different epistatic SNP effects in the different v-eQTL genotype groups and are illustrative only . These lines are constructed ignoring twin structure in the case of the TwinsUK sample and population in the GEUVADIS cohort and do not represent model fit for the analysis performed . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 02510 . 7554/eLife . 01381 . 026Figure 2—figure supplement 9 . ENSG00000164111 ( ANXA5 ) expression is affected by an interaction between two SNPs in both TwinsUK and GEUVADIS cohorts . Expression of ANXA5 is shown , with a separate panel for each v-eQTL ( rs6857766 ) genotype group . Relationship between expression and imputed genotype dosage of the epistasis SNP ( rs12511956 ) is shown to be conditional on v-eQTL genotype . Expression from TwinsUK individuals is shown in the upper panels , GEUVADIS individuals in the lower panels . Best fit lines indicate the different epistatic SNP effects in the different v-eQTL genotype groups and are illustrative only . These lines are constructed ignoring twin structure in the case of the TwinsUK sample and population in the GEUVADIS cohort and do not represent model fit for the analysis performed . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 02610 . 7554/eLife . 01381 . 027Figure 2—figure supplement 10 . ENSG00000137310 ( TCF19 ) expression is affected by an interaction between two SNPs in both TwinsUK and GEUVADIS cohorts . Expression of TCF19 is shown , with a separate panel for each v-eQTL ( rs115523621 ) genotype group . Relationship between expression and imputed genotype dosage of the epistasis SNP ( rs115921994 ) is shown to be conditional on v-eQTL genotype . Expression from TwinsUK individuals is shown in the upper panels , GEUVADIS individuals in the lower panels . Best fit lines indicate the different epistatic SNP effects in the different v-eQTL genotype groups and are illustrative only . These lines are constructed ignoring twin structure in the case of the TwinsUK sample and population in the GEUVADIS cohort and do not represent model fit for the analysis performed . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 02710 . 7554/eLife . 01381 . 028Figure 2—figure supplement 11 . ENSG00000204525 ( HLA-C ) expression is affected by an interaction between two SNPs in both TwinsUK and GEUVADIS cohorts . Expression of HLA-C is shown , with a separate panel for each v-eQTL ( rs114916097 ) genotype group . Relationship between expression and imputed genotype dosage of the epistasis SNP ( rs116012228 ) is shown to be conditional on v-eQTL genotype . Expression from TwinsUK individuals is shown in the upper panels , GEUVADIS individuals in the lower panels . Best fit lines indicate the different epistatic SNP effects in the different v-eQTL genotype groups and are illustrative only . These lines are constructed ignoring twin structure in the case of the TwinsUK sample and population in the GEUVADIS cohort and do not represent model fit for the analysis performed . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 02810 . 7554/eLife . 01381 . 029Figure 2—figure supplement 12 . ENSG00000176531 ( PHLDB3 ) expression is affected by an interaction between two SNPs in both TwinsUK and GEUVADIS cohorts . Expression of PHLDB3 is shown , with a separate panel for each v-eQTL ( rs10409591 ) genotype group . Relationship between expression and imputed genotype dosage of the epistasis SNP ( rs2682547 ) is shown to be conditional on v-eQTL genotype . Expression from TwinsUK individuals is shown in the upper panels , GEUVADIS individuals in the lower panels . Best fit lines indicate the different epistatic SNP effects in the different v-eQTL genotype groups and are illustrative only . These lines are constructed ignoring twin structure in the case of the TwinsUK sample and population in the GEUVADIS cohort and do not represent model fit for the analysis performed . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 02910 . 7554/eLife . 01381 . 030Figure 2—figure supplement 13 . The distance in kilobases from the 246 variants in epistasis to the v-eQTL , plotted against the –log10 p value in 1000 Genomes sample . Using the p value in the replication sample avoids inflation by winners curse . The blue dots are the 57 replicated associations after removing haplotype effects . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 03010 . 7554/eLife . 01381 . 031Table 2 . Effect size estimates and significance for the ten most significant replicated interactions in TwinsUK and GEUVADISDOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 031GeneChrv-eQTLInteracting epistasis SNPInteraction variance in TwinsUKInteraction variance in GEUVADISAdditive variation in GEUVADISp-value in TwinsUKp-value in GEUVADISNUDT29rs10972055rs10814083−0 . 328−0 . 1280 . 3101 . 88 × 10−535 . 43 × 10-22HLA-DQB26rs114183935rs1049130−0 . 337−0 . 1610 . 0991 . 83 × 10−622 . 91 × 10−21HLA-DQB26rs114183935rs9274666−0 . 368−0 . 1190 . 1583 . 45 × 10−181 . 04 × 10−16SPATA2017rs12943759rs11226340 . 3010 . 0780 . 4043 . 12 × 10−691 . 42 × 10−15POU5F16rs116627368rs1156310870 . 3110 . 1160 . 0086 . 95 × 10−346 . 63 × 10−14SERPINB16rs318452rs6940344−0 . 227−0 . 1020 . 1172 . 40 × 10−367 . 66 × 10−14ANXA54rs6857766rs12511956−0 . 411−0 . 1040 . 0563 . 09 × 10−373 . 81 × 10−13TCF196rs115523621rs115921994−0 . 585−0 . 0760 . 2012 . 59 × 10−361 . 48 × 10−11HLA-C6rs114916097rs1160122280 . 1600 . 0770 . 1833 . 35 × 10−182 . 17 × 10−11PHLDB319rs10409591rs2682547−0 . 270−0 . 08580 . 05691 . 67 × 10−144 . 83 × 10−11Effect sizes are reported as the proportion of variance explained by the interaction . Sign of effect size reflects direction of interaction effect: positive implies combined effect of the alternate alleles is an increase in expression greater than predicted by separate additive effects , and negative that it is less . We estimated the proportion of variance explained by the interaction in the GEUVADIS cohort to avoid over-estimating effects because of winner’s curse . As a result , we were able to determine that up to 16% of the variance in gene expression was explained by considering the interaction between the variants , with an average additional variance explained of 4 . 3% ( Table 2; Supplementary file 1B; Figure 3 ) . For the eight genes for which we replicated independent interactions with the v-eQTL , we found that in total up to 10 . 4% of the variance was explained by these multiple interactions , with an average of 5 . 1% . For 24 out of 57 the replicated examples of epistasis , the interaction explains more variance than the additive effects of the SNPs . We show as an example the gene TRIT1 ( Figure 2 ) . The v-eQTL ( rs3131691 ) for TRIT1 lies on the boundary of an ENCODE defined LCL weak enhancer ( Dunham et al . , 2012; Rosenbloom et al . , 2013 ) upstream of the gene , while the SNP in epistasis ( rs230273 ) lies on the boundary of a downstream LCL enhancer region ( Figure 2B ) . The v-eQTL is also 28 bp upstream of a strong eQTL signal ( rs34387655 ) . This eQTL has minor allele frequency ( MAF ) 0 . 08 , and is in high D′ with the v-eQTL ( MAF = 0 . 30 ) , suggesting that the eQTL could be a recent mutation co-segregating with one allele of the v-eQTL . But this eQTL cannot explain the observed interaction , which was still significant when analyzing only major allele homozygotes for the eQTL ( p-value = 0 . 0095 ) . Therefore , we conclude that two causal loci act on the weak enhancer in two different ways; rs34387655 has a direct effect on the enhancer while rs3131691 acts in conjunction with the epistasis variant rs230273 ( or variants in linkage disequilibrium with these SNPs act in these ways ) . 10 . 7554/eLife . 01381 . 032Figure 3 . Variance explained by additive and interacting variants for 57 replicated examples of epistasis in the GEUVADIS cohort . We show the variation explained by the interaction of two SNPs on phenotype , compared to the additive contribution of the SNPs . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 032 The discussion up to this point concerns SNPs in cis with the expressed gene . Looking for examples of trans SNPs ( >5 Mbp from the TSS ) in epistasis with the v-eQTL yielded no hits that replicated in the GEUVADIS cohort . However , using the twin design we were able to address the contribution of long range epistasis by a heritability analysis . Assuming no recombination in the cis region , the proportion of the cis window that dizygotic twins ( DZ ) inherited identically by descent is either 0 , 0 . 5 or 1 and this allows us to perform a linkage analysis to estimate the proportion of variance explained by variants in the cis region , the trans region ( 5 Mbp away from the TSS ) and interactions between the two . We had information about the IBD sharing around 273 of the 508 v-eQTL genes . For 15 of these , interactions between the cis and trans regions explain more than 10% of the variance in expression . For all of these there is greater evidence of cis-trans epistasis affecting expression than an influence of common environment , and for 9 of the 15 the interaction effect was more than the estimated combined direct genetic contribution of both cis and trans variants ( Supplementary file 1C ) . The presence of v-eQTL can be induced by gene–environment interactions , as well as epistasis or haplotype effects . Because our data come from a twin cohort , which includes monozygotic ( MZ ) twin pairs , we have another measure of variability within the dataset: the discordance in expression between MZ twins . Genotype dependent differences in expression within MZ pairs cannot be induced by epistasis or haplotype effects , as both twins share the same genetic background . Therefore , evidence that v-eQTL are also discordant eQTL ( d-eQTL ) would suggest that v-eQTL could also have a GxE explanation , including possibly interactions between the genome and the epigenome ( Martin et al . , 1983; Reynolds et al . , 2007; Figure 4A ) . Using our MZ data , we have tested our 508 v-eQTL for evidence that they are also d-eQTL; using the methods from Storey ( 2002 ) we estimate that 70% of the v-eQTL act in this manner . This suggests that GxE interactions are common amongst these variants ( ‘Materials and methods’ , Figure 4B; Supplementary file 1A ) . In total , 176 of the 508 v-eQTL show significant effects on discordance ( FDR <0 . 05 ) . Of these 176 , we estimate the proportion that are also eQTL as 40 . 3% , less than the proportion of all v-eQTL which act as eQTL . 10 . 7554/eLife . 01381 . 033Figure 4 . Increased discordance within MZ twin pairs identifies GxE interactions . ( A ) We show discordance in expression between MZ twin pairs for the gene BAMBI broken down by v-eQTL genotype ( rs10826519 ) . Discordance is greatest in the GG genotype group ( mean difference between MZ twins is 1 . 12 ) , decreasing with each additional copy of the A allele ( mean discordance is 0 . 85 for GA genotype group , 0 . 60 for AA ) . Since MZ twins are genetically identical , genotype dependent discordance in expression must be a consequence of environment , pointing to GxE . We observe that the SNP also has an effect on the mean level of expression ( p = 5 . 42 × 10−19 ) . ( B ) −log10 p values for genotype dependent discordance in MZ twins against −log10 p values for peak v-eQTL . The blue dots represent points where there is a significant epistasis hit with the v-eQTL , orange where no such interaction was detected . For many of the strong v-eQTL with little evidence of discordance we can identify an epistatic interaction which explains the increase in variance . However , for some loci with strong evidence of genotype dependent MZ discordance we also detect an epistatic interaction , suggesting both epistasis and GxE acts on these genes . DOI: http://dx . doi . org/10 . 7554/eLife . 01381 . 033 By looking at variance between individuals and discordance between monozygotic twins , we mirror an approach which looked at robustness of phenotypes to genetic and environmental influences ( Fraser and Schadt , 2010 ) . In this study of gene expression traits , differences between inbred mouse strains were called ‘genetic robustness QTL’ ( GR-QTL ) . These correspond to our definition of v-eQTL , and the paper discusses how they can be induced by epistatic interactions . The paper also looks at QTL for within strain variance , analogous to our d-eQTL and referred to as ‘environmental robustness QTL’ ( ER-QTL ) , and describe them as induced by gene–environment interactions . They reported finding both GR-QTL and ER-QTL in mice , Arabidopsis and S . cerevisiae . The importance of non-additive variation to explaining missing heritability has been much debated ( Hill et al . , 2008; Zuk et al . , 2012 ) . Here , we were able to report specific examples of interactions explaining noticeable fractions of variation in human gene expression , with in many cases the interaction contributing more than the marginal effects to overall variance . Estimating variance components from pedigrees and twin model studies has concentrated on additive variance , to estimate the narrow sense heritability . The assumption has been that resemblance between related individuals is determined chiefly by additive variation ( Falconer and Mackay , 1996 ) . An overview of analyses of many phenotypes in many organisms concluded that there was little evidence for non-additive variation playing a large role in phenotypic variation ( Hill et al . , 2008 ) . Indeed , the authors provided a theoretical argument that the total contribution of all interacting loci to variance is well approximated by their additive contribution , when the allele frequencies are as predicted by the neutral model . The analysis presented here is powered chiefly to discover common interacting variants , however the result on the neutral model implies there may be many more examples of epistasis which are not statistically detectable without very large samples . Specifically in gene expression , progress has recently been made to move beyond a solely additive view of variation . Becker et al . ( 2012 ) produced evidence for the existence of cis-trans epistasis , though they do not report individual examples which were significant when controlling for all tests and did not consider the contribution of these interactions to phenotypic variation . Further work from Powell et al . ( 2013 ) looked to dissect the phenotypes into dominant and additive components . As with our dissection of cis-trans epistasis , additive genetic variation was most consistently observed , though 960 probes had a dominant component to variation; for a subset of these a non-additive eQTL was proposed . All in all , these global results together with the replicated epistatic interactions presented here suggest a moderate influence of non-additive genetic effects on gene transcription variation . The majority of the interactions are close to each other and to the TSS ( Figure 2—figure supplement 13 ) , consistent with a direct molecular interaction . However , despite physical proximity they are , because of the statistical discovery strategy , in low linkage disequilibrium . There has been discussion in the literature about how interactions between variants affecting fitness can change the linkage disequilibrium structure of a region , by bringing variants which alter the local recombination rate under indirect selection ( Otto and Feldman , 1997 ) . In the case of positive epistasis , where the combined effect on fitness of the deleterious alleles is mitigated by their joint contribution , selection would favour a decrease in the recombination rate between the loci . This was seen in Lappalainen et al . ( 2011 ) : non-synonymous , possibly deleterious , coding mutations together with an eQTL which adjusts expression would be an example of positive epistasis . In support of the theoretical result , such variants were frequently observed in high linkage disequilibrium in their results . In contrast , the approach we take here requires linkage disequilibrium to have broken down between variants in order to distinguish an interaction between two variants from a dominant effect of a single locus . As a consequence , we are powered more to detect epistasis which amplifies the effect of deleterious mutations , rather than positive epistasis as described by Lappalainen et al . ( 2011 ) . Therefore , examples of epistasis of the type they describe would be missed by our methodology ( indeed , the five non-synonymous SNPs we discover to be involved in interactions in the TwinsUK dataset are all predicted by PolyPhen score to be benign with the exception of a one ( rs150369207 ) which is classed as possibly damaging for only one out of nine coding transcripts ) . A recent paper has also looked for evidence of epistasis affecting transcription in humans ( Hemani et al . , 2014 ) , using array expression from whole blood and searching the entire space of all possible pairwise interactions . They discover 501 interactions , affecting expression of 238 genes in 846 samples , and replicate 30 examples in an independent dataset at Bonferroni significance level . The interactions discovered are chiefly cis-trans; of the 501 there are 26 cis–cis interactions and 13 trans–trans . The apparent lower replication rate compared to our study may reflect the greater success that has been seen replicating cis effects than trans effects for standard eQTL ( Grundberg et al . , 2012 ) . Grundberg et al . ( 2012 ) also reported that LCLs ( the tissue used in our study ) showed stronger genetic effects compared to environmental contribution than seen in primary tissues . Finally , RNA-seq has been shown as a more reliable phenotype than array based measures ( Marioni et al . , 2008 ) . We believe all these factors contribute to our success rate in replicating epistatic interactions . In conclusion , we report 26 replicated variance eQTL and 57 replicated cis epistatic interactions , which explain up to 16% of the variance of our phenotypes . In almost a half of cases , more variance is explained by the interaction than by single additive effects . Furthermore , we have also shown substantial evidence for gene by environment interactions . We have shown that a proportion of variation of molecular phenotypes can be ascribed to genetic interactions , and that v-eQTL are a valid way of discovering them . Densely phenotyped cohorts are now commonly collecting such molecular data , and therefore there is considerable scope to look both for more of this type of interactions , and for the particular environments involved in GxE . The ability to find genetic interactions affecting molecular phenotypes also suggests a hypothesis driven path by which genetic interactions underlying more complex traits may be identified . Samples were genotyped on a combination of the HumanHap300 , HumanHap610Q , 1 M-Duo and 1 . 2MDuo 1M Illumnia arrays . Samples were pre-phased using IMPUTE2 ( Howie et al . , 2009 ) with no reference panel , then imputed into the 1000 Genomes Phase 1 reference panel ( interim , data freeze , 10 November 2010 , The 1000 Genomes Project Consortium 2012 ) . Post imputation , SNPs were removed if MAF <0 . 01 or IMPUTE info value <0 . 8 . Samples were prepared for sequencing with the Illumina TruSeq sample preparation kit ( Illumina , San Diego , CA ) according to manufacturer's instructions and were sequenced on a HiSeq2000 machine . Afterwards , the 49-bp sequenced paired-end reads were mapped to the GRCh37 reference genome ( The International Human Genome Sequencing Consortium , 2001 ) with BWA v0 . 5 . 9 ( Li and Durbin , 2009 ) . We use genes defined as protein coding in the GENCODE 10 annotation ( Harrow et al . , 2012 ) , removing genes with more than 10% zero read count . RPKM values were root mean transformed . PEER software ( Parts et al . , 2011 ) was used to remove 50 latent factors; age and body mass index were included when factors were constructed , to prevent removal of important environmental factors . Data were then quantile normalised . GRAMMAR ( Aulchenko et al . , 2007 ) was used to remove correlations between related individuals . Expression of each gene was tested against every SNP within 1 Mbp of the TSS . First , any eQTL effects were removed by regressing expression on the posterior probability of being a heterozygote and the posterior probability of being a minor allele homozygote . The residuals were squared , giving a measure of distance from the mean expression of that genotype class for all individuals . A Spearman rank correlation test between this ‘distance’ and genotype dosage was used to assess evidence of variance effects . A set of five permutations , consistent across all tests to consider linkage disequilibrium structure between SNPs , was applied to the distance residuals and the spearman correlation test was applied as before to estimate the distribution of the test statistic under the complete null hypothesis of no variance effects . An FDR was calculated as the proportion of permuted statistics more significant , divided by 5 . This two stage procedure where relatedness was regressed out separately from v-eQTL mapping was adopted to make the full scan for v-eQTL computationally feasible . The R package lme4 ( Bolker , 2013 ) was used to fit linear mixed models using maximum likelihood to model expression as a function of genetic interactions . The models , with a full description of how the twin structure is captured , are presented in the section ‘Equations’ . A forward stepwise scheme , as used in Lappalainen et al . ( 2013 ) to map standard eQTL , was used to discover independent examples of epistasis . Assuming the K-1 significant examples of epistasis had been discovered , a complete scan of every SNP in the cis window tested for evidence of epistasis with the v-eQTL ( using a likelihood ratio test of Equation 2 nested into Equation 1 , testing the hypothesis cK = 0 ) , conditioned on all previously discovered interactions . If the most significant SNP was Bonferroni significant ( p<1 . 98 × 10−8 ) , the SNP was added to the list and the process continued , otherwise the list was considered complete . This revealed 275 examples of epistasis , affecting expression of 178 genes . To exclude the possibility that significant interactions could be explained by a non-additive genetic effect of the original v-eQTL appearing as epistasis between the v-eQTL and another variant in tight linkage disequilibrium , a further conditional analysis tested the epistasis term conditional on the model it was discovered in and a non-additive effect of the v-eQTL ( testing nested models , Equation 3 and Equation 4 for cK = 0 ) . SNPs which were not Bonferroni significant at the same threshold ( p<1 . 98 × 10−8 ) were removed , leaving 256 epistatic interactions affecting 173 genes . Proportion of variance for linear mixed models was calculated as described in Nakagawa and Schielzeth ( 2012 ) . Scripts to analyse the data are provided in Supplementary material . Denoting individual i , expression by yi , dosage of v-eQTL by Siv , dosage of the kth discovered epistatic SNPs by Sik , probability that the v-eQTL is a heterozygote by Sivhet , and the probability that the v-eQTL is a minor allele homozygote by Sivhom , we have modelled expression in the following ways: ( 1 ) yi=μ+aSiv+∑k=1K−1 ( bkSik+ckSivSik ) +bKSiK+βi+γi+εi ( 2 ) yi=μ+aSiv+∑k=1K−1 ( bkSik+ckSivSik ) +bKSiK+cKSivSiK+βi+γi+εi ( 3 ) yi=μ+ahetSivhet+ahomSivhom+∑k=1K−1 ( bkSik+ckSivSik ) +bKSiK+βi+γi+εi ( 4 ) yi=μ+ahetSivhet+ahomSivhom+∑k=1K−1 ( bkSik+ckSivSik ) +bKSiK+cKSivSiK+βi+γi+εiwhereβi∼N ( 0 , σFAM2 ) γi∼N ( 0 , σMZ2 ) εi∼N ( 0 , σ2 ) To correctly model the twin structure we require that βi = βj when i and j are twins , and γi = γj when i and j are MZ twins ( capturing the increased genetic correlation of MZ twins ) . A variance components model was fitted in the program solar ( Almasy and Blangero , 1998 ) where the covariance matrix for the trait is written:Ω=Πcisσcis2+Πtransσtrans2+Πcis−transσcis−trans2+Iσe2 Πcis and Πtrans are the proportion of cis and trans alleles that twins share inherited identically by descent and Πcis−trans is the Hadamard product of these matrices . Parameters were estimated by maximum likelihood and proportion of variance explained by cis-trans interactions was estimated as:σcis−trans2σcis2+σtrans2+σcis−trans2+σe2 For comparison , the model without cis-trans interactions but with a common environment term was fitted , and the two models compared using likelihood . Maximum expression of the two twins was regressed on minimum expression of the twin pair and genotype of the twin pair to detect whether the relationship between max and min expression was conditional on genotype . Raw RPKM values were root transformed , 20 principal component factors were removed and then the data were quantile normalised . Evidence for v-eQTL and epistasis was calculated as before , with indicator variables for study population ( CEU , YRI , TSI , GBR , FIN ) to control for population effects . Epistasis was assessed for each SNP individually , as LD induced multiple signals and dominance effects had been removed in the TwinsUK sample . To ensure that our results are not caused by heteroskedasticity , we have considered various transformations to remove this issue and found the results to be robust . In particular , of the 131 statistically significant interactions in the GEUVADIS cohort , 126 are also significant when log transformed data is analysed ( a typical way of accounting for heteroskedasticity ) . To eliminate confounding with eQTL variants , an identical forward stepwise cis eQTL scan to that used in Lappalainen et al . ( 2013 ) reported all eQTL significant at p<10−5 in the GEUVADIS dataset . A t test for each reported eQTL assessed significance of the interaction conditional on the v-eQTL , epistasis SNP and the eQTL . If the greatest p value , over all possible eQTL , did not meet the FDR cut-off the SNP was removed from the list of interactions . FDR was calculated using the qvalue package ( Dabney and Storey , 2014 ) in R ( R Development Core Team , 2008 ) using the default settings with the exception that lambda was restricted to lie within the range of the p values to prevent overly lenient correction . The replication dataset together with functions to reproduce the results are provided in Supplementary files 2–4 . Segmentation analysis for LCL cell line GM12878 was downloaded from the UCSC website on 11/6/2013 , url: http://hgdownload . cse . ucsc . edu/goldenPath/hg19/encodeDCC/wgEncodeBroadHmm/wgEncodeBroadHmmGm12878HMM . bed . gz . Sequence data has been deposited at the European Genome-phenome Archive ( EGA , http://www . ebi . ac . uk/ega/ ) under accession number EGAS00001000805 .	Every person has two copies of each gene: one is inherited from their mother and the other from their father . These two copies are often not identical because there can be many different variants of the same gene in the human population . Traits ( such as height , body mass and risk of disease ) vary from one person to the next—and for many traits this variation depends in part on the different gene variants that each person has inherited . Studies seeking to find the differences in DNA that can predict this variation have often assumed that the changes in DNA act on traits independently of the effect of environment and of other genetic variants . In contrast , studies with animals have shown that some genetic variants can interact to produce a bigger ( or smaller ) effect than would be expected from simply ‘adding together’ their individual effects—a phenomenon called epistasis . But how much does epistasis contribute to variation in human traits , if at all ? This question has been much disputed , and is difficult to test , not least because of the sheer number of interactions to assess: tens of millions of changes in DNA have been observed in the human genome , and so there are many more than billions of possible combinations of these changes to investigate . Here , Brown et al . have examined the sequences of all the genes that were expressed in cells taken from a cohort of twins and searched for genetic variants that show these epistatic interactions . By studying gene expression , which can be greatly affected by small changes in the DNA code , Brown et al . were able to identify 508 variants that had a bigger than expected effect on the level of gene expression . This may be a sign that these variants act in combinations: if within one genome a variant increased expression and in another it decreased expression , then this would cause greater variation in gene expression . Further investigation of these 508 variants led to the discovery of 256 examples of epistasis , and 57 of these were replicated in samples from another cohort . Brown et al . calculated that these epistatic interactions explained up to 16% of the variation in gene expression . Furthermore , as well as being involved in epistatic interactions , about 70% of the genetic variants that had an effect on the variation in gene expression were also involved in interactions between genes and the environment . In addition to showing that epistasis contributes to variation in human traits , the work of Brown et al . could help to uncover interactions behind complex traits—beyond the expression level of a gene—that could not previously be investigated .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "genetics", "and", "genomics" ]	2014	Genetic interactions affecting human gene expression identified by variance association mapping
The transition from the juvenile to adult phase in plants is controlled by diverse exogenous and endogenous cues such as age , day length , light , nutrients , and temperature . Previous studies have shown that the gradual decline in microRNA156 ( miR156 ) with age promotes the expression of adult traits . However , how age temporally regulates the abundance of miR156 is poorly understood . We show here that the expression of miR156 responds to sugar . Sugar represses miR156 expression at both the transcriptional level and post-transcriptional level through the degradation of miR156 primary transcripts . Defoliation and photosynthetic mutant assays further demonstrate that sugar from the pre-existing leaves acts as a mobile signal to repress miR156 , and subsequently triggers the juvenile-to-adult phase transition in young leaf primordia . We propose that the gradual increase in sugar after seed germination serves as an endogenous cue for developmental timing in plants . After seed germination , plants undergo two developmental transitions: juvenile-to-adult and adult-to-reproductive ( Bäurle and Dean , 2006 ) . The transition from the juvenile to adult phase is marked by acquisition of reproductive competence and changes in leaf morphology ( Poethig , 2010 ) . The adult to reproductive transition , also known as flowering , transforms the identity of the shoot apical meristem from vegetative into inflorescence . Physiological and genetic studies have demonstrated that both developmental transitions are regulated not only by environmental signals such as day length , light intensity , and ambient temperature , but also by endogenous signals transmitted by plant hormones and age . microRNA156 ( miR156 ) , which targets SQUAMOSA PROMOTER BINDING PROTEIN-LIKE ( SPL ) transcriptional factors , provides an endogenous age cue for developmental timing in plants ( Poethig , 2010 ) . The expression of miR156 decreases over time , with a concomitant rise in SPL level ( Wu and Poethig , 2006; Wang et al . , 2009 ) . Overexpression of miR156 prolongs the juvenile phase , whereas a reduction in miR156 level results in an accelerated expression of adult traits ( Wu and Poethig , 2006; Wu et al . , 2009 ) . SPL promotes the juvenile-to-adult phase transition and flowering through activation of miR172 and MADS-box genes ( Wang et al . , 2009; Wu et al . , 2009; Yamaguchi et al . , 2009; Jung et al . , 2011 ) . Very recently , defoliation experiments and expression analyses demonstrated that the repression of miR156 in the leaf primordia is mediated by a mobile signal ( s ) derived from the pre-existing leaves ( Yang et al . 2011 ) . However , the identity of this signal is still unknown . In addition to being essential as prime carbon and energy sources , sugars also play critical roles as signaling molecules ( Rolland and Sheen , 2005; Smeekens et al . , 2010 ) . In Arabidopsis thaliana , diverse sugar signals are perceived and transduced through a glucose sensor , HEXOKINASE1 ( HXK1 ) . HXK1 exerts its regulatory function through distinct molecular mechanisms including transcriptional activation , translational inhibition , mRNA decay , and protein degradation ( Rolland and Sheen , 2005 ) . Analyses of two catalytic inactive HXK1 alleles further indicate that the signaling activity of HXK1 is uncoupled from its catalytic activity ( Moore et al . , 2003 ) . Recently , a nuclear HXK1 complex has been identified ( Cho et al . , 2006 ) . In this complex , HXK1 binds to two unconventional partners , the vacuolar H+-ATPase B1 ( VHA-B1 ) and the 19S regulatory particle of a proteasome subunit ( RPT5B ) . Since neither VHA-B1 nor RPT5B has DNA binding capacity , the precise molecular mechanism by which this nuclear-localized HXK1 complex regulates gene expression remains unanswered . In addition to the HXK1-dependent pathway , some glucose-responsive genes are regulated through an HXK1-independent pathway . For instance , the expression of the genes encoding chalcone synthase , phenylalanine ammonia-lyase , and asparagine synthase responds to glucose signaling in the absence of HXK1 ( Xiao et al . , 2000 ) . Here , we performed expression and mutant analyses to identify the upstream regulator of miR156 . Our results demonstrate that the expression of miR156 quickly responds to sugar . Sugar reduces miR156 abundance through both transcriptional repression and transcript degradation . Thus , gradual accumulation of sugar after seed germination leads to a reduced level of miR156 , which promotes the juvenile-to-adult phase transition in plants . The transition from juvenile to adult phase in Arabidopsis is accompanied by changes in vegetative morphology . Under long day conditions , the wild type Arabidopsis plants switch from the juvenile to the adult phase from the fifth or sixth leaf . The juvenile leaves are round , smooth on their margins , and barely develop trichomes ( leaf hairs ) on the abaxial side ( lower side ) . By contrast , the adult leaves are elongated , serrated , and produce abaxial trichomes ( Wu et al . , 2009 ) . In the Arabidopsis genome , miR156 is encoded by eight coding loci ( MIR156A–MIR156H ) ( Reinhart et al . , 2002 ) . To understand which locus or loci play important roles within this gene family , we identified all available MIR156 transfer-DNA ( T-DNA ) knockout plants ( Samson et al . , 2002; Alonso et al . , 2003; Woody et al . , 2007; Figure 1A and supplementary file 1A ) . Due to functional redundancy , none of these mutants exhibited visible developmental defects ( data not shown ) . One of the double mutants , mir156a mir156c , displayed a similar , but weak phenotype as the transgenic plant expressing a target mimicry from the constitutively active 35S promoter ( 35S::MIM156 ) , which reduced miR156 activity ( Figure 1D; Franco-Zorrilla et al . , 2007; Todesco et al . , 2010 ) . RNA gel blot demonstrated that the amount of miR156 was moderately decreased in mir156a mir156c in comparison with the wild type ( Figure 1B ) . Accordingly , the transcript levels of two miR156-target genes , SPL3 and SPL9 , were much higher in mir156a mir156c than in the wild type ( Figure 1C ) . 10 . 7554/eLife . 00269 . 003Figure 1 . Phenotypic analyses of the mir156a mir156c double mutant . ( A ) MIR156A and MIR156C genomic regions . Arrowheads mark T-DNA insertion sites . T-DNAs are inserted 137 bp and 218 bp upstream of the stem-loops of MIR156A and MIR156C , respectively . ( B ) Expression of miR156 in the wild type and the mir156a mir156c double mutant . U6 was monitored as loading control . ( C ) Expression of SPL3 and SPL9 in the wild type and the mir156a mir156c double mutant . The expression level in the wild type was set to 1 . 0 . ( D ) Leaf morphology of wild type , mir156a mir156c , and 35S::MIM156 plants . The leaves were detached and scanned . The numbers indicate leaf positions . ( E ) The number of juvenile and adult leaves . n=12 . ( F ) The length-to-width ratio of the blade . Fully expanded leaves were detached and scanned . The length and width of blades were measured . n=12 . Error bars indicate SE . DOI: http://dx . doi . org/10 . 7554/eLife . 00269 . 003 Compared to the wild type , the mir156a mir156c mutant had a shortened juvenile phase . The appearance of abaxial trichomes in mir156a mir156c was accelerated by 2 . 1 plastochrons ( Figure 1E ) . In addition , the length-to-width ratios of the blades in mir156a mir156c were much closer to those of the adult leaves in the wild type ( Figure 1F ) . Furthermore , mir156a mir156c flowered earlier than the wild type ( Figure 1E ) . Taken together , these results indicate that MIR156A and MIR156C have dominant roles within the miR156 family in Arabidopsis . To elucidate the molecular mechanism by which the level of miR156 is regulated by age , we performed time course expression assays on miR156 and the primary transcripts of MIR156A and MIR156C ( pri-MIR156A and pri-MIR156C ) by RNA gel blot and quantitative real-time PCR ( qRT-PCR ) . We collected plants grown under long day conditions for 8 , 9 , and 16 days . As previously reported , the abundance of miR156 gradually declined ( Figure 2A , B; Wu and Poethig , 2006; Wang et al . , 2009 ) . Interestingly , the transcript levels of pri-MIR156A and pri-MIR156C , but not mature miR156 , exhibited damped oscillations with the highest level in the morning and lowest before dark ( Figure 2C , E; Figure 2—figure supplement 1 ) . To test whether this expression pattern is generated by the circadian clock , we grew wild type plants for 5 days in long day conditions , and then transferred them to a constant light condition . After the transfer , the oscillating expression pattern of pri-MIR156A and pri-MIR156C was no longer observed ( Figure 2D , F ) , demonstrating a negligible effect of the circadian clock on miR156 expression . 10 . 7554/eLife . 00269 . 004Figure 2 . Expression of miR156 . ( A and B ) Accumulation of miR156 in 8- , 9- , and 16-day-old long day plants . Expression of miR156 was analyzed by small RNA blot ( A ) and qRT-PCR ( B ) . The plants were collected at Zeitgeber time ( ZT ) 24 . The expression level of miR156 in 8-day-old seedlings was set to 1 . ( C and E ) Expression of pri-MIR156A ( C ) and pri-MIR156C ( E ) . The plants were collected every 4 hr and subjected to qRT-PCR analyses . Black and white boxes indicate dark and light conditions , respectively . ( D and F ) Expression of pri-MIR156A ( D ) and pri-MIR156C ( F ) during the shift from long day ( LD ) to constant light ( CL ) conditions . Five-day-old wild type seedlings were shifted from long day to constant light conditions . The seedlings were collected at ZT 16 , 24 , and 32 . DOI: http://dx . doi . org/10 . 7554/eLife . 00269 . 00410 . 7554/eLife . 00269 . 005Figure 2—figure supplement 1 . Expression pattern of miR156 . Expression of miR156 . The plants were collected every 4 hr and subjected to qRT-PCR analyses . ZT: Zeitgeber time . DOI: http://dx . doi . org/10 . 7554/eLife . 00269 . 005 In addition to the circadian clock , endogenous carbohydrates are also able to trigger the oscillation of RNA transcripts ( Bläsing et al . , 2005 ) . To test this possibility , we carried out sugar treatment assays . Five-day-old seedlings grown in 1/2 Murashige and Skoog ( MS ) liquid media were treated with sugars , including two disaccharides ( maltose and sucrose ) and two hexoses ( glucose and fructose ) . The break-down of maltose results in two glucose molecules , whereas hydrolysis of sucrose produces glucose and fructose . The abundance of pri-MIR156A and pri-MIR156C was greatly reduced after 1 day of treatment with 50 mM sucrose , glucose , or maltose ( Figure 3A ) . A reduction in pri-MIR156A or pri-MIR156C was not detected when the seedlings were treated with the same concentration of mannitol or 3-O-methyl-glucose ( 3-OMG ) , suggesting that the repression of pri-MIR156 by sugars is not due to osmotic stress . Consistent with the reduction in pri-MIR156 levels , mature miR156 was decreased after 1 day of sugar treatment ( Figure 3A; Figure 3—figure supplement 1 ) . Accordingly , the transcript levels of miR156-targeted genes , SPL9 and SPL15 , were markedly increased ( Figure 3B ) . 10 . 7554/eLife . 00269 . 006Figure 3 . Sugar represses miR156 . ( A ) Expression of miR156 , pri-MIR156A , and pri-MIR156C in response to sugar . Five-day-old wild type seedlings in 1/2 Murashige and Skoog ( MS ) liquid media were treated with 50 mM sucrose ( Suc ) , glucose ( Glc ) , fructose ( Fru ) , maltose ( Malt ) , or mannitol ( Man ) for 1 day . ( B ) Expression of SPL9 and SPL15 in response to sugar treatment . Five-day-old wild type seedlings were treated with 50 mM Man or Glc for 1 day . ( C ) pri-MIR156C quickly responds to sugar . Five-day-old wild type seedlings were treated with sugar for 30 min . The expression level in the mannitol-treated samples was set to 1 . ( D ) Expression of pri-MIR156 transcripts . Five-day-old wild type seedlings in 1/2 MS liquid media were treated with 50 mM glucose or mannitol for 1 day . ( E ) Expression of miR156 and pri-MIR156C during sugar starvation . Five-day-old wild type seedlings in 1/2 MS liquid media supplemented with 50 mM sucrose were transferred to 1/2 MS media without sucrose ( MS0 ) . The seedlings were grown for another 2 days and then subjected to expression analyses . Seven-day-old seedlings in 1/2 MS liquid media supplemented with 50 mM sucrose were used as control . ( F ) Expression of other pri-MIRNA transcripts . Five-day-old wild type seedlings in 1/2 MS liquid media were treated with 50 mM glucose or mannitol for 1 day . The expression levels of pri-MIR156 and miR156 were normalized to those of TUBULIN ( TUB ) . In the sugar treatment assays , 50 mM sugars were added at Zeitgeber time 12 . DOI: http://dx . doi . org/10 . 7554/eLife . 00269 . 00610 . 7554/eLife . 00269 . 007Figure 3—figure supplement 1 . Sugar represses miR156 . Accumulation of miR156 in response to sugar . Five-day-old wild type seedlings were treated with sugar for 1 day and subjected to RNA blot analyses . U6 was monitored as an internal control . Sugar treatment started at Zeitgeber time 12 . Man: mannitol; Suc: sucrose; Glc: glucose; Malt: maltose . DOI: http://dx . doi . org/10 . 7554/eLife . 00269 . 007 To monitor how fast miR156 responds to sugar , wild type seedlings were treated with glucose , sucrose , maltose , fructose , or mannitol for 30 min . A reduction of about 40% in pri-MIR156C was observed in the seedlings treated with glucose , sucrose , or maltose , while the level of pri-MIR156C was not altered in those treated with fructose or mannitol ( Figure 3C ) . These results , together with the fact that glucose is the common hydrolytic product shared by sucrose and maltose , suggest that glucose plays a major role in repressing miR156 . To determine whether all the miR156 coding genes are repressed by sugar , we analyzed the expression of their primary transcripts . pri-MIR156G and pri-MIR156H were not readily amplified , probably due to their very low expression level ( data not shown ) . The expression of other pri-MIR156 transcripts except pri-MIR156B was reduced after glucose treatment ( Figure 3D ) . To confirm the role of sugar in miR156 expression , we performed a sugar starvation experiment . Five-day-old wild type seedlings were transferred to 1/2 MS liquid media free of sugar and kept in the dark for 2 days . Compared to the seedlings grown in 1/2 MS liquid media supplemented with sugar under normal light conditions , the sugar-depleted seedlings exhibited a higher expression level of miR156 ( Figure 3E ) . To investigate whether sugar specifically represses miR156 , we analyzed the expression of other miRNA primary transcripts , including pri-MIR159A , pri-MIR159B , and pri-MIR165A . The levels of all these transcripts were not reduced after sugar treatment ( Figure 3F ) . A recent study has shown that the juvenile-to-adult phase transition is mediated by a leaf-derived mobile signal that represses the expression of miR156 in young leaf primordia ( Yang et al . 2011 ) . Given the fact that sucrose is able to move within plants through the vascular tissues ( Truernit 2001 ) and that sucrose as well as its hydrolytic product , glucose , repress the expression of miR156 , we speculated that sugar is a potential candidate for this mobile signal . To test this hypothesis , we first investigated the relationship between sugar content and the level of miR156 in vivo . Under long day conditions , Arabidopsis plants show a rapid life cycle with very short juvenile and adult phases . For this reason , we grew wild type plants under short day conditions to extend the vegetative phase . Then 15-day-old ( in the juvenile phase ) and 60-day-old ( in the adult phase ) plants were collected at Zeitgeber time ( ZT ) 16 . Expression analyses demonstrated that miR156 was highly abundant in 15-day-old plants but less so in 60-day-old plants ( Figure 4A ) . In contrast to this expression pattern , 60-day-old plants exhibited a higher level of glucose , fructose , and sucrose than 15-day-old plants ( Figure 4B ) . These results are consistent with our findings that sugar represses miR156 and indicate an inverse correlation between the level of miR156 and endogenous sugar content in vivo . 10 . 7554/eLife . 00269 . 008Figure 4 . Sugar as a mobile signal to trigger vegetative phase transition . ( A ) Expression of miR156 in 15-day-old and 60-day-old wild type plants grown under short day conditions . ( B ) Sugar measurement . Fifteen-day-old and 60-day-old short day plants were collected at Zeitgeber time 16 . The fructose ( Fru ) , glucose ( Glc ) , and sucrose ( Suc ) content was analyzed by GC-MS and quantified . Significant difference from 15-day-old wild type plants , Student t-test , p<0 . 001 . Error bars indicate SD . n . d . : undetected; FW: fresh weight . ( C ) Seven-day-old wild type Arabidopsis seedlings before and after defoliation . Arrows indicate where the lanolin-sucrose ( Suc ) paste was applied . Scale bar indicates 0 . 5 cm . ( D and E ) Seven-day-old wild type seedlings before and after defoliation . Appearance of the first abaxial trichome ( D ) and the length-to-width ratios of blades ( E ) were measured . n=10 . Significant difference from wild type , Student t-test , p<0 . 001 . Error bars indicate SE . defol: defoliated; Suc: sucrose . ( F ) Expression of miR156 . Seven-day-old wild type seedlings were defoliated and sucrose ( Suc ) or mannitol ( Man ) was applied to the defoliated petioles . The shoot apices were collected for expression analyses 2 days after defoliation . DOI: http://dx . doi . org/10 . 7554/eLife . 00269 . 008 We then performed defoliation assays . The blades of the first two leaves of 7-day-old wild type seedlings were manually removed . Then 50 mM sucrose or mannitol ( as control ) was applied to the petioles of the defoliated leaves ( Figure 4C ) . Consistent with the previous report ( Yang et al . 2011 ) , the removal of the first two leaves resulted in an increased level of miR156 in the shoot apices ( Figure 4F ) . The expression of adult-specific traits was accordingly delayed . Compared to intact plants , the production of abaxial trichomes in the defoliated plants was delayed by 1 . 0 plastochrons ( Figure 4D ) , and the increase in the length-to-width ratio of the lamina was slower ( Figure 4E ) . Sucrose application partially suppressed the delay in the juvenile-to-adult phase transition caused by defoliation . The sucrose-treated plants produced the abaxial trichomes 0 . 8 plastochrons later than intact wild type plants , but 1 . 6 plastochrons earlier than the mannitol-treated plants ( Figure 4D ) . In addition , the length-to-width ratios of the fifth , seventh , and ninth leaves in the sucrose-treated plants were higher than those in the mannitol-treated plants ( Figure 4E ) . In agreement with these phenotypic differences , the expression of miR156 was reduced in the apices of the sucrose-treated plants but not in those treated with mannitol ( Figure 4F ) . To confirm the role of sugar in the juvenile-to-adult phase transition , we analyzed the Arabidopsis cao/chlorina1 ( ch1 ) mutant . A mutation in CAO/CH1 ( At1g44446 ) , which encodes chlorophyll ( Chl ) a oxygenase , causes a reduced level of Chl b and low efficiency of photosynthesis ( Espineda et al . , 1999 ) . Compared to the wild type , the cao/ch1 mutant developed smaller pale green leaves and had a prolonged juvenile phase ( Figure 5—figure supplement 1 ) . The rosette leaves in the cao/ch1 mutant were rounder than those in the wild type plant ( Figure 5A , B ) . Additionally , the appearance of abaxial trichomes in the cao/ch1 mutant was delayed ( Figure 5C ) . Expression analyses indicated that higher levels of miR156 accumulated in the cao/ch1 mutant than in the wild type plant ( Figure 5D ) . 10 . 7554/eLife . 00269 . 009Figure 5 . cao/ch1 mutant impairs vegetative phase transition . ( A ) Leaf morphology of wild type , cao/ch1 , and 35S::MIM156 cao/ch1 plants under long day conditions . The leaves from 15-day-old plants were detached and scanned . The numbers indicate leaf positions . ( B and C ) The length-to-width ratio of the blade ( B ) and the appearance of the first abaxial trichome ( C ) . n=12 . ( D ) Expression of miR156 during development . Wild type plants and cao/ch1 mutants were collected at 7 , 9 , or 12 days after germination under long day conditions . ( E ) Expression of miR156 , pri-MIR156A , and pri-MIR156C . Five-day-old wild type and cao/ch1 mutants in 1/2 Murashige and Skoog ( MS ) liquid media were treated with 50 mM glucose or mannitol for 1 day . The expression levels in the mannitol-treated wild type or cao/ch1 were set to 1 . The treatment was started at Zeitgeber time 12 . **Significant difference from wild type , Student t-test , p<0 . 001 . Error bars indicate SE . DOI: http://dx . doi . org/10 . 7554/eLife . 00269 . 00910 . 7554/eLife . 00269 . 010Figure 5—figure supplement 1 . Phenotype of cao mutant . Plant morphology of wild type ( wt ) , cao/ch1 , and 35S::MIM156 cao/ch1 . Scale bar indicates 1 . 0 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 00269 . 010 To examine whether the delayed phase transition in cao/ch1 depends on miR156 function , we crossed 35S::MIM156 into cao/ch1 . Similarly to 35S::MIM156 , 35S::MIM156 cao/ch1 produced the abaxial trichomes on the first leaf , and the leaves were elongated and serrated ( Figure 5A–C; Figure 5—figure supplement 1 ) . Compared to the wild type , the cao/ch1 mutants exhibited higher glucose sensitivity . Treatment of cao/ch1 seedlings with 50 mM glucose significantly reduced the level of miR156 ( Figure 5E ) . Taken together , we conclude that sugar from the pre-existing leaves acts as a mobile signal to trigger the juvenile-to-adult phase transition through repression of miR156 in the young leaf primordia . miR156 is present in all major plant taxa ( Axtell and Bowman , 2008 ) . To test whether the regulation of miR156 by sugar is evolutionarily conserved , we examined the expression of miR156 in response to sugar in other plants , including Nicotiana benthamiana ( tobacco ) , Physcomitrella patens ( moss ) , and Solanum lycopersicum ( tomato ) . N . benthamiana and S . lycopersicum were grown in 1/2 MS liquid media without sugar . After the first two leaves appeared , the seedlings were treated with 50 mM sucrose for 2 days . The seedlings of N . benthamiana and S . lycopersicum were collected and used for expression analyses . For P . patens , the sugar treatment was conducted during the protonema stage . Compared to those treated with mannitol , the amount of miR156 was greatly reduced in all the sucrose-treated plants ( Figure 6 ) , indicating that repression of miR156 by sugar is evolutionarily conserved . 10 . 7554/eLife . 00269 . 011Figure 6 . Repression of miR156 by sugar is evolutionarily conserved . Expression of miR156 in Physcomitrella patens , Solanum lycopersicum , and Nicotiana benthamiana . The plants were treated with 50 mM sucrose ( Suc ) or mannitol ( Man ) for 2 days . U6 was monitored as the loading control . Treatment was started at Zeitgeber time 12 . DOI: http://dx . doi . org/10 . 7554/eLife . 00269 . 011 To investigate at which level sugar represses miR156 , we performed chromatin immunoprecipitation analyses ( ChIP ) using anti-RNA polymerase II ( anti-Pol II ) antibody , which recognizes the C-terminal heptapeptide repeat of RNA Pol II and has been used to correlate RNA Pol II binding with gene expression . Enrichment of the promoter fragments of MIR156A and MIR156C was compared between the seedlings treated with mannitol and those treated with glucose . As shown in Figure 7A , the promoter fragments ( harboring TATA boxes ) of MIR156A and MIR156C were substantially enriched in the mannitol-treated seedlings , but not in those treated with glucose , indicating that glucose induces transcriptional repression of MIR156A and MIR156C ( Figure 7A ) . 10 . 7554/eLife . 00269 . 012Figure 7 . Sugar promotes the degradation of miR156 primary transcripts . ( A ) Chromatin immunoprecipitation ( ChIP ) analyses . Five-day-old wild type seedlings were treated with 50 mM glucose ( Glc ) or mannitol ( Man ) for 1 day . Anti-Pol II was used for ChIP analyses . The genomic fragments near the MIR156A or MIR156C TATA box were amplified . Relative enrichment was calculated by the ratio of bound DNAs after ChIP to input DNAs . ( B ) Expression of HXK1 in response to glucose . Five-day-old wild type seedlings in 1/2 Murashige and Skoog ( MS ) liquid media were pre-treated with or without actinomycin ( ActD ) for 12 hr . The seedlings were harvested at 0 , 1 , and 3 hr after 50 mM glucose or mannitol was added . The expression level at 0 hr was set to 1 . ( C and D ) Expression of pri-MIR156A ( C ) and pri-MIR156C ( D ) in the wild type and se-3 mutant . Five-day-old wild type seedlings in 1/2 MS liquid media were pre-treated with ActD for 12 hr . The seedlings were then treated with 50 mM glucose or mannitol . The expression levels of pri-MIR156A and pri-MIR156C in the wild type at 0 hr were set to 1 . Sugar treatment was started at Zeitgeber time 12 . DOI: http://dx . doi . org/10 . 7554/eLife . 00269 . 01210 . 7554/eLife . 00269 . 013Figure 7—figure supplement 1 . Effect of CHX on sugar-induced pri-MIR156C degradation . Five-day-old wild type seedlings in 1/2 Murashige and Skoog liquid media were pre-treated with actinomycin-D ( ActD ) for 12 hr . Glucose was added 1 h after 100 µM cycloheximide ( CHX ) . The levels in the mannitol-treated samples ( mock ) were set to 1 . Glucose ( 50 mM ) was added at Zeitgeber time 12 . DOI: http://dx . doi . org/10 . 7554/eLife . 00269 . 01310 . 7554/eLife . 00269 . 014Figure 7—figure supplement 2 . Expression analyses of pri-MIR156A and pri-MIR156C in upf mutants . ( A ) Expression of pri-MIR156A and pri-MIR156C in upf1-5 and upf3-1 mutants . Seven-day-old wild type ( WT ) , upf1-5 , and upf3-1 seedlings were used for expression analyses . ( B ) Glucose response in upf mutants . Five-day-old wild type and upf1-5 seedlings in 1/2 Murashige and Skoog liquid media were pre-treated with actinomycin-D ( ActD ) for 12 h . The transcript level of pri-MIR156C was monitored at 0 , 3 , and 6 hr after glucose ( Glc ) or mannitol ( Man ) treatment . Glucose ( 50 mM ) was added at Zeitgeber time 12 . DOI: http://dx . doi . org/10 . 7554/eLife . 00269 . 014 We next examined the effect of actinomycin-D ( ActD ) , which blocks transcription . To test transcription blocking efficiency , we analyzed the expression of HXK1 , which is rapidly induced by glucose ( Price et al . , 2004 ) . The transcript level of HXK1 was increased about fourfold after 3 h of glucose treatment . By contrast , the expression of HXK1 was not altered in the seedlings treated with glucose and ActD ( Figure 7B ) . The addition of ActD did not affect repression of pri-MIR156C by glucose . The transcript level of pri-MIR156C was reduced by about 75% after 3 hr in the presence of glucose , compared to a 30% reduction in the presence of mannitol ( Figure 7D ) . A similar expression pattern was observed in pri-MIR156A ( Figure 7C ) , suggesting that glucose modulates miR156 expression at the post-transcriptional level through the degradation of pri-MIR156 . To investigate whether the reduction in miR156 primary transcripts after glucose treatment was caused by an increase in the processing efficiency of pri-MIR156 , we performed the glucose treatment assay using the serrate ( se ) mutant which is defective in miRNA biogenesis ( Grigg et al . , 2005; Lobbes et al . , 2006; Yang et al . , 2006; Laubinger et al . , 2008 ) . Similar to the wild type , the amount of pri-MIR156A and pri-MIR156C was markedly decreased in the ActD/glucose-treated se-3 mutant ( Figure 7C , D ) , indicating that glucose regulates the abundance of pri-MIR156 independently of the miRNA processing machinery . HXK1 encodes a glucose sensor that transduces diverse sugar signals . gin2-1 , the HXK1-null mutant ( Moore et al . 2003 ) , exhibited a lower level of miR156 than the wild type ( Figure 8A ) . The expression of miR156 still decreased over time in the gin2-1 mutant ( Figure 8B ) . To test whether the repression of miR156 by sugar is mediated by HXK1 , we compared the glucose response between the wild type and the gin2-1 mutant . The expression of pri-MIR156A and pri-MIR156C was reduced after sugar treatment in both the wild type and the gin2-1 mutant ( Figure 8C , D ) . Similarly , an evident decrease in pri-MIR156C was observed in the gin2-1 seedlings treated with ActD/glucose ( Figure 8E ) . These results suggest that HXK1 plays a role in miR156 expression but is not absolutely required for the repression of miR156 by sugar . 10 . 7554/eLife . 00269 . 015Figure 8 . The role of HXK1 in sugar-induced miR156 repression . ( A ) Expression of miR156 in the 5-day-old wild type ( ecotype Ler ) and gin2-1 mutant . The expression level of miR156 in Ler was set to 1 . ( B ) Time course analyses of miR156 in the gin2-1 mutant . ( C and D ) Expression of pri-MIR156A ( C ) and pri-MIR156C ( D ) in response to glucose in the wild type ( ecotype Ler ) and gin2-1 mutant . Five-day-old seedlings in 1/2 Murashige and Skoog ( MS ) liquid media were treated with 50 mM glucose ( Glc ) or mannitol ( Man ) for 6 hr . The expression level in Ler at 0 h was set to 1 . ( E ) Expression of pri-MIR156C in Ler and gin2-1 . Five-day-old seedlings in 1/2 MS liquid media were pre-treated with actinomycin-D ( ActD ) for 12 hr and then treated with 50 mM glucose or mannitol . The expression level of pri-MIR156C in Ler at 0 hr was set to 1 . Sugar treatment was started at Zeitgeber time 12 . DOI: http://dx . doi . org/10 . 7554/eLife . 00269 . 015 We then performed the sugar treatment assay in the presence of ActD and cycloheximide ( CHX ) , an inhibitor of protein synthesis . The level of pri-MIR156C transcripts was greatly reduced in the ActD-treated samples , but not in those treated with both ActD and CHX ( Figure 7—figure supplement 1 ) , suggesting that sugar-induced pri-MIR156C degradation requires de novo protein synthesis . mRNAs can be degraded through several partially independent pathways , including nonsense-mediated mRNA decay ( NMD ) , 5′-to-3′ mRNA degradation via exonucleases , and 3′-to-5′ mRNA degradation via the exosome . The UP-frameshift ( UPF ) proteins , UPF1 , UPF2 , and UPF3 , are essential for the NMD function in plants ( Arciga-Reyes et al . , 2006; Kurihara et al . , 2009 ) . It has been shown that upf1 and upf3 mutants impair the sugar response and over-accumulate sugar-inducible mRNAs ( Yoine et al . 2006 ) . Therefore , we investigated the role of UPF in the sugar-mediated repression of miR156 . Compared to the wild type , the expression of pri-MIR156A and pri-MIR156C was slightly increased in upf1-5 and upf3-1 mutants ( Figure 7—figure supplement 2A ) . Glucose was still able to repress the accumulation of pri-MIR156C , albeit to a lesser extent ( Figure 7—figure supplement 2B ) , indicating that sugar promotes pri-MIR156 degradation independently of canonical NMD . Based on expression analyses , defoliation experiments , and photosynthetic mutant characterization , we show that sugar acts upstream of miR156 . We propose a model explaining how sugar regulates the juvenile-to-adult phase transition through modulation of miR156 expression as follows . After seed germination , plants start accumulating sugars through photosynthesis . Sucrose , the major transportable sugar , moves from the pre-existing leaves to the young leaf primordia , where its hydrolytic hexose product , glucose , represses the expression of miR156 . As a result , the level of SPL increases and the expression of adult traits is promoted . Identification of sugar as the endogenous developmental timing cue explains the irreversible nature of the age pathway . The level of miR156 is destined to decrease because the gradual accumulation of carbohydrates is inevitable and essential for plant growth and development . In Caenorhabditis elegans , the transitions between the stages of larval development are controlled by the sequential action of two miRNAs , lin-4 and let-7 ( Pasquinelli and Ruvkun , 2002; Moss , 2007; Ambros , 2011 ) . In contrast to miR156 , the expression of these two miRNAs is increased with age . It will be intriguing to examine whether sugar/carbohydrates or nutrients from the diet triggers the upregulation of lin-4 and let-7 in worms . Cellular carbon ( C ) and nitrogen ( N ) are tightly coordinated to sustain optimal plant growth ( Raven et al . , 2004; Zheng , 2009 ) . C compounds including many carbohydrates such as sucrose and glucose are synthesized in the leaf , while N nutrients such as nitrate ( NO3- ) and ammonium ( NH4+ ) are assimilated by the root system . Biochemical and physiological studies have demonstrated long-distance sensing and signaling of the C/N balance in plants . When soil is short of NH4+ and NO3- , photosynthesis in the leaf is inhibited . Whether miR156 and the juvenile-to-adult transition respond to N excess or deficient conditions is another interesting topic awaiting investigation . In addition to the juvenile-to-adult transition , flowering is of great importance for reproductive success in plants . Previous studies revealed that the floral transition is regulated by diverse environmental factors , such as photoperiod , temperature , and light , in combination with the endogenous signal derived from nutritional status . The nutrient-dependent regulation of flowering is likely dependant on the rate of sucrose export from source leaves ( Corbesier et al . , 1998; Sivitz et al . , 2007 ) . This notion is supported by our observations that sugar from pre-existing leaves acts as a long-distance signal to repress the expression of miR156 in young leaf primordia , and that a high level of miR156 delays flowering ( Wang et al . , 2009 ) . Intriguingly , a recent study has demonstrated that INDETERMINATE DOMAIN transcription factor AtIDD8 regulates photoperiodic flowering by modulating sugar transport and metabolism ( Seo et al . , 2011 ) , suggesting that additional sugar-mediated flowering pathways exist . Sugar , produced in mesophyll cells in leaves , is transported from source tissues to sink tissues through vascular bundles ( Kuhn and Grof , 2010; Ayre , 2011 ) . In Arabidopsis , sucrose transporters are involved in loading sucrose into the phloem in source leaves and the uptake of sucrose into the cells of sink tissues such as roots , fruit , and developing leaves ( Williams et al . , 2000 ) . Very recently , the sucrose effluxers , SWEET11 and SWEET12 , which facilitate sucrose efflux into the cell wall of companion cells , have been identified ( Chen et al . , 2011 ) . It is therefore interesting to investigate whether impairment of sucrose transport from leaf cells into the vascular system causes a defect in miR156 expression and developmental transitions . There are several means by which sugar regulates gene expression . For example , sugar decreases the transcript level of rice AMY3 at both the transcriptional and post-transcriptional level . It was shown that destabilization of the mRNAs of AMY3 is mediated by its 3′ untranslated region ( UTR ) ( Chan and Yu , 1998 ) . Similarly , we found that pri-MIR156A and pri-MIR156C are subjected to transcriptional repression as well as transcript degradation in response to glucose . This two-level expression control by sugar might contribute to robust repression of miR156 , which leads to irreversible transition from the juvenile to adult phase in plants . In Arabidopsis , HXK1 is a glucose sensor that transduces diverse aspects of sugar response . For example , the gin2-1 mutant reduces shoot and root growth , delays flowering , increases apical dominance , and alters sensitivity to auxin and cytokinin ( Moore et al . , 2003 ) . However , we did not observe an obvious juvenile-to-adult phase phenotype in the gin2-1 mutant under long day conditions ( data not shown ) . Further studies will determine if the transcriptional repression of miR156 by sugar is mediated by the previously identified nuclear-localized HXK1-VHA-B1-RPT5B complex . The level of miR156 is greatly reduced when plants are treated with both glucose and sucrose . Since these sugars can be easily interconverted , it remains unclear whether the repression of miR156 is hexose or sucrose-dependent . Moreover , based on pharmacological treatment and mutant analyses , we show that sugar is able to trigger the degradation of pri-MIR156A/C independently of the canonical glucose sensor , HXK1 . Thus , investigation of the molecular mechanism by which sugar in particular recognizes pri-MIR156 and promotes their degradation is an important goal for future research . A . thaliana , P . patens , S . lycopersicum , and N . benthamiana were grown at 21°C ( day ) /19°C ( night ) under long day ( 16 hr light/8 hr dark ) or short day ( 8 hr light/16 hr dark ) conditions . White light was provided by a 4:2 mixture of cool white fluorescent lamps ( Lifemax cool daylight 36W/865; Philips Lighting Co . , Shangai , China ) and warm white fluorescent lamps ( Lifemax warm white 36W/830; Philips Lighting Co . ) . Light intensity was 80 µmol/m2/s in long day and 90 µmol/m2/s in short day conditions . mir156a ( SALK_056809 ) , mir156c ( SALK_004679 ) , cao/ch1 , and gin2-1 mutants were ordered from the Arabidopsis Biological Resource Center ( Columbus , OH ) . 35S::MIM156 was described ( Wang et al . , 2008 ) . All treatment assays were carried out under long day conditions . Defoliation assays were performed as described ( Yang et al . , 2011 ) . For the sugar treatment assay , Arabidopsis seeds were sterilized with 20% bleach and germinated in 50 ml 1/2 MS liquid media with shaking at 140 rpm . The seedlings were then transferred to 1/2 MS media supplemented with sugar . For the sugar starvation assay , 5-day-old wild type seedlings grown in 1/2 MS liquid media supplemented with 50 mM sucrose were transferred to 1/2 MS liquid media free of sugar and grown in the dark for 2 days . For the ActD and CHX assay , 20 µg/ml ActD ( Sigma-Aldrich , Beijing , China ) or 100 µM CHX ( Sigma-Aldrich ) was used . P . patens was cultured as described ( Cove et al . , 2009 ) . The plants in the protonema stage were used for the sugar treatment assay . Seedlings of S . lycopersicum and N . benthamiana were treated with 50 mM sucrose for 2 days . Total RNA was extracted with Trizol reagent ( Invitrogen , Life Technologies , Shanghai , China ) . Then 1 µg of total RNA was DNase I-treated and used for cDNA synthesis with an oligo ( dT ) primer . The qRT-PCR primers for SPL3 , SPL9 , SPL15 , and TUB have been described ( Wang et al . , 2008; Wang et al . , 2009 ) . The primer sequences for other genes are shown in supplementary file 1B . A small RNA blot was performed as described ( Wang et al . , 2009 ) . qRT-PCR on mature miR156 was performed according to a published protocol ( Varkonyi-Gasic et al . , 2007 ) . ChIP analysis was performed according to protocol ( Wang et al . , 2009 ) . Crude chromatin extract was pulled down with anti-Pol II antibodies ( Abcam , Hong Kong , China ) . ChIP DNAs were reverse crosslinked and purified using a PCR purification kit ( Qiagen , Shanghai , China ) . A 1 μl sample of DNA was used for real-time PCR analyses . The relative enrichment of each fragment was calculated by the ratio of bound DNAs after ChIP to input DNAs . Wild type plants were grown under short day conditions . Then 15-day-old juvenile or 50-day-old adult plants were collected at ZT 16 . Sugar was measured using 50 mg ( fresh weight ) of tissue . Sample extraction , preparation , and analyses were performed as previously described ( Tan et al . , 2011 ) . The individual sugar was identified based on the retention time and mass spectrometry standards . Quantification was performed by an external standard method .	Like animals , plants go through several stages of development before they reach maturity , and it has long been thought that some of the transitions between these stages are triggered by changes in the nutritional status of the plant . Now , based on experiments with the plant Arabidopsis thaliana , Yu et al . and , independently , Yang et al . have provided fresh insights into the role of sugar in ‘vegetative phase change'—the transition from the juvenile form of a plant to the adult plant . The new work takes advantage of the fact that vegetative phase change is controlled by two genes that encode microRNAs ( MIRNAs ) . Arabidopsis has eight MIR156 genes and both groups confirmed that supplying plants with sugar reduces the expression of two of these—MIR156A and MIR156C—while sugar deprivation increases their expression . Removing leaves also leads to upregulation of both genes , and delays the juvenile-to-adult transition . Given that this effect can be partially reversed by providing the plant with sugar , it is likely that sugar produced in the leaves—or one of its metabolites—is the signal that triggers the juvenile-to-adult transition through the reduction of miR156 levels . Yu and co-workers confirmed that sugar also reduces the expression of MIR156 in tobacco , moss , and tomato plants , suggesting that this mechanism is evolutionarily conserved . Consistent with the work of Yang and colleagues , Yu and co-workers revealed that sugar is able to reduce the transcription of MIR156A and MIR156C genes into messenger RNA . Moreover , they showed that sugar can also suppress MIR156 expression by promoting the breakdown of MIR156A and MIR156C primary messenger RNA transcripts . The work of Yu et al . and Yang et al . has thus provided key insights into the mechanisms by which a leaf-derived signal controls a key developmental change in plants . Just as fruit flies use their nutritional status to regulate the onset of metamorphosis , and mammals use it to control the onset of puberty , so plants use the level of sugar in their leaves to trigger the transition from juvenile to adult forms .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology", "developmental", "biology" ]	2013	Sugar is an endogenous cue for juvenile-to-adult phase transition in plants
Osteoclasts are bone-resorbing cells essential for skeletal remodeling . However , over-active osteoclasts can cause bone-degenerative disorders . Therefore , the level of NFATc1 , the master transcription factor of osteoclast , must be tightly controlled . Although the activation and amplification of NFATc1 have been extensively studied , how NFATc1 signaling is eventually resolved is unclear . Here , we uncover a novel and critical role of the orphan nuclear receptor Nur77 in mediating an NFATc1 self-limiting regulatory loop to prevent excessive osteoclastogenesis . Nur77 deletion leads to low bone mass owing to augmented osteoclast differentiation and bone resorption . Mechanistically , NFATc1 induces Nur77 expression at late stage of osteoclast differentiation; in turn , Nur77 transcriptionally up-regulates E3 ubiquitin ligase Cbl-b , which triggers NFATc1 protein degradation . These findings not only identify Nur77 as a key player in osteoprotection and a new therapeutic target for bone diseases , but also elucidate a previously unrecognized NFATc1→Nur77→Cblb—•NFATc1 feedback mechanism that confers NFATc1 signaling autoresolution . Bone is a dynamic organ that replaces itself every 10 years in humans ( Office of the Surgeon General ( US ) , 2004 ) . It needs to maintain an optimal density to carry out important support , movement , and protective functions in organisms . The density of bone is tightly regulated by the bone-forming osteoblasts and the bone-resorbing osteoclasts . Osteoblasts come from the mesenchymal precursor cells , whereas osteoclast precursors come from monocyte–macrophage lineage cells . Upon RANKL signaling , osteoclast precursor cells fuse and become multi-nucleated giant cells that can degrade both the organic and inorganic tissues of the bone . Osteoclast dysregulation has been associated with several human skeletal diseases such as osteoporosis , rheumatoid arthritis , and cancer metastasis to the bone ( Novack and Teitelbaum , 2008 ) . Drugs that can inhibit osteoclast differentiation , activity , or survival have been shown to be effective against these diseases ( Drake et al . , 2008; Body et al . , 2010 ) . NFATc1 is a key transcriptional switch that activates osteoclastogenesis . Ectopic NFATc1 expression alone in osteoclast precursors is sufficient to produce mature osteoclasts , whereas NFATc1 deletion blocks the ability of the precursors to differentiate into osteoclasts ( Takayanagi et al . , 2002 ) . NFATc1 binds to its response elements containing a consensus sequence of GGAAA , and its target genes in osteoclasts include cathepsin K , CLC-7 chloride channel , vacuolar proton pump subunit Atp6v0d2 , etc ( Kuroda and Matsuo , 2012 ) . NFATc1 has been shown to auto-amplify during osteoclast differentiation , and this auto-amplification process has been suggested to be important for osteoclast lineage commitment ( Asagiri et al . , 2005 ) . However , so far very few studies have dealt with whether and how NFATc1 signaling is attenuated upon its initial activation . As a result , despite the crucial functions of NFATc1 in osteoclastogenesis , the mechanisms for how NFATc1 signaling is resolved to prevent excessive osteoclast differentiation are still incompletely understood . Nur77 ( encoded by Nr4a1 ) , also known as nerve growth factor IB ( NGFIB ) , TR3 or NAK-1 , is an orphan nuclear receptor in the Nr4a family , which also includes Nurr1 ( Nr4a2 ) and Nor-1 ( Nr4a3 ) . Unlike other nuclear receptors whose functions are mainly modulated by their ligands , Nur77 is mostly regulated at the transcriptional and post-transcriptional levels ( Zhao and Bruemmer , 2010 ) . Nur77 can bind to NurRE or NBRE as monomer , homodimer , or heterodimer ( Philips et al . , 1997 ) . Nur77 has been implicated in a variety of physiological processes , including thymocyte negative selection , hypothalamic-pituitary adrenal axis , chronic inflammation , and vascular smooth muscle cell proliferation ( Hsu et al . , 2004 ) . Nonetheless , it is unknown whether Nur77 can directly regulate skeletal homeostasis or bone cell differentiation . In this study , we uncover a novel and important role of Nur77 in NFATc1 protein degradation during osteoclastogenesis and bone resorption , thus revealing a previously unrecognized mechanism that is essential for the resolution of NFATc1 signaling in which NFATc1 exerts self-limitation via an NFATc1→Nur77→Cblb—•NFATc1 negative feedback loop . To examine the expression pattern of Nur77 during osteoclast differentiation , we treated bone marrow-derived osteoclast precursor cells with RANKL for 4 days ( Figure 1A ) . A time course analysis showed that Nr4a1 mRNA started to rise on day 2 during osteoclastogenesis ( Figure 1B ) . The other two members of the NR4A family , Nr4a2 and Nr4a3 , were either not expressed or expressed at much lower levels ( Figure 1B ) . To determine if there is any potential regulatory role of Nur77 in osteoclastogenesis , we compared osteoclast differentiation cultures from bone marrow hematopoietic progenitors of Nur77 knockout ( Nur77 KO ) mice and WT littermate controls . The results revealed an enhanced osteoclast differentiation in Nur77 KO cultures shown by the higher expression of osteoclast differentiation markers such as tartrate-resistant acid phosphatase ( Trap ) , cathepsin K ( Ctsk ) , calcitonin receptor ( Calcr ) , carbonic anhydrase 2 ( Car2 ) , as well as the master osteoclastogenic transcription factor Nfatc1 ( Figure 1C ) . Consistent with these results , we have also observed more and larger mature osteoclasts in the differentiation cultures ( Figure 1D ) , as well as higher resorptive activity ( Figure 1D ) . The expression of pro- and anti-apoptotic genes was comparable , indicating an unaffected osteoclast apoptosis ( Figure 1E ) . In contrast , osteoblast differentiation from Nur77 KO bone marrow mesenchymal progenitors was unaltered , shown by the similar induction of osteoblast markers such as collagen , type I , alpha 1 ( Col1a1 ) , and Osteocalcin ( Figure 1F ) . These results suggest that Nur77 may specifically suppress osteoclastogenesis during bone remodeling . 10 . 7554/eLife . 07217 . 003Figure 1 . Nur77 deletion increases osteoclastogenesis and bone resorption . ( A ) A schematic diagram of ex vivo bone marrow osteoclast differentiation . ( B ) Expression of Nur77 , Nurr1 , and Nor-1 encoding genes ( Nr4a1 , Nr4a1 , Nr4a3 ) during a time course of RANKL-induced osteoclast differentiation ( n = 3 mice ) . Nfatc1 mRNA level was similar to Nr4a1 mRNA level on day 2–3 . ( C–D ) Osteoclast differentiation was enhanced in Nur77 KO cultures compared to WT control cultures . ( C ) Expression of osteoclast differentiation markers on day 3 ( n = 4 mice ) . ( D ) Representative images of the TRAP-stained osteoclast differentiation cultures . Mature osteoclasts were identified as multinucleated TRAP+ ( purple ) cells on day 9 . Scale bar , 25 μm . Quantification of osteoclast size , osteoclast number per area , and osteoclast-resorptive activity by calcium release from bone plate to culture medium is shown ( n = 4 mice in triplicate cultures ) . Oc , osteoclast . ( E ) Osteoclast apoptosis was unaltered , quantified by the expression of apoptosis genes on day 9 of osteoclast differentiation cultures ( n = 4 ) . ( F ) Osteoblast differentiation was unaltered in Nur77 KO cultures , measured by the expression of osteoblast markers ( n = 4 ) . ( G–O ) Nur77 KO mice exhibited bone loss . Tibiae from Nur77 KO mice or WT littermate controls ( 3 month old , male , n = 6 ) were analyzed by μCT . ( G ) Representative images of the trabecular bone of the tibial metaphysis ( top ) ( scale bar , 10 μm ) and the entire proximal tibia ( bottom ) ( scale bar , 1 mm ) . ( H–O ) Quantification of trabecular bone volume and architecture . ( H ) BV/TV , bone volume/tissue volume ratio . ( I ) BS , bone surface . ( J ) BS/BV , bone surface/bone volume ratio . ( K ) Tb . N , trabecular number . ( L ) Tb . Th , trabecular thickness . ( M ) Tb . Sp , trabecular separation . ( N ) Conn . D . , connectivity density . ( O ) SMI , structure model index . ( P ) Serum CTX-1 bone resorption marker was increased ( 3 month old , male , n = 6 ) . ( Q ) Serum P1NP bone formation marker was unaltered ( 3 month old , male , n = 6 ) . ( R–S ) Bone histomorphometry ( 3-month-old , male , n = 6 ) . ( R ) Quantification of osteoclast surface ( Oc . S/B . S ) and osteoclast number ( Oc . N/B . Ar ) . ( S ) Quantification of osteoblast surface ( Ob . S/B . S ) and osteoblast number ( Ob . N/B . Ar ) . B . S , bone surface; B . Ar , bone area . Error bars , SD . DOI: http://dx . doi . org/10 . 7554/eLife . 07217 . 003 To determine if Nur77 is a physiologically significant regulator of bone , we next examined the in vivo skeletal phenotype of Nur77 KO mice . MicroCT analysis revealed that male Nur77 KO mice had a low bone mass compared to WT male littermate controls ( Figure 1G ) , illustrated by a 36% lower bone volume/tissue volume ratio ( BV/TV ) ( Figure 1H ) , 19% less bone surface ( BS ) ( Figure 1I ) , 8% greater bone volume/bone surface ratio ( BV/BS ) ( Figure 1J ) , 14% less trabecular number ( Tb . N ) ( Figure 1K ) , 8% less trabecular thickness ( Tb . Th ) ( Figure 1L ) , and 19% more trabecular separation ( Tb . Sp ) ( Figure 1M ) . This resulted in a 19% decrease in connectivity density ( Conn . D . ) ( Figure 1N ) and a 40% increase in the Structure Model Index ( Wei et al . , 2014 ) , which quantifies the 3D structure for the relative amount of plates ( SMI = 0 , strong bone ) and rods ( SMI = 3 , fragile bone ) ( Figure 1O ) . Cortical thickness was reduced by 7% ( WT = 0 . 1125 ± 0 . 003 mm; KO = 0 . 1048 ± 0 . 004 mm; p = 0 . 02; n = 6 ) whereas tibia length was unaltered . Female Nur77 KO mice showed a similar phenotype with a 19% lower trabecular BV/TV ( WT = 0 . 2025 ± 0 . 03; KO = 0 . 1632 ± 0 . 02; p = 0 . 03 ) and a 5% decreased cortical thickness ( WT = 0 . 108 ± 0 . 003 mm; KO = 0 . 102 ± 0 . 001 mm; p = 0 . 04 ) ( 3 month old , n = 6 ) . ELISA analyses showed that the serum bone resorption marker C-terminal telopeptide fragments of the type I collagen ( CTX-1 ) was 3 . 7-fold higher in Nur77 KO mice ( Figure 1P ) , whereas the serum bone formation marker N-terminal propeptide of type I procollagen ( P1NP ) was unchanged ( Figure 1Q ) . Consistent with these observations , histomorphometry of the femur showed that osteoclast surface and osteoclast number were significantly increased in Nur77 KO mice ( Figure 1R ) , whereas osteoblast surface and osteoblast number ( Figure 1S ) were unaltered . Together , these results suggest that Nur77 deletion causes low bone mass primarily through increasing osteoclastogenesis and bone resorption . Because Nur77 has been implicated to regulate many other cell types ( Hsu et al . , 2004 ) , we next performed bone marrow transplantation experiments to examine whether Nur77 regulation of bone resorption stems from the intrinsic effects in the hematopoietic/osteoclast lineage or non-autonomous effects from other tissues or cell types such as osteoblasts , osteocytes , or the neuroendocrine system . Two complimentary sets of bone marrow transplantation were performed , and serum bone markers were assessed two months later . In the first set , we harvested donor bone marrow cells from both WT and Nur77 KO mice , and transplanted them to irradiated WT recipient mice ( Figure 2A ) . The results showed that WT mice receiving Nur77 KO bone marrow cells exhibited significantly higher CTX-1 levels than the control group ( Figure 2B ) , but unaltered P1NP levels ( Figure 2C ) , suggesting that Nur77 KO hematopoietic lineage was sufficient to elevate bone resorption . In the second set , we transplanted WT donor bone marrow cells into either Nur77 KO or WT control recipient mice ( Figure 2D ) . Nur77 KO mice receiving WT bone marrow cells showed normalized CTX-1 levels similar to the control group ( Figure 2E ) , with also similar P1NP levels ( Figure 2F ) , suggesting that WT bone marrow can completely rescue the osteoclast defects in the Nur77 KO mice and thus other tissues/cell types play only a minor role if any . The results from these two experiments indicate that Nur77 regulation of bone resorption is intrinsic to the hematopoetic lineage . 10 . 7554/eLife . 07217 . 004Figure 2 . Nur77 regulation of bone resorption is intrinsic to the hematopoietic lineage . ( A–F ) Bone marrow transplantation . Bone marrow cells from 2-month-old male donor mice were transplanted into 2-month-old irradiated male recipient male mice ( n = 5 ) and analyzed 3 months later at 5 month old . ( A–C ) Transplantation of Nur77 KO donor bone marrow cells into WT recipients conferred elevated bone resorption compared to WT control donor bone marrow cells . ( A ) A schematic diagram . ( B ) Serum CTX-1 . ( C ) Serum P1NP . ( D–F ) Transplantation of WT donor bone marrow cells into Nur77 KO recipients rescued the bone resorption to a level similar to the WT control recipients . ( D ) A schematic diagram . ( E ) Serum CTX-1 . ( F ) Serum P1NP . ( G ) Expression of RANKL and OPG , as well as RANKL/OPG ratio , in Nur77 KO ex vivo osteoblast differentiation cultures compared to WT control cultures ( n = 3 ) . ( H ) Expression of Rankl and Opg , as well as Rankl/Opg ratio , in osteocytes from femur shaft in Nur77 KO mice compared to WT control mice ( n = 3 ) . ( I ) Expression of pro-osteoclastogenic cytokines in bone marrow cells from Nur77 KO mice compared to WT control mice ( n = 3 ) . Error bars , SD . DOI: http://dx . doi . org/10 . 7554/eLife . 07217 . 004 In the bone milieu , osteoblasts and osteocytes provide RANKL and the RANKL decoy receptor OPG to stimulate and inhibit osteoclast differentiation , respectively ( Boyce and Xing , 2007 ) . Thus , we assessed whether RANKL and OPG levels were different between Nur77 KO and WT control mice . We first compared osteoblast differentiation cultures derived from Nur77 KO or WT control mice and found that there was no difference in the expression of Rankl or Opg , resulting in a comparable Rankl/Opg ratio ( Figure 2G ) . Recently osteocytes , the mature long living osteoblasts embedded in the bone matrix , have been shown to provide the majority of RANKL and OPG to osteoclasts ( Nakashima et al . , 2011; Xiong et al . , 2011 ) . Hence , we compared Rankl and Opg expression in femur shafts that contain mainly osteocytes . Our results showed that there was no difference in Rankl expression ( Figure 2H ) ; however , Nur77 KO osteocytes express a markedly higher level of Opg ( Figure 2H ) , leading to a significantly lower Rankl/Opg ratio ( Figure 2H ) . The lower Rankl/Opg ratio in the Nur77 KO mice is unlikely to be the cause of the augmented osteoclastogenesis , but presumably an attempt of osteocytes to suppress the osteoclast over-activation . These results , coupled with unaltered in vitro osteoblast differentiation ( Figure 1F ) and in vivo bone formation ( Figure 1Q , S ) , suggest that osteoblast/osteocyte are not major contributors to the excessive osteoclastogenesis and bone resorption in Nur77 KO mice . Inflammatory cytokines in the bone microenvironment can also promote osteoclast differentiation ( Zupan et al . , 2013 ) . Thus , we collected bone marrow cells from mouse femurs for gene expression assessment . No significant difference was found in the expression of Tnfα , IL1β , or IL6 ( Figure 2I ) . Moreover , it has been shown that serum levels of TNFa , IL1 , and IL6 were unchanged in mice injected with either Nur77-expressing lentivirus or Nur77 siRNA ( Hu et al . , 2014 ) , further suggesting that the augmented osteoclastogenesis in Nur77 KO mice is not due to differential levels of these inflammatory cytokines . The above findings , together with the fact that Nur77 KO osteoclast precursors exhibited enhanced osteoclast differentiation independent of the bone and neuroendocrine environment ( Figure 1C–D ) , indicate that Nur77 regulation of osteoclastogenesis is mainly intrinsic and cell-autonomous to the osteoclast lineage . Given that Nur77 exerts functions within the osteoclast itself , we decided to investigate whether Nur77 affects RANKL signaling pathways . RANKL binding to RANK receptor on osteoclast precursor cells activates AP1 transcription factor via c-Jun phosphorylation , as well as NFκB transcription factor via IκBα degradation , which in turn induces and initiates the autoamplification of NFATc1 , the master transcriptional switch of osteoclastogenesis ( Kuroda and Matsuo , 2012 ) . Although Nfatc1 mRNA level was 2-fold higher in Nur77 KO osteoclast differentiation cultures on day 3 ( Figure 1C ) , we found that NFATc1 protein level was sevenfold higher in Nur77 KO cultures compared to WT control cultures on day 3 ( Figure 3A ) . It has been reported that NFATc1 protein is degraded on day 3–4 during osteoclast differentiation , despite the continuously rising Nfatc1 mRNA levels ( Kim et al . , 2010 ) . Indeed , our time course analysis showed that in WT cultures , NFATc1 protein was elevated on day 2 but then rapidly down-regulated on day 3 and day 4 ( Figure 3A ) . In contrast , in Nur77 KO cultures , the initial increase of NFATc1 protein was sustained and NFATc1 protein remained high on day 3 and day 4 ( Figure 3A ) . We then compared c-Jun phosphorylation and IκBα degradation in the osteoclast differentiation cultures upon RANKL stimulation , but did not observe any significant difference between Nur77 KO and WT control cultures ( Figure 3B ) , which is in agreement to the similar NFATc1 protein induction on day 1 and day 2 ( Figure 3A ) . Moreover , there is no Nur77 binding sequence in a 4-Kb region of NFATc1 promoter . These results indicate that Nur77 regulation of NFATc1 protein level resides downstream of transcription . 10 . 7554/eLife . 07217 . 005Figure 3 . Nur77 inhibits osteoclast differentiation by promoting NFATc1 degradation . ( A ) NFATc1 protein levels during a time course of osteoclast differentiation from the bone marrow cells of Nur77 KO mice or WT control mice . Left , representative western blot image . Right , quantification of NFATc1/β-actin ratio . ( B ) c-Jun phosphorylation and IκBα degradation post RANKL treatment in osteoclast differentiation cultures from the bone marrow cells of Nur77 KO mice or WT control mice . P-c-Jun , phosphorylated c-Jun; t-c-Jun , total-c-Jun . ( C ) Effects of MG132 on NFATc1 protein levels in Nur77 KO or WT bone marrow osteoclast differentiation cultures . Cells were treated with 25 μM MG132 for 6 hr 3 days after RANKL stimulation . Left , representative western blot image . Right , quantification of NFATc1/β-actin ratio . ( D ) Effects of Nur77 over-expression on NFATc1 protein and mRNA levels . HEK293 cells were transfected with NFATc1 , together with either Flag-Nur77 or GFP control . Left , representative western blot image with quantification of NFATc1/β-actin ratio . Right , relative Nfatc1 mRNA . ( E ) Effects of Nur77 over-expression on NFATc1 transcriptional output ( n = 3 ) . HEK293 cells were transfected with NFATc1 and its luciferase reporter , together with increasing amount of Nur77 . ( F ) MG132 abolished the effects of Nur77 over-expression on NFATc1 transcriptional output ( n = 3 ) . HEK293 cells were treated with MG132 ( 25 μM ) 1 day after transfection for 6 hr before harvesting . All data are representative of at least three experiments . Error bars , SD . DOI: http://dx . doi . org/10 . 7554/eLife . 07217 . 005 It has been shown that the decrease in NFATc1 protein levels at later stage of osteoclast differentiation is due to ubiquitin-mediated protein degradation; and that MG132 , a proteasome inhibitor , can restore NFATc1 protein to a similar level on day 2 ( Kim et al . , 2010 ) . To examine whether the differences in NFATc1 levels between Nur77 KO and WT mice were due to protein degradation , we treated osteoclast differentiation cultures with MG132 on day 3 . As our result shows , MG132 treatment increased NFATc1 protein level in WT cultures to a level similar to Nur77 KO cultures; and MG132 treatment could no longer further increase NFATc1 protein level in the Nur77 KO cultures ( Figure 3C ) . In line with these observations , our co-IP experiments reveal that Nur77 ( either endogenous or flag-tagged ) does not directly interact with NFATc1 ( not shown ) , suggesting that Nur77 does not directly modulate NFATc1's localization or activity . These results indicate that Nur77 deletion elevates NFATc1 protein levels by suppressing ubiquitin degradation pathway . As a complementary gain-of-function approach , we tested whether Nur77 over-expression could promote NFATc1 protein degradation . We transfected HEK293 cells with NFATc1 together with Nur77 or a GFP control , and then quantified Nfatc1 mRNA and protein levels . The result shows that Nur77 over-expression significantly decreased NFATc1 protein levels ( Figure 3D , left ) without altering Nfatc1 mRNA levels ( Figure 3D , right ) . Consistent with the lower Nur77 protein abundance , Nur77 over-expression also dosage-dependently reduced the NFATc1 transcriptional output from a luciferase reporter driven by NFATc1 response elements ( Figure 3E ) . This Nur77 reduction of NFATc1 activity was completely abolished by MG132 ( Figure 3F ) , indicating that it was mediated by ubiquitin degradation pathway . The mRNA expression of Nr4a1 and Nfatc1 in osteoclasts is comparable on day 2–3 ( Figure 1B ) , supporting that Nur77 regulation of NFATc1 is relevant in the osteoclast . These findings further support the notion that Nur77 promotes NFATc1 protein degradation . In the ubiquitin degradation pathway , E3 ligases are responsible for substrate specificity and ubiquitination regulation . We next searched for E3 ligases that could be responsible for NFATc1 degradation in osteoclasts . It has been reported that Cbl-b , an E3 ligase in the Cbl family , is a major contributor to the ubiquitin-mediated down-regulation of NFATc1 at late stage of osteoclast differentiation ( Kim et al . , 2010 ) . Therefore , we tested the hypothesis that Nur77 may promote NFATc1 degradation by inducing Cbl-b . We found that Cblb expression was significantly lower in Nur77 KO osteoclast differentiation cultures compared to WT control cultures on day 2 and day 4 ( Figure 4A ) . Conversely , Nur77 over-expression in HEK293 cells significantly increased Cblb expression ( Figure 4B ) . Importantly , a truncated Nur77 mutant in which the DNA binding domain ( DBD ) was deleted could no longer up-regulate Cblb , suggesting that Nur77 induction of Cblb transcription depends on its DNA binding ability ( Figure 4B ) . The functional connection between Nur77 and Cbl-b is further supported by the similar bone phenotype in Nur77 KO mice ( Figure 1 ) and Cblb KO mice ( Nakajima et al . , 2009 ) , including increased ex vivo osteoclast differentiation and in vivo bone resorption , but unaltered bone formation , leading to lower bone mass . 10 . 7554/eLife . 07217 . 006Figure 4 . Nur77 transcriptionally up-regulates E3 ligase Cbl-b . ( A ) Cblb expression during a time course of osteoclast differentiation from the bone marrow cells of Nur77 KO mice or WT control mice ( n = 3 ) . ( B ) Nur77 over-expression increased Cblb mRNA in a DNA-binding-dependent manner . HEK293 cells were transfected with vector control , WT Nur77 , or a mutant Nur77 with a deletion of the DNA binding domain ( DBD ) ( n = 3 ) . ( C ) Nur77 activated Cbl-b promoter via NurRE . HEK293 cells were transfected with Nur77 , together with a luciferase vector control or a luciferase reporter driven by 1 Kb Cbl-b promoter containing either a WT NurRE or a mutant NurRE ( n = 3 ) . Inset shows the mutations in the two mutant reporters . ( D ) ChIP assay of Nur77 binding and H3K4me3 levels at the endogenous Cbl-b promoter . HEK293 cells were transfected with Flag-Nur77 , Nur77 binding were detected with anti-Flag antibody and compared with IgG control antibody ( n = 3 ) . ( E–H ) CRISPR/Cas9 deletion of NurRE in the endogenous Cbl-b promoter abolished Nur77 regulation of Cbl-b and NFATc1 . ( E ) A schematic representation of CRISPR/Cas9 gRNAs and their target locus in the Cbl-b promoter . ( F–H ) Effects of Nur77 over-expression on Cbl-b mRNA ( F ) ( n = 4 ) , NFATc1 protein ( G ) , and NFATc1 transcriptional output ( H ) ( n = 3 ) in two independent HEK293 CRISPR mutant clones and WT controls . Error bars , SD . DOI: http://dx . doi . org/10 . 7554/eLife . 07217 . 006 To investigate whether Cblb is a direct Nur77 target gene , we tested whether Nur77 could transcriptionally activate the Cbl-b promoter . We cloned a 1-kb segment of Cbl-b promoter upstream of a luciferase reporter and tested its expression in a transient transfection assay in HEK293 cells . The result showed that co-transfection with Nur77 significantly up-regulated the Cbl-b promoter activity by 2 . 2-fold ( Figure 4C ) , suggesting that Nur77 is able to directly activate Cblb transcription . Bioinformatic analyses revealed a pair of motifs that may comprise a Nur77 response element ( NurRE ) in the Cbl-b promoter at ∼600 bp upstream of the transcription start site ( Figure 4C , inset ) . To examine whether these putative NurRE motifs are important for Nur77 induction of Cbl-b promoter , we mutated each NurRE motif to derive mutant-1 and mutant-2 luciferase reporters ( Figure 4C , inset ) . Both mutant reporters exhibited a significantly compromised ability to be activated by Nur77 compared to the WT reporter ( Figure 4C ) , indicating that both NurRE motifs are functionally required . To determine whether Nur77 can bind to the endogenous Cbl-b NurRE , we performed chromatin immunoprecipitation ( ChIP ) assay . Nur77 was found to be enriched at the NurRE region , leading to transcription activation shown by the presence of H3K4Me3 histone mark at the transcription start site ( Figure 4D ) . These results suggest that Nur77 can directly induce Cbl-b transcription by binding to a NurRE in the Cbl-b promoter . We next sought to elucidate whether Nur77 induction of Cbl-b is functionally required for Nur77 down-regulation of NFATc1 protein . Instead of deleting Cblb , we designed a more prudent strategy to specifically disrupt the NurRE region in the endogenous Cbl-b promoter using CRISPR/Cas9 genome editing tool , thus more precisely dissecting the functional interaction among Nur77 , Cbl-b , and NFATc1 ( Figure 4E ) . Compared with WT control cells , the ability of Nur77 to increase Cblb mRNA ( Figure 4F ) , as well as to decrease NFATc1 protein level ( Figure 4G ) and transcriptional output ( Figure 4H ) , was significantly attenuated in two independent CRISPR mutant clones . These results provide strong evidence that Nur77 promotes NFATc1 protein degradation by directly inducing the transcription of Cbl-b E3 ligase . Since Nur77 expression consistently rises during osteoclastogenesis ( Figure 1B ) , we hypothesize that there is an upstream regulator that induces Nur77 transcription upon RANKL signaling activation . Interestingly , we found several NFATc1 response elements in the Nur77 promoter region , suggesting that NFATc1 up-regulates Nur77 to initiate a negative feedback loop . To test whether NFATc1 itself is sufficient to increase Nur77 expression independent of other RANKL signaling pathways , we performed transfection assays to over-express NFATc1 . Compared to a GFP negative control , NFATc1 over-expression significantly increased Nur77 expression in both HEK293 cells and mouse myoblast C2C12 cells ( Figure 5A ) . Conversely , treatment of osteoclast differentiation cultures with cyclosporin A , a calcineurin inhibitor that suppresses NFATc1 activity , dosage-dependently decreased Nur77 expression ( Figure 5B ) . 10 . 7554/eLife . 07217 . 007Figure 5 . NFATc1 induces Nur77 transcription to elicit a self-limiting loop . ( A ) NFATc1 over-expression increased Nur77 mRNA . Mouse myoblast cell line C2C12 or human embryonic kidney cell line HEK293 were transfected with NFATc1 or GFP control . ( B ) NFATc1 inhibition by Cyclosporin A dosage-dependently decreased Nur77 mRNA . Osteoclast differentiation cultures were treated with Cyclosporin A on day 2 for 24 hr ( n = 4 ) . ( C ) NFATc1 over-expression enhances Nur77 promoter activity . HEK293 cells were transfected with a luciferase vector control or a luciferase reporter driven by a 0 . 8 Kb Nur77 promoter , together with NFATc1 or GFP control ( n = 3 ) . ( D ) ChIP assay of NFATc1 binding and H3K4me3 level at the endogenous Nur77 promoter in RAW264 . 7 mouse macrophage cell line with or without 2 day RANKL stimulation . ( E ) A working model of an NFATc1 self-limiting loop in which NFATc1 elicits its own degradation by inducing Nur77 and consequently Cbl-b to resolve NFATc1 signaling . Error bars , SD . DOI: http://dx . doi . org/10 . 7554/eLife . 07217 . 007 We next examined whether NFATc1 could directly activate Nur77 promoter . We cloned a 0 . 8 Kb Nur77 promoter region upstream of a luciferase reporter and tested its inducibility by NFATc1 by transient transfection . Compared to a GFP negative control , NFATc1 over-expression significantly elevated the luciferase output ( Figure 5C ) . Moreover , ChIP assay showed that RANKL treatment of osteoclast differentiation cultures markedly increased NFATc1 binding to the endogenous Nur77 promoter ( Figure 5D ) , leading to activated Nur77 transcription as shown by the higher level of H3K4Me3 histone mark at the transcription start site ( Figure 5D ) . Together , these results indicate that Nur77 is a direct transcriptional target of NFATc1 , and thus revealing a key mechanism for how NFATc1 resolves its own signaling to prevent excessive osteoclastogenesis via an NFATc1→Nur77→Cblb—•NFATc1 self-limiting loop ( Figure 5E ) . To further determine whether Cbl-b is required for the anti-osteoclastogenic function of Nur77 in vivo , we conducted genetic experiments by comparing Nur77 Cblb double knockout mice ( DKO ) with Nur77 KO mice , Cblb KO mice , and WT littermate controls . Ex vivo bone marrow osteoclast differentiation assay showed that Nur77 deletion could no longer further enhance osteoclast differentiation in the absence of Cbl-b; furthermore , Cblb KO cultures and Nur77 Cblb DKO cultures showed a similar enhanced osteoclastogenesis as Nur77 KO cultures compared with WT control cultures ( Figure 6A–E ) . In accordance to this ex vivo finding , in vivo analyses showed that when Cbl-b is absent , Nur77 deletion could no long further elevate serum bone resorption ( Figure 6F ) or reduce bone mass ( Figure 6G–K ) ; Cblb KO mice and Nur77 Cblb DKO mice showed a similar high bone resorption and low bone mass phenotype as Nur77 KO mice compared with WT controls ( Figure 6F–K ) . These findings demonstrate that Cbl-b deletion fully recapitulates Nur77 deletion , and Cbl-b deletion completely abolishes Nur77 regulation of osteoclastogenesis and bone resorption . Therefore , this in vivo genetic evidence strongly supports Cbl-b as a major and essential mediator of Nur77 function in the osteoclast lineage . 10 . 7554/eLife . 07217 . 008Figure 6 . Cbl-b deletion abolishes Nur77 regulation of osteoclastogenesis and bone resorption . Nur77 Cblb DKO mice were compared with Nur77 KO mice , Cblb KO mice , and WT littermate control mice ( 3 month old , male , n = 4 ) . ( A–E ) Ex vivo bone marrow osteoclast differentiation ( n = 4 mice in triplicate cultures ) . ( A ) Representative images of the TRAP-stained osteoclast differentiation cultures . Mature osteoclasts were identified as multinucleated TRAP+ ( purple ) cells on day 9 . Scale bar , 25 μm . ( B ) Osteoclast size . ( C ) Osteoclast number per area . ( D ) Osteoclast-resorptive activity by calcium release from bone plate to culture medium . ( E ) Expression of osteoclast differentiation markers on day 3 . ( F ) ELISA analysis of serum CTX-1 bone resorption marker ( n = 4 ) . ( G–K ) μCT analysis of trabecular bone parameters ( n = 4 ) . ( G ) BV/TV , bone volume/tissue volume ratio . ( H ) BS , bone surface . ( I ) Tb . N , trabecular number . ( J ) Tb . Sp , trabecular separation . ( K ) Conn . D . , connectivity density . Statistical analyses were performed with ANOVA followed by the post-hoc Tukey pairwise comparisons . Error bars , SD . DOI: http://dx . doi . org/10 . 7554/eLife . 07217 . 008 In this study , we have identified the nuclear receptor Nur77 as a critical negative regulator of osteoclastogenesis and bone resorption , revealing its novel bone protective role . Nur77 deletion causes elevated bone resorption and bone loss in mice . Moreover , we have also unraveled a previously unrecognized mechanism for how Nur77 attenuates NFATc1 signaling at late stage of osteoclast differentiation . Nur77 transcriptionally up-regulates Cbl-b E3 ligase to trigger NFATc1 protein degradation , so that NFATc1 signaling can be resolved in a timely fashion to prevent excessive osteoclastogenesis and bone resorption ( Figure 5E ) . When osteoclast differentiation is enhanced in a culture , typically all RANKL-induced osteoclastogenic genes will be elevated , including Nfatc1 , because the gene expression analysis is a population-based assay ( Figure 1C ) . In addition to induction by RANKL-activated upstream signaling , NFATc1 also auto-amplifies itself at mRNA level ( Asagiri et al . , 2005 ) , which makes it harder to discern the origin of NFATc1 regulation in osteoclast cultures . Thus , we subsequently examined the intrinsic ability of Nur77 to regulate NFATc1 via transfection in a heterologous cell type such as HEK293 cells , in the absence of the other RANKL signaling in osteoclast cultures . We show that Nur77 can decrease NFATc1 protein levels without affecting Nfatc1 mRNA levels ( Figure 3D ) . Also , there is no Nur77 binding sequence in NFATc1 promoter , further support that Nur77 does not directly regulate NFATc1 transcription . Moreover , our co-IP experiment show that Nur77 ( either endogenous or a flag-tagged ) does not directly interact with NFATc1 , suggesting that Nur77 does not directly modulate NFATc1 localization or activity . As a result , we conclude that Nur77 mainly regulates NFATc1 protein degradation . In searching for the Nur77-induced NFATc1-targeting E3 ubiquitin ligase , we have considered both of the two members in the Cbl family—Cbl and Cbl-b . Findings from our study and previous studies indicate that Cbl-b , but not Cbl , is the major regulator in osteoclast due to the following reasons: ( 1 ) previous studies and our data show that Cbl-b KO mice exhibit enhanced osteoclast differentiation , higher bone resorption and lower bone mass that is similar to the phenotype we observed in Nur77 KO mice ( Nakajima et al . , 2009 ) ( Figure 6 ) . In contrast , adult Cbl KO mice have no obvious bone phenotype ( Chiusaroli et al . , 2003 ) . This supports that Cbl-b , but not Cbl , is a physiologically significant regulator of osteoclast in vivo . ( 2 ) Nur77 over-expression induces the expression of Cbl-b , but not Cbl . ( 3 ) A complete NurRE exists only in Cbl-b promoter , but not in Cbl promoter . ( 4 ) Nur77 KO osteoclast cultures show a significant lower level of Cbl-b , but not Cbl . ( 5 ) Finally , our in vivo genetic data comparing Nur77 Cblb DKO mice with Cblb KO mice show that Cbl-b deletion completely abolishes the enhanced osteoclast differentiation , elevated bone resorption , and reduced bone mass in Nur77 KO mice , demonstrating that Cbl-b is the major and essential mediator of Nur77 regulation of osteoclastogenesis , which is not compensated by Cbl . In the process of studying the role of Nur77 in NFATc1 regulation and osteoclast differentiation , we inadvertently discovered that Nur77 is not only a regulator of NFATc1 but also a transcriptional target of NFATc1 , thus revealing a negative feedback loop where NFATc1 induces its own degradation by up-regulating Nur77 and Cbl-b . This mechanism is crucial for proper cellular differentiation and function , since the resolution of signaling is just as important as its initiation and amplification . A breach in the NFATc1→Nur77→Cblb—•NFATc1 regulatory loop , exemplified by the Nur77 KO mice , will cause pathologically elevated NFATc1 levels during late stage of osteoclastogenesis and send osteoclasts into overdrive . To the best of our knowledge , we are the first group to propose an NFATc1 self-limiting regulatory mechanism . Therefore , NFATc1 exerts three functions to control osteoclastogenesis . In addition to the previously recognized roles in its auto-amplification and activating osteoclast genes to initiate the differentiation , NFATc1 also plays a role in its auto-resolution to cease the differentiation ( Figure 5E ) . Our findings will pave the road for future investigations to examine whether this NFATc1→Nur77→Cblb—•NFATc1 negative feedback loop may be widely applicable to NFATc1 regulation of other cellular processes such as T cell activation and cancer development . Identification of novel osteoclast signaling pathways provides insights into potential new therapeutic options to treat bone-degenerative diseases by inhibiting osteoclast activity . Current clinically approved osteoclast inhibitors , such as bisphosphonates and denosumab ( anti-RANKL antibody ) , may cause severe side effects such as osteonecrosis of the jaw ( ONJ ) ( Khan et al . , 2009; Boquete-Castro et al . , 2015 ) . These side effects could stem from a variety of factors including target specificity . The whole body Nur77 KO mice , however , provides an example where despite the wide spread expression of a gene , it still can be targeted due to the differential sensitivity in different tissues . Although Nur77 has been implicated in numerous physiological functions in vitro , most of these functions did not hold true in vivo based on a general lack of phenotype in Nur77 KO mice ( Lee et al . , 1995; Chao et al . , 2013 ) . Until recently , Nur77 KO mice largely appear healthy and normal , and only exhibit clinical deficiencies under severe stress ( Chao et al . , 2009; Palumbo-Zerr et al . , 2015 ) or with the deletion of an additional NR4A gene ( Mullican et al . , 2007 ) . This may be partially due to functional redundancy among NR4A family members so that the compensation by Nurr1 and Nor1 masks the effects of Nur77 loss . Interestingly , we found that Nur77 is the predominant NR4A member in the osteoclast lineage with little or no Nurr1 or Nor1 expression ( Figure 1B ) , explaining the critical role of Nur77 in osteoclastogenesis so that the effects on bone is evident by Nur77 deletion alone . This creates an exciting opportunity for selective drug targeting and precision medicine with minimal side effects . Most recently , Tontonoz et al . have uncovered a muscle protective role of Nur77 as mice deficient in Nur77 alone exhibit reduced muscle mass and myofiber size ( Tontonoz et al . , 2015 ) . Therefore , Nur77 activation may represent a promising therapeutic strategy for musculoskeletal degenerative diseases with dual benefits on muscle and bone . The crucial regulation of NFATc1 protein degradation by Nur77 and Cbl-b suggests that it may be therapeutically beneficial to accelerate RANKL signaling resolution during osteoclastogenesis . Indeed , defects in the components of ubiquitin and proteasome system have been implicated in diseases including cancer and neurodegenerative disorders ( Popovic et al . , 2014 ) . Bortezomib , a peptide inhibitor of proteasome , has been approved for clinical usage in pathological settings such as refractory multiple myeloma ( Richardson et al . , 2005 ) . The discovery of pathway-specific ubiquitin–proteasome activators , however , is somewhat lagging behind . Nonetheless , oleuropein , a small molecule proteasome activator , has been shown to delay replicative senescence of human embryonic fibroblast ( Katsiki et al . , 2007 ) . In addition , oleuropein treatment has been shown to inhibit osteoclast formation and suppress the loss of trabecular bone in ovariectomized mice ( Hagiwara et al . , 2011 ) , giving hope that small molecules that selectively activate the protein degradation pathway may be a promising future therapeutic strategy for skeletal and other diseases . Nur77 KO mice ( Lee et al . , 1995 ) in a C57BL/6 and 129SvJ hybrid background was originally generated by Jeffrey Milbrandt at Washington University School of Medicine and kindly provided by Orla Conneely at Baylor College of Medicine . Cblb KO mice in a mixed background were originally generated by Josef Penninger ( Bachmaier et al . , 2000 ) . Mice were fed with standard chow ad libitum and kept on a 12-hr light , 12-hr dark cycle . Nur77 KO mice were bred with Cblb KO mice to generate Nur77 Cblb double heterozygous mice , which were then bred to generate littermates for Nur77 KO , Cblb KO , Nur77 Cblb DKO , and WT control . All experiments were conducted using littermates . Bone marrow transplantation was performed as described ( Wan et al . , 2007; Krzeszinski et al . , 2014 ) . Briefly , bone marrow cells from 2-month-old male donor ( WT or Nur77 KO ) were intravenously transplanted via retro orbital injection into 2-month-old male recipients ( WT or Nur77 KO ) that were irradiated at lethal dose ( 1000 roentgen ) ; the mice were analyzed 3 month post transplantation . Our established protocol for lethal irradiation and bone marrow transplantation achieves >95% repopulation of donor cells in the recipient mice , as measured by the percentage of CD45 . 1 vs CD45 . 2 . Sample size estimate was based on power analyses performed using SAS 9 . 3 TS X64_7PRO platform at the UTSW Biostatistics Core . With the observed group differences and the relatively small variation of the in vivo measurements , n = 4 and n = 3 will provide >90% and >80% power at type I error rate of 0 . 05 ( two-sided test ) , respectively . All protocols for mouse experiments were approved under number 2008-0324 by the Institutional Animal Care and Use Committee of UTSW . μCT was performed using a Scanco μCT-35 instrument ( Scanco Medical , Brüttisellen , Switzerland ) as described ( Wei et al . , 2010 ) . Histomorphometry were performed as described ( Wan et al . , 2007; Wei et al . , 2011 ) . Serum CTX-1 bone resorption marker and P1NP bone formation marker were measured with RatLaps EIA kit and Rat/Mouse PINP EIA kit ( Immunodiagnostic Systems , Tyne & Wear , United Kingdom ) , respectively . To analyze osteocyte gene expression , mouse femur was cut off at both ends to allow marrow cells to be flushed out with media . It was then soaked in PBS and spun down to remove residual marrow cells , and snap frozen in liquid nitrogen , stored at −80°C until RNA extraction . Osteoclasts were differentiated from bone marrow cells as described ( Wan et al . , 2007 ) . Briefly , hematopoietic bone marrow cells were purified with a 40-μm cell strainer , cultured for 16 hr with 5 ng/ml MCSF ( R&DSystems , Minneapolis , Minnesota ) in α-MEM containing 10% FBS . Floating cells were then collected and differentiated with 40 ng/ml of M-CSF in α-MEM containing 10% FBS for 3 days ( day −3 to day 0 ) , then with 40 ng/ml of MCSF , and 100 ng/ml of RANKL ( R&D Systems ) for 3–9 days ( day 0 to day 9 ) . Mature osteoclasts were identified as multinucleated ( >3 nuclei ) TRAP+ cells on day 9 . Osteoclast differentiation and apoptosis were quantified by the RNA expression of osteoclast markers and apoptosis genes on day 3 and 6 , respectively , using RT-QPCR analysis . For osteoclast-resorptive function analyses , osteoclast differentiation was conducted in OsteoAssay bone plates ( Lonza , Basel , Switzerland ) , and osteoclast activity was quantified as calcium release from bone into culture medium using a colorimetric calcium detection kit ( Abcam , Cambridge , Massachusetts ) . Osteoblast differentiation from bone marrow cells was performed as previously described ( Wei et al . , 2012 , 2014 ) . Briefly , bone marrow cells were cultured for 4 days in MSC media ( Mouse MesenCult Proliferation Kit , StemCell Technologies ) , then differentiated into osteoblast with α-MEM containing 10% FBS , 5 mM β-glycerophosphate , and 100 μg/ml ascorbic acid for 9 days . Antibodies for NFATc1 , total-c-Jun and IκBα , as well as cyclosporine A were purchased from Santa Cruz Biotechnology ( Dallas , Texas ) . Phospho ( ser73 ) -c-Jun antibody was from Cell Signaling ( Beverly , Massachusetts ) . Anti-Histone H3 ( tri methyl K4 ) antibody was from Abcam . Antibodies for Flag and β-actin were from Sigma ( St . Louis , Missouri ) . MG132 was from Fisher ( Pittsburgh , Pennsylvania ) . Western blot and ChIP assays were performed as previously described ( Wan et al . , 2007; Krzeszinski et al . , 2014 ) . NFATc1 expression plasmid was purchased from Open Biosystems ( Lafayette , Colorado ) . Human Flag-Nur77 expression plasmid was kindly provided by Orla Conneely lab . Nur77-ΔDBD expression plasmid was constructed by deleting the amino acid residues 270–335 from the WT Nur77 expression plasmid . RNA was reverse transcribed into cDNA using an ABI High Capacity cDNA RT Kit ( Life Technologies , Carlsbad , California ) and analyzed using real-time quantitative PCR ( SYBR Green ) in triplicate . All RNA expression was normalized by the ribosomal gene L19 . Cbl-b-promoter-luc-WT and Nur77-promoter-luc were constructed by cloning 1-Kb and 0 . 8-Kb segment upstream of transcription start site into pGL4 luciferase vector . Cbl-b-luc-Mut1 and Cbl-b-luc Mut2 were created by introducing mutations to three residues in each NurRE region of Cbl-b promoter , using the QuikChange II XL Site-Directed Mutagenesis Kit ( Agilent Technologies , Santa Clara , California ) . NFATc1 transcriptional activity was quantified using pNFAT-Luc reporter ( Agilent Technologies ) . For transient transfection , a luciferase reporter was co-transfected into HEK293 cells with expression plasmids for β-gal and factors to be tested using FuGENE HD reagent ( Roche , Basel , Switzerland ) . Vector alone or a GFP expression plasmid served as a negative control . Luciferase activity was measured 48 hr later and normalized by β-gal activity . All transfection experiments were performed in triplicates and repeated for at least three times . Plasmids for gRNA cloning and hCas9 expression were from Addgene ( Cambridge , Massachusetts ) . Oligos for gRNA were designed to target upstream and downstream of the NurRE in the Cbl-b promoter and cloned into gRNA vector according to the instruction form George Church Laboratory . Both vectors for gRNAs and the expression plasmids for hCas9 and GFP marker were co-transfected into HEK293 cells . GFP+ cells were sorted into 96-well plates at 1 cell/well 48 hr later . Each clone was expanded; genomic DNA was amplified by PCR and genotyped by sequencing . Two independent clones with NurRE deletion were compared to WT control . All statistical analyses were performed with Student's t-test and represented as mean ± standard deviation unless noted otherwise . For in vivo experiments with ≥3 groups , statistical analyses were performed with ANOVA followed by the post-hoc Tukey pairwise comparisons . The p values were designated as: * , p < 0 . 05; , p < 0 . 01; * , p < 0 . 005; **** , p < 0 . 001; n . s . non-significant ( p > 0 . 05 ) .	Bones are constantly remodeled in response to the stresses of everyday life . Cells called osteoclasts break down old or damaged bone and cells called osteoblasts make new bone . In healthy bones , the work of these two types of cells is well balanced . But in bone-weakening diseases like osteoporosis and certain bone cancers this balance is disturbed and the osteoclasts become overly active , leading to weak and thin bones . Some drugs can help block the development of osteoclasts and help reduce bone loss in these diseases , but they may cause unwanted side effects . A better understanding of the processes that maintain a healthy balance of osteoblasts and osteoclasts could help scientists develop better treatments with fewer side effects . Scientists have already learned that a protein called NFATc1 turns on the production of osteoclasts . But no one knew how NFATc1 is turned off in healthy bone to prevent the excessive growth of osteoclasts and too much bone turnover . Now , Li et al . have identified a protein called Nur77 as an important regulator of NFATc1 by examining genetically engineered mice that lack Nur77 . These modified mice had more osteoclasts and thinner bones than normal mice . Further experiments used radiation to wipe out the bone marrow of normal mice , who then received bone marrow transplants from mice that lacked Nur77 . After the transplant , the normal mice showed bone loss . When the experiment was reversed , and Nur77-lacking mice received bone marrow from normal mice , their bone loss was alleviated . This indicates that Nur77 acts in the bone marrow cells to control osteoclasts and skeletal health . Li et al . found that Nur77 cannot control the expression of the gene that encodes NFATc1 or directly bind to the NFATc1 protein . Instead , Nur77 increases the production of an enzyme that breaks down the NFATc1 protein . Unexpectedly , the experiments also found that NFATc1 turns on the expression of Nur77 . This means that NFATc1 essentially regulates itself by increasing its own breakdown when NFATc1 levels increase . This helps to explain how osteoclast production is normally kept in check , and may suggest new strategies for treating bone diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2015	Nur77 prevents excessive osteoclastogenesis by inducing ubiquitin ligase Cbl-b to mediate NFATc1 self-limitation
Human γ-secretase is an intra-membrane protease that cleaves many different substrates . Aberrant cleavage of Notch is implicated in cancer , while abnormalities in cutting amyloid precursor protein lead to Alzheimer's disease . Our previous cryo-EM structure of γ-secretase revealed considerable disorder in its catalytic subunit presenilin . Here , we describe an image classification procedure that characterizes molecular plasticity at the secondary structure level , and apply this method to identify three distinct conformations in our previous sample . In one of these conformations , an additional transmembrane helix is visible that cannot be attributed to the known components of γ-secretase . In addition , we present a γ-secretase structure in complex with the dipeptidic inhibitor N-[N- ( 3 , 5-difluorophenacetyl ) -L-alanyl]-S-phenylglycine t-butyl ester ( DAPT ) . Our results reveal how conformational mobility in the second and sixth transmembrane helices of presenilin is greatly reduced upon binding of DAPT or the additional helix , and form the basis for a new model of how substrate enters the transmembrane domain . γ-Secretase clears the anchors of type-I membrane proteins that are left behind in the membrane after shedding of their ectodomain . The substrate specificity of this intra-membrane protease is remarkably relaxed . It will cleave a wide range of substrates , as long as they form a single hydrophobic transmembrane helix , and the remaining ectodomain is not too large ( Struhl and Adachi , 2000 ) . As a manifestation of its promiscuous substrate specificity , γ-secretase also cleaves its substrate in different positions . For two of its most studied substrates , the Notch receptor and the amyloid precursor protein ( APP ) , γ-secretase performs an initial endopeptidase-like ε-cleavage , which is followed by carboxypeptidase-like trimming , called γ-cleavage . Cleavage of APP leads to secretion of β-amyloid ( Aβ ) peptides into the extracellular environment . Abundant deposits of Aβ peptides in the brain , clinically known as β-amyloid plaques , are a defining characteristic of Alzheimer's disease ( AD ) . Variability in the position of both ε– and γ-cleavages results in Aβ peptides with lengths ranging from 36 to 49 residues , with Aβ40 being the most common form . Longer peptides seem to be more prone to aggregation , and increased ratios of Aβ42/Aβ40 are thought to play a role in AD pathogenesis ( Tanzi and Bertram , 2005 ) . Because of its central role in Aβ generation , γ-secretase is an attractive target for treatment of AD . However , a clinical trial with the γ-secretase inhibitor semagacestat had to be interrupted prematurely due to strong side effects , including skin cancer , weight loss and a faster decline of the cognitive skills of patients receiving the highest dose of the drug ( De Strooper , 2014 ) . Probably , global inhibition of the complex to reduce the formation of Aβ-peptides is undesirable , as this may also affect other pathways such as Notch signaling . Therefore , the development of modulators of γ-secretase activity that leave ε-cleavage intact but stimulate γ-cleavage has been suggested as an alternative ( Wolfe , 2012 ) . Moreover , the development of specific inhibitors of Notch cleavage may be beneficial for the treatment of cancer . However , such developments are currently hindered by a lack of quantitative insights into the mechanism of γ-secretase proteolysis . The γ-secretase complex consists of four essential , integral membrane proteins: presenilin ( PS ) , nicastrin , anterior pharynx defective 1 ( Aph-1 ) , and presenilin enhancer 2 ( Pen-2 ) ( De Strooper , 2003; Kimberly et al . , 2003 ) . PS provides the two essential aspartates that form the proteolytic active site where both ε and γ-cleavages occur . More than 200 missense mutations in the gene for PS have been linked to an early-onset , familial form of Alzheimer's disease ( FAD ) ( ( De Strooper et al . , 2012 ) and http://www . alzforum . org/mutations ) . In humans , two genes encode two homologous forms of PS ( PS1 and PS2 ) , but most FAD-derived mutations target PS1 . Because there are also two forms of Aph-1 ( Aph-1a and Aph-1b ) , four possible complexes may form , of which nicastrin:Aph-1a:PS1:Pen-2 is the most abundant . Nicastrin has a large , heavily glycosylated ectodomain , which is thought to play a role in substrate recognition ( Shah et al . , 2005 ) . Aph-1 has been proposed to play a scaffolding role ( Lee et al . , 2004 ) . Pen-2 is essential for proteolytic activity of the mature complex , and facilitates auto-proteolysis of PS1 in a long cytosolic loop between its sixth and seventh transmembrane helices ( TM6 and TM7 ) ( Thinakaran et al . , 1996 ) . We recently solved the cryo-EM structure of a nicastrin:Aph-1a:PS1:Pen-2 complex to a resolution of 3 . 4 Å ( Bai et al . , 2015a ) . This structure allowed building a near-complete atomic model , and revealed how Aph-1 and Pen-2 hold a remarkably flexible PS1 subunit underneath the nicastrin ectodomain . In particular , TM2 , the cytoplasmic side of TM6 and the long linker between TM6 and TM7 of PS1 were largely disordered . The presence of the linker could only be inferred from fuzzy densities in 2D class averages , whereas the approximate position of TM2 could only be inferred from a 7 Å low-pass filtered map . This low-pass filtered map also showed a rod-shaped density in the cavity between TM2 , TM3 and TM5 , but the identity of this density could not be determined . Although the sample used for structure determination did show proteolytic activity for APP ( Lu et al . , 2014 ) , the active site appeared to be in an inactive conformation as the two catalytic aspartates were too far apart to catalyze proteolysis . In this paper , we set out to gain further insights into the proteolytic mechanism of γ-secretase by using two complementary approaches to sample the conformational landscape of its catalytic subunit . With the aim of trapping the complex in a more defined conformation , we solve a structure in complex with the non-transition-state analogue inhibitor DAPT , which is a precursor in the development of semagacestat ( Dovey et al . , 2001 ) . In addition , we describe how masked cryo-EM image classification combined with subtraction of part of the signal from the experimental images allows visualizing molecular dynamics of the catalytic subunit in its apo-state at the secondary structure level . The resulting four structures represent a significant step towards understanding how this protease cleaves its many substrates . A powerful method of dealing with structural heterogeneity in cryo-EM data sets is to 'focus' refinement on a defined region of the protein complex of interest . In this approach one masks out part of the reference during 3D refinement , thereby effectively ignoring structural variability in less interesting parts . For example , we used masked refinements to deal with variability in the relative orientations of ribosomal subunits ( Amunts et al . , 2014; Wong et al . , 2014 ) . Similarly , masked multi-reference refinement may be used as a clustering tool , i . e . to separate experimental particle images based on differences in a defined region of interest . We refer to this approach as masked 3D classification . However , masked 3D classifications aimed at analyzing the conformational landscape of γ-secretase in its apo-state were unsuccessful . An initial data set of 400 , 000 particles gave rise to a 3 . 5 Å map . Using different masks on the transmembrane domain , masked 3D classification consistently yielded only a single class showing good density . Although this approach did result in the selection of 160 , 000 particles from which we could calculate a better 3 . 4 Å map , it did not reveal the nature of conformational freedom within the catalytic subunit ( Bai et al . , 2015a ) . A fundamental problem with masked refinements is that one compares projections of a partial map with experimental projections of the entire particle ( Figure 1 ) . This leads to inconsistencies in the comparisons that underlie the refinement procedure . For example , one might want to focus classification on the part of the particle that is depicted in red in Figure 1A , and ignore any variations in the yellow part . Masking away the yellow part from the reconstruction in masked refinements ( Figure 1C ) yields reference projections that only contain the red part ( Figure 1F ) . However , each experimental image ( Figure 1D ) contains signal coming from the entire particle , i . e . from both the yellow and the red parts . Therefore , the yellow part of the signal will act as an additional source of noise in the comparison between the experimental image and the masked reference projection . It will depend on the signal-to-noise ratio in the original image and on the size of the part of the complex that is masked away , whether this additional noise will affect the refinement . For large particles , high signal-to-noise ratios in the data make masked refinements relatively robust , but even for ribosomes masked refinements of the small subunit proved much more difficult than for the large subunit ( Wong et al . , 2014 ) . 10 . 7554/eLife . 11182 . 003Figure 1 . Masked classification with residual signal subtraction . ( A ) A 3D model of a complex of interest . ( B ) The part of the complex one would like to ignore in masked classification ( V1 ) is shown in yellow . ( C ) The part of the complex one would like to focus classification on ( V2 ) is shown in red . ( D ) An experimental particle image is assumed to be a 2D projection of the entire complex in panel A that is affected by the contrast transfer function ( CTF ) , and to which experimental noise ( N , shown in grey ) has been added . ( E ) A CTF-affected 2D projection of V1 . ( F ) A CTF-affected 2D projection of V2 . Previous approaches to masked classification in RELION ( Amunts et al . , 2014; Wong et al . , 2014 ) compared experimental particles ( panel D ) , with reference projections of only V2 ( panel F ) . This results in inconsistent comparisons . ( G ) In the modified masked classification approach , one subtracts the CTF-affected 2D projection of V1 ( panel E ) from the experimental particle ( panel D ) . This results in a modified experimental particle image that only contains experimental noise and the CTF-affected projection of V2 , so that comparison with the reference projection in panel F becomes consistent . DOI: http://dx . doi . org/10 . 7554/eLife . 11182 . 003 To reduce the inconsistency in image comparison , we explored a modification of the masked classification approach . Using the example in Figure 1 , if the noisy experimental image were to contain only signal from the red part , then a masked refinement would be consistent . To emulate this situation , we subtract projections of the yellow part of the reconstruction ( Figure 1B , E ) from every experimental image . This requires knowledge about the relative orientation of each particle , which is obtained from a consensus refinement of the entire data set against a single , unmasked reference . Also , we apply the CTF of each experimental image to the projection of the yellow part prior to the subtraction . The resulting , modified experimental particle image ( Figure 1G ) then ideally only contains the signal from the red part of the particle , and the only inconsistency in the image comparison is the original experimental noise . Therefore , using the subtracted experimental particle images in masked refinements that are focused on the red part of the signal will be better than using the original images . Because the PS1 subunit showed the highest level of disorder in our high-resolution structure , we decided to perform masked classification on PS1 with subtraction of the signal from the rest of the complex . Since the total molecular weight of the ordered part of PS1 in our previous map was less than 30 kDa , we reasoned that the remaining signal in the subtracted experimental images would probably not be strong enough to allow the determination of their relative orientations . Therefore , we performed masked classification on the set of 400 , 000 particles , while keeping all orientations fixed at the values determined in the refinement of the 3 . 5 Å consensus map . Classification into eight classes yielded three majority classes that showed good density ( Figure 2 ) . Five smaller classes gave suboptimal reconstructions , and the particles from these classes were discarded . The ( original , non-subtracted ) particles from the good classes were then subjected to separate 3D auto-refinement runs ( Scheres , 2012 ) , all of which were started from the same 40 Å low-pass filtered reference to avoid model bias . The resulting maps were very similar in the nicastrin and Aph-1 subunits , but obvious differences were present in the PS1 and Pen-2 subunits . The maps had resolutions in the range of 4 . 0–4 . 3 Å , which allowed reliable main-chain tracing , but left density for many side chains less well defined ( Table 1 , Figure 3 , Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 11182 . 004Figure 2 . Masked classification with signal subtraction on PS1 . ( A ) From a masked 3D classification run on the PS1 subunit , the three largest classes ( in cyan and blue , labeled class 1–3 ) showed good density . The five smallest classes ( in grey ) were ignored in further analyses as they showed suboptimal density . ( B ) Superposition of classes 1 and 2 and of classes 2 and 3 reveals the different orientations of TM6 in all three classes ( indicated with red arrows ) , and the fact that TM2 ( indicated with a red dashed box ) is only ordered in class 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11182 . 00410 . 7554/eLife . 11182 . 005Figure 2—figure supplement 1 . Cross-refinement of the masked classification results . A 10Å low-pass filtered version of the map from class 3 ( left ) was refined against the particle subset identified for class 1 . The resulting map ( right ) is undistinguishable from the original map from class 1 , indicating that artifacts due to model bias did not play a role . DOI: http://dx . doi . org/10 . 7554/eLife . 11182 . 00510 . 7554/eLife . 11182 . 006Figure 2—figure supplement 2 . Masked classification on Aph-1 . ( A ) From a masked 3D classification run on the Aph-1 subunit with subtraction of the residual signal , all eight classes look very similar . ( B ) A superposition of all eight maps . DOI: http://dx . doi . org/10 . 7554/eLife . 11182 . 00610 . 7554/eLife . 11182 . 007Figure 2—figure supplement 3 . Reproducibility of masked classification on PS1 . Resulting maps from masked classifications on PS1 with subtraction of the residual signal using ( A ) six instead of eight classes; ( B ) a different random seed; and ( C ) ten classes and a slightly different mask from the run shown in Figure 2 . The classes similar to the three largest classes shown in Figure 2 are highlighted in the same cyan and blue colors . DOI: http://dx . doi . org/10 . 7554/eLife . 11182 . 00710 . 7554/eLife . 11182 . 008Figure 2—figure supplement 4 . Masked classification with simulated data . ( A ) Three examples of experimental particle images . ( B ) Three examples of simulated particle images . In total , 50 , 000 images of particles from Class 1 and 50 , 000 images of particles from Class 2 were simulated ( also see Methods ) . ( C ) Resulting maps for a masked classification with signal subtraction on the catalytic subunit using two references . ( D ) Distribution of the simulated particles in the resulting classes: 93% of the particles were clustered correctly . DOI: http://dx . doi . org/10 . 7554/eLife . 11182 . 00810 . 7554/eLife . 11182 . 009Figure 3 . Three classes from the apo-state ensemble . ( A ) The reconstructed density for the three classes . Nicastrin is shown in green , Aph-1 in orange , PS1 in cyan , and Pen-2 in yellow . α-Helical density that is unaccounted for by the γ-secretase model is shown in purple . The same color code is used throughout this paper . ( B ) Schematic representation of the γ-secretase atomic models . TMs of PS1 are numbered . The active site aspartates are shown in red . The purple dashed box highlights TM2 of PS1 , which is only ordered in class 1 . The purple arrows indicate the orientation of TM6 , which is different in each class . DOI: http://dx . doi . org/10 . 7554/eLife . 11182 . 00910 . 7554/eLife . 11182 . 010Figure 3—figure supplement 1 . Fourier shell correlations for the three apo-state classes . ( A ) FSC between the refined atomic models and the maps . ( B ) FSC between reconstructions from independently refined halves of the data sets . DOI: http://dx . doi . org/10 . 7554/eLife . 11182 . 01010 . 7554/eLife . 11182 . 011Table 1 . Refinement and model statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 11182 . 011Class1Class2Class3DAPTData collection Particles63 , 87379 , 26366 , 72051 , 366 Pixel size ( Å ) 1 . 41 . 41 . 41 . 4 Defocus range ( μm ) 0 . 7–3 . 20 . 7–3 . 20 . 7–3 . 20 . 6–2 . 8 Voltage ( kV ) 300300300300 Electron dose ( e-/Å−2 ) 40404040Map features Density TM2+--+ Cα-Cα distance D257– D385 ( Å ) 9 . 512 . 79 . 18 . 0 Conformation Pen-2ininoutin α-helical density++--Model composition Non-hydrogen atoms10 , 4439 , 9229 , 91610 , 543 Protein residues1 , 3151 , 2451 , 2471 , 329Refinement Resolution ( Å ) 4 . 14 . 04 . 34 . 2 Map sharpening B-factor ( Å2 ) −100−100−130−130 Fourier Shell Correlation0 . 82360 . 88180 . 80500 . 8602 Rfactor0 . 29170 . 30280 . 28160 . 3426Rms deviations Bonds ( Å ) 0 . 01120 . 00950 . 01200 . 0093 Angles ( ° ) 1 . 73521 . 60831 . 79221 . 6186Model geometry Molprobity score3 . 172 . 893 . 163 . 22 Clashscore ( all atoms ) 17 . 359 . 3113 . 2618 . 15 Good rotamers ( % ) 89 . 690 . 986 . 688 . 9Ramachandran plot Favored ( % ) 85 . 184 . 484 . 284 . 3 Allowed ( % ) 10 . 912 . 311 . 511 . 8 Outliers ( % ) 4 . 03 . 34 . 33 . 9 As a control for model bias we performed a cross-refinement , where the 10 Å low-pass filtered map from class 3 was used as initial reference for refinement of the particles assigned to class 1 ( Figure 2—figure supplement 1 ) . The initial reference did not influence convergence , as the cross-refinement yielded a map that was indistinguishable from the one obtained for class 1 in Figure 3 . As a negative control , we performed masked classification with signal subtraction using eight classes on the Aph-1 subunit ( Figure 2—figure supplement 2 ) . The density for this subunit in the consensus map was very well defined , indicating that this subunit is much more rigid than PS1 . In this case , the eight resulting classes attracted similar numbers of particles and all eight classes gave rise to very similar reconstructions . Finally , in a third control to test reproducibility we performed multiple different masked classifications on the PS1 subunit ( Figure 2—figure supplement 3 ) . We used different numbers of classes ( six and ten instead of eight ) , a different random seed , or slightly different masks on PS1 . In all cases , although the class distributions and the structural details varied , the resulting classes revealed similar differences for TM2 and TM6 . In addition , we tested our method on a simulated set of images containing a mixture of projections from the maps of classes 1 and 2 in Figure 3 . For these simulations we used similar signal-to-noise ratios , CTF parameters and orientational distributions as observed in our experimental data set ( also see Methods ) . Masked classification with signal subtraction on the PS1 subunit correctly identified 93% of the simulated particles ( Figure 2—figure supplement 4 ) . Comparison of the three structures that were identified in the experimental data set using masked classification on the PS1 subunit ( Table 1 , Videos 1–2 ) explained observations made in the consensus structure . In the high-resolution structure , density for the cytoplasmic side of TM6 was weak , while density for TM2 was only visible after applying a 7 Å low-pass filter . This agrees with the observation that TM2 is only ordered in class 1 , whereas TM6 adopts different orientations in all three classes . Control classifications with variations in the number of classes , random seeds or masks revealed even more variations in the conformation of TM6 ( e . g . see Figure 2—figure supplement 3 ) . Probably , TM6 adopts a wide range of conformations in solution and our classification merely provides discrete snapshots of a continuum . 10 . 7554/eLife . 11182 . 012Video 1 . A morph between the atomic models from class 1 and 2 of the apo-state ensemble . DOI: http://dx . doi . org/10 . 7554/eLife . 11182 . 01210 . 7554/eLife . 11182 . 013Video 2 . A morph between the atomic models from class 2 and 3 of the apo-state ensemble . DOI: http://dx . doi . org/10 . 7554/eLife . 11182 . 013 Despite the fact that we did not focus our classification on Pen-2 , we also observe significant variations in the conformation of Pen-2 . Compared to classes 1 and 2 , Pen-2 in class 3 has rotated away from PS1 . This rotation concurs with a large conformational change in PS1 , where TM3 and TM4 rotate in the same direction as Pen-2 , TM5 and TM6 move towards the extracellular space , and TM6 rotates towards TM7 ( see Video 2 ) . The rotation and upward motion of TM6 positions the two catalytic aspartates within close enough distance of each other to potentially catalyze proteolysis ( although we do not see density for the aspartate side chains ) . When the high-resolution consensus map was low-pass filtered to 7 Å , besides density for TM2 a second , unidentified rod-shaped density was also observed in the cavity formed by TM2 , TM3 and TM5 of PS1 ( Bai et al . , 2015a ) . A similar density that could not be attributed to any of the known γ-secretase components is also visible in classes 1 and 2 , but not in class 3 . The rod-shaped density is best defined in class 1 , where it shows clear features of α-helical pitch . We modeled this density as an α-helix with an almost 90-degree kink at the extracellular side of the transmembrane domain ( Figure 4 ) . The kink of this helix is in close proximity of the loop of residues 240–244 of nicastrin , and then extends into the membrane through the cavity formed by TM2 , TM3 and TM5 of PS1 , until it disappears just before reaching the active site . The density closest to the active site looks less helical , and in this region we modeled it as an extended chain . The entire cavity in which the helix is present is lined with residues from TM2 , TM3 and TM5 of PS1 that have been implicated in FAD . The cryo-EM density was not of sufficient quality to allow the identification of this helix . Mass spectrometry analysis suggested the presence of multiple proteins with a single predicted transmembrane helix in our sample , and the presence of three of these proteins was further confirmed by Western blot analysis ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 11182 . 014Figure 4 . Helical density in class 1 of the apo-state ensemble . ( A–C ) Three different views of the map and the atomic model are shown . The kinked α-helix that is unaccounted for by the γ-secretase model is shown in purple . PS1 residues that interact with this helix are labeled . DOI: http://dx . doi . org/10 . 7554/eLife . 11182 . 01410 . 7554/eLife . 11182 . 015Figure 4—figure supplement 1 . Mass spectrometry and Western blot analyses . ( A ) The lower part of a Coomassie blue-stained SDS-PAGE gel of the γ-secretase sample that was also used for cryo-EM imaging was cut out and submitted for mass spectrometric analysis . Besides known γ-secretase components ( bold text ) , several proteins with a single predicted transmembrane helix were also identified . ( B ) Western blot analyses confirmed the presence of at least three of the identified proteins in the sample . Arrowheads indicate their expected molecular weights . DOI: http://dx . doi . org/10 . 7554/eLife . 11182 . 015 In order to gain further insights into the plasticity of the catalytic subunit , we also performed cryo-EM single-particle analysis on γ-secretase in complex with DAPT ( Figure 5 , Figure 5—figure supplement 1 ) . Despite collecting a comparable amount of micrographs as we did for the apo-state complex , 2D and 3D classification approaches selected less than 20% of the complexes as suitable for high-resolution reconstruction . This contrasts with a selection of approximately 40% for the apo-state data set . Because the overall appearance of the micrographs for both data sets was similar , it could be that either DAPT or the dimethyl sulfoxide ( DMSO ) in which DAPT was dissolved interfered with the structural integrity of the complex . Nonetheless , from the selected 51 , 366 particles we calculated a 4 . 2 Å map , which was of sufficient quality to build a reliable main-chain model , although the density for many side chains was less clear . For these data , masked classifications with signal subtraction revealed only a single class with good density for the transmembrane helices , and this class did not show any helical-like density in the cavity between TM2 , TM3 and TM5 of PS1 ( Figure 5—figure supplement 2 ) . 10 . 7554/eLife . 11182 . 016Figure 5 . Cryo-EM structure of γ-secretase in complex with DAPT . ( A ) The reconstructed density for the entire complex . Density attributed to DAPT is shown in blue . ( B ) Schematic representation of the atomic model . TMs of PS1 are numbered . ( C ) Two approximately orthogonal close-ups of the DAPT-binding site . Residues that interact with DAPT , as well as His163 and Glu280 are labeled . DOI: http://dx . doi . org/10 . 7554/eLife . 11182 . 01610 . 7554/eLife . 11182 . 017Figure 5—figure supplement 1 . Fourier shell correlations for the DAPT-bound structure . ( A ) FSC curves of the refined atomic model versus the map it was refined against ( in black ) ; of a model refined in the first of the two independently refined half-maps versus that same map ( in red ) ; and of a model refined in the first of the two independent half-maps versus the second half-map ( in green ) . ( B ) FSC between the two independently refined half-maps . DOI: http://dx . doi . org/10 . 7554/eLife . 11182 . 01710 . 7554/eLife . 11182 . 018Figure 5—figure supplement 2 . Masked classification with signal subtraction on the DAPT-bound data . Using either masks around the PS1 subunit ( top row ) or the entire transmembrane domain ( bottom row ) , masked classification with signal subtraction revealed only a single majority class ( pink ) that showed good density for the transmembrane helices ( corresponding to 23% and 26% of the particles , respectively ) . In these maps , no helical-like density was observed in the cavity between TM2 , TM3 and TM5 of PS1DOI: http://dx . doi . org/10 . 7554/eLife . 11182 . 01810 . 7554/eLife . 11182 . 019Figure 5—figure supplement 3 . Newly ordered elements in the DAPT-bound structure . Two orthogonal views of a cartoon representation of the transmembrane domain of the DAPT-bound structure are shown . Those parts of the atomic model that were built in this structure but were disordered in the high-resolution consensus structure of the apo-state are shown in red . The density attributed to the DAPT molecule is shown in blue . DOI: http://dx . doi . org/10 . 7554/eLife . 11182 . 01910 . 7554/eLife . 11182 . 020Figure 5—figure supplement 4 . Similarity between the DAPT-bound structure of PS1 and class 1 . Two orthogonal views of an overlay of the DAPT-bound PS1 structure ( in cyan ) and class 1 of the apo-state ensemble ( in grey ) are shown . The catalytic aspartates in the DAPT-bound structure are shown in red . DOI: http://dx . doi . org/10 . 7554/eLife . 11182 . 020 Upon inhibitor binding , there are no prominent changes in Aph-1 and nicastrin , and Pen-2 is in a very similar conformation as in classes 1 and 2 of the apo-state ensemble . The largest changes occur in PS1 , which is much better ordered in the complex with DAPT . TM2 and its linkers with TM1 and TM3 , the cytoplasmic ends of TM3 and TM6 , and part of the long linker between TM6-7 all become ordered upon DAPT binding ( Figure 5—figure supplement 3 ) . The linker between TM1 and TM2 sticks partially into the transmembrane region to fold back out again to connect to TM2 . This helix contains 32 residues and runs at an angle of approximately 40 degrees with the plane of the membrane . At the cytoplasmic side of the membrane a short linker connects TM2 to TM3 , and TM3 extends 7 residues longer than in the apo-state . Interestingly , apart from local changes in the cytoplasmic sides of TM2 and TM6 , the overall conformation of PS1 in the DAPT-bound structure is very similar to class 1 from the apo-state ensemble , and the two structures overlap with a root mean squared deviation ( r . m . s . d . ) of 0 . 4 Å between 265 Cα atom pairs ( Figure 5—figure supplement 4 ) . Near the active site , TM6 displays a strong kink at Pro264 , which positions its cytoplasmic side underneath a cavity formed by TM2 , TM3 , TM5 , TM6 and TM7 . This cavity contains the only peak of strong density in the transmembrane domain that is unaccounted for by the protein model ( Figure 5C ) . It breaks up into disconnected pieces at 4 . 5 standard deviations above the mean , which is somewhat weaker than the surrounding transmembrane helices in PS1 , which break up at approximately 6 standard deviations above the mean . The size of this density is consistent with the expected size of a single DAPT molecule . Therefore , we tentatively assign this density to the inhibitor , although at the limited resolution of our map we cannot determine its exact orientation or conformation . The pocket where the inhibitor binds is very hydrophobic , which is as anticipated given the hydrophobic nature of the DAPT molecule . In particular , Met146 and Met233 , as well as Trp165 , Phe283 and Gly384 seem to be involved in interactions with the inhibitor . Except for Phe283 , all these residues have been targeted for mutations in FAD patients . The location of the inhibitor right next to the active site is also in good agreement with previous observations that the DAPT binding site is distinct from , but in close proximity of the active site ( Kornilova et al . , 2003; Morohashi et al . , 2006 ) . Combined with a small movement of TM7 towards TM6 , the kink in TM6 brings the Cα atoms of the two catalytic aspartates within 8 . 0 Å of each other , which may again be close enough to facilitate catalysis . Also the conformation of the highly conserved 433PAL435 motif , which is important for γ-secretase activity ( Wang et al . , 2004 , 2006 ) , changes upon DAPT binding . A large part of the linker between TM6 and TM7 remains invisible , but at the cytoplasmic end of TM6 part of this linker becomes ordered as it passes beneath the inhibitor ( Figure 5C ) . There , residues 280-283 form a single α-helical turn , which is stabilized by a hydrogen bond between Glu280 and His163 , while Phe283 interacts with the inhibitor . Mutation of Glu280 into an alanine ( the so-called Paisa mutation ) is by far the most common cause of FAD ( http://www . alzforum . org/mutations ) . Residues 285–288 form a short β-strand before the density of the linker disappears just three residues before the auto-proteolytic cleavage site between Thr291 and Met292 . The density then reappears at residue 378 , where residues 378–381 form a second β-strand that hydrogen bonds with the first . Although many proteins employ functionally important flexibility at the level of secondary structure , few experimental techniques exist for the study of this dynamics . Nuclear magnetic resonance ( NMR ) is a powerful tool for the characterization of structural ensembles in dynamic complexes , but its applicability is typically limited to proteins with a molecular weight below 40– 50 kDa . For complexes larger than several hundred thousand daltons , dynamic changes in tertiary and quaternary structure have been studied by cryo-EM image classification , in particular with the recent advent of direct electron detectors and improved computer algorithms ( Bai et al . , 2015b; Dashti et al . , 2014 ) . However , for complexes that are too large for NMR , the characterization of molecular dynamics within individual protein domains has typically been restricted to computer simulations . The procedure for masked cryo-EM image classification combined with residual signal subtraction fills part of this gap in experimental techniques . The idea to subtract part of the signal from experimental cryo-EM images is not new . Michael Radermacher and colleagues subtracted partial projections from cryo-EM images to study symmetry mismatches in bacteriophage ϕ29 ( Morais et al . , 2003 ) and flaviviruses ( Zhang et al . , 2007 ) ; Steven Ludtke and colleagues used image subtraction in the e2ligandclassify . py program to separate ribosomes with and without secY channels ( Park et al . , 2014 ) ; Hongwei Wang and colleagues used a modified version of RELION to subtract projections of NSF rings to analyze a symmetry mismatch and structural variability in the SNAP-SNARE complex ( Zhou et al . , 2015 ) ; we used an iterative image subtraction method in RELION to improve the density of a flexible domain of the spliceosomal U4/U6 . U5 tri-snRNP complex ( Nguyen et al . , 2015 ) ; and most recently , Huiskonen and colleagues also used RELION to subtract viral capsid densities in order to visualize an RNA polymerase bound inside the virus ( Ilca et al . , 2015 ) . The procedure described here provides an easily accessible and generally applicable tool for signal subtraction coupled to masked refinements and/or classifications of single-particle data . Its successful application to γ-secretase demonstrates its potential for complexes that are considered to be relatively small for cryo-EM structure determination , and shows that it allows separation of protein structures that differ only in the orientation and position of a few α-helices . Application of the masked classification approach to the data set of apo-state γ-secretase particles revealed a range of different conformations for TM6 . This conformational flexibility leads to a variation in the distance between the two aspartates that form the active site . Interestingly , the class where the aspartates are closest together also shows a markedly different conformation of Pen-2 and a re-arrangement of TM3 , TM4 and TM5 in PS1 . Pen-2 is required for autocatalytic maturation and protease activity of γ-secretase . Binding of Pen-2 to the complex activates the active site , and binding of Pen-2 was observed to have an allosteric effect on TM6 ( Takeo et al . , 2012 ) . However , in the absence of PS1 structures without Pen-2 bound , it will probably remain unclear whether the changes in Pen-2 and PS1 observed here are relevant to this allosteric activation mechanism . Our analysis of the structure in complex with DAPT provides complementary insights into the conformational freedom of γ-secretase . Upon binding of the inhibitor , the catalytic subunit undergoes a marked rigidification . TM2 and its linkers to TM1 and TM3 become ordered , and so do the cytoplasmic half of TM6 and part of the linker between TM6 and TM7 . Around the active site , the conformations of the kink in TM6 and the conformation of the long linker between TM6 and TM7 are noticeably different from the apo-state consensus structure . Maturation of the γ-secretase complex requires auto-proteolytic cleavage at Thr290 ( or alternatively at Val292 or Met297 ) . The observed kink in TM6 may expedite the U-turn that is required to position the auto-proteolytic cleavage site back into the active site . Moreover , auto-proteolytic cleavage is predicted to occur in an α-helix spanning residues 280–300 of the linker . In the DAPT-bound structure , a single helical turn starts at Glu280 , but residues 285–288 form a short β-sheet with the end of the linker that connects to TM7 . The auto-proteolytic site is still flexible in this structure , as the density disappears after Tyr288 . Since auto-proteolysis appeared to be complete in our sample ( Lu et al . , 2014 ) , it could be that residues 280–300 do form a helical structure prior to self-cleavage . Alternatively , it could be that the inhibitor specifically alters the secondary structure in the linker , for example through its observed interaction with Phe283 . TM2 of PS1 is well ordered in both the DAPT-bound structure and in class 1 of the apo-state , whereas it is invisible in the other apo-state classes . This helix only seems to be ordered when something is bound in the large cavity that is lined with FAD-derived mutations between TM2 , TM3 and TM5 . In the inhibitor-bound structure this cavity contains the density that we attribute to DAPT , while density for a kinked α-helix is visible in class 1 of the apo-state . Although the density for the kinked α-helix was not of sufficient quality for unambiguous identification , mass spectrometry and Western blot analyses suggest that a mixture of different proteins with a single transmembrane helix may be present in our sample . Four of the proteins that were identified by mass spectrometry had also previously been observed to co-purify with γ-secretase: TMP21/p23 , p24a , Vamp-8 and Sec22B ( Wakabayashi et al . , 2009 ) . TMP21 was also observed to be a component of the γ-secretase complex that acts as a negative regulator of γ-cleavage , while leaving ε-cleavage intact ( Chen et al . , 2006 ) . We hypothesize that the kinked α-helix in our structure arises from a mixture of co-purified proteins in our sample that bind to the γ-secretase complex in a manner that mimics substrate binding . This hypothesis is in good agreement with previous biochemical observations about an 'initial substrate-binding site' that is distinct from the active site ( Beher et al . , 2003; Das et al . , 2003; Esler et al . , 2002; Tian et al . , 2002 ) . Photolabeling experiments suggest that the initial substrate-binding site partially overlaps with the DAPT-binding site ( Kornilova et al . , 2003; Morohashi et al . , 2006 ) , and mutational analysis identified TM2 and TM6 to be involved in substrate binding ( Watanabe et al . , 2010 ) . Experiments with photoaffinity probes based on α-helical substrate-like inhibitors showed that DAPT could not displace a 10-residue long helical probe , suggesting spatially separated binding sites of the substrate and the inhibitor . This however was not the case for a 13-residue long peptide , for which addition of DAPT led to a strong reduction in photolabeling . Because this peptide also prevented labeling of a transition-state mimicking photoprobe , the longer α-helical probe probably also interacts with the active site ( Kornilova et al . , 2005 ) . Our hypothesis explains these data well . A superposition of the kinked α-helix on top of the DAPT structure shows that after the kink , the α-helix extends for 10 residues into the transmembrane domain , before its four C-terminal residues overlap with the DAPT-binding site and almost reach the active site ( Figure 6A ) . 10 . 7554/eLife . 11182 . 021Figure 6 . A hypothesis for substrate binding . ( A ) A superposition of the kinked α-helix from class 1 and the DAPT-density and atomic model for PS1 from the DAPT-bound structure . The lower end of the kinked helix and the DAPT density overlap . The recently identified residues that interact with a phenylimidazole-like γ-secretase modulator are shown with spheres . ( B ) Ensemble of NMR-models for an Aβ42 peptide in an aqueous solution of fluorinated alcohols . ( C ) Hypothetical model for how APP ( one of the NMR models is shown in pink ) binds to γ-secretase . DOI: http://dx . doi . org/10 . 7554/eLife . 11182 . 02110 . 7554/eLife . 11182 . 022Figure 6—figure supplement 1 . Predicted transmembrane regions . APP and proteins that were identified by mass spectrometry to be present in the cryo-EM sample contain a large , positively charged residue ( boxed ) next to the N-terminus of their predicted transmembrane helix ( indicated with a purple cylinder ) . TMP21 is the product of the TMED10 gene; P24a is the product of the TMED2 gene . The products of the TMED1 , TMED4 , TMED9 , VAMP2 and VAMPB genes were observed after additional mass-spectrometric analysis of specific gel bands . DOI: http://dx . doi . org/10 . 7554/eLife . 11182 . 022 Furthermore , the NMR structure of the Aβ42 peptide in an aqueous solution of fluorinated alcohols shows a strikingly similar 90-degree kink in its α-helical structure , which is positioned right after Lys699 ( Figure 6B ) . A superposition of the NMR model on the kinked α-helix in our structure places its N-terminus flat on the extracellular side of the membrane , while the cleavage site that would result in the formation of Aβ42 peptides is placed in close proximity to the active site ( Figure 6C ) . Interestingly , all the proteins that were both identified in our sample by mass spectrometry and that had also previously been observed to co-purify with γ-secretase ( Wakabayashi et al . , 2009 ) contain either an arginine or a lysine just before their predicted transmembrane helix ( Figure 6—figure supplement 1 ) . It is therefore tempting to speculate that the large , positively charged residue together with the kink in the α-helix act as an anchor to delimit the cleavage position in the substrate . This would explain why mutations in Lys699 have a marked effect on the length of the Aβ cleavage products ( Kukar et al . , 2011 ) . In addition , the kink in the α-helix is positioned next to the recently mapped binding site of a phenylimidazole-like γ-secretase modulator ( highlighted with spheres in Figure 6A ) , which could explain how this modulator affects cleavage in the distant active site through changes in the substrate conformation ( Takeo et al . , 2014 ) . It could also explain why replacement of Lys699 or increasing the positive charge at Gly700 has been observed to block or attenuate the effects of γ-secretase modulators ( Jung et al . , 2014 ) . In conclusion , we show that masked cryo-EM image classification combined with the subtraction of part of the signal from the experimental images allows separation of structures that differ at the secondary structure level . Application of this approach to our previously described data set of γ-secretase in its apo-state reveals distinct conformations for TM2 and TM6 of PS1 . In one of the identified structures , we observe a kinked α-helix in the cavity between TM2 , TM3 and TM5 of PS1 that cannot be attributed to any of the γ-secretase components . Mass spectrometry and Western blot analyses suggest the presence of a mixture of proteins with a single transmembrane helix that co-purified with the complex , one of which is a known modulator of γ-secretase cleavage . These results are complemented with a cryo-EM structure of the complex bound to the dipeptidic inhibitor DAPT . Binding of the inhibitor leads to a marked reduction in the flexibility of the catalytic subunit . Together , our results form the basis for a hypothesis that substrate enters the transmembrane domain through the cavity formed by TM2 , TM3 and TM5 of PS1 . The disordered nature of TM2 in the absence of substrate or inhibitor suggests that this helix may act as a lateral gate through which the transmembrane helix of the substrate enters the cavity . The observation that the inhibitor-bound structure closely resembles the class with the kinked α-helix suggests that the inhibitor and the substrate stabilize a similar conformation from the apo-state ensemble . DAPT would then act as an inhibitor by blocking access of the substrate to the active site . An appealing route to confirm our hypothesis , and to gain further mechanistic insights , is the determination of additional cryo-EM structures in complex with different substrates , substrate analogues , inhibitors or other modulators of activity . Masked classification with signal subtraction will be a useful tool in this endeavor , as understanding how different factors shift the equilibrium of conformational states in this flexible enzyme will be key to increase our understanding of its functioning . The sample preparation procedure and imaging conditions for the apo-state data were described previously ( Bai et al . , 2015a ) . To prepare the complex with DAPT , we incubated 20 μl of the same γ-secretase sample for 20 min at 4°C with 0 . 2 μl of a 10 mM solution of DAPT in 100% DMSO ( yielding an approximate concentration of 4 μM γ-secretase , 100 μM DAPT and 1% DMSO ) . Subsequently , aliquots of 3 μl were applied to previously glow-discharged holey carbon grids ( Quantifoil Au R1 . 2/1 . 3 , 300 mesh ) , and flash frozen in liquid ethane using an FEI Vitrobot . The imaging conditions were kept identical as for the apo-state data . In brief , zero-energy loss images were recorded manually on an FEI Titan Krios microscope at 300 kV , using a slit width of 20 eV on a GIF-Quantum energy filter . A Gatan K2-Summit detector was used in super-resolution counting mode at a calibrated magnification of 35 , 714× ( yielding a pixel size of 1 . 4 Å ) , and a dose rate of ~2 . 5 electrons/Å2/s ( ~5 electrons/pixel/s ) . Exposures of 16 s were dose-fractionated into 20 movie frames . Defocus values in the DAPT-bound data set ranged from 0 . 6–2 . 8 µm . Similar image processing procedures were employed for the apo-state and the DAPT-bound data sets . We used MOTIONCORR ( Li et al . , 2013 ) for whole-frame motion correction , CTFFIND4 ( Rohou and Grigorieff ) for estimation of the contrast transfer function parameters , and RELION-1 . 4 ( Scheres , 2012 ) for all subsequent steps . References for template-based particle picking ( Scheres , 2015 ) were obtained from 2D class averages that were calculated from a manually picked subset of the micrographs . A 20 Å low-pass filter was applied to these templates to limit model bias . All low-pass filters employed were cosine-shaped and fell to zero within 2 reciprocal pixels beyond the specified frequency . To discard false positives from the picking , we used initial runs of 2D and 3D classification to remove bad particles from the data . The selected particles were then submitted to 3D auto-refinement , particle-based motion correction and radiation-damage weighting ( Scheres , 2014 ) . The resulting 'polished particles' were used for masked classification with subtraction of the residual signal as described in the main text , and the original particle images from the resulting classes were submitted to a second round of 3D auto-refinement . All 3D classifications and 3D refinements were started from a 40 Å low-pass filtered version of the high-resolution consensus structure . Fourier Shell Coefficient ( FSC ) curves were corrected for the effects of a soft mask on the FSC curve using high-resolution noise substitution ( Chen et al . , 2013 ) . Reported resolutions are based on gold-standard refinement procedures and the corresponding FSC=0 . 143 criterion ( Scheres and Chen , 2012 ) . Prior to visualization , all density maps were corrected for the modulation transfer function ( MTF ) of the detector , and then sharpened by applying a negative B-factor that was estimated using automated procedures ( Rosenthal and Henderson , 2003 ) . For the apo-state data set , the template-based algorithm picked 1 . 8 million particles from 2 , 925 micrographs , and 412 , 272 particles were selected after initial 2D and 3D classification . Subsequent 3D auto-refinement and particle polishing yielded a 3 . 5 Å map with fuzzy densities in the transmembrane region . Masked classification into eight classes with subtraction of the residual signal yielded three classes with good density as described in the main text . Poor reconstructed density was observed in the other five classes . Separate 3D auto-refinements of the corresponding particles in the original data set for the three best classes gave rise to reconstructions to 4 . 0– 4 . 3 Å resolution ( also see Figures 2–3 , Table 1 ) . For the DAPT-bound state , 1 . 4 million particles were picked from 2 , 206 micrographs , and initial classification selected 271 , 361 particles . After particle polishing , this subset gave rise to a 4 . 3 Å resolution map with relatively poor density in the transmembrane domain . Application of the masked classification procedure with residual signal subtraction into eight classes identified a single class with good density . After 3D auto-refinement , the corresponding 51 , 366 particles gave a map with a resolution of 4 . 2 Å , which showed improved density in the transmembrane domain . To expedite application of the modified classification procedures proposed in this paper by others , we describe these steps in more detail . The mask for masked classification on the PS1 subunit ( the red part of the signal in Figure 1 ) was generated by converting the atomic model of the PS1 subunit from the high-resolution consensus structure ( including a poly-alanine model for TM2 ) into a density map using the program e2pdb2mrc . py from EMAN2 ( Tang et al . , 2007 ) . This map was then converted into a soft-edged mask using relion_mask_create . A mask around the entire γ-secretase complex , including the belt of fuzzy density from the amphipols , was generated using standard auto-masking from the RELION post-processing procedure . Subtraction of the PS1 mask from the mask of the entire complex using relion_image_handler yielded a mask containing only nicastrin , Aph-1 , Pen-2 and the amphipol belt ( the yellow part of the signal in Figure 1 ) . This mask was applied to the 3 . 5 Å map that was calculated from a consensus refinement using all 400 thousand selected apo-state particles . The resulting masked map ( yellow . mrc ) was used for subtraction of the signal from the experimental particles as outlined in Figure 1 . The corresponding program has been available in RELION from release 1 . 3 onwards and is used as follows: relion_project --i yellow . mrc --subtract_exp --angpix 1 . 4 --ctf --ang Refine3D/run1_data . star --o newparticles The Refine3D/run1_data . star file was produced by the consensus refinement and contains the orientation and CTF parameters of all 400 thousand particles . The command generates a new particle image stack called newparticles . mrcs and a new STAR-file with all relevant metadata called newparticles . star . The latter is used directly as input in the masked classification run , which may be launched from the RELION GUI using standard inputs , and providing the mask around the PS1 subunit as 'Reference mask' on the Optimisation tab . Because of the small size of the PS1 subunit , we chose to set the 'Perform image alignment' option on the Sampling tab to 'No' . To generate the simulated particle images described in Figure 2—figure supplement 4 , we also used the relion_project program: relion_project --i Refine3D/run1_class001 . mrc --osimulated_run1 --ctf --angpix 1 . 4 --angRefine3D/run1_data . star --add_noise --model_noiseRefine3D/run1_model . star By providing the data . star file ( option --ang ) and final map ( option --i ) from the refinements of the classes described in Figure 3 , we generated simulated particles with similar orientational distributions and CTF parameters as those observed for the experimental data . By using the estimated power spectra of the noise for the experimental images ( as provided through the --model_noise option ) , also the simulated spatial frequency-dependent signal-to-noise ratios are similar to those in the experimental data . Model building for the three apo-state classes and the DAPT-bound structure was started from the coordinates that were built in our 3 . 4 Å apo-state consensus structure ( PDB ID: 5A63 ) . Nearly all of the residues from Aph-1 and nicastrin fitted well into the four maps , but parts of PS1 and Pen-2 had to be manually adjusted in COOT ( Emsley et al . , 2010 ) . Building of TM2 and the lower part of TM6 in PS1 was started from idealized α-helices , and sequence assignment of TM2 was guided by comparison with the structure of PSH ( PDB ID: 4HYC ) and by recognizable side chain features for Phe and Tyr residues . All models were refined in REFMAC ( Murshudov et al . , 1997 ) , using modified procedures for cryo-EM maps ( Brown et al . , 2015 ) and secondary structure restraints generated by ProSMART ( Nicholls et al . , 2014 ) . Overfitting of the atomic model for the DAPT-bound structure was monitored by refining the model in one of the half-maps from the gold-standard refinement approach , and testing the resulting model against the other half-map ( Amunts et al . , 2014 ) . The same relative weight between the EM-density and geometric terms that resulted in good fits to the density without overfitting for the DAPT-bound structure was used for the final refinement of all four structures . Protein samples ( 5 μg purified γ-secretase ) were resolved by sodium dodecyl sulphate polyacrylamide gel electrophoresis ( SDS-PAGE ) on 4–12% Bis-Tris gels ( Life Technologies , Carlsbad , CA ) using MES running buffer ( Formedium , UK ) for 32 min at 200V . The gel was stained with InstantBlue protein stain ( Expedeon , San Diego , CA ) for direct protein visualization . Gel slices were prepared for mass spectrometric analysis using the Janus liquid handling system ( PerkinElmer , UK ) . Briefly , the excised protein gel pieces were placed in a well of a 96-well microtitre plate and destained with 50% v/v acetonitrile and 50 mM ammonium bicarbonate , reduced with 10 mM DTT , and alkylated with 55 mM iodoacetamide . After alkylation , proteins were digested with 6 ng/µl trypsin ( Promega , UK ) overnight at 37°C . The resulting peptides were extracted in 2% v/v formic acid , 2% v/v acetonitrile . The digest was analysed by nano-scale capillary LC-MS/MS using an Ultimate U3000 HPLC ( ThermoScientific Dionex , San Jose , USA ) to deliver a flow of approximately 300 nl/min . A C18 Acclaim PepMap100 5 µm , 100 µm x 20 mm nanoViper ( ThermoScientific Dionex , San Jose , USA ) , trapped the peptides prior to separation on a C18 Reprosil-pur 3 µm , 75 µm x 105 mm PicoCHIP ( New Objectives , MA , USA ) . Peptides were eluted with a gradient of acetonitrile . The analytical column outlet was directly interfaced with a hybrid linear quadrupole fourier transform mass spectrometer ( LTQ Orbitrap XL , ThermoScientific , San Jose , USA ) . Data-dependent analysis was carried out , using a resolution of 30 , 000 for the full MS spectrum , followed by five MS/MS spectra in the linear ion trap . LC-MS/MS data were then searched against a protein database ( UniProt KB ) using the Mascot search engine programme ( Matrix Science , UK ) ( Perkins et al . , 1999 ) . Database search parameters were set with a precursor tolerance of 10 ppm and a fragment ion mass tolerance of 0 . 8 Da . One missed enzyme cleavage was allowed and variable modifications for oxidized methionine , carbamidomethyl cysteine , pyroglutamic acid , phosphorylated serine , threonine and tyrosine were included . MS/MS data were validated using the Scaffold programme ( Proteome Software Inc . , USA ) ( Keller et al . , 2002 ) . All data were additionally interrogated manually . Protein samples ( 1 . 6 μg purified γ-secretase ) were resolved by SDS-PAGE on 4–12% Bis-Tris gels ( Life Technologies ) using MES running buffer ( Formedium ) for 32 min at 200V . The gel was transferred to Immobilon-P 0 . 45 µm PVDF membrane ( Millipore , Germany ) in 1X transfer buffer ( Life Technologies ) supplemented with 20% methanol for 1 hr at 65V . The membrane was subsequently blocked for 1 hr at room temperature in 5% bovine serum albumin/Tris-buffered saline and Tween-20 ( TBST ) prior to incubation with indicated primary antibodies ( anti-TMP21 , ab133771 , 1:500 , Abcam , UK; anti-VAMP-8 , ab76021 , 1:5000 Abcam , UK; anti-Miner1 , 13318-AP , 1:500 , Proteintech , UK ) overnight at 4°C . The membrane was washed with TBST and incubated with horseradish peroxidase-linked goat anti-rabbit IgG ( NA934VS , GE Healthcare ) for 1 hr at room temperature . The membrane was washed extensively with TBST and target proteins detected on FUJI Medical X-Ray Super RX film ( 100 NIF 18 x 24 , Fujifilm , UK ) using Amersham ECL Western Blotting Detection Reagent ( GE Healthcare , UK ) .	An enzyme called gamma-secretase cuts other proteins in cells into smaller pieces . Like most enzymes , gamma-secretase is expected to move through several different three-dimensional shapes to perform its role , and identifying these structures could help us to understand how the enzyme works . One of the proteins that is targeted by gamma-secretase is called amyloid precursor protein , and cutting this protein results in the formation of so-called amyloid-beta peptides . When gamma-secretase doesn't work properly , these amyloid-beta peptides can accumulate in the brain and large accumulations of these peptides have been observed in the brains of patients with Alzheimer's disease . Earlier in 2015 , a group of researchers used a technique called cryo-electron microscopy ( cryo-EM ) to produce a three-dimensional model of gamma-secretase . This revealed that the active site of the enzyme , that is , the region that is used to cut the other proteins , is particularly flexible . Now , Bai et al . – including many of the researchers from the earlier work – studied this flexibility in more detail . For the experiments , gamma-secretase was exposed to an inhibitor molecule that stopped it from cutting other proteins . This meant that the structure of gamma-secretase became more rigid than normal , which made it possible to collect more detailed structural information using cryo-EM . Bai et al . also developed new methods for processing images to separate the images of individual enzyme molecules based on the different shapes they had adopted at the time . These methods make it possible to view a mixture of very similar enzyme structures that differ only in a small region of the protein ( in this case the active site ) . In the future , it would be useful to repeat these imaging experiments using a range of different molecules that alter the activity of gamma-secretase . Furthermore , the new image processing methods developed by Bai et al . could be used to study flexibility in the shapes of other proteins .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2015	Sampling the conformational space of the catalytic subunit of human γ-secretase
Social norms can promote cooperation by assigning reputations to individuals based on their past actions . A good reputation indicates that an individual is likely to reciprocate . A large body of research has established norms of moral assessment that promote cooperation , assuming reputations are objective . But without a centralized institution to provide objective evaluation , opinions about an individual’s reputation may differ across a population . In this setting we study the role of empathy–the capacity to form moral evaluations from another person’s perspective . We show that empathy tends to foster cooperation by reducing the rate of unjustified defection . The norms of moral evaluation previously considered most socially beneficial depend on high levels of empathy , whereas different norms maximize social welfare in populations incapable of empathy . Finally , we show that empathy itself can evolve through social contagion . We conclude that a capacity for empathy is a key component for sustaining cooperation in societies . Widespread cooperation among unrelated individuals in human societies is puzzling , given strong incentives for exploitative cheating in well-mixed populations ( Ohtsuki et al . , 2006 ) . Theories of cooperation based on kin selection , multilevel selection , and reciprocal altruism ( Nowak , 2006 ) provide some insight into the forces driving prosocial behavior , but in human societies cultural forces appear to be of much greater importance ( Gintis et al . , 2003; Buckholtz and Marois , 2012 ) . One possible explanation rooted in cultural norms is that humans condition their behavior on moral reputations: the decision to cooperate depends on the reputation of the recipient , which itself depends on the recipient’s previous actions ( Leimar and Hammerstein , 2001; Nowak and Sigmund , 2005 ) . Altruistic behavior , for instance , may improve an individual’s reputation and confer the image of a valuable member of society , which attracts cooperation from others in future interactions ( Nowak and Sigmund , 1998 ) . Game theory has been used to study how reputations might facilitate cooperation in a population engaged in repeated social interactions , such as the Prisoner’s Dilemma or the Donation Game ( Rapoport et al . , 1965; Nowak and Sigmund , 2005 ) . In the simplest analysis an individual’s reputation is binary , either ‘good’ or ‘bad’ , and the strategy of a potential donor depends on the recipient’s reputation ( Ohtsuki and Iwasa , 2004 ) – for example , cooperate with a good recipient and defect against a bad recipient . A third-party observer then updates the reputation of the donor in response to her action towards a recipient . Reputation updates are governed by a set of rules , known as a social norm , which prescribe how an individual’s reputation depends on her actions during social interactions . A common simplification in models of moral reputations is that all reputations are both publicly known and fully objective ( e . g . Nowak and Sigmund , 1998; Pacheco et al . , 2006; Ohtsuki et al . , 2009; Sasaki et al . , 2017 ) . This means that all individuals know the reputations of all members of the society , and personal opinions about each individual’s reputation do not differ . This is a reasonable assumption if there is a central institution that provides objective moral evaluation , or if opinions regarding reputations homogenize rapidly through gossip ( Nowak and Sigmund , 2005 ) . But these conditions are rare in human populations , and opinions about reputations typically differ among individuals – for instance , because observers use different moral evaluation rules , or because of divergent observation histories , or errors . In these cases a single focal individual may have different reputations in the eyes of distinct observers , resulting in much lower rates of sustained cooperation ( Okada et al . , 2017; Hilbe et al . , 2018 ) . Moral relativity – that is , when an individual’s reputation depends on the observer – introduces an interesting and overlooked ambiguity in how an observer should evaluate a donor interacting with a recipient . One approach is to assume that the observer can refer only to her own opinion of the recipient’s reputation , when evaluating a donor . We call this an ‘egocentric’ judgment , because the observer makes moral evaluations solely from her own perspective ( Figure 1a ) . Alternatively , an observer can perform a moral evaluation that accounts for the recipient’s reputation in the eyes of the donor ( Figure 1b ) . This ‘empathetic’ case requires that the observer take the perspective of another person , which assumes some capacity for recognizing the relativity of moral status . Psychological studies implicate empathy as potent driver of prosocial and cooperative behavior in human societies ( Eisenberg and Fabes , 1990; Batson et al . , 1997; Batson and Moran , 1999; Decety et al . , 2016 ) . The cognitive capacity to intentionally adopt the subjective perspective of another individual is known as a key component of empathetic behavior ( Davis , 1983 ) . This so-called ‘perspective-taking’ component of empathy is in turn related to the theory of mind , or the ability to attribute mental states to explain and predict the behavior and emotions of other individuals ( Premack and Woodruff , 1978; Hughes and Dunn , 1998 ) . Empathetic perspective-taking generally develops between infancy and pre-school years , with at least some components learned from parents ( Krevans and Gibbs , 1996; Knafo et al . , 2008; Farrant et al . , 2012 ) . And yet empathetic behavior is not universal , as even adults often fail to empathize , especially in interactions with unfamiliar social or different cultural groups ( Cikara et al . , 2014 ) . In the context of social dilemmas , it has been suggested that empathy might play a role in evaluating the ‘fairness’ of opponents’ actions and predicting their strategies ( Singer and Fehr , 2005 ) . However , the role of empathy for moral evaluation of social behavior has not been thoroughly studied . In particular , there is currently no formal way to analyze the role of empathetic perspective-taking in game-theoretic models of human cooperation . Here we work to resolve the ambiguity of subjective moral judgment by introducing the concept of empathy into game-theoretic analyses of cooperation . We treat empathy E as the probability that an observer will form moral evaluations from the perspective of another person ( Figure 1c ) . First we investigate the effects of empathy on the level of sustained cooperation under simple social norms , while players update their strategies . Next we consider evolution of empathy itself using the formalism of adaptive dynamics; and we determine conditions under which empathy will evolve and remain evolutionarily stable . We consider a population of individuals who can choose between cooperation or defection in a sequence of pairwise , one-shot donation games . In a given game the donor must choose whether or not to cooperate with the other player . If a donor cooperates she pays the cost of an altruistic act c , while the recipient receives the benefit b>c; if the donor defects she incurs no cost , and the recipient receives no benefit . The donation game is therefore a special case of the prisoner’s dilemma ( Rapoport et al . , 1965 ) characterized by the payoff matrix ( b−c−cb0 ) . The decision to cooperate or defect depends on the donor’s strategy S= ( p , q ) , which prescribes an action conditioned on the reputation of the recipient . Here p and q denote the probability that the donor will cooperate with a ‘bad’ or a ‘good’ recipient , respectively . As is common in the game-theoretic literature with reputations ( Sasaki et al . , 2017 ) , we assume that both p and q are in {0 , 1} , and so we focus on three strategies: Always Cooperate ALLC , S= ( 1 , 1 ) ) ; Always Defect ( ALLD , S= ( 0 , 0 ) ) ; and Discriminate ( DISC , S= ( 0 , 1 ) ) , which cooperates when paired with a recipient with a good reputation and defects against a recipient with a bad reputation . We also relax this assumption and report qualitatively similar results for the continuous strategy space ( p , q ) ∈[0 , 1]2 . ( We neglect anti-discriminators S= ( 1 , 0 ) , which can never achieve high frequency . ) Players’ reputations in the eyes of each member of the society are updated according to a social norm . In general , the update rule prescribed by a social norm can depend on the entire history of donor-recipient interactions , including the reputations of all interacting parties ( Santos et al . , 2018 ) . Complex rules of moral evaluation , however , require high cognitive ability and effort that seem unrealistic in many real-world social interactions . Moreover , relatively simple ‘second-order’ norms of moral assessment , which update a donor’s reputation based solely on the donor’s action and the recipient’s reputation , tend to outperform more complex social norms ( Santos et al . , 2018 ) . We consider second-order social norms , which can be encoded by a binary matrix Ni⁢j . The row-index i indicates donor’s action , i=1 for defect or i=2 for cooperate; and the column-index j indicates reputation of the recipient , j=1 for bad or j=2 for good . We focus on the four second-order norms that are most prominent in the literature: Stern Judging ( GBBG ) , Simple Standing ( GBGG ) , Scoring ( BBGG ) , and Shunning ( BBBG ) . For example , under Stern Judging ( SJ ) or Simple Standing ( SS ) an observer will assign a good reputation to a donor who punishes a recipient with a bad reputation , by defection . Whereas under Shunning ( SH ) or Scoring ( SC ) an observer will assign a bad reputation to a donor who defects against any recipient , regardless of the recipient’s reputation . Following ( Sasaki et al . , 2017 ) we also allow for errors in strategy execution and in observation: a cooperative act is erroneously executed as defection with probability e1 , while an observer erroneously assigns a bad reputation instead of a good reputation , and vice versa , with probability e2 . The broad consensus in the literature is that Stern Judging is the most efficient norm for promoting cooperation , along with widespread adoption of the discriminator strategy . This result is robust to variation in strategy exploration rates ( Santos et al . , 2016a ) , population sizes and error rates ( Santos et al . , 2016b ) , and it even extends to the realm of more complex norms of third and fourth order ( Santos et al . , 2018 ) . Pacheco et al . ( 2006 ) have additionally shown that Stern Judging is the norm most likely to evolve in a group-structured population , because it maximizes the collective payoff of the group . Prior studies of cooperation and moral assessment ( Nowak and Sigmund , 1998; Pacheco et al . , 2006; Ohtsuki et al . , 2009; Sasaki et al . , 2017; Santos et al . , 2016a; Santos et al . , 2016b ) have assumed that reputations are objective and common knowledge in the population – meaning that opinions about reputations do not differ among individuals . Here we relax this assumption and allow individuals to differ in their opinions about one another . This reveals an under-appreciated subtlety in the application of norms for updating reputations . Namely , when an observer updates the reputation of a donor interacting with a recipient , the ‘reputation of the recipient’ could be considered either from the observer’s own perspective , or from the donor’s perspective . Under a purely egotistical application of a social norm , the ‘recipient reputation’ means the reputation in the eyes of the observer , who is forming an assessment of the donor . In this case the observer either ignores , or is unaware of , the donor’s view of the recipient . This case corresponds to E=0 in our analysis , the no-empathy model of moral assessment . However , we also analyze the possibility of empathetic moral assessment , E>0 , whereby the observer may account for donor’s view of the recipient’s reputation when assessing the donor . In the extreme case E=1 , for example , the observer always uses the donor’s view of the recipient’s reputation when applying the social norm to update the donor’s reputation . In general , the parameter E∈[0 , 1] determines the probability that an observer uses the donor’s view of the recipient’s reputation , as opposed to her own view , when applying the social norm to update the donor’s reputation ( see Figure 1 and Equations 5–8 in Materials and methods ) . To analyze how empathy influences cooperation we first examine strategy evolution with a fixed degree of empathy 0≤E≤1 . We use the classic replicator-dynamic equations ( Taylor and Jonker , 1978; Nowak and Sigmund , 2004 ) that describe how the frequencies of strategies ( ALLD , ALLC , and DISC ) evolve over time in an infinite population of players’ strategies reproducing according to their payoffs . To simplify analysis , in the infinite-population model we assume that reputation frequencies reach equilibrium before strategies are updated – that is , the timescale of reputation updating is faster than that of strategy evolution ( see Materials and methods ) . For each of the four most common norms we find bi-stable dynamics ( Sasaki et al . , 2017 ) . That is , depending on the initial conditions the population will evolve to one of two stable equilibria: a monomorphic population of pure defectors , which supports no cooperation , or a population of cooperative ( non-ALLD ) strategies that supports some degree of cooperation . How does empathy influence the prospects for cooperation ? Under the Scoring norm , strategic evolution does not depend on the degree of empathy , because this norm ignores the recipient’s reputation when updating a donor’s reputation . For the other norms considered , however , empathy tends to increase cooperation . In particular , the basin of attraction towards the stable equilibrium that supports cooperation ( green regions in Figure 2 ) is always larger when players are more empathetic – meaning that when E is larger , there is a larger volume of initial conditions in the strategy space that lead to the stable equilibrium supporting cooperation . In the case of Shunning and Stern judging , the stable equilibrium that supports cooperation consists of a monomorphic population of discriminators ( Figure 2 ) . Not only is the basin of attraction towards this equilibrium larger when a population is more empathetic , but so too is the equilibrium frequency of cooperative actions increased by greater degrees of empathy ( Figure 2 and Figure 2—figure supplement 1 ) . And so empathy increases the frequency of outcomes that support cooperation , and also increases the frequency of cooperation at these outcomes . In the case of Simple Standing the stable equilibrium that supports cooperation consists of a mix of ALLC and DISC strategists . The discriminator frequency at this equilibrium increases with empathy as ( 1 ) fZ∗=1s2⁢ ( 1-e2 ) -s+1-ε⁢ ( 1-ε+s⁢ ( 2-ε-e2 ) -1 ( 1-E ) ⁢ ( ε-e2 ) ) where ε= ( 1-e1 ) ⁢ ( 1-e2 ) +e1⁢e2 and s=b/c , until it reaches fZ∗=1 . The rate of cooperative play at this mixed equilibrium shows only a weak dependence on the degree of empathy ( Figure 2—figure supplement 1 ) . Aside from the stable equilibria discussed above , for all four norms there is also an unstable equilibrium , with some portion of the population playing ALLD and some portion playing DISC . The frequency of discriminators at this unstable equilibrium is ( 2 ) fZ=cb⁢ ( ε-e2 ) ⁢ ( 1+ ( 1-E ) ⁢γNorm ) , whereγSH=s2 ( 1−e2 ) −s ( 1+e2 ( ε−e2 ) ) +e2e2 ( s2−1 ) +E ( s2 ( 1−e2 ) −s+e2 ) , γSJ= ( s2+1 ) ( 1−e2 ) −s−s ( 1−e2 ) ( ε−e2 ) E ( ( s2+1 ) ( 1−e2 ) −s ) −1+e2+s/2 , γSS= ( s2+1 ) ( 1−e2 ) −s−s ( ε−e2 ) ( 1−e2 ) E ( ( s2+1 ) ( 1−e2 ) −s ) + ( s2−1 ) ( 1−e2 ) . When E=1 this expression coincides with the expressions found by Sasaki et al . ( 2017 ) . This result reflects the sense in which previous studies that assumed no variation in personal opinions about reputations are mathematically equivalent to always taking another person’s perspective ( E=1 ) . In a finite population the frequencies of strategies do not evolve towards a fixed stable equilibrium , but rather continue to fluctuate , irrespective of initial conditions , due to demographic stochasticity . To study the impact of empathy on cooperation in this setting we undertook Monte Carlo simulations . In this model , successful strategies spread through social contagion: a strategy is copied with the probability 1/ ( 1+exp⁡ ( -w⁢[Π1-Π0] ) ) , where w is the selection strength , and Π1 and Π0 are payoffs of two randomly selected individuals ( Traulsen et al . , 2007; Traulsen et al . , 2010 , see Materials and methods ) . In addition to these imitation dynamics , player strategies also change via random exploration at a rate μ . For the sake of simplicity , we assumed that the timescale at which games are played and payoffs are acquired is much faster than the timescales of imitation , exploration , and reputation dynamics , so that each individual plays many games selection and mutation take place ( see Materials and methods ) . Empathy tends to increase mean levels of cooperation in finite populations under stochastic dynamics ( Figure 3 ) , similar to our findings in an infinite population . The effects of empathy are pronounced: the stationary mean frequency of cooperation ranges from near zero to near unity , in response to increasing the value of the empathy parameter E . For high values of empathy , Stern Judging is the most efficient social norm at promoting cooperation , followed by Simple Standing and Scoring . This rank ordering of social norms is consistent with the prior literature ( Santos et al . , 2016a; Santos et al . , 2016b; Santos et al . , 2018 ) . However we find a striking reversal from the established view of social norms when individuals are less empathetic . As E→0 Scoring promotes the most cooperation , while Stern Judging and Simple Standing engender less cooperation . And so the level of empathy strongly influences the amount of cooperation that evolves , and it even changes the ordering of which social norms are best at promoting cooperation . In particular , Stern judging is the most socially beneficial norm ( Pacheco et al . , 2006; Santos et al . , 2016a; Santos et al . , 2016b; Santos et al . , 2018 ) only when individuals account for subjectivity in moral assessment , or when individuals are forced to agree with one another through a centralized institution of moral assessment . We have seen that empathy promotes cooperation in finite populations with reputation-conditional strategies . However , empathy is not inevitable and not universal in humans ( Cikara et al . , 2014 ) . It remains unclear if empathy itself can evolve to high levels , and whether a population of empathetic individuals can resist invasion from egocentric moral evaluators . In the following analysis we assume that the degree of empathy in moral evaluation can be observed , inferred or learned , and can therefore evolve through social contagion ( imitation dynamics ) ( Cushman et al . , 2017 ) , similar to how social norms are learned ( Buckholtz and Marois , 2012 ) . Alternatively , an individual’s capacity for empathetic observations may have a genetic component evolving via Darwinian selection . We analyze the evolution of empathy using the framework of adaptive dynamics ( Geritz et al . , 1998 ) . Assuming rare mutations to the continuous empathy trait E∈[0 , 1] , we calculate the invasion fitness of an invader EI in an infinite resident population with empathy ER by comparing their expected payoffs . We report pairwise-invisibility plots and investigate the evolutionary stability of singular points E∗ , where the gradient of invasion fitness ∂⁡W⁢ ( ER , EI ) /∂⁡EI ( evaluated at EI=ER ) vanishes . To support our analytic treatment we also perform Monte Carlo simulations in finite populations subject to demographic stochasticity , where empathy evolves through social copying according to individual payoffs , similar to strategy evolution under imitation dynamics ( Traulsen et al . , 2007 ) . Evolution can often favor empathy , depending upon the social norm and the initial conditions . To study empathy dynamics , we initially assume that the population is monomorphic for the discriminator strategy . In the case of the Shunning norm , then , there is a single , repulsive singular value of empathy ( Figure 4e ) at ( 3 ) ESH∗=e2c/b+e2-1⁢ ( 1-c/bε-e2 ) +c/bε-e2 . Such a population is bistable . If the initial level of empathy exceeds ESH∗ the population will evolve towards complete empathy ( E=1 ) and the discriminator strategy will remain stable; but if the initial level of empathy is less than ESH∗ the population will evolve towards complete egocentrism ( E=0 ) , at which point the discriminator strategy is no longer stable ( Figure 2g ) and the population will be replaced by pure defectors . The singular value ESH∗ decreases as the benefit of cooperation b/c increases , permitting a larger space of initial conditions that lead to the evolution of complete empathy ( Figure 5c ) . And so , in summary , under the Shunning norm long-term strategy and empathy co-evolution will tend towards a completely empathetic population of discriminators , especially when the benefits to cooperation are high; or , alternatively , evolution will lead to complete population-wide defection . Similar dynamics occur under the Stern Judging norm . In this case , starting from a monomorphic population of discriminators , there are two singular values for E: an evolutionary repeller ESJ∗<1/2 and attractor ESJ∗>1/2 ( Figure 4a , b ) given by ( 4 ) ESJ∗=12±14+ ( ( 1−c/b ) ( e2+ε−1−c/b ) ( ε−e2 ) 2 ) . Provided empathy initially exceeds the repulsive value evolution will favor increasing empathy towards the attractive value , and the population of discriminators will remain stable . Increasing b/c again favors the evolution of empathy , as it increases the value of the locally stable ESJ∗>1/2 and also the range of initial values that that lead to ESJ∗>1/2 through fixation of small mutations ( Figure 5a ) . However , if empathy starts below the repulsive value , selection will favor evolution toward the attractive singular point at ESJ∗=0 , which no longer supports DISC as a stable equilibrium in strategy space ( Figure 2a ) . And so , in summary , under Stern Judging co-evolution of strategies and empathy will tend towards a highly empathetic population of discriminators , especially when the benefits to cooperation are large; or , alternatively , evolution will lead to all defectors and empathy will thereafter drift neutrally . The evolution of empathy is more complicated under Simple Standing . Assuming the population consists of discriminators there is a single evolutionarily stable and globally attractive singular point ESS∗ ( Figure 4c ) . The value of empathy at this singular point is larger when the benefits of cooperation are larger ( Figure 5b ) . However once this value of empathy is reached , the strategic equilibrium at pure discriminators is no longer stable , and the population will instead be replaced by a mix of DISC and ALLC strategists , under replicator dynamics . This new strategic equilibrium will , in turn , lower the singular value of ESS∗ under adaptive dynamics , which again changes the equilibrium balance of DISC and ALLC strategists . Long-term strategy-empathy co-evolution will continue in this fashion , with both ALLC and DISC present in the population , until the singular value of empathy reaches ESS∗=0 ( see Figure 5—figure supplement 1 ) . The strategic equilibrium at this point lies near the boundary of two basins of attraction ( Figure 2d ) and is vulnerable to invasion by pure defectors in a finite population . And so , in summary , while the exact dynamics will depend on the time scales of empathy and strategy evolution , Simple Standing cannot sustain empathy over the long term as both these components of personality co-evolve , eventually resulting in a population of pure defectors . Empathy has long been associated with prosocial behavior and altruism in humans . Much of the existing literature focuses on the emotional component of empathy – linkage of emotional states between individuals , emotional contagion ( Hatfield et al . , 1993 ) and empathy-induced helping ( Cialdini et al . , 1997; May , 2011 ) . For instance , there is substantial evidence that the effect of ‘self-other merging’ provides moral motivation to cooperate ( Batson et al . , 1997; Batson et al . , 1995; Batson and Moran , 1999 ) and contributes to the resolution of public-goods dilemmas ( Batson , 1994 ) . Empathy is not a unitary construct , however , and besides the purely emotional reaction to the states of other individuals there is the cognitive ability to understand another person’s psychological perspective ( Davis , 1983; Smith , 2006 ) . Very little is known about the origins of empathy in relation to cooperative behavior , although some research suggests that the capacity for emotional empathy evolved in the context of parental care ( de Waal , 2008 ) . Even less is known about social evolution and selective forces operating on empathy in modern societies . Even if empathy promotes altruistic behavior , why should empathetic perspective-taking itself evolve and be stable against the invasion of morally egocentric individuals ? Here we have studied the role of empathetic perspective-taking in a game-theoretic context of moral evaluation , where individuals make moral judgments from their own subjective perspectives . Studying the impact of empathy in this context is critical to understanding cooperative behavior in modern , highly-connected societies that generally lack a centralized institution of objective moral assessment . Social norms specify the rules of moral evaluation . It is well known that moral reputations can sustain high levels of cooperation if individuals discriminate between the ‘good’ and the ‘bad’ . Social norms themselves likely emerge from individual beliefs of what reputations should be assigned to defectors and cooperators in distinct social situations . While some studies assume that social reputations are absolute – for instance due to shared information , public monitoring , institutions and gossip – our study draws attention to the potential for disagreements on reputations that arise from errors or different observation histories . The same individual can have different reputations in the eyes of distinct observers; in other words , moral evaluations are not absolute , and social reputation is relative . When monitoring of social interactions is private , cooperation is much harder to evolve and sustain – as reflected by the results of two recent studies by Hilbe et al . ( 2018 ) and Okada et al . ( 2017 ) . Both of those studies analyzed models with private , but egocentric , moral evaluation corresponding to E=0 in our analysis . Our model of private evaluation makes the additional assumption that each observer evaluates a donor based on a different social interaction , along the lines of Okada et al . ( 2018 ) . Our key finding is that high levels of cooperation can be sustained , even with private monitoring of reputations , provided individuals recognize moral relativity and are capable of making moral judgments from another person’s perspective ( E>0 ) . Egocentric evaluation leads to unjustified or irrational defection , because a person perceived as ‘bad’ by the observer might actually appear ‘good’ in the eyes of the donor who’s action is being evaluated , or vice versa . This point is particularly striking in the case of Stern Judging , the norm that assigns a ‘good’ reputation only to individuals who cooperate with other ‘good’ players and defect against ‘bad’ ( Kandori , 1992; Pacheco et al . , 2006 ) . Despite being the most efficient norm at promoting cooperation in empathetic societies , Stern Judging performs very poorly in egocentric populations . On the other hand , Scoring – the norm that does not take into account the recipient’s reputation at all – is immune to the effects of empathy and dominates in societies with egocentric moral evaluation rules . Finally , we have shown that empathetic perspective-taking can evolve through cultural copying , and remain evolutionarily stable if a society is governed by Stern Judging or Shunning . Once these societies evolve empathy , individuals performing egocentric evaluations of observed social behavior will be rewarded less than their empathetic peers , and this remains true even if strategies are allowed to co-evolve with empathy . However , we have also seen that egocentric and uncooperative societies are nevertheless possible evolutionary outcomes . In populations governed by Stern Judging , Shunning and Scoring this outcome represents an alternative locally ( though not globally ) attractive stable state in the strategy-empathy phase space . In the case of Simple Standing , the egocentric and uncooperative outcome is the only long-term stable outcome as both empathy and strategies are allowed to evolve . Our study raises a number of questions to be addressed in future work on empathy , norms , and the evolution of cooperation . Whereas we have studied empathy as a fixed trait , an individual’s tendency for empathetic evaluation might instead depend in a non-linear way on the current make-up of strategies in the population . Another question involves the competition of social norms for moral evaluation – a topic that has been studied in some contexts , such as when errors do not occur ( Uchida et al . , 2018 ) , or in the presence of population structure ( Masuda , 2012; Pacheco et al . , 2006 ) . Perhaps an even more fundamental question is whether and how population-wide social norms can evolve from individual moral beliefs to begin with . It is unclear whether social contagion or individual-level Darwinian selection is sufficient to establish a hierarchy of norms governing how individuals update each others’ reputations in a population . We have shown that the norms that promote the most cooperation change depending on the capacity for empathetic perspective-taking , but should we also expect different norms to evolve under empathetic and egocentric modes of judgment ? For instance , populations characterized by fully empathetic moral judgment might be conducive to the evolution of selfish norms that indiscriminately assign ‘bad’ reputations to evade costly cooperation without being punished , while models with private egocentric evaluation may lead to the evolution of more cooperative norms , such as Scoring or Stern Judging ( Yamamoto et al . , 2017; Uchida et al . , 2018 ) . Such questions about the origin of and competition between moral norms remain outstanding . In addition to the deterministic replicator-dynamics analysis of strategy evolution , we performed a series of individual-based simulations to measure mean levels of cooperation under continuous influx of mutations in the strategy space ( Santos et al . , 2016a ) . We assume that all individuals follow the same social norm and are characterized by the same value of empathy , E . The population consists of N individuals , each with its own strategy and its own subjective list of reputations . Each generation , any given individual interacts with all other members of the society in three different roles: once as a donor , once as a recipient , and once as an observer . First , each individual plays a single round of the donation game with all other members of the society according to her strategy S= ( p , q ) and the subjective reputation of the recipient , also taking into account the implementation error e1 . Here p and q denote the probabilities that a donor cooperates with a ‘bad’ ( B ) and ‘good’ ( G ) recipient , respectively . The act of cooperation fails with the probability e1 ( defection always succeeds ) . The cumulative payoff is then assigned to each individual , with the benefit of cooperation fixed at b and the cost of a cooperative act c . To update their list of subjective reputations based on the social norm Ni⁢j , each player then chooses to observe a single interaction per donor ( that is , with a randomly chosen recipient ) , again taking into account subjective reputation of the recipient either in the eyes of the donor ( probability E ) or the eyes of the observer ( with a probability 1-E ) . The newly assigned reputation is reversed with the probability e2 , representing observation errors . For the sake of simplicity , we assume that all reputations are updated simultaneously after all donor-recipient interactions have taken place . We model selection and drift of strategies as a process of social contagion implemented as a pairwise comparison process . Following the reputation-updating step , a random pair of individuals is chosen; the first individual adopts the strategy of the second with the probability 1/ ( 1+exp⁡ ( -w⁢[Π1-Π0] ) ) , where w is the selection strength , and Π1 and Π0 are payoffs of the two earned within the last generation . In our simulations of populations with N=100 individuals , we used w=1 . 0 . Finally , each individual is subject to random strategy exploration , in which a new random strategy is adopted with a small probability μ ( Santos et al . , 2016a ) . The simulation is initialized with random strategies and random lists of subjective reputations . We recorded the mean rate of cooperation averaged over 150 , 000 generations in 50 replicate populations , which is reported in Figure 3 . Let gi⁢j be the frequency of ‘good’ individuals in the sub-population i as seen by individuals belonging to the sub-population j , where i and j correspond either to resident ( i , j=0 ) or invader ( i , j=1 ) sub-population . Working in the limit of negligible invader frequencies , and assuming that the population consists only of DISC strategists , for Stern Judging norm we have: ( 10 ) g00=E0 ( g00ε+ ( 1−g00 ) ( 1−e2 ) ) + ( 1−E0 ) ( g002ε+g00 ( 1−g00 ) ( e2+1−ε ) + ( 1−g00 ) 2 ( 1−e2 ) ) ;g01=E1 ( g00ε+ ( 1−g00 ) ( 1−e2 ) ) + ( 1−E1 ) ( g01g00ε+g01 ( 1−g00 ) e2+ ( 1−g01 ) g00 ( 1−ε ) + ( 1−g01 ) ( 1−g00 ) ( 1−e2 ) ) ;g10=E0 ( g01ε+ ( 1−g01 ) ( 1−e2 ) ) + ( 1−E0 ) ( g01g00ε+g01 ( 1−g00 ) e2+ ( 1−g01 ) g00 ( 1−ε ) + ( 1−g01 ) ( 1−g00 ) ( 1−e2 ) ) . Here ε= ( 1-e1 ) ⁢ ( 1-e2 ) +e1⁢e2 , and E0 and E1 are empathy values of resident and invader sub-population . For Simple Standing norm , the relative frequencies of good individuals are: ( 11 ) g00=E0 ( g00ε+ ( 1−g00 ) ( 1−e2 ) ) + ( 1−E0 ) ( g002ε+g00 ( 1−g00 ) ( 1−e2 ) + ( 1−g00 ) g00e2+ ( 1−g00 ) 2 ( 1−e2 ) ) ;g01=E1 ( g00ε+ ( 1−g00 ) ( 1−e2 ) ) + ( 1−E1 ) ( g01g00ε+g01 ( 1−g00 ) e2+ ( 1−g01 ) g00 ( 1−e2 ) + ( 1−g01 ) ( 1−g00 ) ( 1−e2 ) ) ;g10=E0 ( g01ε+ ( 1−g01 ) ( 1−e2 ) ) + ( 1−E0 ) ( g01g00ε+g01 ( 1−g00 ) ( 1−e2 ) + ( 1−g01 ) g00e2+ ( 1−g01 ) ( 1−g00 ) ( 1−e2 ) ) . Likewise , for the Shunning norm: ( 12 ) g00=E0 ( g00ε+ ( 1−g00 ) ( e2 ) ) + ( 1−E0 ) ( g002ε+g00 ( 1−g00 ) e2+ ( 1−g00 ) g00e2+ ( 1−g00 ) 2e2 ) ;g01=E1 ( g00ε+ ( 1−g00 ) e2 ) + ( 1−E1 ) ( g01g00ε+g01 ( 1−g00 ) e2+ ( 1−g01 ) g00e2+ ( 1−g01 ) ( 1−g00 ) e2 ) ;g10=E0 ( g01ε+ ( 1−g01 ) e2 ) + ( 1−E0 ) ( g01g00ε+g01 ( 1−g00 ) e2+ ( 1−g01 ) g00e2+ ( 1−g01 ) ( 1−g00 ) e2 ) . Under Scoring , the frequencies of ‘good’ individuals do not depend on empathy: ( 13 ) g=e21−ε+e2 . We then calculate the expected payoffs of individuals in resident and invader sub-populations: ( 14 ) {Π0=b⁢ ( 1-e1 ) ⁢g00-c⁢ ( 1-e1 ) ⁢g00;Π1=b⁢ ( 1-e1 ) ⁢g10-c⁢ ( 1-e1 ) ⁢g01 . These payoffs are used to generate pairwise invasibility plots in Figure 4 . Singular points are found by setting ∂⁡ ( Π1-Π0 ) ∂⁡E1=0 and setting E0=E1 . To verify the ESS results of the adaptive-dynamics calculations we performed a series of Monte-Carlo simulations in finite populations of N=100 individuals . The simulation routine is largely the same as for strategy evolution , except that in this case we fixed the strategy at DISC and allowed E to evolve via constant influx of small mutations . Each generation , empathy of an individual changes via mutation at a rate μE=0 . 005 . Since empathy is a continuous parameter , we draw the mutational deviation δ⁢E from a normal distribution centered around δ⁢E0=0 with a standard deviation σ=0 . 01 . Selection for empathy is modeled by choosing five random pairs of individuals and assuming that in each pair the first individual copies the empathy value E1 of the second with the probability 1/ ( 1+exp⁡ ( -w⁢[Π1-Π0] ) ) , where Π1 and Π0 are their payoffs .	When meerkats have pups , they employ an individual to stand guard and warn the others of potential dangers and predators , putting their own life at risk . What seems like a selfless act is actually a common behavior found throughout the animal kingdom . But rather than acting out of concern for another ones wellbeing , it is considered to be an altruistic behavior towards kin , where an individual sacrifices its own reproductive success for the sake of the reproductive fitness of its entire clan . In human societies , however , people often act altruistically towards unrelated individuals and have developed sophisticated systems of moral evaluation to decide who is worthy of cooperation and likely to reciprocate a favor . In other words , individuals will only help those who have a good reputation for being altruistic themselves . However , for this system to work , reputations need to be public knowledge , and societies need to agree on everyones reputations . But what happens when opinions about an individual's reputation are private and vary across a population ? Now , Radzvilavicius et al . wanted to find out whether altruism can emerge when people have different opinions about each others moral reputations . To do so , they used a so-called evolutionary game theory a mathematical description of how strategies change in a population over time . In their model , each individual could decide if they wanted to pay a personal cost to create a benefit for another individual . Each participant decided whether to act altruistically based on the reputation of the recipient; observers could update the individuals reputation based on their behavior . The mathematical model revealed that when people are more empathetic and able to put themselves in someone elses shoes , altruism tends to spread over time . When people take into account different opinions and form moral judgements from another person's perspective , the population can sustain a higher level of cooperation . Moreover , the capacity for taking another person's perspective can itself evolve and remain stable in a population meaning that those individuals who evaluate each other empathetically tend to do better , and empathy spreads through social influence . These findings can help us understand how empathy might have evolved in societies that value reputation as a means of reciprocity . A next step could be to test the theory developed by Radzvilavicius et al . in manipulative experiments , or to compare the theory to field data on reputations and behavior in online interactions .	[ "Abstract", "Introduction", "Model", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology" ]	2019	Evolution of empathetic moral evaluation
The prolyl-3 , 4-dihydroxylase Ofd1 and nuclear import adaptor Nro1 regulate the hypoxic response in fission yeast by controlling activity of the sterol regulatory element-binding protein transcription factor Sre1 . Here , we identify an extra-ribosomal function for uS12/Rps23 central to this regulatory system . Nro1 binds Rps23 , and Ofd1 dihydroxylates Rps23 P62 in complex with Nro1 . Concurrently , Nro1 imports Rps23 into the nucleus for assembly into 40S ribosomes . Low oxygen inhibits Ofd1 hydroxylase activity and stabilizes the Ofd1-Rps23-Nro1 complex , thereby sequestering Ofd1 from binding Sre1 , which is then free to activate hypoxic gene expression . In vitro studies demonstrate that Ofd1 directly binds Rps23 , Nro1 , and Sre1 through a consensus binding sequence . Interestingly , Rps23 expression modulates Sre1 activity by changing the Rps23 substrate pool available to Ofd1 . To date , oxygen is the only known signal to Sre1 , but additional nutrient signals may tune the hypoxic response through control of unassembled Rps23 or Ofd1 activity . Eukaryotic cells require oxygen and must adapt to changes in its supply to maintain homeostasis . Research on mechanisms underlying oxygen homeostasis has focused mainly on hypoxia-inducible factor ( HIF ) signaling in mammals ( Semenza , 2012 ) . HIFs are heterodimeric transcription factors whose activity is regulated by molecular oxygen through the prolyl hydroxylase domain ( PHD ) enzymes . These enzymes hydroxylate HIF proline residues to promote HIF degradation using 2-oxoglutarate ( 2OG ) and oxygen as co-substrates and Fe ( II ) as a co-factor ( Kaelin and Ratcliffe , 2008 ) . The PHDs belong to a large family of non-heme , 2OG/Fe ( II ) -dependent oxygenases whose members catalyze a broad set of reactions in eukaryotes and bacteria ( Hausinger , 2015 ) . Despite this diversity , the active site of most 2OG oxygenases is located in a highly-conserved double-stranded beta-helix fold with the Fe ( II ) coordinated by a catalytic triad ( HxD/E…H ) ( McDonough et al . , 2010 ) . Since oxygen is a co-substrate for this enzyme family , they can act as sensors of cellular oxygen supply , making them ideal regulators for pathways requiring oxygen ( Ratcliffe , 2013 ) . Lipid synthesis is highly oxygen-consumptive with yeast ergosterol synthesis requiring 12 molecules of dioxygen ( Espenshade and Hughes , 2007 ) . As a result , oxygen supply and sterol synthesis are coupled through the transcription factor Sre1 in the fission yeast Schizosaccharomyces pombe . Sre1 is a sterol regulatory element-binding protein ( SREBP ) homolog required for adaptation to low oxygen environments ( Hughes et al . , 2005 ) . Notably , Sre1 is conserved across fungi , and homologs in pathogenic fungi are required for virulence since the mammalian host environment is hypoxic ( Chang et al . , 2007; Willger et al . , 2008; Bien and Espenshade , 2010 ) . In S . pombe , Sre1 is synthesized as an ER membrane-bound precursor that is cleaved in the Golgi to release its N-terminal transcription factor domain ( Sre1N ) in response to sterol depletion during hypoxia ( Hughes et al . , 2005; Porter et al . , 2010; Stewart et al . , 2011 ) . Sre1N upregulates genes required for hypoxic growth and its own transcription . Upon reoxygenation , Sre1N signaling is rapidly down-regulated by the 2OG oxygenase Ofd1 ( Hughes and Espenshade , 2008 ) . Ofd1 interacts with Sre1N in the presence of oxygen to inhibit DNA-binding and accelerate degradation ( Lee et al . , 2011; Porter et al . , 2012 ) . In contrast to HIF regulation by PHDs , oxygen regulation of Sre1N does not require its hydroxylation by Ofd1 . Rather , under hypoxia Ofd1 preferentially binds the nuclear import adaptor Nro1 ( Lee et al . , 2009 ) . Binding to Nro1 prevents Ofd1 binding to Sre1N , allowing Sre1N to activate hypoxic gene expression . This oxygen-regulated switch in Ofd1 binding requires oxygenase activity . However , the mechanism by which oxygen elicits this change in binding remains unclear as does the precise role for Ofd1 enzyme activity in regulating Sre1N . Previous work in budding yeast identified a role for the Ofd1 and Nro1 homologs Tpa1 and Ett1 , respectively , in translation termination . These studies found that ribosomes from cells lacking Tpa1 read through stop codons with increased frequency relative to wild-type cells ( Keeling et al . , 2006 ) . While Ofd1 and Nro1 function in opposition in the context of Sre1N regulation , ett1Δ cells share the same ribosomal read-through defect as tpa1Δ cells ( Henri et al . , 2010; Rispal et al . , 2011 ) . Importantly , Tpa1 oxygenase activity is required to rescue this phenotype , suggesting that the ribosome read-through defect is mediated by a Tpa1 enzyme substrate . More recently , several studies identified the small ribosomal protein uS12 as an enzyme substrate for Ofd1 , Tpa1 , and the homologs in human ( OGFOD1 ) and fly ( Sudestrada1 ) ( Loenarz et al . , 2014; Singleton et al . , 2014; Katz et al . , 2014 ) . uS12 is an essential and universal ribosomal protein that functions in translation fidelity ( Sharma et al . , 2007 ) . While the human and fly enzymes catalyze prolyl-3-hydroxylation of uS12 P62 , the fungal homologs catalyze 3 , 4-dihydroxylation of P62 . Prolyl-3-hydroxylation was implicated in translation fidelity , but the function of uS12 dihydroxylation in fungi is unknown . The finding that uS12 is an Ofd1 substrate categorizes Ofd1 as a ribosomal oxygenase ( ROX ) ( Ge et al . , 2012 ) . This small but growing enzyme family modifies ribosomal proteins from the large and small subunit in both eukaryotes and prokaryotes . For many ROXs , the function of the hydroxylated product remains unknown , similar to dihydroxylated uS12 P62 in fungi . Here , we independently identified uS12 , known as Rps23 in S . pombe , as a binding partner and substrate of Ofd1 and investigated the role that P62 dihydroxylation plays in fission yeast . We confirmed that Ofd1 dihydroxylates Rps23 and further report that Ofd1 and Rps23 form a complex with Nro1 . This complex functions to transport newly synthesized Rps23 to the nucleus while simultaneously facilitating the dihydroxylation of P62 . In addition , Ofd1 activity regulates formation of the Ofd1-Rps23-Nro1 complex such that decreased enzyme activity under low oxygen stabilizes the complex . As a result , unassembled Rps23 regulates Sre1N signaling by sequestering Ofd1 in an oxygen-dependent manner , thereby coupling hypoxic gene expression to rates of ribosomal synthesis . Finally , we identified a conserved Ofd1 binding sequence shared by all known Ofd1 binding partners . This study outlines a new paradigm for control of hypoxic adaptation , assigns a second function to uS12 hydroxylation , and defines a distinct mechanism by which an oxygenase functions as a cellular oxygen sensor . To identify conserved enzyme substrates of Ofd1/OGFOD1 , we performed a yeast two-hybrid screen for human OGFOD1 binding partners using a human fetal brain cDNA library as prey . We isolated uS12 ( coded by the RPS23 gene ) as a binding partner of OGFOD1 and confirmed that the interaction is conserved between Ofd1 and Rps23 in fission yeast ( Figure 1A; Figure 1—figure supplement 1A ) . To investigate whether Ofd1 binds directly to Rps23 , we used an in vitro GST pull-down assay . In contrast to free GST , full-length GST-Rps23 bound Ofd1 ( Figure 1B , lanes 2–3 ) , indicating direct binding between Ofd1 and Rps23 . Next , we mapped the region of Rps23 required for binding to Ofd1 . Rps23 contains an evolutionarily divergent N-terminal extension domain ( aa 1–46 ) followed by a conserved C-terminal globular domain ( aa 47–143 ) ( Smith et al . , 2008 ) . We fused GST to Rps23 truncations and found that Ofd1 binds to the N-terminal extension domain ( Figure 1B , lanes 4–5 ) . Furthermore , Rps23 aa 1–23 were necessary and sufficient to capture Ofd1 ( Figure 1B , lanes 6–7 ) . This region of Rps23 ( Figure 1C , colored magenta ) is buried in assembled 40S subunits and thus inaccessible to Ofd1 based on crystal structures of budding yeast ribosomes ( Ben-Shem et al . , 2011 ) . Therefore , Ofd1 binds to Rps23 prior to its assembly into the 40S subunit in the nucleus . Since unassembled Rps23 represents only 2% of total Rps23 in cells ( Figure 1—figure supplement 1–1B , lanes 1 and 4 ) , we used ribosome-depleted lysates to test if Ofd1 binds Rps23 in vivo . Immunopurified wild-type Ofd1 bound unassembled Rps23 in the presence of the crosslinker DSP , but failed to bind Rps5 , a ribosomal protein of similar size and charge ( Figure 1D , lanes 7–8 ) . In addition , the Fe ( II ) -binding mutant Ofd1 H142A D144A was unable to pull down Rps23 ( Figure 1D , lanes 11–12 ) , indicating that an intact Ofd1 active site is required to bind Rps23 . Since oxygenase function is required for Ofd1-Rps23 binding , we hypothesized that Rps23 is a substrate of Ofd1 . We identified Rps23 P62 as a promising target for hydroxylation based on earlier work that reported a +16 Dalton ( Da ) mass shift for mammalian RPS23 aa 61–68: QPNSAIRK ( Louie et al . , 1996 ) . P62 is essential and part of a highly conserved loop that projects into the decoding center of the ribosome upon anticodon binding ( Sharma et al . , 2007 ) . We confirmed that the +16 Da mass shift occurred on this proline residue by incubating Ofd1 with purified 6xHis-RPS23 and performing MS/MS that targeted the QPNSAIR peptide ( Figure 1—figure supplement 1–1C ) . In light of subsequent studies showing that Ofd1 and its fungal homologs dihydroxylate P62 ( Loenarz et al . , 2014 ) , we developed a stable isotope labeling with amino acids in cell culture ( SILAC ) assay to quantify Rps23 P62 hydroxylation . Wild-type cells were labeled with heavy ( H ) lysine and ofd1Δ cells with light ( L ) lysine , mixed , and processed to enrich for Rps23 from assembled 40S subunits . Following digestion by LysC , H/L peptide ratios were measured by LC-MS/MS ( Figure 1E ) . We detected only dihydroxylated P62 in wild-type cells , consistent with published findings ( Loenarz et al . , 2014 ) . Analysis of the wild-type-ofd1Δ SILAC pair revealed a complete loss of dihydroxylated P62 in ofd1Δ cells , accompanied by the appearance of unmodified P62 ( Figure 1F ) . We found similar results for the SILAC pair of wild-type and enzyme-dead ofd1 H142A D144A ( Figure 1F ) . Collectively , these data show that Ofd1 binds unassembled Rps23 ( aa 1–23 ) and independently demonstrate that Ofd1 catalyzes Rps23 dihydroxylation . While Rps23 is the only known enzyme substrate of Ofd1 , we previously showed that Ofd1 directly binds Nro1 , a negative regulator of Ofd1 and a nuclear import adaptor ( Lee et al . , 2009; Yeh et al . , 2011 ) . To determine if a common Ofd1 binding site exists , we aligned the sequences of Rps23 ( aa 1–23 ) and Nro1 ( aa 1–30 ) required for binding to Ofd1 . We identified a stretch of identical and chemically similar residues ( Figure 2A ) and mutated each amino acid to aspartate in the context of either GST-Rps23 FL ( Figure 2B ) or GST-Nro1 aa 1–30 ( Figure 2C ) to test if these residues are required for binding to Ofd1 . Mutating the N-terminal basic residue , proline , glycine , or leucine completely abolished binding to Ofd1 for both Rps23 and Nro1 , while mutating a non-conserved residue adjacent to the proline had an intermediate effect ( Figure 2B–C , lanes 4–8 ) . Downstream mutations revealed slight differences between the Ofd1 binding sequences: mutations in conserved alanine residues resulted in a severe reduction in Rps23-Ofd1 binding , while the corresponding mutations in Nro1 only partially disrupted binding ( Figure 2B–C , lanes 10–11 ) . For both Rps23 and Nro1 though , mutations in the most C-terminal amino acids displayed only mild defects in Ofd1 binding relative to upstream residues ( Figure 2B–C , lanes 13–15 ) . These similarities between the Ofd1 binding sites in Rps23 and Nro1 argue for a conserved Ofd1 binding sequence . Given that both Rps23 and Nro1 contain an Ofd1 binding site , we hypothesized that these proteins compete for binding to Ofd1 and predicted that down-regulation or deletion of one binding partner would increase binding between Ofd1 and the other partner . While Nro1 is not essential ( Lee et al . , 2009 ) , Rps23 is required for cell viability due to its role in translation ( Sharma et al . , 2007 ) . Two genes – rps23+ and rps2302+ – code for identical Rps23 proteins . While deletion of either gene results in a modest decrease in total Rps23 levels , levels of unassembled Rps23 are dramatically reduced by ~90% ( Figure 1—figure supplement 1B ) . rps23Δ and rps2302Δ cells thus represent severe knock-downs with respect to this Ofd1 binding partner . To test if Nro1 and unassembled Rps23 compete for binding to Ofd1 , we immunopurified Ofd1 from wild-type , nro1Δ , rps23Δ , and rps2302Δ lysates depleted of ribosomes and analyzed the bound fraction for Nro1 and Rps23 . In wild-type cells treated with crosslinker , Ofd1 bound both Rps23 and Nro1 , but not Rps5 ( Figure 3A , lane 6 ) . Surprisingly , Ofd1 failed to pull down Rps23 in nro1Δ cells ( Figure 3A , lane 8 ) , indicating that Ofd1-Rps23 binding requires Nro1 . In addition , Ofd1 binding to Nro1 was weakened in rps23Δ and rps2302Δ cells ( Figure 3A , lanes 9–10 ) . In the absence of crosslinker , immunopurified Ofd1 bound less Nro1 in wild-type cells and the Ofd1-Rps23 interaction was barely detected ( Figure 3B , lane 6 ) . However , Ofd1-Nro1 binding was reduced in rps23Δ and rps2302Δ cells relative to wild-type cells across both conditions ( Figure 3A–B; lanes 6 , 9–10 ) . Consistent with this , Nro1 bound less Ofd1 in rps23Δ cells and , to a lesser extent , rps2302Δ cells relative to wild-type when immunopurified from crosslinked samples ( Figure 3C , lanes 6 , 9–10 ) . Unexpectedly , Nro1 showed strong binding to Rps23 in wild-type , ofd1Δ , rps23Δ , and rps2302Δ cells , independent of crosslinker ( Figure 3C–D , lanes 6–7 and 9–10 ) . Nro1-Rps23 binding was specific to Rps23 since Nro1 did not pull down Rps5 ( Figure 3C–D , lanes 6–10 ) . We consistently observed reduced unassembled Rps5 in rps23Δ and rps2302Δ cells , indicating that mechanisms exist to coordinate ribosomal protein expression . These results show that Rps23- and Nro1-binding to Ofd1 is interdependent rather than competitive , suggesting that they form a complex with Ofd1 . To test this further , we assayed Nro1 and Rps23 binding in vitro using a GST pull-down assay . Full-length GST-Rps23 directly bound purified Nro1 ( Figure 3E , lane 3 ) . Using Rps23 truncations , we then found that Rps23 contains at least two distinct binding sites for Nro1 because both GST-Rps23 aa 1–46 and GST-Rps23 aa 47–143 bound Nro1 ( Figure 3E , lanes 4 and 6 ) . However , the Ofd1 and Nro1 binding sites on Rps23 do not overlap as GST-Rps23 aa 1–23 failed to bind Nro1 ( Figure 3E , lane 5 ) . These in vitro binding studies , together with the immunopurifications , demonstrate that Ofd1 , Rps23 , and Nro1 form a complex in cells . The discovery that Nro1 is required for Ofd1 to bind to Rps23 in vivo ( Figure 3A ) led us to hypothesize that Nro1 functions in Rps23 hydroxylation . To test this , we examined Rps23 P62-containing peptides from wild-type and nro1Δ cells using SILAC . Interestingly , less than 10% of the detected dihydroxylated P62 signal originated from nro1Δ cells ( Figure 4A ) , indicating that only 10% of the Rps23 is dihydroxylated in the absence of Nro1 compared to 100% in wild-type cells . Furthermore , we detected both unmodified and monohydroxylated P62 peptides with low H/L ratios , indicating that these peptides originated from nro1Δ cells ( Figure 4A ) . To determine the relative abundance of the unmodified and monohydroxylated forms in cells lacking Nro1 , we first measured the relative levels of unmodified Rps23 in nro1Δ and ofd1Δ cells using SILAC . We found that nro1Δ cells contributed 25% of the unmodified P62 signal , meaning that 32% of Rps23 P62 is unmodified in nro1Δ ribosomes . Consequently , monohydroxylated Rps23 P62 represents ~60% of the total Rps23 in these cells . ( Figure 4B ) . The substantial fraction of monohydroxylated P62 in nro1Δ cells shows that the efficient dihydroxylation of Rps23 P62 requires Nro1 . We tested if Nro1 participates directly in the reaction by reconstituting the reaction in vitro . Purified Rps23 and Ofd1 were incubated alone or in the presence of purified Nro1 and then analyzed by Tandem Mass Tag ( TMT ) -labeled LC-MS/MS to quantify P62 hydroxylation . The ratio of Ofd1:Rps23:Nro1 ( 1:10:10 ) replicated the physiological ratio of Ofd1:unassembled Rps23:Nro1 in cells ( data not shown ) . Consistent with the in vivo results , addition of Nro1 to the in vitro reaction increased Rps23 P62 dihydroxylation by 1 . 6 fold ( Figure 4C ) . The increase in dihydroxylated Rps23 was accompanied by a corresponding decrease in monohydroxylated Rps23 , suggesting a role for Nro1 in facilitating the second hydroxylation event . The presence of Nro1 also resulted in a small but significant reduction in the levels of the unmodified P62 substrate compared to the reaction lacking Nro1 . Together , these in vivo and in vitro results demonstrate that complete Rps23 dihydroxylation requires Nro1 . Thus far , our data shows that Ofd1 dihydroxylates unassembled Rps23 in a complex with Nro1 . Since Nro1 is structurally similar to karyopherins and functions as a nuclear import adaptor for Ofd1 ( Yeh et al . , 2011 ) , we tested if Nro1 plays a role in Rps23 localization . Wild-type cells that express Rps23-GFP from the rps2302 locus showed Rps23-GFP localized to the nucleus by live-cell imaging ( Figure 5A ) . The Rps23-GFP fusion protein does not assemble into functional 40S subunits ( data not shown ) and localized to the nucleus rather than being exported to the cytosol . In nro1Δ cells , Rps23-GFP was excluded from the nucleus despite only a minor reduction in Rps23-GFP expression ( Figure 5B ) , indicating that Nro1 is required for nuclear import of unassembled Rps23 . However , alternate pathways for Rps23 import must exist since nro1Δ cells are viable . Furthermore , Ofd1-dependent hydroxylation of Rps23 P62 is not required for nuclear import of Rps23 . In both ofd1Δ and ofd1 H142A D144A cells , Rps23-GFP localizes to the nucleus , as does Rps23 P62A-GFP in ofd1+ cells . ( Figure 5A ) . These data indicate that Rps23 enters the nucleus regardless of hydroxylation state , and suggests that nuclear import of Rps23 is not blocked under hypoxic conditions . Because both Rps23 and Ofd1 require Nro1 for nuclear localization , we hypothesized that Nro1 imports Rps23 and Ofd1 together as a complex . Therefore , we predicted that Ofd1 might be mislocalized in rps23Δ cells due to the dramatic reduction in unassembled Rps23 and consequent reduction in complex formation ( Figure 3A–D ) . mCherry-Ofd1 expressed from the endogenous locus localized primarily to the nucleus in wild-type cells , consistent with published data ( Figure 5C ) ( Hughes and Espenshade , 2008 ) . Deletion of rps23+ resulted in loss of nuclear localization with Ofd1 mislocalized throughout the cell ( Figure 5C ) . This supports our hypothesis that Ofd1 and Rps23 traffic together to the nucleus with Nro1 . Inhibition of Crm1-dependent nuclear export by leptomycin B ( LMB ) in rps23Δ cells restored Ofd1 nuclear localization , indicating that Ofd1 still traffics to the nucleus in these cells but at a reduced rate ( Figure 5C ) . However , LMB treatment failed to restore nuclear localization of mCherry-Ofd1 in nro1Δ cells , consistent with the requirement of Nro1 for Ofd1 nuclear import ( Figure 5C ) ( Yeh et al . , 2011 ) . Collectively , these data suggest that unassembled Rps23 is imported into the nucleus in a complex with Nro1 and Ofd1 and that Rps23 dihydroxylation likely occurs in conjunction with nuclear import . In addition to their roles in Rps23 hydroxylation and nuclear import , Ofd1 and Nro1 are key regulators of Sre1 signaling and hypoxic adaptation in fission yeast ( Hughes and Espenshade , 2008; Lee et al . , 2009 ) . Ofd1 inhibits signaling by binding Sre1N , which both prevents Sre1N DNA-binding and accelerates its degradation ( Lee et al . , 2011 ) . Under low oxygen , Ofd1 binds Nro1 , freeing Sre1N to activate gene expression . Given that Ofd1-Nro1 binding requires Rps23 ( Figure 3A–D ) , we hypothesized that unassembled Rps23 functions as a positive regulator of Sre1N like Nro1 . To test if Rps23 regulates Sre1 , we first analyzed growth of rps23Δ and rps2302Δ cells in the presence of the hypoxia-mimetic cobalt . In fission yeast , cobalt inhibits ergosterol synthesis , resulting in reduced ergosterol levels and the accumulation of methylated sterol intermediates ( Lee et al . , 2007 ) . Cobalt likely disrupts ergosterol synthesis by causing defects in Fe-dependent enzymes since overexpression of erg25+ , an Sre1N target gene and Fe-dependent enzyme , rescues growth on cobalt . Consequently , Sre1 activity is required for growth in the presence of cobalt and sre1Δ cells show severe growth defects on plates supplemented with cobalt chloride ( Stewart et al . , 2011 ) . Growth on cobalt thus reflects the cell’s capacity to activate Sre1 . Like nro1Δ cells , rps23Δ and rps2302Δ cells failed to grow on cobalt chloride ( Figure 6A ) . Consistent with this , rps23Δ and rps2302Δ cells , like nro1Δ cells , showed decreased Sre1N activation and reduced precursor levels under low oxygen ( Figure 6B , lanes 7–12 ) . In contrast , ofd1Δ cells resemble wild-type cells with respect to Sre1N levels despite the role of Ofd1 as a negative regulator of Sre1N ( Figure 6B , lanes 1–2 and 5–6 ) . Ofd1-independent mechanisms act upstream of Ofd1 to regulate the transport and cleavage of the Sre1 precursor to generate Sre1N ( Hughes and Espenshade , 2008; Stewart et al . , 2011 ) . The failure of rps23Δ and rps2302Δ cells to grow on cobalt was due to defects in Sre1N production rather than general defects in translation since expression of Sre1N from a plasmid rescued growth on cobalt ( Figure 6C ) . Furthermore , rps23Δ and rps2302Δ cells failed to upregulate the Sre1N target gene hem13+ in the absence of oxygen , but hypoxic activation of the Sre1N-independent gene erg1+ was normal ( Figure 6—figure supplement 6–1A ) . Finally , the rps23 mutant was the only ribosomal gene identified in both fission yeast deletion collection screens for cobalt-sensitivity ( Ryuko et al . , 2012; Burr et al . , 2016 ) , and we confirmed the specificity of the cobalt phenotype ( Figure 6—figure supplement 6–1B ) . These two screens failed to identify rps2302Δ cells as cobalt-sensitive; however , we were unable to verify the deletion of rps2302+ by PCR in the deletion collection strain ( data not shown ) . These data demonstrate that Rps23 is a specific positive regulator of Sre1 signaling . To test whether Rps23 regulates Sre1 after proteolytic activation , we used strains that express only soluble Sre1N from the sre1 locus , bypassing the requirement for cleavage ( Hughes and Espenshade , 2008 ) . Low oxygen stabilized Sre1N in wild-type cells ( Figure 6D , lanes 1–2 ) , but rps23Δ and rps2302Δ cells failed to accumulate Sre1N under low oxygen ( Figure 6D , lanes 4 and 6 ) . Deletion of ofd1+ in rps23Δ and rps2302Δ cells restored Sre1N expression to wild-type levels under low oxygen ( Figure 6D , lanes 10 and 12 ) , indicating that the defects in Sre1N accumulation are the result of increased inhibition by Ofd1 and further demonstrating that rps23Δ and rps2302Δ cells are not defective in Sre1N translation . The model for Ofd1 inhibition of Sre1N predicts that Ofd1 acts on Sre1N directly , but direct binding has not been demonstrated and the Ofd1 binding site on Sre1N has not been identified . Previous yeast two-hybrid studies mapped the Ofd1 binding region to Sre1 aa 271–340 ( Lee et al . , 2011 ) . Alignment of this region with the Ofd1 binding sites in Rps23 and Nro1 identified Sre1 aa 286–297 as a putative Ofd1 binding site ( Figure 6E ) . To test if Sre1N binds Ofd1 directly , we performed an in vitro binding assay and found that GST-Sre1 aa 271–340 bound purified Ofd1 ( Figure 6F , lane 3 ) . Mutational analysis showed that Sre1N-Ofd1 binding required Sre1N residues 286–293 and 297 ( Figure 6F , lanes 4–11 and 15 ) . In contrast , individually mutating Sre1N aa 294–296 impaired Ofd1 binding but did not abolish the interaction ( Figure 6F , lanes 12–14 ) . These data indicate that Sre1N binds Ofd1 directly through a conserved motif shared with Rps23 and Nro1 . Collectively , these experiments demonstrate that unassembled Rps23 functions with Nro1 as a positive regulator of Sre1N activity by binding Ofd1 and preventing its binding to Sre1N . Our data indicate that Ofd1 binds independently to both a complex of Rps23-Nro1 and Sre1N , a dimeric , basic helix-loop-helix leucine zipper transcription factor ( Párraga et al . , 1998 ) . Given that these proteins share a consensus Ofd1 binding sequence , Ofd1 likely forms a dimer , which is consistent with structural studies ( Figure 7A ) ( Kim et al . , 2010; Henri et al . , 2010; Horita et al . , 2015 ) . Deletion of nro1+ reduces Ofd1-Rps23-Nro1 complex formation ( Figure 3A ) and inhibits Sre1N signaling ( Lee et al . , 2009 ) , indicating that Rps23-Nro1 and Sre1N dimer compete for binding to Ofd1 . Finally , Rps23 P62 hydroxylation by Ofd1 requires oxygen ( Loenarz et al . , 2014 ) . Based on these data , we proposed a model in which Ofd1 is free to bind Sre1N and inhibit transcriptional activity under normoxia . Under hypoxia , Ofd1 hydroxylation of Rps23 is inhibited , sequestering Ofd1 in complex with Rps23-Nro1 and freeing Sre1N to activate hypoxic gene expression ( Figure 7A ) . Consistent with this , a 90% reduction in unassembled Rps23 ( rps23Δ cells; Figure 1—figure supplement 1–1B , lanes 4–5 ) resulted in impaired Sre1N signaling under low oxygen due to increased inhibition by Ofd1 ( Figure 6D ) . rps23Δ cells fail to produce sufficient Rps23 to effectively sequester Ofd1 under hypoxia . Based on this model , we hypothesized that elevated levels of unassembled Rps23 would further sequester Ofd1 and stabilize Sre1N in the presence of oxygen . We overexpressed Rps23 in 2XSRE-ura4+ reporter cells and analyzed growth on plates lacking uracil ( Figure 7B ) ( Lee et al . , 2009 ) . In this reporter strain , sre1N and sre1Δ cells failed to grow in the absence of uracil when transformed with the empty vector control ( Figure 7B , spots 1 and 3 ) . However , Rps23 overexpression in sre1N cells , but not sre1Δ cells , supported growth in the absence of uracil , indicating activation of Sre1 signaling ( Figure 7B , spots 2 and 4 ) . Furthermore , overexpression of GST-Rps23 in sre1N cells led to the accumulation of Sre1N under normoxia compared to control cells ( Figure 7C ) . GST-Rps23 serves as a proxy for unassembled Rps23 given its ability to bind both Ofd1 and Nro1 ( Figures 1B and 3E ) while being unable to assemble into functional 40S subunits: cells that express GST-Rps23 from the rps2302 locus are not viable in an rps23Δ background ( Figure 7—figure supplement 1A ) . These data support the sequestration model by demonstrating that changes in Rps23 levels affect Sre1 signaling . Next , we tested if the Ofd1-binding sites in Nro1 and Rps23 mediate the sequestration of Ofd1 . We mutated the Ofd1-binding site at several conserved residues in Nro1 and Rps23 and expressed these mutants in nro1Δ sre1N and rps2302Δ sre1N cells , respectively . These cells carrying empty vector failed to grow in the presence of cobalt ( Figure 7D ) , consistent with the fact that both nro1Δ sre1N and rps2302Δ sre1N cells were unable to activate Sre1N under hypoxia due to Ofd1-dependent inhibition ( Lee et al . , 2009; Figure 6D , lanes 5–6 , 11–12 ) . We predicted that expression of wild-type Nro1 and Rps23 would rescue growth on cobalt by sequestering Ofd1 , but that Nro1 and Rps23 mutants that fail to bind Ofd1 would consequently fail to restore growth . nro1Δ sre1N cells expressing wild-type Nro1 grew on cobalt while cells expressing the Ofd1-binding mutants - Nro1 P6D , G8D , and A11D – phenocopied empty vector controls ( Figure 7D ) . Consistent with these cobalt growth defects , cells expressing Nro1 P6D , G8D , and A11D failed to activate Sre1N under low oxygen despite equal Nro1 expression levels ( Figure 7—figure supplement 7–1B , lanes 3–10 ) . Expression of Nro1 L15D , a mutant predicted to retain Ofd1-binding ( Figure 2 ) , partially rescued both growth on cobalt and Sre1N accumulation under hypoxia ( Figure 7D; Figure 7—figure supplement 1B , lanes 4 , 12 ) . rps2302Δ sre1N cells expressing wild-type rps2302+ grew on cobalt ( Figure 7D ) . Unexpectedly , although Rps23 P4D , G6D , and A9D are defective for Ofd1-binding in vitro ( Figure 2B ) , rps2302Δ sre1N cells expressing these mutants showed partial to full recovery on cobalt relative to cells expressing wild-type Rps23 or Rps23 L13D , a mutant that binds Ofd1 in vitro . Furthermore , the Rps23 mutants except for G6D restored Sre1N activity under low oxygen ( Figure 7—figure supplement 1C ) . Reduced expression of Rps23 G6D may explain its inability to fully rescue Sre1N levels ( Figure 7—figure supplement 1D ) . These findings indicate that although the Ofd1-binding sites in Nro1 and Rps23 behave similarly in vitro , these sequences do not contribute equally to Ofd1 sequestration in vivo , with sequestration more dependent on Nro1 residues . Our model also predicts that Ofd1 oxygenase activity is crucial for the oxygen-dependent sequestration of Ofd1 by Rps23-Nro1 and hypoxia stabilizes the Ofd1-Rps23-Nro1 complex due to the inability of Ofd1 to modify Rps23 P62 . Ofd1 requires both 2OG and molecular oxygen as co-substrates for hydroxylation so we treated wild-type cells with dimethyloxalylglycine ( DMOG ) , a cell permeable competitive inhibitor of Ofd1 that is metabolized to a 2OG analog , to test if enzyme activity affects Ofd1 binding to Rps23 and Nro1 ( Jaakkola et al . , 2001 ) . Ofd1 immunopurified from cells treated with DMOG bound more Nro1 and Rps23 compared to vehicle-treated cells , in both the absence and presence of crosslinker ( Figure 7E , lanes 5–8 ) . Next , we examined whether Ofd1-Rps23-Nro1 complex stability is oxygen-dependent . Ofd1 immunopurified from wild-type cells bound more Rps23 and Nro1 in the absence of oxygen compared to cells grown in normoxia ( Figure 7F , lanes 3–4 ) . Together , these experiments show that the Ofd1-Rps23-Nro1 complex is stabilized under conditions that prevent P62 hydroxylation and support the sequestration model for Ofd1-dependent regulation of Sre1N . In this study , we report three major findings: ( 1 ) Nro1 functions in a new nuclear import pathway for the small ribosomal protein uS12/Rps23; ( 2 ) Nro1 is the first protein adaptor for prolyl dihydroxylation; and ( 3 ) Rps23 has an extra-ribosomal function essential for the fission yeast hypoxic response . Unassembled Rps23 sits at the center of these findings , and our data support the following model for how Rps23 is incorporated into ribosomes: Nro1 binds newly synthesized Rps23 in the cytosol; Ofd1 then binds Rps23-Nro1 and dihydroxylates P62 during nuclear import; Nro1 releases dihydroxylated Rps23 in the nucleus where it is incorporated into assembling 40S ribosomal subunits . Multiple lines of evidence support this model . Rps23 binds Nro1 in vitro , the interaction does not require Ofd1 , and binding is stable in the absence of chemical crosslinker ( Figure 3 ) , suggesting that Rps23 and Nro1 bind first . The order of complex assembly is further supported by the fact that Nro1 is required for Ofd1 binding to Rps23 ( Figure 3A ) , and conversely Rps23 promotes Ofd1 binding to Nro1 ( Figure 3C ) . Consistent with this , efficient dihydroxylation of Rps23 requires Nro1 in vivo and in vitro ( Figure 4 ) . Finally , Rps23 nuclear localization requires Nro1 ( Figure 5A ) , and Ofd1 nuclear import decreases in the absence of the Rps23-Nro1 complex ( Figure 5C ) , indicating that import and dihydroxylation are coordinated and coincident . It remains to be determined how the complex dissociates in the nucleus and whether Ofd1-dependent hydroxylation plays a role . Ofd1 binding to Rps23 and Nro1 is mediated through a shared binding sequence in vitro ( Figure 2 ) . Given previous work showing that the budding yeast enzyme crystallizes as a dimer and exists in solution as a reversible monomer-dimer mix ( Henri et al . , 2010; Kim et al . , 2010; Horita et al . , 2015 ) , we modeled Ofd1 as a dimer in the complex and initially assumed that the Ofd1-binding sites in Rps23 and Nro1 contribute equally and independently to binding Ofd1 ( Figure 7A ) . However , cellular studies indicate that these binding sequences are not functionally equivalent , and that sequestration of Ofd1 is more dependent on the conserved residues in Nro1 compared to Rps23 ( Figure 7D ) . These findings highlight the need for future high resolution structural studies on the Ofd1-Rps23-Nro1 complex to determine both complex stoichiometry and how the Ofd1-binding sequences contribute to complex formation . In a second key finding , Nro1 functions as an adaptor for Ofd1-dependent dihydroxylation in addition to its role as a nuclear importer . The mechanism by which Nro1 facilitates dihydroxylation of Rps23 is unclear . Previous work found that Nro1 is structurally similar to the α subunit of human collagen prolyl-4-hydroxylase that functions in substrate recruitment ( Rispal et al . , 2011 ) . Indeed , the study’s authors predicted that Nro1 may serve a similar role for Ofd1 . We report a small but significant reduction in unmodified Rps23 P62 in reactions incubated with Nro1 which supports Nro1 functioning in substrate recruitment ( Figure 4C ) . However , our finding that addition of Nro1 increased production of dihydroxylated P62 at the expense of monohydroxylated P62 suggests that Nro1 facilitates the second hydroxylation event and may act by preventing release of monohydroxylated Rps23 from Ofd1 . The stereochemistry of the monohydroxylated species that accumulates in nro1Δ cells is unknown . Previous studies showed that prolyl-3-hydroxylation is sufficient to restore the defects in translation termination caused by loss of the Ofd1 homolog Tpa1 , suggesting that loss of this post-translational modification underlies the translation defects ( Loenarz et al . , 2014 ) . Thus , we hypothesize that prolyl-3-hydroxylation of P62 is dramatically reduced in nro1Δ cells given the translation termination defect reported for budding yeast ett1Δ cells ( Henri et al . , 2010; Rispal et al . , 2011 ) . Since Ett1 and Nro1 are homologs , we predict that ett1Δ cells will show a similar loss of dihydroxylated Rps23 P62 as seen in nro1Δ cells . The requirement of Nro1 for Rps23 P62 hydroxylation explains why ett1Δ and tpa1Δ cells share the same read-through phenotype , while ofd1Δ and nro1Δ cells display opposing Sre1N phenotypes . The third finding in this study is that unassembled Rps23 plays a central role in control of the hypoxic response in fission yeast . While Rps23 P62 hydroxylation in fungi is required for translation fidelity ( Loenarz et al . , 2014 ) , we demonstrate that Rps23 has an extra-ribosomal function in the regulation of Sre1N and hypoxic gene expression . The Ofd1 hydroxylation reaction is oxygen-dependent , and this property confers oxygen regulation to the stability of the Ofd1-Rps23-Nro1 complex . The Rps23-Nro1 complex functions as a positive regulator of Sre1N by sequestering the negative regulator Ofd1 under low oxygen or DMOG-treatment ( Figure 7E–F ) . We show that manipulation of unassembled Rps23 expression activates or represses Sre1N signaling in vivo ( Figures 6 and 7 ) . Unlike the HIF system , this sequestration mechanism allows Sre1N activity to be regulated indirectly by oxygenase activity without requiring Sre1N to be an Ofd1 substrate . While the Ofd1-Rps23 enzyme-substrate relationship is conserved in humans ( Loenarz et al . , 2014 ) , we do not have evidence that OGFOD1 and RPS23 function in the hypoxic response in mammals , perhaps because metazoans lack an obvious Nro1 homolog . To date , oxygen is the only known nutrient that signals to Sre1N . However , Sre1N activates anabolic pathways such as sterol synthesis that require carbon and other nutrients ( Todd et al . , 2006 ) . Given that maintenance of cellular homeostasis requires coordination of nutrient availability and metabolism , other regulatory inputs likely exist . The unassembled ribosomal protein pool is highly dynamic and subject to multiple nutrient inputs ( Lam et al . , 2007; Gasch et al . , 2000; Schawalder et al . , 2004 ) , making Rps23 an ideal regulatory molecule . We speculate that in response to nutrient availability , rates of Rps23 synthesis couple nutrient supply to Sre1N activity . Under nutrient-deprivation , reduced Rps23 synthesis dampens Sre1N activity by failing to sequester Ofd1 , thereby efficiently matching lipid synthesis to nutrient supply . Conversely , high rates of Rps23 synthesis support maximal Sre1N activity by sequestering Ofd1 . Cells with different levels of Rps23 synthesis still respond to hypoxia through the oxygen-dependent sequestering of Ofd1 ( Figure 6B ) , but the dynamic range of this response will be tuned by nutrient supply through Rps23 . In addition , Nro1 may similarly bind other Ofd1 substrates , allowing for integration of multiple signals . Current studies are focused on testing this model . Ofd1 is a member of a large family of 2OG oxygenases that catalyze a diverse set of reactions ranging from histone demethylation for regulation of transcription to halogenation for antibiotic synthesis ( Hausinger , 2015 ) . Here , we describe a mechanism by which the requirement for oxygen as a substrate confers the ability to sequester the oxygenase from other binding partners under low oxygen conditions . These enzymes also require 2OG as a co-substrate , raising the possibility that 2OG can signal to Sre1N and control the hypoxic response . Finally , our studies define a new mechanism for oxygen-sensing and suggest that other 2OG-oxygenase family members may function through a similar mechanism to regulate as yet unidentified , non-substrate targets . Indeed , several 2OG oxygenases are also ribosomal oxygenases that could similarly leverage their substrates for oxygen-dependent regulation of cell function ( Ge et al . , 2012 ) . Common lab reagents were obtained from either Sigma or Thermo Fisher Scientific . Oligonucleotides were provided by Integrated DNA Technologies ( Coralville , IA ) . Heavy lysine ( 13C6/15N2 ) was purchased from Cambridge Isotope Laboratories ( Tewksbury , MA ) . DMOG was from Frontier Scientific Inc ( Logan , UT ) . Protease inhibitors ( 0 . 5 μM PMSF , 10 μg/ml leupeptin , and 5 μg/ml pepstatin ) were included in buffers where indicated . Strains in this study are described in Supplementary file 1 and were generated using standard techniques ( Bähler et al . , 1998; Alfa and Cold Spring Harbor Laboratory , 1993 ) . All fission yeast strains were validated by PCR -sequencing and western blotting . Haploid S . pombe cells were cultured at 30°C in rich medium ( 0 . 5% [w/v] yeast extract ( BD Biosciences , San Jose , CA ) , 3% [w/v] glucose , 225 μg/ml each of uracil , adenine , leucine , histidine , and lysine ) to 1 × 107 cells/ml unless otherwise indicated ( Moreno et al . , 1991 ) . Minimal medium constitutes Edinburgh Minimal Medium ( MP Biomedical , Santa Ana , CA ) plus supplements ( 225 μg/ml each of uracil , adenine , leucine , histidine , and lysine ) . sre1N ( aa 1–440 ) under control of the constitutive CaMV promoter was described in Hughes et al . ( 2005 ) . rps2302+ and rps2302 point mutants were cloned into a constitutive adh1-driven expression vector with leu2+ marker , derived from pART1 ( McLeod et al . , 1987 ) . GST and GST-rps2302 were placed under the control of the inducible nmt1 promoter by insertion into the BamHI/NotI restriction sites of pSLF172 ( Forsburg and Sherman , 1997 ) . nro1+ under control of the CaMV promoter was used as a template to generate nro1 point mutants and was described in Yeh et al . ( 2011 ) . Transformations were performed by electroporation unless stated otherwise . Secondary antibodies were obtained from LI-COR ( Lincoln , NE ) : IRDye800CW/IRDye680RD mouse and rabbit IgG , IRDye 680RD Detection Reagent . Commercially obtained antibodies include mouse anti-GST monoclonal ( RRID:AB_291280 , MMS-112R-500 , Covance , Princeton , NJ ) , mouse anti-GFP monoclonal ( RRID:AB_390913 , 1814460 , Roche Applied Science , Penzberg , Germany ) , mouse anti-Rps23 monoclonal ( RRID:AB_2180354 , SJ-K2 , Santa Cruz , Dallas , TX ) , mouse anti-Rps5 monoclonal ( RRID:AB_2713966 , A-8 , Santa Cruz ) , and mouse anti-actin monoclonal ( RRID:AB_626632 , C4 , Santa Cruz ) . Lab-made antibodies are rabbit polyclonal antisera and were previously described: anti-Ofd1 ( GST-Ofd1 full-length antigen; Hughes and Espenshade , 2008 ) , anti-Nro1 ( 6xHis-Nro1 full-length antigen; Lee et al . , 2009 ) , anti-Dsc5 ( 6xHis-Dsc5 aa 251–427 antigen; Stewart et al . , 2012 ) , and anti-Sre1 ( 6xHis-Sre1 aa 1–260 antigen; RRID:AB_2713965 , Hughes et al . , 2005 ) . Yeast two-hybrid screen ( >3 fold library coverage ) was conducted at Duke University using a human fetal brain cDNA library . RPS23 represented 80% of positive clones ( 56/70 ) . Confirmation of screen hits was performed following Matchmaker Gal4 Two-Hybrid System user manual ( Clontech , Mountain View , CA ) using the yeast two-hybrid S . cerevisiae strain AH109 ( James et al . , 1996 ) . Briefly , bait ( Gal4_BD empty vector , Gal4_BD-Ofd1 or Gal4_BD-OGFOD1 ) and prey ( Gal4_AD empty vector , Gal4_AD-Rps23 , or Gal4_AD-RPS23 ) plasmids were co-transformed into AH109 cells . Transformants were equally divided into two halves , then plated on control ( SD/-Leu/-Trp ) or reporter ( SD/-Ade/-His/-Leu/-Trp/X-α-Gal ) plates . Plates were incubated at 30°C for 8 days and images were acquired using a flatbed scanner in transmitted light mode at a resolution of 600 dpi . Strains were cultured in rich medium to 0 . 5–1 . 0 × 107 cells/ml . Cells ( 1 × 108 ) were pelleted and lysed by vortexing with glass beads ( 0 . 5 mm ) for 10 min at 4°C in 1% [v/v] NP-40 , 50 mM Hepes-HCl pH 7 . 4 , 100 mM NaCl , 1 . 5 mM MgCl2 , and protease inhibitors . Lysates were centrifuged at 20 , 000 × g for 1 min at 4°C and the resulting supernatant was designated the whole cell lysate ( WCL ) . Ribosomes were pelleted from WCL by centrifugation at 80 , 000 rpm for 16 min at 4°C using an Optima TLX ultracentrifuge and TLA100 rotor ( Beckman Coulter , Brea , CA ) . WCL ( 5 μg ) and ten-fold by volume supernatant were loaded on 16% acrylamide gels . Rps23 detected in the ribosome-cleared supernatant by immunoblot was considered unassembled and normalized to Dsc5 . Rps23 fractions were calculated as the mean of three biological replicates ± SEM and analyzed by paired , one-tailed t test . GST fusion protein expression vectors were generated by cloning gene sequences into the bacterial expression vector pGEX-4T3 ( GE Healthcare Life Sciences , Chicago , IL ) using BamHI/NotI restriction sites . Fusion protein expression was induced in E . coli BL21-CodonPlus ( DE3 ) -RIPL competent cells ( Agilent , Santa Clara , CA ) with 0 . 1 mM IPTG at 37°C for 2 hr in LB medium . Cells were sonicated in lysis buffer ( 25 mM phosphate pH 6 . 8 , 250 mM NaCl , 2 mM DTT , and protease inhibitors ) prior to the addition of 1/10 vol 10% [v/v] Triton X-100 . Lysates were incubated for 10 min at 4°C then centrifuged at 20 , 000 × g for 10 min . MagneGST particles ( 1 . 5 μl/reaction; Promega , Madison , WI ) were blocked with 1% [w/v] BSA diluted in lysis buffer for 30 min at room temperature then incubated for 1 hr at room temperature with saturating amounts of cleared cell lysates . Following 3 washes in 25 mM phosphate pH 6 . 8 , 250 mM NaCl , 1% [v/v] TWEEN 20 , and protease inhibitors , the particles were resuspended in 200 μl containing 84 . 5 nM Ofd1 or 84 . 5 nM Nro1 diluted in lysis buffer plus 1% [v/v] Triton X-100 for 1 hr at 4°C . Ofd1 and Nro1 were purified under native conditions as previously described ( Yeh et al . , 2011 ) . Particles were washed twice in lysis buffer plus 1% [v/v] TWEEN 20 ( no DTT ) and once in 50 mM Tris-HCl pH 7 . 5 , 100 mM NaCl prior to elution with 10 mM reduced glutathione in 50 mM Tris-HCl pH 7 . 5 , 100 mM NaCl for 30 min at room temperature . Eluates were analyzed by immunoblotting . Structures were downloaded from the PDB ( Berman et al . , 2003 ) and analyzed using the PyMOL Molecular Graphics System ( Schrödinger , LLC ) . Pairwise and multiple sequence alignments were generated using EMBOSS- WATER and T-coffee respectively ( Rice et al . , 2000; Notredame et al . , 2000 ) . Ofd1 and Nro1 immunopurifications ( IPs ) were performed as previously described ( Lee et al . , 2009 ) with modifications . Briefly , 1 × 108 cells ( Nro1 IP ) or 2 × 108 cells ( Ofd1 IP ) were pelleted , washed in PBS ( 137 mM NaCl , 2 . 7 mM KCl , 10 mM Na2HPO4 , 2 mM KH2PO4 , pH 7 . 4 plus protease inhibitors ) and resuspended in 1 ml PBS plus 2 mM DSP ( Pierce , Waltham , MA ) or vehicle ( 8% DMSO unless stated otherwise ) for 5 min to crosslink proteins . The reaction was quenched by addition of 1 M Tris-HCl pH 7 . 5 to a final concentration of 20 mM Tris . Ribosome-cleared lysates were generated as described above and 1 . 0–1 . 5 mg protein in 600 μl volume was incubated with affinity-purified antibodies ( 8 ng antibody/μg protein ) and 30 μl Protein A agarose beads ( Repligen , Waltham , MA ) for 2 hr at 4°C . Beads were washed three times and resuspended in SDS lysis buffer ( 10 mM Tris-HCl pH 6 . 8 , 100 mM NaCl , 1% [w/v] SDS , 1 mM EDTA , 1 mM EGTA ) plus protease inhibitors prior to boiling ( 95°C , 5 min ) and immunoblotting . Strains were cultured for ≥15 generations in SILAC medium ( Edinburgh minimal medium plus 75 mg/l each of leucine , histidine , adenine , uracil , and 30 mg/l heavy or light lysine ) as described ( Fröhlich et al . , 2013 ) . Pelleted cells were resuspended in lysis buffer ( 50 mM Hepes-HCl pH 7 . 4 , 100 mM NaCl , 1% [v/v] NP-40 , 1 . 5 mM MgCl2 ) plus protease inhibitors and 1 mM DTT , and lysed by vortexing with glass beads ( 0 . 5 mm ) . Lysates were cleared by centrifugation at 20 , 000 × g for 10 min and heavy-labeled or 1:1 mixtures of heavy- and light-labeled proteins were layered over a 10% sucrose cushion ( 20 mM Tris-HCl pH 7 . 2 , 500 mM KCl , 5 mM MgCl2 , 10% [w/v] sucrose ) plus protease inhibitors . Lysates were centrifuged at 41 , 000 rpm for 6 hr in the SW 41 Ti Rotor ( Beckman Coulter ) and ribosome pellets were solubilized in 150 μl of 6 M ultrapure urea , 3 M LiCl , 50 mM KCl , and 5 μM BME ( pH adjusted to 4 . 5 ) overnight at 4°C using micro stir bars . Lysates were cleared by centrifugation at 72 , 000 rpm for 1 hr in the TLA100 rotor , and proteins were precipitated from the supernatant overnight with the addition of two volumes of 20% [w/v] TCA . Precipitated proteins were washed with 1:1 [v/v] ether:ethanol and air-dried prior to resuspension in 8 M ultrapure urea , 150 mM Tris-HCl pH 8 . 5 . Samples were separated by gel electrophoresis using 16 . 5% acrylamide gels ( Bio-Rad , Hercules , CA ) and stained with Super Blue ( Protea , Morgantown , WV ) . 15–20 kDa gel bands were excised and analyzed by LC-MS/MS . Gel pieces were destained and rehydrated in 150 μl 0 . 01 µg/µl LysC ( Wako Chemicals , Japan ) in 25 mM triethylammonium bicarbonate ( TEAB ) then covered with an additional 150 μl of 25 mM TEAB and digested overnight at 37°C . Peptides were extracted from the gel pieces with 50% [v/v] acetonitrile , 0 . 1% [v/v] trifluoroacetic acid ( TFA ) and evaporated to dryness in a speed vac . Samples were rehydrated in 0 . 1% [v/v] TFA and loaded on Oasis HLB µElution solid phase extraction plates ( Waters , Milford , MA ) and desalted with two 100 μl aliquots of 0 . 1% [v/v] TFA followed by 100 μl of 10 mM TEAB . Each sample was then step fractionated under basic conditions with 10% , 25% , and 75% [v/v] acetonitrile in 10 mM TEAB yielding three fractions for each of the samples for LC/MS/MS analysis . These fractions were then dried and brought up in 10 μl of 2% [v/v] acetonitrile , 0 . 1% [v/v] formic acid . The LC-MS/MS analysis was performed on a Q-Exactive HF mass spectrometer ( Thermo Scientific , Waltham , MA ) with a nanoACQUITY nano flow chromatography system ( Waters ) . The samples were trapped at 5 μl/min then eluted onto a 20 cm x 75 µm i . d . C18 column for a 90 min gradient at 300 nl/min . The mass spectrometer settings were 240 , 000 resolution for MS and 35 , 000 resolution for MS2 with target of 3e6 for MS and 1e5 for MS2 . Maximum injection times were set to 60 and 300 millisec respectively and a normalized collision energy of 28 was used . The data from the three fractions for each sample were combined and searched against the RefSeq2015 Schizosaccharomyces pombe using the Mascot ( Matrix Science , London , UK ) search engine running through Proteome Discoverer v . 1 . 4 ( Thermo Scientific ) . The precursor mass tolerance was set to 15 ppm and the fragment tolerance was set to 0 . 03 Da . For the search settings , the following modifications were set as variable: deamidated ( N , Q ) , oxidation ( P , M ) , dioxidation ( P ) , and K8 Heavy ( K ) . The K8 Heavy modification adds a mass of 8 . 014 Da , and SILAC pairs within four ppm mass were considered for calculating SILAC ratios . All SILAC data were filtered for unique , high confidence peptide spectrum matches ( PSMs ) . For quantification , missing quan values were replaced with the minimum intensity and single-peak quan channels were used . To quantify Rps23 P62 hydroxylation between SILAC pairs , first the labeling efficiency ( e ) was calculated using the median H/L of 40S protein PSMs ( j=1 , … , m ) from the heavy sample alone , sj: ( 1 ) e=s ( m+1 ) /2/ ( 1+s ( m+1 ) /2 ) , for ordered values of sj . Next , the relative abundance of Rps23 protein in the SILAC pair was calculated using non-P62-containing H/L ratios from Rps23 PSMs ( k=1 , … , q ) , represented by rk . These values were adjusted for labeling efficiency to give rak: ( 2 ) rak=abs[1e∙rk1+rk/1-1e∙rk1+rk] P62-containing peptide ratios ( pl ) were then separated into three groups based on P62 modification state ( unmodified , monohydroxylated , and dihydroxylated ) and each group was analyzed independently . Ratios were normalized for Rps23 protein level to give pnl using the median of ( 2 ) : ( 3 ) pnl=pl/ra ( q+1 ) /2 , for ordered values of rak . Normalized values were adjusted for labeling efficiency as described above to give pal and these ratios were log2 transformed . Significance was tested using Mann-Whitney , in which the log2 ( pal ) values were compared against a theoretical set ( log2 ( snj ) ) where H = L calculated from sj values corrected for labeling efficiency ( saj ) : ( 4 ) snj= ( saj/1+saj ) ∙0 . 51- ( saj/1+saj∙0 . 5 ) To calculate the %H and %L for unmodified , monohydroxylated , and dihydroxylated P62 , the fractional heavy ( Hl ) value of pnl was determined and corrected for labeling efficiency: ( 5 ) Hl=1 e ∙[ pnl1+pnl ] %H and %L were reported as the median from ( 5 ) multiplied by 100 , with %L equal to 100 – ( %H ) . To determine the relative abundance of the unmodified and monohydroxylated forms in nro1Δ cells , we assumed that all Rps23 P62 is unmodified in ofd1Δ cells . This assumption was based on the finding that over 99 . 9% of the dihydroxylated P62 in the wild-type-ofd1Δ SILAC pair originated from the heavy-labeled wild-type cells ( Figure 1F ) . In addition , analysis of the ofd1Δ alone sample detected only unmodified Rps23 P62 . The percent unmodified in nro1Δ cells is therefore approximated by 100 ÷ ( H/L ) , where H/L is the median ratio of unmodified P62 in the ofd1Δ ( H ) - nro1Δ ( L ) SILAC pair . From this analysis , we found that unmodified P62 accounts for 32% of Rps23 in nro1Δ ribosomes . We used the same logic to calculate the percentage of dihydroxylated P62 in nro1Δ cells using the wild-type-nro1Δ SILAC pair and assuming wild-type cells contain only dihydroxylated P62 . Given conservation of mass and the expectation that P62 exists in one of only three states , we then estimated the relative percentage of monohydroxylated Rps23 P62 in nro1Δ cells as 100 - ( % unmod ) – ( % di ) , leading to the conclusion that approximately 60% of Rps23 P62 is monohydroxylated . S . pombe rps23+ cDNA was cloned into pMAL-c5X expression vector ( NEB , Ipswich , MA ) , and the plasmid was transformed into BL21-CodonPlus ( DE3 ) -RIPL cells . The transformants were grown to OD600 = 0 . 6 at 37°C and then induced with 0 . 4 mM IPTG overnight at 18°C . Cells were harvested by centrifugation and cell pellets were stored at −80°C . Bacterial pellets were resuspended in lysis buffer ( 20 mM Tris-HCl pH 8 , 150 mM NaCl ) and lysed by French press . Cleared lysates were loaded onto an amylose resin column ( NEB ) and MBP-Rps23 was eluted with 50 mM Tris-HCl pH 7 . 5 , 150 mM NaCl , 10 mM maltose , 1 mM EDTA , and 0 . 5% [v/v] Triton X-100 . Fractions containing MBP-Rps23 were pooled and dialyzed in 20 mM Tris-HCl pH 8 , 1 mM EDTA , and loaded onto a Mono Q anion exchanger ( GE Healthcare Life Sciences ) . Sample was eluted with 20 mM Tris-HCl pH 8 , 1 M NaCl , and purified MBP-Rps23 was collected in the flow-through , concentrated , and stored in 20 mM Tris-HCl pH 8 , 150 mM NaCl . For TMT analysis , reactions were performed in 20 μl volumes with Ofd1 ( 0 . 5 μM ) and MBP-Rps23 ( 5 μM ) diluted into lysis buffer ( 20 mM Tris HCl pH 7 . 5 , 150 mM NaCl ) with and without Nro1 ( 5 μM ) . Ofd1 and Nro1 were natively purified as previously described ( Yeh et al . , 2011 ) . Reactions contained 20 mM Tris HCl pH 7 . 0 , 0 . 1% [w/v] BSA , 0 . 5 mM DTT , 4 mM ascorbate , 0 . 15 mM FeSO4 , and 0 . 3 mM 2OG . After incubation at 37°C for 1 hr , the reaction was terminated by the addition of 1 . 25 μl 2 M HCl and analyzed by LC-MS/MS . For the initial identification of hydroxylated P62 , the reaction volume was scaled up to 50 μl . Ofd1 ( 5 . 2 μM ) and purified human 6xHis-RPS23 ( 79 μM ) were diluted into the lysis buffer along with 0 . 1% [w/v] BSA , 0 . 5 mM DTT , 4 mM ascorbate , 0 . 15 mM FeSO4 , and 0 . 3 mM 2OG . The reaction was incubated overnight at 37°C . Reaction samples were LysC-digested and labeled with TMT reagents ( Thermo Fisher Scientific ) for 1 hr at room temperature . Samples were step fractionated with four basic , reversed phase fractions of 5% , 15% , 25% , and 75% [v/v] acetonitrile in 10 mM TEAB , and run on a Q Exactive HF mass spectrometer ( Thermo Scientific ) . The mass spectrometer settings were 120 , 000 resolution for MS and 60 , 000 resolution for MS2 over a 90 min gradient , targeting 3e6 ions for MS and 1e5 for MS2 . Maximum injection times were 100 millisec for MS and 200 millisec for MS2 . A stepped normalized collision energy 30/35 was used for fragmentation . The isolation window was set to 1 . 2 Da with a 0 . 5 offset , and the instrument was set to fragment the top 15 peptides by data dependent analysis with a dynamic exclusion of 10 s . An inclusion list ( masses in Da: 457 . 95331 , 458 . 28107 , 463 . 28510 , 463 . 61322 , 468 . 61679 , 468 . 94397 ) was inserted to give preference to the QPNSAIRK peptide over the highest abundance peptides . The masses in the inclusion list represented the various modification states of P62 ( unmodified , monohydroxylated , and dihydroxylated ) , as well as the TMT label and deamidated N ( +0 . 98 Da ) . To increase the signal for the dihydroxylated P62 peptide , the 75% fraction was re-run with the settings ‘Do Not Pick Others . ’ Data was searched against the RefSeq2015 database Schizosaccharomyces pombe using the Mascot ( Matrix Science ) search engine running through Proteome Discoverer v . 1 . 4 ( Thermo Scientific ) with one missed cleavage allowed and a tolerance of 10 ppm MS and 0 . 03 Da MS2 . In order to compare the relative abundance of Rps23 P62 hydroxylation in the absence and presence of Nro1 , first the relative abundance of Rps23 protein between the two samples was calculated using non-P62-containing peptides . PSMs with unique QuanResultIDs were used for quantification . The intensities from each PSM ( j=1 , … , m ) , corresponding to the samples with and without Nro1 , represented by s1 and s2 respectively , were summed and divided by one another to give a correction factor ( f ) for Rps23 protein levels:f=∑j=1ms1 , j∑j=1ms2 , j Next , the P62-containing PSMs ( k=1 , … , q ) ( p1 , k and p2 , k ) were separated into three groups based on P62 modification status ( unmodified , monohydroxylated , and dihydroxylated ) and each group was analyzed independently . The ratios p1 , kp2 , k were normalized to Rps23 protein abundance and log2 transformed:log2p1 , kp2 , k∕f Finally , the Wilcoxon signed rank test was used to determine if the median of these ratios differed significantly from zero , the null hypothesis . Indicated strains were cultured in rich medium to 4–7 × 106 cells/ml . When indicated , cells were incubated with 92 . 5 nM Leptomycin B ( Cell Signaling , Danvers , MA ) and vehicle ( 0 . 5% [v/v] ethanol ) in rich medium for 1 hr . For live-cell imaging , 5 × 106 cells were pelleted , resuspended in 15 μl rich medium , and 1 μl spotted on untreated glass slides immediately prior to imaging . Cells were imaged on a Zeiss Axio Imager M2 upright fluorescence microscope ( Oberkochen , Germany ) . Images were captured using a Hamamatsu ORCA-ER digital camera ( Japan ) and iVision software ( BioVision , Milpitas , California ) , and brightness/contrast settings were adjusted using ImageJ such that the range was identical across the strains under comparison . Yeast strains were cultured and processed as previously described ( Hughes et al . , 2005 ) with hypoxic conditions maintained using an In vivo2 400 workstation ( Biotrace , Inc , Leeds , UK ) . Briefly , cells grown to 1 × 107 cells/ml were resuspended in oxygenated or deoxygenated rich medium at 5 × 106 cells/ml for 3 hr then harvested and flash froze in liquid nitrogen . Cell pellets were lysed in 27 mM NaOH , 1% [v/v] 2-mercaptoethanol followed by TCA precipitation . Protein pellets were resuspended in SDS lysis buffer , and 20–100 μg protein was loaded for gel electrophoresis and immunoblotting . For alkaline phosphatase ( Roche , Basel , Switzerland ) treatment , 20 μg protein was diluted in 50 mM Tris-HCl pH 8 . 5 and incubated at 37°C for 1 hr prior to gel loading . Immunoblotting was performed as described in Hughes et al . ( 2005 ) with modifications . Unless indicated otherwise , samples were processed as described for low oxygen assays . Following electrophoresis , gels were transferred to nitrocellulose membranes using Trans-Blot Turbo transfer system ( Bio-Rad ) . Membranes were blocked with 5% [w/v] milk in PBS-T ( 137 mM NaCl , 2 . 7 mM KCl , 10 mM Na2HPO4 , 2 mM KH2PO4 , 0 . 05% ( v/v ) Tween 20 , pH 7 . 4 ) then incubated in primary antibody followed by an appropriate secondary ( IRDye800CW or IRDye680RD mouse or rabbit IgG ) and scanned using LI-COR Odyssey CLx imaging system . For immunopurification samples , IRDye 680RD Detection Reagent was added to primary antibody incubation . Strains grown on rich medium agar for two days were diluted in water to 1 . 6 × 106 cells/ml and 3 μl were spotted on plates containing rich medium and rich medium supplemented with 1 . 6 mM CoCl2 . Plates were incubated at 30°C for the indicated time . For random spore analysis , strains were mated for two days on malt-extract plates at room temperature prior to glusulase treatment ( 0 . 5% [v/v] ) for 16 hr . Spores were then diluted to 1 . 6 × 106 cells/ml and spotted on rich medium or rich medium plus nourseothricin ( 100 µg/ml ) and G418 ( 100 µg/ml ) . Total RNA preparation from S . pombe and RT-qPCR analysis have been described previously ( Shao and Espenshade , 2014 ) . Briefly , total RNA was isolated using RNA STAT-60 ( amsbio , Cambridge , MA ) . cDNA was synthesized using oligo d ( T ) 23VN primers ( NEB ) . The tested genes were quantified by real-time PCR using SYBR Green qPCR master mix ( Promega ) . tub1+ served as the internal control to calculate the relative expression across different samples . Error bars represent ± SEM of fold changes from three biological replicates . Human RPS23 cDNA was cloned into the pProEX HTb bacterial expression vector ( Invitrogen ) and protein expression was induced in E . coli BL21-CodonPlus ( DE3 ) -RIPL competent cells ( Agilent ) with 0 . 6 mM IPTG at 37°C for 2 hr in LB medium . Cells were pelleted and solubilized in lysis buffer ( 6 M Guanidine HCl , 0 . 5 M NaCl , and 20 mM Tris-HCl pH 8 . 0 ) ( 10 ml/L bacterial culture ) at room temperature for 2 hr . Lysates were cleared by centrifugation in a JA20 rotor ( Beckman Coulter ) at 16 , 000 rpm for 40 min and incubated with Ni-NTA agarose ( 1 ml resin/10 ml lysate; Qiagen , Hilden , Germany ) on a rotator for 2 hr at room temperature . Lysates plus resin were transferred to columns and washed 3x with lysis buffer . 6xHis-RPS23 was eluted 3x with lysis buffer adjusted to pH 3 . 5 and dialyzed overnight into 25 mM phosphate pH 6 . 8 , 0 . 25 M NaCl , 2 mM DTT , and 1 mM PMSF . In vitro hydroxylation reaction with Ofd1 and 6xHis-RPS23 was digested with 1 μg trypsin ( 0 . 2 μg/μl in 50 mM acetic acid ) for 2 hr at 37°C and quenched with the addition of TFA to a final concentration of 0 . 1% [v/v] . Peptides were analyzed using LTQ Orbitrap Velos MS ( Thermo Fisher Scientific ) and parent masses ( 393 . 22 Da and 401 . 21 Da ) were selected by SIM scan . Peptides were searched against the RefSeq 2012 database Homo sapiens using the Mascot V2 . 2 . 6 ( Matrix Science ) search engine running through Proteome Discoverer v1 . 3 ( Thermo Scientific ) with fragment match tolerance set to 0 . 03 Da .	Animals , plants , and fungi need oxygen to release energy within their cells and for other chemical reactions . Enzymes that use oxygen typically become less active when less oxygen is available , and this makes them well suited to help cells sense oxygen . These enzymes include oxygenases , some of which modify proteins by adding oxygen to specific sites in a reaction called hydroxylation . Oxygenases control how mammals adapt to low levels of oxygen – a condition referred to as hypoxia . These enzymes achieve this by hydroxylating a protein – specifically a transcription factor – that turns on genes for survival in low oxygen . Cells quickly destroy the hydroxylated transcription factor but when oxygen is limiting , it remains unmodified . This means that , rather than being destroyed , the transcription factor binds DNA , and activates genes that keep the cells alive and growing in low oxygen . In fission yeast , an oxygenase called Ofd1 controls the activity of a transcription factor called Sre1 . Yeast requires Sre1 to grow when oxygen is limiting . Exactly how Ofd1 regulates Sre1 is unknown , but the mechanism is different from that in mammals because regulation of gene expression does not need Sre1 to be hydroxylated . Now , Clasen et al . report that Ofd1 actually hydroxylates another protein called Rps23 . This protein is one of about 80 that form the cell’s protein-building machinery , the ribosome . It turns out that , before Rps23 becomes part of the ribosome , it binds Ofd1 in a complex with other proteins . The multi-protein complex then acts to hydroxylate and transport Rps23 into the nucleus , where ribosomes are built and where the cell stores its DNA . When little oxygen is around , Ofd1 cannot hydroxylate Rps23 . This stops the complex from falling apart and traps Ofd1 away from the transcription factor Sre1 . When not bound by Ofd1 , Sre1 is free to turn on genes that allow growth at low levels of oxygen . Finally , Clasen et al . show that more unassembled Rps23 means less Ofd1 is available to inhibit Sre1 , which controls the yeast cell’s response to hypoxia . Humans have proteins similar to Ofd1 and Rps23 . As such , this pathway for sensing oxygen in yeast may occur in humans too . Further work is now needed to explore if other enzymes that hydroxylate ribosomal proteins work in a similar way .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2017	Prolyl dihydroxylation of unassembled uS12/Rps23 regulates fungal hypoxic adaptation
Parkinson’s disease is a progressive neuropathological disorder that belongs to the class of synucleinopathies , in which the protein alpha-synuclein is found at abnormally high concentrations in affected neurons . Its hallmark are intracellular inclusions called Lewy bodies and Lewy neurites . We here report the structure of cytotoxic alpha-synuclein fibrils ( residues 1–121 ) , determined by cryo-electron microscopy at a resolution of 3 . 4 Å . Two protofilaments form a polar fibril composed of staggered β-strands . The backbone of residues 38 to 95 , including the fibril core and the non-amyloid component region , are well resolved in the EM map . Residues 50–57 , containing three of the mutation sites associated with familial synucleinopathies , form the interface between the two protofilaments and contribute to fibril stability . A hydrophobic cleft at one end of the fibril may have implications for fibril elongation , and invites for the design of molecules for diagnosis and treatment of synucleinopathies . Parkinson’s disease ( PD ) is a neurodegenerative disorder characterized by the presence of Lewy bodies ( LB ) and Lewy neurites ( LN ) . Spillantini et al . ( 1997 ) identified fibrils formed by the presynaptic protein alpha-synuclein ( α-Syn , 140 residues , ~14 kD ) as the main component of these human brain inclusions ( Spillantini et al . , 1998; Spillantini et al . , 1997 ) . Certain α-Syn fibril forms can seed LB-like and LN-like inclusions in cell culture and intra-neuronal aggregation of mouse α-Syn in vivo ( Luk et al . , 2009; Thakur et al . , 2017; Volpicelli-Daley et al . , 2014 ) . In addition , abnormal α-Syn produces neuronal cell inclusions and axonal spheroids , as well as oligodendrocytic aggregates , known as glial cytoplasmic inclusions , found abundantly in Multiple System Atrophy ( MSA ) ( Arima et al . , 1998; Tu et al . , 1998 ) , which makes α-Syn fibrils an important target for the development of diagnostic tools and therapeutic strategies for PD and related synucleinopathies . Despite α-Syn fibrils , other forms of α-Syn might also be involved in neurodegeneration , such as an oligomeric α-Syn intermediate ( Danzer et al . , 2007; Lashuel et al . , 2002; Outeiro et al . , 2008; Vicente Miranda et al . , 2017; Villar-Piqué et al . , 2016; Winner et al . , 2011 ) , or the process of fibril aggregation itself ( Oueslati et al . , 2010; Reynolds et al . , 2017; Taschenberger et al . , 2012 ) . Fibrils of α-Syn show significant fibril strain polymorphism ( Peelaerts et al . , 2015 ) . Several factors point to α-Syn as an important player in the onset of PD: ( i ) six known point mutations in the α-Syn gene ( SNCA ) are associated with familial forms of synucleinopathies: A30P ( Krüger et al . , 1998 ) , E46K ( Zarranz et al . , 2004 ) , H50Q ( Appel-Cresswell et al . , 2013 ) , G51D ( Lesage et al . , 2013 ) , A53E ( Pasanen et al . , 2014 ) , and A53T ( Polymeropoulos et al . , 1997 ) ; ( ii ) animal models suggest a role of α-Syn in the etiology of PD , Dementia with Lewy Bodies ( DLB ) , and MSA ( Feany and Bender , 2000; Hashimoto et al . , 2003; Periquet et al . , 2007; Tyson et al . , 2017 ) ; ( iii ) individuals with duplications or triplications of the α-Syn gene exhibit overexpression of α-Syn and develop PD ( Ibáñez et al . , 2004; Singleton et al . , 2003 ) . Two related proteins , β-synuclein ( β-Syn ) and γ-synuclein ( γ-Syn ) , with sequence homology to α-Syn , have been described ( Clayton and George , 1998; Jakes et al . , 1994; Stefanis , 2012 ) . β-Syn and α-Syn share the greatest aminoacid sequence homology , with β-Syn lacking 12 amino acids ( residues 71 to 82 ) within the non-amyloid component region ( NAC; residues 61–95 in α-Syn ) ( Giasson et al . , 2001; Uéda et al . , 1993 ) . In synucleins , regions with the highest homologies are located in the structurally heterogeneous , amino-terminal half ( residues 10–84 in α-Syn ) composed of 5 to 6 imperfect repeats with the consensus sequence KTKEGV ( Der-Sarkissian et al . , 2003 ) . In contrast , the carboxyl terminus is highly negatively charged and unstructured ( Chen et al . , 2007; Vilar et al . , 2008 ) . A number of post-translational modifications have been described for α-Syn including phosphorylation ( Anderson et al . , 2006; Fujiwara et al . , 2002; Paleologou et al . , 2010 ) , acetylation ( Iyer et al . , 2016; Maltsev et al . , 2012 ) , ubiquitination ( Hasegawa et al . , 2002 ) , and C-terminal truncation ( Anderson et al . , 2006; Crowther et al . , 1998 ) . C-terminal truncation of α-Syn occurs normally in vivo , under physiological conditions and it has been shown to promote fibrillization ( Crowther et al . , 1998; Li et al . , 2005; Liu et al . , 2005; Wang et al . , 2016 ) . In turn , truncated forms of α-Syn play a role in inducing Lewy body formation ( Dufty et al . , 2007; Li et al . , 2005; Prasad et al . , 2012 ) , suggesting that truncation by proteolysis may be important in the pathological process . In vivo studies investigating α-Syn aggregation demonstrated that activation of the inflammasome and more specifically caspase-1 , the enzymatic component of the inflammasome , leads to the production of an α-Syn fragment truncated at aspartic acid 121 ( D121 ) ( Wang et al . , 2016 ) . This C-terminally-truncated α-Syn form ( α-Syn ( 1-121 ) ) aggregates more rapidly than full-length α-Syn ( including disease-associated mutants ) , and its production is associated with cell toxicity . Furthermore , the use of VX-765 , a pro-drug that produces a specific inhibitor of caspase-1 in vivo ( Wannamaker et al . , 2007 ) , improved survival of a neuronal cell model of PD ( Wang et al . , 2016 ) , and reduced neurodegeneration in a transgenic mouse model of MSA ( Bassil et al . , 2016 ) , suggesting an important role of α-Syn ( 1-121 ) in cellular toxicity in both , cell cultures as well as a mouse model . To this date , high resolution structures of α-Syn fibrils are limited to the results of a micro-electron diffraction ( microED ) study of two small segments of the protein ( Rodriguez et al . , 2015 ) and a solid-state NMR structure obtained from ~5 nm diameter , single protofilaments ( Tuttle et al . , 2016 ) , in addition to solid state NMR studies at the secondary structure level ( Bousset et al . , 2013; Kim et al . , 2009; Vilar et al . , 2008 ) , and X-ray diffraction studies of shorter segments of α-Syn ( Li et al . , 2014 ) , or α-Syn bound to other molecules ( De Genst et al . , 2010; Gruschus et al . , 2013; Rao et al . , 2010; Ulmer et al . , 2005; Xie et al . , 2010; Yagi-Utsumi et al . , 2015; Zhao et al . , 2011 ) . Here , we report the atomic structure of α-Syn ( 1-121 ) fibrils determined by cryo-electron microscopy ( cryo-EM ) . The structure allows conclusions about the organization of α-Syn fibrils at near-atomic resolution , suggest mechanisms for fibril formation and growth , and allows conclusions on fibril stability . Several preparations of recombinant human α-Syn fibril were screened by negative stain transmission electron microscopy ( TEM; Figure 1—figure supplement 1 ) . These included fibrils formed by full length α-Syn ( Figure 1A ) , α-Syn phosphorylated at serine 129 , N-terminally acetylated , and C-terminal truncated α-Syn comprised of residues 1–119 ( α-Syn ( 1-119 ) ) , 1–121 ( α-Syn ( 1-121 ) ) , or 1–122 ( α-Syn ( 1-122 ) ) . The diameters of the α-Syn fibrils produced varied from 5 nm to approximately 10 nm when studied by negative stain TEM . The fibrils formed by α-Syn ( 1-121 ) were straight , between 20 and 500 nm long and the only ones of consistent diameters of 10 nm ( Figure 1B , Figure 1—figure supplement 1E ) . This fibrillar form α-Syn ( 1-121 ) has been described as an aggregation-prone species resulting from α-Syn truncation by caspase-1 ( Wang et al . , 2016 ) . The recombinantly produced α-Syn ( 1-121 ) used here showed a similarly aggressive aggregation profile . Preparations of α-Syn ( 1-121 ) fibrils were quick-frozen in the holes of fenestrated carbon coated cryo-electron microscopy ( cryo-EM ) grids , and imaged with a Titan Krios 300kV cryo-EM instrument , equipped with a Quantum-LS energy filter and a K2 Summit direct electron detector . Helical image processing of recorded cryo-EM movies produced a 3D reconstruction of the α-Syn ( 1-121 ) fibril at an overall resolution of 3 . 4 Å ( Figure 1C and D , Figure 1—figure supplement 2 , Figure 2 , and Video 1 ) . Our 3D map shows that fibrils are formed by two protofilaments , each of 5 nm in diameter ( Figure 1 ) . These lack C2 symmetry , but are related by an approximate 21 screw symmetry , akin to the symmetry exhibited by the paired helical filaments of tau ( Fitzpatrick et al . , 2017 ) and by amyloid-ß ( 1-42 ) filaments ( Gremer et al . , 2017 ) . α-Syn ( 1-121 ) fibrils are therefore polar , meaning that both protofibrils are aligned into the same direction . The position of a given ß-sheet in a protofilament is produced by the rotation of 179 . 5° of one sheet around its axis ( helical twist ) , followed by a vertical translation of 2 . 45 Å ( helical rise ) . This ß-sheet arrangement results in a spacing of 4 . 9 Å between α-Syn subunits in successive rungs of a single protofilament ( Figure 1C and D ) . The quality of the EM map allowed an atomic model of the region between residues L38 and V95 to be built . Each α-Syn ( 1-121 ) molecule comprises eight in-register parallel β-strands ( i . e . residues 42–46 ( β1 ) , 48–49 ( β2 ) , 52–57 ( β3 ) , 59–66 ( β4 ) , 69–72 ( β5 ) , 77–82 ( β6 ) , 89–92 ( β7 ) , and 94- ( ~102 ) ( β8 ) ) , which are interrupted by glycine residues ( i . e . G41 before β1 , G47 between β1 and β2 , G51 between β2 and β3 , G67 and G68 between β4 and β5 , G73 between β5 and β6 , G84 and G86 between β6 and β7 , and G93 between β7 and β8 ) or an arch ( i . e . E57-K58 between β3 and β4 ) ( Figure 1A , F and G ) . The β-strands β2-β7 wind around a hydrophobic intra-molecular core composed of only alanine and valine residues and one isoleucine ( i . e . V48 , V49 , V52 , A53 , V55 , V63 , A69 , V70 , V71 , V74 , A76 , V77 , A78 , I88 , A89 , A90 , A91 ) . Considering that these hydrophobic clusters are maintained along the fibril , they are likely to contribute to the stability of the protofilament . The hydrophobic core is surrounded by two hydrophilic regions ( i . e . ( i ) : Q79 , T81 , and ( ii ) : T72 , T75 , T54 , T59 , and E61 ) both still within the core of the structure ( Figure 3 ) . While most of these side chains form so-called side chain hydrogen bond ladders ( Nelson et al . , 2005; Riek , 2017 ) , the second hydrophilic region comprising four threonine residues and a negatively charged glutamic acid side chain surrounds a tunnel filled with some ordered molecules of unknown nature , as evidenced by an additional density ( Figure 1—figure supplement 3D ) . The less well defined β1 and β8 strands are attached to the core , while the first 37 N-terminal residues and the last ~20 C-terminal residues of α-Syn ( 1-121 ) are not visible in the 3D reconstruction ( Figure 1E and Figure 1—figure supplement 2A ) , indicating a disordered structure in line with quenched hydrogen/deuterium exchange – solution-state NMR ( H/D exchange NMR ) and limited proteolysis ( Vilar et al . , 2008 ) , which showed these terminal segments to be unprotected in nature . Together with our results , this suggests that approximately 40 residues of both the N- and C-terminal ends of full-length human α-Syn are flexible , and surround the structured core of the fibril with a dense mesh of disordered tails , similar to the ‘fuzzy coat’ recently described in the cryo-EM tau structure ( Fitzpatrick et al . , 2017 ) . Two β-sheets ( one from each protofilament ) interact at the fibril core via a hydrophobic steric zipper-geometry comprised of β-strand β3 ( i . e . residues G51-A56 ) . As a consequence , two α-Syn molecules per fibril layer are stacked along the fibril axis ( Figure 2B and C ) . The side chains of residues A53 and V55 form the inter-molecular surface contributing to the interface between the two protofilaments , which is further stabilized by a surface-exposed salt bridge between E57 and H50 that might be sensitive to pH , as an unprotected histidine has a pK of ~6 . 2 ( Figure 1—figure supplement 3H ) . The same structure with a steric zipper topology was found in micro-crystals of the peptide comprising residues G47-A56 ( Rodriguez et al . , 2015 ) . Interestingly , the β-strand β6 that is sandwiched between β-strands β2/β3 and β7 is also aligned with a neighboring molecule but shifted by one monomer along the fibril axis , as shown in Figure 1G and Figure 2—figure supplement 1 . Thus , hetero and homo steric zippers are both present in the 3D structure . Of these , the homo steric zipper at the inter-molecular interface has an extensive and well-packed β-strand interface , forming a very densely packed fibril . This stacking generates an asymmetric fibril with two distinct ends . Furthermore , the hydrophobic core of the fibril is composed of β-strands that interact with each other in a half-stacked zipper topology , contrasting with the hydrophilic core comprised of β-strands β4 and β5 , which are non-stacked ( Figure 1G and Figure 2—figure supplement 1 ) . The latter confirms previous results from site-directed spin labeling experiments , which show that the region including residues 62–67 at the beginning of the NAC region , has a pronounced lack of stacking interactions ( Chen et al . , 2007 ) . The outer surface of the ordered region of the fibrils is mostly hydrophilic , with a few exceptions ( i . e . L38 , V40 , V82 , A85 , A90 , F94 , V95 ) ( Figure 3A ) . The side chain of V66 should probably not be classified as surface exposed because of its interaction with β-strand β8 ( Figure 1—figure supplement 2A ) . If we ignore the influence of the non-polar alanine residues due to the small size of their side chains , the surface of the fibrils has two highly hydrophobic regions formed by residues L38 and V40 , and by residues F94 and V95 . Other interesting properties of the surface are the salt bridge formed by the side chains of E46 and K80 ( Figure 1—figure supplement 3G ) and the rather highly positive clustering of K43 , K45 , K58 , H50 that requests the binding of a counter-ion , as it is supported by an observed density ( Figure 1—figure supplement 3C ) . Six familial mutations in α-Syn are known to be associated with PD and other synucleinopathies ( i . e . A30P , E46K , H50Q , G51D , A53E , and A53T ) . Of these , all but A30P are located in the heart of the core of the fibril structure presented here ( Figure 1A and E ) . E46 forms a salt bridge with K80 ( Figure 1—figure supplement 3G ) . The mutation of the glutamic acid E46 to a positively charged lysine in an E46K mutant would thus induce a charge repulsion between β-strands β1 and β6 , likely destabilizing this α-Syn fibril structure ( Tuttle et al . , 2016 ) . The familial PD/DLB-causing mutation E46K was found to enhance phosphorylation in mice ( Mbefo et al . , 2015 ) , and its toxic effect was increased by the triple-K mutation ( E35K , E46K , E61K ) in neuronal cells ( Dettmer et al . , 2017 ) . Previous high-resolution structures of α-Syn only included small peptides or single protofilaments ( Rodriguez et al . , 2015; Tuttle et al . , 2016 ) . Our 3D map suggests structural contributions of some familial mutations to fibril stability , since H50 , G51 and A53 are all involved in the inter-molecular contact between the two β-sheets from adjacent protofilaments at the core of the here studied α-Syn ( 1-121 ) fibrils . Mutation of the positively charged histidine 50 into a polar , uncharged glutamine in the H50Q mutant would likely interfere with the salt bridge established between residues E57 and H50 ( Figure 1—figure supplement 3H ) . Adding to the absent side-chain of glycine 51 a negatively charged aspartic acid in mutant G51D , or transforming the small side-chain of alanine A53 into a larger threonine in mutant A53T , would likely disrupt the steric zipper interaction between the two protofibrils , whereby the A53T mutation would in addition change the highly hydrophobic surface at the zipper to partly hydrophilic one . In our α-Syn ( 1-121 ) fibril structure , A53 is part of a hydrophobic pocket that defines the interaction of protofilaments and likely contributes to fibril stability as the hydrophobic interactions exist along the fibril axis . Mutations at the core of this α-Syn fibril would compromise the formation of the structure presented here . This suggests that a different fibril structure ( i . e . fibril strain ) could be formed from α-Syn containing the above discussed familial PD mutations . Several features of our structure , such as non-functional hydrophobic surface patches ( Figure 3 ) , a hydrophilic tunnel ( Figure 1—figure supplement 3D ) , and a positively charged side chain arrangement like the one comprised of residues K43 , K45 , K58 , H50 ( Figure 1—figure supplement 3C ) are not found in functional amyloid structures such as that of HET-s ( Wasmer et al . , 2008 ) . However , similar structural characteristics have been previously observed for pathological tau filaments obtained from Alzheimer’s disease brains where ( i ) : lysine and tyrosine residues play a similarly stabilizing role in the interface region of two protofilaments of the straight filaments ( SF ) , and ( ii ) : the area in the center of the protofilaments is dominated by hydrophilic residues ( Fitzpatrick et al . , 2017 ) . It is plausible that these structural features might arise because folding to form the amyloid fibril structure is dictated by the need to bury the maximum number of hydrophobic side-chains as efficiently as possible , as is also the case for the Aβ ( 1-42 ) amyloid fibrils ( Gremer et al . , 2017 ) . The artificial , highly toxic , but not synucleinopathy-related mutant E57K ( Winner et al . , 2011 ) is interesting to mention in the context of the 3D structure presented , because E57 is also at the inter-molecular interface ( Figure 2 ) . The presence of a positive lysine side chain at this position in the E57K mutant would significantly interfere with the formation of the interface and even the amyloid fibril ( Winner et al . , 2011 ) . Indeed , this mutant was designed in a successful structure-based attempt to interfere with amyloid fibril formation ( at least under some conditions ) ( Winner et al . , 2011 ) . Furthermore , both in a lentivirus-rat system as well as in a transgenic mouse model , the E57K mutant formed a significant amount of oligomers and was highly toxic , resulting in a large decay of TH-sensitive neurons in the substantia nigra of rats and a motor phenotype reminiscent of PD in mice ( Winner et al . , 2011 ) . Thus , the artificial mutant E57K can be regarded as a ‘familial PD-like’ mutation both from the in vivo and from the structure/mechanism-based point of view . Full-length α-Syn subunits in a fibril studied by NMR ( [Tuttle et al . , 2016] , PDB 2N0A ) were found to be in a roughly similar secondary structure arrangement as in the here reported structure of α-Syn ( 1-121 ) ( Figure 4A ) , even though the primary structure and the side-chain interactions of our here reported structure are very different from the NMR structure . Most importantly , the fibrils used for the NMR study were only approximately 5 nm wide , which corresponds to the diameter of a single protofilament . The larger diameter of our fibrils , 10 nm , results from the interaction between two protofilaments , which allowed us to hypothesize on the nature of α-Syn ( 1-121 ) protofilament interactions . Fibrils of 5 to 10 nm in diameter found in substantia nigra samples from the brain of PD patients , ( Crowther et al . , 2000 ) , cingulate cortex of patients with DLB ( Spillantini et al . , 1998 ) , cerebral cortex of PD patients ( Kosaka et al . , 1976 ) , and in-vitro aggregated samples ( Bousset et al . , 2013 ) . Crowther et al . ( 2000 ) had already suggested that the 10 nm filaments are the result of the interaction between 5 nm protofilaments . An important difference between our here reported structure and the NMR structure reported by Tuttle et al . ( 2016 ) is the orientation of residue A53 . The mutation A53T is associated with early onset PD . In our structure , residue A53 faces the interface between the two protofibrils and thereby likely contributes to fibril stability . In contrast , Tuttle et al . ( 2016 ) reported in their NMR structure A53 to point towards the hydrophobic core of the one observed individual protofilament , which may explain the lack of 10 nm fibrils in their sample . However , it is also noted here that the NMR study by Tuttle et al . ( 2016 ) showed a significant disagreement among the ten lowest-energy NMR structures for residues 51–67 [Figure 3d in Tuttle et al . ( 2016 ) ] , indicating a lower confidence for those residues in the NMR structure . Our here reported cryo-EM map has the side-chains for those residues pointing into the opposite direction as reported in the Tuttle et al . ( 2016 ) structure . Our structure includes a serine residue at position 87 ( Figure 1—figure supplement 3E ) , which is one of the several phosphorylation sites in α-Syn , in addition to Y125 , S129 , Y133 and Y135 ( Oueslati et al . , 2012; Paleologou et al . , 2010 ) . S87 is the only phosphorylation site located within the NAC region . The previous solid-state NMR structure of α-Syn placed the side chain of this residue towards the inside of the protofilament core , leading to the assumption that phosphorylation of S87 might be the only modification occurring at a region not accessible in the fibrillar state . However , in our cryo-EM structure , S87 faces the outside of the fibril and hence remains accessible for disease-associated modification in α-Syn fibrils . We also observed the arrangement of G47 and A78 described by Tuttle et al . ( 2016 ) , which was proposed to favor the interaction between residues E46 and K80 and allow them to form a stable salt bridge between two consecutive α-Syn monomers ( Figure 1—figure supplement 3G ) . The conservation of the geometry adopted by these residues confirms their role in facilitating backbone-backbone interactions . In addition , our structure also confirms that residues A69 and G93 ( and likely G68 ) help to stabilize the distal loop in a protofilament ( Figure 1—figure supplement 3F ) . A microED structure obtained from crystals produced from a 10-residue peptide simulating the core of α-Syn fibrils ( PreNAC , from 47 to 56; Figure 4B ) and including a threonine instead of an alanine at position 53 ( i . e . A53T ) , also proposed that residue 53 forms the hydrophobic core within a protofilament ( Rodriguez et al . , 2015 ) . In addition , the microED model suggested that the interaction between adjacent protofilaments would occur through residues 68 to 78 ( referred to as NACore ) ( Rodriguez et al . , 2015 ) . However , their short peptides did not include most residues responsible for the α-Syn monomer topology that we observed . Instead , our cryo-EM structure reveals that the PreNAC is responsible for the interaction between protofilaments , and places the NACore at the very center ( i . e . the core ) of a single protofilament . Our 3D structure allows us to hypothesize a mechanism for fibril elongation ( fibril growth ) . Because two different stacking modes are present ( i . e . the half-stack at the intermolecular interface and the stacking of β-strand β6 ) , the two ends of the fibrils are distinct , suggesting an end-dependent growth of the fibrils , as documented and also suggested for other amyloids ( Lührs et al . , 2005 ) . One end of the fibril includes a hydrophobic cleft formed between β-strands β2/ β3 on one side and β7 on the other side ( residues V49 , V52 , A88 , I89 ) , providing a hydrophobic entry point for the next incoming molecule , with the matching segment consisting of 5 hydrophobic residues ( V74-V82 , Figure 5 ) . This suggests that the initial binding event of fibril elongation might be a hydrophobic interaction involving residues V74-V82 . This peptide segment is the central part of the NAC region and strong experimental evidence suggests that it is critical for fibril formation ( Giasson et al . , 2001 ) . In addition , it has been shown that β-synuclein , which lacks residues V74 to V82 , is incapable of forming fibrils ( Giasson et al . , 2001 ) . It is intriguing to speculate that a small molecule binding into this hydrophobic cleft could be a potent fibril elongation inhibitor or tracer , with the potential to be applied in PD and other synucleinopathies . Finally , the inter-molecular stacking may also play a role in fibril elongation , since the zipper interaction is of hydrophobic nature . Furthermore , it is likely that fibril growth alternates between the two protofilament structures at the level of monomer addition . Failure thereof may result in the growth of a single protofilament with little stability , yielding a dynamic on- and off-binding of monomers and larger oligomers , which has been observed for other amyloid fibril systems ( Carulla et al . , 2005 ) . In conclusion , we present the structure of recombinant α-Syn ( 1-121 ) fibrils determined at a resolution of 3 . 4 Å by cryo-EM . Our structure encompasses nearly the complete protein ( residues 38 to 95 ) , and includes the NAC region ( residues 61 to 95 ) of α-Syn . We determined that various residues associated with familial forms of PD and other synucleinopathies are located in the interacting region between two protofilaments , suggesting their involvement in fibril formation and stabilization . The cryo-EM structure presented here reveals how two protofilaments interact to form a fibril , and how the NAC region contributes to protofilament formation and stability . Our structure also presents novel insights into how several PD-relevant mutations of α-Syn would compromise the structure of this fibril , suggesting that in the case of certain familial forms of PD , a different structure of α-Syn than this fibril strain might be involved . Our findings on protofilament interaction and our hypothesis on the mechanism of fibril elongation invite for the design of molecules for diagnostics or treatment of synucleinopathies . Recombinant full-length α-Syn was expressed from the pRT21 expression vector in BL21 ( DE3 ) competent Escherichia coli ( E . coli ) . For N-terminal acetylation of α-Syn , cells were pre-transfected by pNatB vector coding for the N-terminal acetylase complex ( plasmid kindly provided by Daniel Mulvihill , School of Biosciences , University of Kent , Canterbury , UK ) ( Johnson et al . , 2010 ) . C-terminally truncated forms of α-Syn ( 1-119 ) , α-Syn ( 1-121 ) , and α-Syn ( 1-122 ) were expressed in BL21-DE3-pLysS competent E . coli . Purification of α-Syn strains was performed by periplasmic lysis , ion exchange chromatography , ammonium sulfate precipitation , and gel filtration chromatography as previously described ( Huang et al . , 2005; Luk et al . , 2009 ) . Polo like kinase 2 ( PLK2 ) was expressed in BL21-DE3-pLysS competent E . coli , isolated via its His-tag and immediately used to phosphorylate purified α-Syn . This was followed by standard ion exchange and gel filtration chromatography to separate phosphorylated from non-phosphorylated α-Syn . Endotoxins were removed from all α-Syn strains by Detoxi-Gel Endotoxin Removing Gel ( Thermo Scientific ) usually in one run or until endotoxin levels were below detection level . The sequence of the expressed α-Syn strains was verified by tryptic digestion followed by MALDI mass spectrometry ( MS ) or HPLC/ESI tandem MS for total mass was performed . Purity and monodispersity was determined by Coomassie blue or Silver staining of the SDS PAGE gel and analytical ultracentrifugation and the concentration was determined by the bicinchoninic acid ( BCA ) assay ( Thermo Scientific ) with bovine serum albumin as a standard . Dialyzed and lyophilized α-Syn ( 1-121 ) was prepared by dialyzing the purified protein in a 2 kD Slide-A-Lyzer unit ( Thermo Scientific , for max . 3 ml ) against HPLC-water ( VWR ) . 500 µg protein aliquots were pipetted into 1 . 5 ml tubes , frozen on dry ice , and lyophilized for 2 hr using an Eppendorf concentrator ( Eppendorf ) . Lyophilized samples were stored at −80°C until use . Fibrils were prepared by dissolving dialyzed and lyophilized , recombinant α-Syn protein at 5 mg/mL in incubation buffer ( DPBS , Gibco; 2 . 66 mM KCL , 1 . 47 mM KH2PO4 , 137 . 93 mM NaCl , 8 . 06 mM Na2HPO4-7H2O pH 7 . 0–7 . 3 ) . Reactions of 200 µL per tube were incubated at 37°C with constant agitation ( 1 , 000 rpm ) in an orbital mixer ( Eppendorf ) . Reactions were stopped after 5 days , sonicated ( 5 min in a Branson 2510 water bath ) , aliquoted , and stored at −80°C until use . The presence of amyloid fibrils was confirmed by thioflavin T fluorimetry and high molecular weight assemblies were visualized by gel electrophoresis . Cryo-EM grids were prepared using a Vitrobot Mark IV ( ThermoFisher Scientific ) with 95% humidity at 4°C . Amyloid fibrils ( 3 µL aliquots ) were applied onto glow-discharged , 300 mesh , copper Quantifoil grids . After blotting , grids were plunge frozen in liquid ethane cooled by liquid nitrogen . Samples were imaged on a Titan Krios ( ThermoFisher Scientific ) transmission electron microscope , operated at 300 kV and equipped with a Gatan Quantum-LS imaging energy filter ( GIF , 20 eV energy loss window; Gatan Inc . ) . Images were acquired on a K2 Summit electron counting direct detection camera ( Gatan Inc . ) in dose fractionation mode ( 50 frames ) using the Serial EM software ( Mastronarde , 2005 ) at a magnification of 165 , 000× ( physical pixel size 0 . 831 Å ) and a total dose of ~69 electrons per square angstrom ( e-/Å2 ) for each micrograph . Micrographs were drift-corrected and dose-weighted using MotionCor2 ( Zheng et al . , 2017 ) through the Focus interface ( Biyani et al . , 2017 ) . Additional data collection parameters are detailed in Table 1 . Helical reconstruction was carried out with the RELION 2 . 1 software ( Scheres , 2012 ) , using methods described in He and Scheres ( 2017 ) . Filaments were manually selected using the helix picker in RELION 2 . 1 . Filament segments were extracted using a box size of 280 pixels ( 233 Å ) and an inter-box distance of 28 pixels . A total of 18 , 860 segments were extracted from 792 fibrils manually picked from 118 micrographs ( Table 1 ) . 2D classification was carried out with a regularization value of T = 10 , and 2D class averages with a clear separation of β-strands were selected for further data processing . Power spectra of 2D class averages show the layer line at 1/ ( 4 . 9 Å ) with peak intensities on both sides of the meridian ( Bessel order n = 1 ) . This is the result of an approximate 21 screw symmetry between α-Syn subunits on the two protofilaments ( Figure 1—figure supplement 2 ) . Segments assigned to the best 2D classes were used for 3D classification using a regularization value of T = 8 and with optimization of the helical twist and rise . For both 3D classification and refinement , a helical_z_percentage parameter of 10% was used , which defines the size of the central part of the intermediate asymmetrical reconstruction that is used to apply real-space helical symmetry ( He and Scheres , 2017 ) . An initial reconstruction was calculated using a cylinder generated via the helix toolbox in RELION 2 . 1 as initial model . This reconstruction was low-pass filtered to 60 Å and employed as the initial model for a 3D classification with a single class ( K = 1 ) and T = 20 , an approach that allowed the successful reconstruction of amyloid filaments ( Fitzpatrick et al . , 2017 ) . The handedness of the reconstruction was determined by comparison with atomic force microscopy images , which showed left-coiled surface patterns for the fibrils . Refinement was carried out by the auto-refine procedure with optimization of helical twist and rise . This resulted in a structure with overall resolution of 3 . 8 Å . Post-processing with a soft-edge mask and an estimated map sharpening B-factor of −82 . 6 Å gave a map with a resolution of 3 . 4 Å ( by the FSC 0 . 143 criterion ) . An estimation of local resolution was obtained using RELION 2 . 1 and a local-resolution-filtered map was calculated for model building and refinement . A model of the α-Syn ( 1-121 ) fibril was built into the Relion local resolution-filtered map using COOT ( Emsley and Cowtan , 2004 ) , with the PDB ID 2N0A as an initial model for the early interpretation of the map . The structure helped to determine the directionality of the protein chain and facilitated the assignment of densities in the map to specific residues . However , due to the large differences between the NMR structure and our EM map , major rebuilding was necessary . The high quality of the EM map allowed us to unambiguously build residues 38–95 . A comparison was also carried out between our structure and X-ray structures of α-Syn fragments 69–77 ( PDB ID 4RIK ) , 68–78 ( PDB ID 4RIL ) and 47–56 ( PDB ID 4ZNN; with the mutation A53T ) . The structure ( 10 monomers , 5 on each protofilament ) was refined against the RELION local resolution-filtered map with PHENIX real space refine ( Afonine et al . , 2013 ) . Rotamer , Ramachandran restraints , and ‘NCS’ constraints were imposed , and two B-factors per residue were used during refinement . For validation , we randomized the coordinates ( with a mean shift of 0 . 3 Å ) and refined ( using the same settings ) against one of the refinement half-maps ( half-map 1 ) . We then calculated the FSC between that model ( after refinement against half-map 1 ) and half-map 1 , as well as the FSC between the same model and half-map 2 ( against which it was not refined ) . The lack of large discrepancies between both FSC curves indicates no overfitting took place .	People with Parkinson’s disease have damaged cells in a part of the brain involved in movement , learning and reward-seeking behaviors . These cells contain blob-like aggregates that contain abnormally high amounts of a protein called alpha-synuclein . It is generally believed that , within these blobs , this protein clusters together into small needles called fibrils . Discerning the structure of a fibril could help researchers to understand both how alpha-synuclein damages brain cells and how diseases like Parkinson’s spread . Biophysicists have attempted to reveal the fibril structure previously . But many of these efforts only looked at short segments of the alpha-synuclein protein . Researchers still need more detailed imagery of the fibrils to confirm previous findings regarding their architecture and ultimately to identify ways to counteract the damage they cause . Guerrero-Ferreira et al . used a technique called cryo-electron microscopy to capture images of frozen fibrils made from a version of human alpha-synuclein that readily aggregates and that is only slightly shorter than the full-length protein . Processing these high-resolution images with computer software then revealed a three-dimensional model of the fibril structure , in which fine details are clearly visible . In the fibril , the proteins cluster to form a helix , similar to a flight of stairs . Each turn of the helix is formed by two alpha-synuclein molecules , facing each other but rotated by almost 180 degrees from one another . The three-dimensional model displays which parts of the protein lie at the core of the helix and thereby stabilize the fibril structure . Guerrero-Ferreira et al . speculate that fibrils may also take alternative forms because common alpha-synuclein mutations , which correlate with disease , would destabilize the observed helical structure . In the future , researchers may be able to use the features of this three-dimensional model to help design molecules that would make the fibrils detectable via medical imaging . This could help doctors to diagnose people with Parkinson’s disease at an earlier stage . Further research is also needed to understand where and how fibrils form , if differences in fibril structures exist within or between patients , possibly leading to different sub-classes of the disease , and how such fibrils interact with and possibly damage human brain cells .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "neuroscience" ]	2018	Cryo-EM structure of alpha-synuclein fibrils
Amyloid plaques , consisting of deposited beta-amyloid ( Aβ ) , are a neuropathological hallmark of Alzheimer’s Disease ( AD ) . Cerebral vessels play a major role in AD , as Aβ is cleared from the brain by pathways involving the cerebrovasculature , most AD patients have cerebrovascular amyloid ( cerebral amyloid angiopathy ( CAA ) , and cardiovascular risk factors increase dementia risk . Here we present a notable advance in vascular tissue engineering by generating the first functional 3-dimensioinal model of CAA in bioengineered human vessels . We show that lipoproteins including brain ( apoE ) and circulating ( high-density lipoprotein , HDL ) synergize to facilitate Aβ transport across bioengineered human cerebral vessels . These lipoproteins facilitate Aβ42 transport more efficiently than Aβ40 , consistent with Aβ40 being the primary species that accumulates in CAA . Moreover , apoE4 is less effective than apoE2 in promoting Aβ transport , also consistent with the well-established role of apoE4 in Aβ deposition in AD . Alzheimer’s Disease ( AD ) is the leading cause of senile dementia with over 44 million affected persons and an economic burden of over $600 billion ( Mayeux and Stern , 2012 ) . In addition to AD’s neuropathological hallmarks of amyloid plaques consisting of deposited Aβ peptides and neurofibrillary tangles consisting of hyperphosphorylated tau proteins , 60–90% of AD brains have evidence of cerebral amyloid angiopathy ( CAA ) , cerebral small vessel disease and microvascular degeneration ( Attems and Jellinger , 2014 ) . Apolipoprotein ( apo ) E , which in the brain is secreted primarily from astrocytes , is the principal lipid carrier within the central nervous system ( CNS ) and , in humans , exists as three isoforms , namely APOE2 , APOE3 , and APOE4 ( Lane-Donovan and Herz , 2017 ) . ApoE has well-established effects on Aβ metabolism with apoE4 being detrimental , apoE3 neutral and apoE2 protective . ApoE is also hypothesized to contribute to cerebrovascular dysfunction ( Zlokovic , 2013 ) . As the major routes by which Aβ is cleared from the brain involve the cerebrovasculature ( Ueno et al . , 2014 ) , understanding the vascular contributions to dementia is of great interest ( Snyder , 2015 ) . Cardiovascular risk factors including type two diabetes mellitus ( T2DM ) , hypertension , hypercholesterolemia , obesity and stroke increase AD risk ( Duron and Hanon , 2008 ) , yet the mechanisms by which cardiovascular health impacts brain function remain poorly understood . Interestingly , epidemiological studies suggest that AD risk may be attenuated by high levels of circulating high-density lipoprotein cholesterol ( HDL-C ) , which is also highly associated with reduced cardiovascular disease ( CVD ) risk ( Zuliani et al . , 2010 ) . Specifically , levels of apoA-I , the major HDL-associated protein , positively correlate with Mini-Mental State Examination ( MMSE ) and Cognitive Ability Screening Instrument ( CASI ) scores ( Merched et al . , 2000 ) and high serum HDL-C levels ( >55 mg/dl ) in cognitively normal elderly individuals is associated with significantly reduced risk ( HR 0 . 4 ) of AD even after adjusting for APOE genotype and vascular risk factors including obesity and T2DM ( Reitz et al . , 2010 ) . In symptomatic AD patients , plasma apoA-I levels negatively correlate with hippocampal and whole brain volume as well as mean entorhinal cortical thickness ( Hye et al . , 2014 ) , and decreased levels of serum apoA-I can discriminate AD from non-demented age-matched control subjects ( Shih et al . , 2014 ) . As HDL and apoE have several potent vasoprotective functions including reducing inflammation , increasing vascular tone through promoting endothelial nitric oxide ( NO ) synthase activity , and suppressing vascular adhesion molecule expression ( Stukas et al . , 2014b; Sacre et al . , 2003 ) , an important goal is to understand how plasma-derived circulating ( i . e . HDL ) and brain-derived ( i . e . apoE ) lipoproteins might affect Aβ metabolism in cerebral vessels . However , lipoprotein metabolism in mice and humans are substantially different , as the major circulating lipoprotein in mice is HDL whereas low-density lipoprotein ( LDL ) , which increases CVD risk , is the major circulating lipoprotein in humans ( Yin et al . , 2012 ) . AD animal model studies may therefore not always take into account the inherent vascular resilience of mice compared to humans , and thus may have limits to their translational relevance . Additionally , static in vitro models of human ECs cultured with or without astrocytes do not replicate the complex cellular interactions and extracellular matrix of a native vessel . To address some of the key limitations of existing experimental models to investigate the role of lipoproteins on Aβ metabolism at the vessel , we generated three-dimensional ( 3D ) bioengineered human vessels using a scaffold-directed dynamic pulsatile flow bioreactor system , where primary human ECs and smooth muscle cells ( SMC ) were cultivated in the absence or presence of human astrocytes to generate bipartite or tripartite vessels , respectively , which display the histological features of native peripheral and cerebral arteries . Here we demonstrate the utility of this novel experimental platform to investigate how human lipoproteins affect Aβ transport through and accumulation within physiologically relevant bioengineered human vessels . Bipartite bioengineered vessels were fabricated by sequentially seeding primary human myofibroblasts and ECs isolated from umbilical cords into a tubular woven scaffold consisting of polyglycolic acid ( PGA ) , polycaprolactone ( PCL ) and polylactate ( PLA ) , measuring 15 mm long and 2 mm in diameter ( Figure 1 ) . After 4 weeks in culture , Haematoxylin-Eosin staining demonstrated the formation of a dense and homogenous tissue on the luminal side of the scaffold ( Figure 2a ) . The tissue was composed of cells and extracellular matrix , as demonstrated by the presence of collagen by Picrosirius staining ( Figure 2b ) , and confirmed by immunohistochemical detection of collagen IV and laminin in the extracellular matrix ( Figure 2c–d ) . Immunohistochemical staining also demonstrated multiple layers of α-smooth muscle actin ( α-SMA ) positive cells on the inner side of the scaffold and a monolayer of CD31 positive ECs lining the bioengineered vascular lumen ( Figure 2e–f ) . Integrity of the endothelial barrier was functionally assessed by injecting Evans blue dye into the bioreactor circulation loop . As expected , Evans blue penetrated into tissue prepared without EC , whereas it was excluded two weeks after EC seeding , demonstrating a functionally tight endothelial barrier ( Figure 2g ) . In addition to cord cells , bioengineered vessels could also be fabricated with brain derived SMC ( hBSMC ) and microvascular cortical EC ( hBMEC ) . Immunohistological staining confirmed that brain-derived and cord-derived primary human cells form similar structures in bioengineered vessels ( Figure 2—figure supplement 1a ) , and function similarly with respect to tissue Aβ accumulation and transport ( Figure 2—figure supplement 1b–e ) . Accumulation of Aβ within cerebral vessel walls , known as CAA , is a common pathological feature in AD ( Attems and Jellinger , 2014 ) . To determine if CAA can form in our bioengineered vascular model , we injected monomeric Aβ40 or Aβ42 on the anteluminal side of the vessel to mimic native conditions where Aβ is predominantly produced by neurons . Both ELISA ( Figure 3a , b ) and 6E10 immunostaining ( Figure 3c–f ) confirmed dose-dependent retention of Aβ40 and Aβ42 within the bioengineered vascular wall 48 hr after injection ( white ) , which could be distinguished from autofluorescence by residual scaffold material ( shaded in blue ) . As Aβ fibrillization within the vessel wall is an important feature of CAA , we also stained vessels using Thioflavin-S ( Thio-S ) , and confirmed dose-dependent fibrillization in the bioengineered vessels ( Figure 3e–f ) . We further characterized the time course of Aβ accumulation in bioengineered vessels after anteluminal injection of 1 μM monomeric Aβ40 or Aβ42 . Both ELISA quantification and 6E10 immunostaining revealed Aβ deposition by 2 hr after injection , after which Aβ levels remained stable for up to 72 hr ( Figure 3g–h ) . Interestingly , quantification of beta-sheet formation in bioengineered tissue lysates with Thioflavin-T also revealed increasing signal over time ( Figure 3i–j ) , biochemically confirming increased Aβ fibrillization into beta-sheet structures within bioengineered vessels after seeding of Aβ monomers . Finally , extracellular deposition of amyloid was confirmed by immunofluorescent staining for collagen IV , α-SMA , and Aβ . Specifically , confocal microscopy of immunofluorescent staining against amyloid fibrils using the OC antibody demonstrated that fibrillary Aβ accumulates outside of the cells ( Figure 3—figure supplement 1a–b ) , whereas 6E10 staining revealed both extracellular accumulation and intracellular vesicular deposition ( Figure 3—figure supplement 1b–f ) . Accumulation of Aβ was quantified in tissues engineered from brain- or cord- derived cells , and although deposition of Aβ40 and Aβ42 tended to be higher in tissues engineered from brain cells , the differences were not significant ( Figure 2—figure supplement 1b–c ) . These data provide compelling support that 3D bioengineered human vessels can be used as an in vitro model of CAA . As a major route of Aβ egress from the brain is direct transport across the cerebral vessel into the circulation ( Ueno et al . , 2014 ) , we next evaluated the suitability of our bioengineered vessels to analyze ‘brain-to-blood’ Aβ transport by injecting Aβ in the anteluminal tissue chamber compartment of bipartite vessels or scaffold-only controls and measuring the level of Aβ recovered in the circulating medium over 4 hr . ELISA quantification revealed that both Aβ40 and Aβ42 were transported at a slower rate in bioengineered vessels , whereas Aβ freely diffused across scaffold-only controls ( Figure 3k–l ) . These data demonstrate the feasibility of bioengineered human vessels to study Aβ recovery in the circulation . Transport of Aβ was also analysed in tissues engineered from brain- or cord- derived cells and demonstrated no significant differences ( Figure 2—figure supplement 1d–e ) . Due to the greater ease of obtaining the quantities of primary human cells required to generate over 300 independent tissues used in this study , the remaining experiments were performed with vessels engineered from cord-derived EC and SMC , as they are structurally and functionally equivalent to vessels engineered with brain-derived cells . Genetic variation in APOE represents the most common genetic risk factor for AD , with APOE4 being detrimental and APOE2 protective ( Zlokovic , 2013 ) . To test whether apoE regulates vascular Aβ accumulation or transport into the circulation in our engineered bipartite vessels , we injected Aβ40 or Aβ42 monomers in the absence or presence of recombinant apoE of different isoforms into the tissue chamber at a molar ratio 25:1 to mimic the relative concentrations in brain cerebrospinal fluid ( CSF ) ( Deane et al . , 2008 ) . Although neither apoE isoform significantly altered the rate of recovery of Aβ into the circulation medium over 4 hr ( Figure 4a–b ) , we observed a significant and selective decrease of the amount of Aβ42 deposited into the bioengineered tissue after 24 hr with apoE2 co-injection ( Figure 4c–d ) . As several epidemiological studies associate circulating HDL levels with reduced AD risk ( Zuliani et al . , 2010 ) , and we recently demonstrated that a single intravenous injection of reconstituted HDL ( rHDL ) acutely lowers soluble brain Aβ levels in APP/PS1 mice ( Robert et al . , 2016 ) , we reasoned that circulating HDL might promote Aβ recovery into the circulation and reduce its accumulation in bioengineered vessels . To test this hypothesis , 200 μg/ml of HDL isolated from normolipidemic young donors were perfused through bioengineered vessels immediately after injecting Aβ into the anteluminal space . Over 4 hr , the levels of Aβ42 and Aβ42 recovered in the circulation medium were slightly increased in the presence of HDL , but this did not reach significance ( Figure 4e , f ) . After 24 hr , we observed a strong trend toward decreased tissue levels of accumulated Aβ40 and significantly lower accumulated Aβ42 in the presence of HDL ( Figure 4g–h ) . We then tested for a functional interaction between apoE and HDL by analyzing Aβ transport and tissue accumulation after injecting either beneficial recombinant apoE2 or detrimental recombinant apoE4 into the anteluminal tissue chamber , in the presence or absence of HDL injected into the circulating medium . Importantly , the combination of anteluminal apoE2 and circulating luminal HDL led to significantly increased Aβ40 and Aβ42 transport over 4 hr compared to either Aβ alone or Aβ with apoE2 or HDL alone ( Figure 5a–b ) . Consistent with our previous observations at 24 hr , the levels of Aβ40 accumulated in the tissue were not significantly affected by apoE2 , HDL , or both apoE and HDL ( Figure 5c ) , however , the levels of accumulated Aβ42 were significantly reduced by apoE2 , HDL , and the combination of both apoE2 and HDL ( Figure 5d ) . Interestingly , the combination of anteluminal apoE4 and circulating luminal HDL significantly increased both Aβ40 and Aβ42 transport over 4 hr compared to either Aβ alone or Aβ with apoE4 ( Figure 5e–f ) , and concomitantly the level of Aβ42 accumulated in the tissue at 24 hr was significantly lower in the presence of both apoE4 and HDL compared to Aβ42 alone or Aβ42 and apoE4 ( Figure 5h ) . These results strongly support a cooperative role between brain apoE and circulating HDL to preferentially clear Aβ across the vasculature , and suggest that one beneficial role of circulating HDL is to functionally counteract against apoE4 in this Aβ transport assay . Because bipartite vessels consisting of ECs and SMCs have the anatomy of peripheral rather than cerebral vessels , we extended the translational relevance of our bioengineered vessels by incorporating human primary astrocytes on the antelumen ( Figure 1b ) to mimic cerebral vessels . As the diameter of our vessels is approximately 2 mm prior to cell seeding and further contains SMC rather than pericytes , our bioengineered tripartite vessels were specifically designed to resemble larger human leptomeningeal and penetrating arteries rather than the cerebral microvasculature . Tripartite bioengineered vessels are therefore composed of EC lining the lumen , several layers of SMC , and a layer of astrocytes in the antelumen as confirmed by immunofluorescent staining against CD31 , α-SMC actin , and GFAP , respectively ( Figure 6a–c ) . Interestingly , higher magnification images revealed GFAP positive protrusions penetrating into the tissue ( Figure 6d ) , suggestive of astrocyte endfeet . Immunofluorescent staining confirmed the expression of aquaporin four and NDRG2 by astrocytes in bioengineered vessels ( Figure 6—figure supplement 1a–b ) . Importantly , we observed that peripherally-derived ECs ( HUVEC ) , when cultured in the presence of astrocytes and under flow conditions , expressed high level of the tight-junction proteins ZO-1 and ZO-2 ( Figure 6e–f ) as well as the blood brain barrier ( BBB ) tight junction protein Claudin 5 ( Figure 6g ) and the specific BBB transporter Glut-1 ( Figure 6h ) , demonstrating that the EC phenotype of a HUVEC is reprogrammed to become brain-like in the presence of astrocytes . Comparison of Glut-1 expression between bipartite , tripartite , umbilical cord and brain tissues confirmed reprograming of HUVEC in engineered vessels , with expression levels comparable to native brain vessels , whereas native umbilical cords lack detectable Glut-1 ( Figure 6—figure supplement 2 ) . The integrity of the endothelial barrier in tripartite vessels was functionally assessed by injecting Evans blue into the circulation loop and compared to bipartite vessels . As expected , Evans blue was excluded in both tripartite and bipartite vessels ( Figure 6j ) . Barrier integrity was further assessed by measuring permeability of 4 kDa or 40 kDa FITC-dextran . Tripartite tissues demonstrated less extravasation of both 4 kDa and 40 kDa than bipartite tissues , although these were not significantly different from each other , whereas , as expected , both tissues are significantly less permeable than unseeded scaffold ( Figure 6—figure supplement 3 ) . The functionality of EC and astrocytes within tripartite bioengineered vessels was evaluated by measuring nitric oxide ( NO ) produced by EC and native apoE secretion by astrocytes , which were genotyped as APOE3/E3 . To determine NO production , vessels were incubated with either 10 nM of acetylcholine ( Ach ) or 200 μg/ml of HDL for 60 min before measuring conversion of L-3H-arginine to L-3H-citruline . A significant increase was observed after both treatments , whereas , in the presence of the specific NO synthase inhibitor L-NG-nitroarginine methyl ester ( L-NAME ) , the conversion was blocked ( Figure 6k ) . For apoE secretion , the brain-penetrant Liver-X-Receptor ( LXR ) agonist GW3965 ( 1 μM ) was circulated through the lumen of the vessels 72 hr before collecting medium . ELISA quantification demonstrated that GW3965 significantly stimulated the secretion of native apoE in the tripartite vessels ( Figure 6l ) . These data confirmed that our tripartite bioengineered vascular tissue has both structural and functional characteristics of native cerebral arteries . Of note , the limited availability of primary human astrocytes with distinct APOE genotypes restricted our subsequent experiments in tripartite vessels to APOE3/E3 . As the combination of recombinant apoE injected into the anteluminal tissue chamber and HDL injected into the circulating medium significantly increased Aβ40 and Aβ42 transport and reduced Aβ42 tissue accumulation in bipartite vessels ( Figures 4 and 5 ) , we tested for functional synergy of these lipoproteins in tripartite cerebral vessels by treating the vessels with GW3965 for 72 hr to stimulate native apoE secretion from astrocytes and perfusing HDL through the lumen immediately after Aβ injection . Consistent with our observations in bipartite vessels , neither GW3965 alone nor HDL alone modified the rate of Aβ40 or Aβ42 transport through tripartite vessels over 4 hr ( Figure 7a , b ) . However , the combination of GW3965 and HDL together resulted in a significantly increased initial rate of Aβ42 transport without affecting Aβ40 ( Figure 7a , b ) . Interestingly , relative to baseline tripartite conditions , neither GW3965 , HDL , nor both GW3965 and HDL affected accumulation of Aβ40 or Aβ42 within tripartite tissues after 24 hr ( Figure 7c , d ) , which differs from our observations in bipartite vessels ( Figure 5 ) . To further understand the discrepancy between bipartite and tripartite vessels for Aβ42 tissue accumulation , we hypothesised that basal levels of native apoE secreted from astrocytes in tripartite vessels might already be sufficient to reduce Aβ accumulation . Western blotting confirmed that tripartite tissue lysates had significantly more apoE than bipartite tissue under baseline conditions ( Figure 7e ) . We then directly compared Aβ transport and tissue accumulation between bipartite and tripartite vessels and observed increased Aβ42 but not Aβ40 recovery into circulation medium in tripartite compared to bipartite vessels ( Figure 7f , g ) . With respect to tissue accumulation , Aβ42 levels were significantly lower and Aβ40 levels showed a trend toward lower levels in tripartite compared to bipartite vessels ( Figure 7h , i ) . Notably , Aβ42 levels in tripartite vessels were similar to those observed in bipartite vessels to which recombinant apoE2 was added . We next hypothesised that pre-aggregated amyloid fibrils might accumulate to form CAA in tripartite vessels . To test this , we compared Aβ transport and accumulation by injecting either monomeric or pre-aggregated Aβ40 or Aβ42 on the anteluminal side of tripartite vessels , and observed significantly slower transport of Aβ40 and Aβ42 fibrils compared to monomers ( Figure 8a–b ) . Furthermore , tissue accumulation was significantly increased after injection of fibrils compared to monomers ( Figure 8a–b ) . To test whether native astrocyte-secreted apoE might reduce aggregation of pre-formed Aβ40 and Aβ42 fibrils ( Castellano et al . , 2011 ) , we quantified Aβ levels in bipartite and tripartite tissues 24 hr after injection of pre-formed Aβ40 and Aβ42 fibrils and observed no significant difference in Aβ accumulation ( Figure 8c–d ) . Together , these data suggest that astrocyte-derived apoE specifically facilitates transport of soluble Aβ from the brain . The profound socioeconomic impact of AD has stimulated extensive research in the last decade , with increased attention on the contribution of cerebrovascular dysfunction in dementia and cognitive decline ( Snyder , 2015; Raz et al . , 2016 ) . It is clearly critical to elucidate the mechanisms that regulate Aβ egress from the human brain , as well as understand how to protect cerebrovascular health during aging , yet there are significant barriers toward mechanistic experimentation in a human context . Almost all of our current knowledge about Aβ egress through cerebral vessels has been obtained from animal models , primarily in mice genetically engineered to express human APP , which enables the progressive accumulation of Aβ and β-amyloid to be studied . Despite the tremendous wealth of knowledge generated from animal models , there are considerable challenges in translating these results into humans . For example , many AD risk genes have roles in various aspects of lipid metabolism , the most important of these being APOE ( Giri et al . , 2016 ) , yet the innate physiological differences between murine and human lipoprotein metabolism ( Getz and Reardon , 2012 ) may limit the predictive power of mouse model studies . Although it is well known that humans have three APOE allelic variants compared to the single Apoe allele in mice , and targeted replacement mice are available that express human APOE , there are also other important metabolic distinctions between mice and humans . For example , the primary circulating lipoprotein in rodents is HDL , which , due to its multiple vasoprotective functions , bestows upon mice a natural resilience to cardiovascular diseases such as atherosclerosis . By contrast , the major circulating lipoprotein in humans is LDL , which is mechanistically linked to vascular dysfunction and cardiovascular disease ( Barter et al . , 2007 ) . Despite the wide recognition of the importance of cardiovascular risk factors to AD pathogenesis , they cannot easily or routinely be incorporated into rodent AD models . In vitro studies using human cells therefore represent a useful alternative approach , yet most studies of cerebrovascular function use monotypic cultures of brain ECs , which do not mimic the complexity of cell-cell and/or cell-matrix interactions found in the native vessel . As an improvement , EC and astrocytes , EC and SMC , and EC and pericytes have been co-cultured , however , this is almost always under static culture conditions ( Di et al . , 2009; Cho et al . , 2007; Bicker et al . , 2014; Navone et al . , 2013; Man et al . , 2008; Hatherell et al . , 2011; Bussolari et al . , 1982; Cucullo et al . , 2007 ) . More recent studies developed an EC and astrocyte co-culture model using a complex flow system , but this model did not allow histological analysis or cell-ECM interactions to be assessed ( Cucullo et al . , 2007 ) . Considerable recent advances in tissue engineering technology have helped to develop microfluidic systems ( i . e . ‘organ on a chip’ ) that recapitulate the 3D complexity of the BBB , primarily to produce small vessel structures like capillaries ( Prabhakarpandian et al . , 2013; Herland et al . , 2016; Booth and Kim , 2012; Griep et al . , 2013; Cho et al . , 2015 ) . However , as CAA preferentially forms in larger arteries , we aimed to produce 3D in vitro models of functional human vessels that retain the anatomical and functional properties of native human cerebral arteries . In particular , the pulsatile native-like flow environment possible in our 3D models represents a major advantage over studies using static cell culture . Notably , bioengineered peripheral artery equivalents similar to our bipartite model are clinically used to replace damaged vessels in patients with structural cardiovascular diseases ( Schmidt et al . , 2006 ) . Here we demonstrate the utility of bioengineered vessels to investigate how lipoproteins on the brain side or blood side of the vessel affect Aβ transport and development of CAA , using both bipartite and tripartite engineered vessels . Importantly , the generation of a functional , 3D cerebrovascular model using primary human ECs , SMCs and astrocytes represents a major advance in the bioengineering field . To represent human pial , leptomeningial and penetrating cerebral arteries , we reduced the internal diameter of the scaffold to 2 mm compared to previous engineered vessel equivalents ( Schmidt et al . , 2006; Robert et al . , 2013a ) , while retaining the histologically confirmed architecture of native vessels . Specifically , our bipartite vessels contained a monolayer of ECs forming a hollow lumen , surrounded by multiple layers of α-SMA and secreted extracellular matrix components including collagen and laminin , which demonstrate the in situ functionality of the component cells in the tissue . As a major function of the endothelium is to form a tight barrier between the blood and the interstitial vascular tissue , we also demonstrated the structural integrity of the endothelial layer in bioengineered vessels by demonstrating their impermeability to Evans blue and FITC-Dextran , and confirmed EC function by demonstrating secretion of NO . Furthermore , our completely novel tripartite vessel possessed a layer of GFAP positive astrocytes on the anteluminal side . The formation of structures resembling astrocyte end-feet and secretion of apoE that retains its response to LXR stimulation represent key structural and functional features of astrocytes in healthy cerebral vessels , further supporting the validity of our bioengineered tissue as a valuable new model for mechanistic studies of the human cerebrovasculature . Finally , we observed that the presence of astrocytes induced expression of brain-specific tight junction proteins and transporters in tripartite vessels generated from ECs from a peripheral source , offering the potential to understand how vascular context may reprogram EC phenotype in future studies . Numerous studies have demonstrated that the accumulation and aggregation of Aβ within the muscular layer of arteries and arterioles represents a key step in the development of CAA ( Biffi and Greenberg , 2011; Love , 2009 ) . We show that our platform can be used to investigate how human lipoproteins on each side of the BBB regulate vascular function and Aβ transport and accumulation . As apoE is the major apolipoprotein expressed in the central nervous system , brain-derived apoE would affect cerebrovascular function from the anteluminal side . By contrast , as apoA-I is produced only in liver and intestine , HDL is found in the circulation and would affect cerebrovascular function from the lumen . Although lipid-free apoA-I can be transported into the brain and is present in CSF ( Stukas et al . , 2014a ) , there is thus far no evidence that mature HDL might cross the BBB . Our results support a functional cooperation between brain apoE and circulating HDL to promote clearance of Aβ through the cerebral vessel by mechanisms that remain to be fully elucidated . That HDL and apoE consistently affected Aβ42 more than Aβ40 suggests that Aβ40 may be less amenable to lipoprotein-mediated transit across and removal from the vascular wall compared to Aβ42 , i . e . Aβ40 is more prone to being retained in the vessel compared to Aβ42 . These data are consistent with the observation that Aβ40 is the predominant species found in human CAA ( Yamada , 2015 ) even though Aβ42 is reported to be essential for CAA development in mice ( McGowan et al . , 2005 ) . Our results are also consistent with the hypothesized effects of apoE isoform on vascular function , as we demonstrate that recombinant apoE2 promotes more Aβ42 clearance than recombinant apoE4 . Importantly , our platform now provides an opportunity to evaluate potential therapeutic strategies to facilitate Aβ clearance , including approaches that target HDL or apoE . Our study , nevertheless , has several limitations . A major limitation is the availability of primary human astrocytes with distinct APOE genotypes , although future studies using standardized astrocytes derived from stem cells could be one possible solution . Similarly , how well recombinant apoE resembles native apoE will require additional investigation especially concerning its lipidation status . Furthermore although previous in vivo studies showed that apoE diminishes Aβ clearance in an isoform-specific manner ( Deane et al . , 2008 ) , these analyses were restricted to the transport of preformed Aβ-apoE complexes only and could not differentiate between transport occurring at large vessels versus capillaries . By contrast , our experiments evaluated functional Aβ and apoE interactions under a variety of conditions . Murine and human ECs also differentially clear Aβ with a 30-fold increase of Aβ uptake in mouse cells ( Qosa et al . , 2014 ) . Together , these observations might explain the differences observed between previously published murine data and our human model . A further limitation is that we used readily obtainable umbilical cord cells due to the slower growth rate and supply of primary cerebrovascular cells . Although one could argue that cord-derived cells do not reflect the physiology of the brain vasculature , we clearly demonstrate that HUVEC become reprogrammed to express selective BBB marker proteins when cultured in tripartite bioengineered vessels and that the histological structure of the vessels as well as Aβ accumulation and transport were similar between brain and cord cellular origins . Another limitation is that HDL was obtained from healthy young donors . As HDL functions can be compromised by aging , cardiovascular disease and T2DM ( Riwanto and Landmesser , 2013 ) , it will be important in the future to understand how HDL purified from aged cognitively healthy individuals , AD subjects , or patients with cardiovascular risk factors may affect cerebrovascular function and Aβ accumulation especially in combination with apoE . It should be noted that the concentration of HDL used in the present study is estimated to be 7-fold lower compared to normolipidomic individuals . HDL concentration within the plasma is typically expressed based on its cholesterol content , with normal levels between 40 and 60 mg/dl , which could roughly be translated to 140 mg protein/dL , similar to apoA-I plasma concentration ( Koren et al . , 1985 ) . We circulated 200 μg/mL HDL in our experiments , similar to what has previously been published for other in vitro studies ( Datta et al . , 2001; Robert et al . , 2013a ) . Although previous in vitro studies have also used similar Aβ levels ( Takamatsu et al . , 2014; Xu et al . , 2013 ) , the Aβ dose used in this study corresponds to a supra physiological concentration of Aβ compared to human brain , CSF or cerebral interstitial fluid , a limitation imposed by the detection limits of our assay ( Brody et al . , 2008; Seubert et al . , 1992; Herukka et al . , 2015 ) As well , although we could demonstrate that our bioengineered vessels are able to produce NO under physiologically relevant stimuli , our current scaffolding materials are too stiff to permit measurement of vascular compliance . Engineering improvements that explore alternative materials and enable scalable production of bioengineered vessels are also important avenues for future studies . Scalable higher throughput methods are particularly important , as each vessel takes approximately 4 weeks to mature . In conclusion , we demonstrate for the first time the feasibility to engineer dynamic 3D human artery equivalents to investigate fundamental AD-relevant cerebrovascular processes in vitro , including CAA and the role of lipoproteins in its prevention . Our experimental platform combines the native-like multilayer 3D architecture of both peripheral and cerebral arterial walls with pulsatile flow profiles through a functional lumen , in a manner that permits delivery and sampling of experimental substrates from either side of the bioengineered vessel . By extending the frontier of vascular tissue engineering into diseases involving the cerebrovasculature , our unique cerebral vessel may facilitate in vitro investigations with higher predictive value for human pathologies than current cellular or animal model approaches , as almost all components of the platform can be experimentally manipulated . For example , drug development and delivery studies can be performed , and the effects of hypertension can be studied by modulating factors such as flow rate and pressure of the circulating media . Plasma or immune cells from specific patient groups can be evaluated for effects on cerebrovascular function , including pre- and post-intervention analyses , to better understand the interactions between cardiovascular factors ( i . e . T2DM , hypercholesterolemia , hypertension ) and brain factors ( i . e . APOE genotype ) . Finally , this platform may also facilitate the discovery of blood biomarkers for central nervous system indications . Further advances in bioengineered cerebral vessels to include other brain cells may ultimately improve translational relevance and provide a valuable complement to in vivo studies in animal models . All experiments were conducted under an approved clinical protocol ( UBC Clinical Ethics Research Board H13-02719 ) after obtaining written informed consent from all subjects . Human umbilical vein endothelial cells ( HUVEC ) and human umbilical cord myofibroblasts ( UCMFB ) were isolated as described ( Robert et al . , 2013a ) . HUVEC were isolated using the instillation method , where umbilical veins were filled with a solution of collagenase ( 2 mg/ml , Collagenase A , Roche ) in serum-free DMEM ( Invitrogen , ThermoFischer Scientific , Waltham , MA ) . After 20 min at 37°C , Advanced DMEM ( Gibco , ThermoFischer Scientific ) supplemented with 1% L-glutamine , 0 . 05% Penicillin/Streptavidin ( Pen/Strep ) and 10% FBS ( Invitrogen ) was flushed through the lumen and the cell suspension was centrifuged at 1 , 200 rpm for 5 min . HUVEC were expanded in endothelial growth medium ( EGM−2 ) ( LONZA Inc . , Switzerland , supplemented with vascular endothelial growth factor ( VEGF ) , human recombinant insulin-like growth factor-1 ( hrIGF-1 ) , human epidermal growth factor ( hEGF ) , amphotericin-B , hydrocortisone , ascorbic acid , heparin , and 2% foetal bovine serum ( FBS ) ) up to passage 7 . After HUVEC isolation , the remaining vessels were minced into small pieces ( ~2–3 mm ) and incubated at room temperature ( RT ) without medium under sterile laminar flow for 25–30 min to ensure physical attachment of UCMFB . Advanced DMEM supplemented with 1% L-glutamine , 0 . 05% Pen/Strep and 10% FBS was subsequently added to the minced vessels and adherent cells were expanded up to passage 8 . Primary mature astrocytes ( Sciencell ) were cultivated in astrocyte media ( Sciencell , Carlsbad , CA ) supplemented with astrocyte growth factor , 0 . 05% Pen/Strep and 2% FBS ( Sciencell ) up to passage 5 . Primary cerebral SMC ( Sciencell ) were cultivated in Advanced DMEM supplemented with 1% L-glutamine , 0 . 05% Pen/Strep and 10% FBS up to passage 5 . Primary cortical microvascular EC ( Cell Systems , Kirkland , WA ) were cultivated in complete EGM−2 up to passage 4 . Bioengineered constructs were fabricated using a dynamic , semi-pulsatile flow bioreactor system . Tubular biodegradable scaffolds ( length 1 . 5 cm and inner diameter 2 mm ) were produced as described ( Robert et al . , 2013a; Robert et al . , 2017 ) with minor modifications . Briefly , non-woven polyglycolic acid ( PGA , Biomedical Structure , Warwick , RI ) meshes ( thickness: 1 mm and density: 70 mg/cc ) were dip-coated with polycaprolactone ( PCL ) and polylactate ( PLA ) by dipping PGA mesh in a solution of 1 . 75% ( w/w ) PCL/PLA/tetrahydrofuran ( THF ) solution ( Sigma-Aldrich , St . Louis , MO ) , shaped into tubes using heat , and externally coated with a 10% PCL/THF ( w/w ) solution . Scaffolds were sterilized by immersion in 70% ethanol for 30 min followed by three PBS washes and then immersion in advanced DMEM supplemented with 10% FBS for at least 12 hr . UCMFB were seeded at density of 2−3 × 106 cells/cm2 on the inner surface of the scaffold using fibrin ( fibrinogen 10 mg clottable protein/ml PBS and thrombin 100–10 mU/ml PBS ) as a cell carrier that was added directly to the scaffold , then incubated under static conditions for a minimum of 3 days before exposure to dynamic flow . The flow of nutrient medium ( Advanced DMEM supplemented with 10% FBS , 1% L-glutamine and 0 . 05% Pen/Strep ) was directed through the lumen of the bioreactor circulation loop to mimic blood flow for a minimum of one week . Vascular intermediates were then seeded with HUVEC ( 1 × 106 cells/cm2 ) and cultivated first in static conditions for a minimum of 5 days in EGM−2 supplemented as above . For bipartite vessels , after the static phase , vascular grafts were placed back in the bioreactor for 14 additional days with increasing medium flow to a final rate of 10 ml/min by the 10th day ) . For tripartite vessels , after the static phase after HUVEC addition , primary astrocytes were seeded ( 1 × 106 cells/cm2 ) using fibrin as a cell carrier as above on the antelumen side of the tissue . After 5 min at RT , grafts were placed under flow conditions with EGM-2 supplemented as above in the circulation chamber and complete astrocyte medium in the tissue chamber for 14 additional days with increasing medium flow to 10 ml/min by the 10th day . Recombinant Aβ40 and Aβ42 peptides ( California Peptide Research , Salt Lake City , UT ) were dissolved in hexafluoroisopropanol ( HFIP , Sigma-Aldrich ) . The HFIP was removed by evaporation overnight and stocks were stored at −20°C . On the day of the assay , soluble monomers were prepared by reconstituting the peptide film in DMSO to 5 mM , diluted further to 100 µM in RPMI without FBS . 100 μl of Aβ solution was injected in the tissue chamber containing 900 μl of DMEM ( Gibco ) without FBS to the desired concentration using a syringe under flow conditions . For fibrils , after reconstitution in RPMI Aβ40 and Aβ42 were incubated at 37°C for 48 hr . Fibrilization was confirmed by dot blot with fibril antibody ( OC AB2286 EMD Millipore 1:1000 , not shown , RRID: AB_1977024 ) . For luminal recovery , 100 μl circulating medium was collected at the indicated time . All experiments were conducted under an approved clinical protocol ( UBC Clinical Ethics Research Board H14-03357 ) . Upon receipt of written informed consent , 100 ml of fasted blood was collected from normolipidemic healthy donors into vacutainer tubes containing EDTA . Plasma HDL ( 1 . 063–1 . 21 g/ml ) was isolated by sequential potassium bromide gradient ultracentrifugation as described ( Robert et al . , 2013b ) . The purity of the HDL preparations was verified by sodium dodecyl sulfate-polyacrylamide gel electrophoresis ( SDS-PAGE ) followed by Coomassie blue staining to ensure no LDL or albumin contamination ( not shown ) . Total protein concentration was assessed using the BCA assay ( Thermofisher Scientific ) . Recombinant apoE2 and apoE4 were commercially purchased and solubilized following the manufacture’s instructions ( ABCAM , apoE2 ab55210 , apoE3 ab123764 and apoE4 ab50243 ) . Secretion of endogenous apoE from human astrocytes in tripartite vessels was induced by injecting 0 . 8 μM GW3965 in the circulation medium 72 hr prior to addition of Aβ . ApoE concentrations were measured using ELISA as previously described ( Fan et al . , 2016 ) . Briefly , ELISA plates were coated overnight with anti-human apoE mAB E276 antibody ( MabTech , Cincinnati , OH , RRID: AB_1925746 ) at 1 . 55 μg/mL in PBS at 4°C , washed two times with PBST ( 0 . 05% Tween 20 in PBS ) , and blocked with 0 . 1% Blocker A ( MesoScale Discovery , Rockville , MA ) in PBST . After 1 hr incubation at RT and two washes with PBST , medium or human recombinant ApoE standard ( MabTech ) were added to each well . After 1 hr at RT and two subsequent PBST washes , biotinylated anti-human apoE monoclonal antibody E887 ( MabTech , RRID: AB_1925729 ) was added to each well at a concentration of 0 . 5 μg/mL in blocking buffer . After 1 hr at RT , plates were washed before adding QuantaBlue Substrate ( Pierce , ThermoFischer Scientific ) working solution ( 9 parts of Substrate Solution to one part Stable Peroxide Solution ) . Fluorescence was read after 15 min at RT on an EnSpire 2300 Multilabel Plate Reader ( 325Ex/420Em ) . Evans blue ( Sigma-Aldrich ) was injected at a final concentration of 0 . 5% in the circulation loop of the bioreactor for 10 min followed by continuous PBS washing for 20 min . Vessels were cut open longitudinally and en face preparations were analysed macroscopically with photo documentation . Restriction of paracellular transport was determined by measuring FITC dextran extravasation to the tissue chamber as described ( Gaillard et al . , 2001 ) . Briefly 250 µg/ml of 4 kDa or 40 kDa FITC-dextran ( Sigma-Aldrich ) was circulated through the lumen of bipartite , tripartite tissues or scaffold only . After 1 hr tissue media was collected , fluorescent was read at RT on an EnSpire 2300 Multilabel Plate Reader ( 492Ex/518Em ) and the permeability coefficient ( Papp ) was calculated using the following equation: Papp= ( dQ/dt ) * ( 1/AC060 ) where dQ/dt is the amount of FITC-dextran transported per minute ( ng/min ) , A is the surface area of the tissue ( cm [Attems and Jellinger , 2014] ) , C0 is the initial concentration of FITC-Dextran ( ng/ml ) and 60 is the conversion from minutes to seconds . NO synthesis was measured as described ( Robert et al . , 2013a ) using a commercial NOS activity assay kit ( Caymenchemical , Ann Arbor , MI ) . Briefly , a 2–3 mm ring of vascular tissue was mechanically ground in 150 μl of ice cold homogenized buffer ( 25 mM Tris-Cl , pH7 . 4 , 1 mM EDTA and 1 mM EGTA ) and centrifuged 15 min at 4°C at 10 , 000 g . The supernatant was aliquoted and incubated in reaction buffer containing 25 mM Tris-Cl , pH7 . 4 , 0 . 25 mM EDTA , 0 . 6 mM CaCl2 , 1 mM NADP , 200 nM calmodulin , 3 µM tetrahydrobiopterin , 1 µM flavin adenine dinucleotide , 1 µM flavin adenine monoucleotide and 0 . 2 µCi L-3H-arginine; PerkinElmer , Waltham , MA ) in the presence of 10 nM acetylcholine ( Ach ) , 0 . 2 mg/ml HDL or 1 mM L-NG-nitroarginine methyl ester ( L-NAME ) . After 60 min at 37°C , the reaction was stopped by adding 400 μl of stop buffer ( 50 nM HEPES , 5 mM EDTA , pH 5 . 5 ) . The solution was loaded onto an ion exchange column equilibrated with stop buffer to separate L-3H-citruline from L-3H-arginine . Scintillation mix ( Ultimate Gold , PerkinElmer ) was added to the supernatant and counted using LS6500 β-counter ( Beckman Coulter , Brea , CA ) . The percent citrulline formed was calculated as follows: % conversion = ( cpm reaction-cpm background ) /cpm total *100 . For standard histology , bioengineered vessels were fixed in formalin ( Thermo Fisher Scientific ) for 24 hr , dehydrated through a series of graded ethanol series in a tissue processor ( Sakura , Torrance , CA ) , embedded in paraffin and sectioned at 7 µm thickness . For staining , sections were deparaffinised in 3 baths of Xylene and rehydrated through a graded ethanol series ( 100 , 90 , 80% and 70% , for 1 min ) . Sections were stained using Haematoxylin and Eosin ( Sigma-Aldrich ) and Picrosirius ( ABCAM , Canada ) following the manufactures’ instructions . For immunohistochemistry and Thioflavin-S staining , vessels were washed twice in PBS , cryopreserved in O . C . T . embedding matrix , and processed on a cryotome to generate 20 µm sections that were stored at –80°C until further analysis . Sections were rehydrated in PBS for 2 × 10 min before fixing in 4% paraformaldehyde ( PFA ) for 20 min at RT . After one Tris-HCl ( 0 . 5 mM pH 7 . 6 ) and two PBS washes , sections were blocked for 30 min in 5% goat serum and 1% BSA in PBS . For immunofluorescence , sections were incubated overnight at 4°C with specific antibodies against CD31 ( WM59 Biolegend , San Diego , CA , 1:50 , RRID: AB_314328 ) , von Willebrand factor ( Sigma-Aldrich , 1:200 , RRID: AB_259543 ) , α-SM-actin ( 1A4 Sigma-Aldrich , 1:200 , RRID: AB_476856 ) , collagen IV ( EMD Millipore , 1:100 , RRID: AB_2276457 ) , laminin ( Abcam , 1:200 , RRID: AB_298179 ) , claudin 5 ( 4C3C2 ThermoFisher Scientific , 1:50 , RRID:AB_2533200 ) , ZO-1 ( 1A12 ThermoFisher Scientific , 1:50 , RRID: AB_2533147 ) , ZO-2 ( ThermoFisher Scientific , 1:50 , RRID: AB_2533976 ) , Aβ 1–16 ( 6E10 ThermoFisher Scientific , 1:50 , RRID: AB_2565328 ) and Aβ fibrils ( OC AB2286 EMD Millipore 1:200 , RRID: AB_1977024 ) . After three additional PBS washes , sections were incubated for 45 min at RT with anti-rabbit or anti-mouse Alexa-488 or Alex-594 secondary antibodies ( Invitrogen ) with DAPI . Sections were finally washed three times in PBS and mounted in Prolong Diamond antifade ( ThermoFisher Scientific ) . For Thioflavin-S staining , sections were rehydrated and stained in 1% Thioflavin-S ( Sigma-Aldrich ) for 10 min at RT in the dark , washed five times with PBS and mounted in Vectashield ( Vector ) . Brighfield and fluorescent images were acquired with an inverted microscope ( Zeiss , Germany ) or a SP8 confocal microscope ( Leica , Canada ) . Luminal medium was collected from the circulation chamber and 5 mm tissue rings were crushed and lysed in RIPA buffer ( 10 mM Tris pH 7 . 4 , 150 mM NaCl , 1 . 0% NP-40 , 1 . 0% sodium deoxycholate , 0 . 1% SDS and cOmplete protease inhibitor with EDTA ( Roche , Switzerland ) ) . Aβ40 ( KHB3442 , Life Tech , ThermoFischer Scientific ) and Aβ42 ( KHB3482 , Life Tech ) were quantified using commercial ELISAs and normalized to total protein concentration , measured by BCA . Aβ fibrillization was measured as described ( Truran et al . , 2016 ) . Briefly , 10 μl of RIPA homogenized tissue was mixed with 90 μl of 20 μM Thioflavin-T ( Sigma-Aldrich ) in 150 mM NaCl with 5 mM HEPES , pH 7 . 4 in black 96-well plates . Formation of fibrillar β-amyloid pleated sheets was monitored by excitation at 440 nm and measuring emission intensity at 490 nm using an Infinite M2000 Pro plate reader ( Tecan ) . Human cortical brain tissues were previously obtained from the Brain and Tissue Bank , University of Maryland School of Medicine under the UBC clinical protocol ( C04-0595 ) . Human umbilical cords were obtained from the British Columbia Women’s Hospital , Vancouver , BC , Canada . Statistical comparisons between different groups were performed using Student T-test , one way ANOVA with Dunnett post test , or two way ANOVA with Sidak multi comparison test . Data were obtained from at least four independently generated bioengineered vessels and graphically represented as mean ±standard error of the mean ( SEM ) . P-values of <0 . 05 were considered statistically significant . All statistical analyses were performed using GraphPad Prism-5 software ( RRID:SCR_002798 ) .	Alzheimer’s disease causes gradual loss of memory and difficulties in learning . The brains of patients with the disease show several abnormalities including deposits of a peptide molecule called beta-amyloid that is known to be toxic to nerve cells . This peptide can also cause damage to the brain by accumulating within the muscular walls of large blood vessels , a condition known as cerebral amyloid angiopathy ( CAA ) and is present in most Alzheimer’s disease patients . A group of molecules known as lipoproteins , which transport fats throughout body fluids , are thought to be involved in the process by which beta-amyloid leaves the brain . Apolipoprotein E ( apoE ) is one such molecule and it is made in the brain by cells called astrocytes . There are three different versions of apoE that are associated with different levels of risk of developing Alzheimer’s disease . Other lipoproteins , such as high-density lipoprotein , which is present in the blood , may also play a role in clearing beta-amyloid proteins from the brain . However , it has been difficult to investigate the roles of these lipoproteins in Alzheimer’s disease because current test-tube models do not fully mimic the composition of human brain blood vessels or show how they work . Robert et al . have used a tissue engineering approach to generate the first three-dimensional model of human brain blood vessels that can reproduce cerebral amyloid angiopathy . To make the model , different types of human cells similar to those found in real blood vessels and astrocytes were grown under conditions that resemble real-life conditions , including mimicking blood flow through the engineered vessels . Having established that the engineered vessels behaved similarly to normal blood vessels , Robert et al . used them to test whether lipoproteins helped to clear beta-amyloid proteins from the vessels . These experiments showed that a form of apoE that protects against Alzheimer’s disease was more effective in transporting beta-amyloid proteins across the walls of blood vessels than other forms of apoE . Further experiments showed that high-density lipoprotein in the blood and apoE on the brain side of the vessel work together to help transport beta-amyloid into the vessels . Together , these findings show that the model of CAA developed by Robert et al . provides a valuable new tool for exploring how this condition develops . The model could also be used more widely in the future , for example , to study how to deliver new drugs that could help treat Alzheimer’s disease into the brain .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Clearance of beta-amyloid is facilitated by apolipoprotein E and circulating high-density lipoproteins in bioengineered human vessels
Reactive oxygen species ( ROS ) -dependent signaling pathways from chloroplasts and mitochondria merge at the nuclear protein RADICAL-INDUCED CELL DEATH1 ( RCD1 ) . RCD1 interacts in vivo and suppresses the activity of the transcription factors ANAC013 and ANAC017 , which mediate a ROS-related retrograde signal originating from mitochondrial complex III . Inactivation of RCD1 leads to increased expression of mitochondrial dysfunction stimulon ( MDS ) genes regulated by ANAC013 and ANAC017 . Accumulating MDS gene products , including alternative oxidases ( AOXs ) , affect redox status of the chloroplasts , leading to changes in chloroplast ROS processing and increased protection of photosynthetic apparatus . ROS alter the abundance , thiol redox state and oligomerization of the RCD1 protein in vivo , providing feedback control on its function . RCD1-dependent regulation is linked to chloroplast signaling by 3'-phosphoadenosine 5'-phosphate ( PAP ) . Thus , RCD1 integrates organellar signaling from chloroplasts and mitochondria to establish transcriptional control over the metabolic processes in both organelles . Cells of photosynthesizing eukaryotes are unique in harboring two types of energy organelles , the chloroplasts and the mitochondria , which interact at an operational level by the exchange of metabolites , energy and reducing power ( Noguchi and Yoshida , 2008; Cardol et al . , 2009; Bailleul et al . , 2015 ) . Reducing power flows between the organelles through several pathways , including photorespiration ( Watanabe et al . , 2016 ) , malate shuttles ( Scheibe , 2004; Zhao et al . , 2018 ) and transport of photoassimilate-derived carbon rich metabolites from chloroplasts to mitochondria . At the signaling level , the so-called retrograde signaling pathways originating from the organelles influence the expression of nuclear genes ( de Souza et al . , 2017; Leister , 2019; Waszczak et al . , 2018 ) . These pathways provide feedback communication between the organelles and the gene expression apparatus in the nucleus to adjust expression of genes encoding organelle components in accordance with changes in the developmental stage or environmental conditions . Reactive oxygen species ( ROS ) , inevitable by-products of aerobic energy metabolism , play pivotal roles in plant organellar signaling from both chloroplasts and mitochondria ( Dietz et al . , 2016; Noctor et al . , 2018; Waszczak et al . , 2018 ) . Superoxide anion radical ( O2˙ – ) is formed in the organelles by the transfer of electrons from the organellar electron transfer chains ( ETCs ) to molecular oxygen ( O2 ) . In illuminated chloroplasts , superoxide anion formed from O2 reduction by Photosystem I ( PSI ) is converted to hydrogen peroxide ( H2O2 ) which is further reduced to water by chloroplastic H2O2-scavenging systems during the water-water cycle ( Asada , 2006; Awad et al . , 2015 ) . Chloroplastic ROS production can be enhanced by application of methyl viologen ( MV ) , a chemical that catalyzes shuttling of electrons from PSI to O2 ( Farrington et al . , 1973 ) . The immediate product of this reaction , O2˙ – , is not likely to directly mediate organellar signaling; however , H2O2 is involved in many retrograde signaling pathways ( Leister , 2019; Mullineaux et al . , 2018; Waszczak et al . , 2018 ) . Organellar H2O2 has been suggested to translocate directly to the nucleus ( Caplan et al . , 2015; Exposito-Rodriguez et al . , 2017 ) , where it can oxidize thiol groups of specific proteins , thereby converting the ROS signal into thiol redox signals ( Møller and Kristensen , 2004; Nietzel et al . , 2017 ) . One recently discovered process affected by chloroplastic H2O2 is the metabolism of 3'-phosphoadenosine 5'-phosphate ( PAP ) . PAP is a toxic by-product of sulfate metabolism produced when cytoplasmic sulfotransferases ( SOTs , e . g . , SOT12 ) transfer a sulfuryl group from PAP-sulfate ( PAPS ) to various target compounds ( Klein and Papenbrock , 2004 ) . PAP is transported to chloroplasts where it is detoxiﬁed by dephosphorylation to adenosine monophosphate in a reaction catalyzed by the adenosine bisphosphate phosphatase 1 , SAL1 ( Quintero et al . , 1996; Chan et al . , 2016 ) . It has been proposed that oxidation of SAL1 thiols directly or indirectly dependent on chloroplastic H2O2 inactivates the enzyme , and accumulating PAP may act as a retrograde signal ( Estavillo et al . , 2011; Chan et al . , 2016; Crisp et al . , 2018 ) . ROS are also produced in the mitochondria , for example by complex III at the outer side of the inner mitochondrial membrane ( Cvetkovska et al . , 2013; Ng et al . , 2014; Huang et al . , 2016; Wang et al . , 2018 ) . Blocking electron transfer through complex III by application of the inhibitors antimycin A ( AA ) or myxothiazol ( myx ) enhances electron leakage and thus induces the retrograde signal . Two known mediators of this signal are the transcription factors ANAC013 ( De Clercq et al . , 2013 ) and ANAC017 ( Ng et al . , 2013b; Van Aken et al . , 2016b ) that are both bound to the endoplasmic reticulum ( ER ) by a transmembrane domain . Mitochondria-derived signals lead to proteolytic cleavage of this domain . The proteins are released from the ER and translocated to the nucleus where they activate the mitochondrial dysfunction stimulon ( MDS ) genes ( De Clercq et al . , 2013; Van Aken et al . , 2016a ) . MDS genes include the mitochondrial alternative oxidases ( AOXs ) , SOT12 , and ANAC013 itself , which provides positive feedback regulation and thus enhancement of the signal . Whereas multiple retrograde signaling pathways have been described in detail ( de Souza et al . , 2017; Leister , 2019; Waszczak et al . , 2018 ) , it is still largely unknown how the numerous chloroplast- and mitochondria-derived signals are integrated and processed by the nuclear gene expression system . Nuclear cyclin-dependent kinase E is implicated in the expression of both chloroplastic ( LHCB2 . 4 ) and mitochondrial ( AOX1a ) components in response to perturbations of chloroplast ETC ( Blanco et al . , 2014 ) , mitochondrial ETC , or H2O2 treatment ( Ng et al . , 2013a ) . The transcription factor ABI4 is also suggested to respond to retrograde signals from both organelles ( Giraud et al . , 2009; Blanco et al . , 2014 ) , although its significance in chloroplast signaling has recently been disputed ( Kacprzak et al . , 2019 ) . Mitochondrial signaling via ANAC017 was recently suggested to converge with chloroplast PAP signaling based on similarities in their transcriptomic profiles ( Van Aken and Pogson , 2017 ) . However , the mechanistic details underlying this convergence remain currently unknown . Arabidopsis RADICAL-INDUCED CELL DEATH1 ( RCD1 ) is a nuclear protein containing a WWE , a PARP-like [poly ( ADP-ribose ) polymerase-like] , and a C-terminal RST domain ( RCD1-SRO1-TAF4 ) ( Overmyer et al . , 2000; Ahlfors et al . , 2004; Jaspers et al . , 2009; Jaspers et al . , 2010a ) . In yeast two-hybrid studies RCD1 interacted with several transcription factors ( Jaspers et al . , 2009 ) including ANAC013 , DREB2A ( Vainonen et al . , 2012 ) , and Rap2 . 4a ( Hiltscher et al . , 2014 ) via the RST domain ( Jaspers et al . , 2010b ) , and with the sodium transporter SOS1 ( Katiyar-Agarwal et al . , 2006 ) . In agreement with the numerous potential interaction partners of RCD1 , the rcd1 mutant demonstrates pleiotropic phenotypes in diverse stress and developmental responses ( Jaspers et al . , 2009 ) . It has been identified in screens for sensitivity to ozone ( Overmyer et al . , 2000 ) , tolerance to MV ( Fujibe et al . , 2004 ) and redox imbalance in the chloroplasts ( Heiber et al . , 2007; Hiltscher et al . , 2014 ) . RCD1 was found to complement the deficiency of the redox sensor YAP1 in yeast ( Belles-Boix et al . , 2000 ) . Under standard growth conditions , the rcd1 mutant displays differential expression of over 400 genes , including those encoding mitochondrial AOXs ( Jaspers et al . , 2009; Brosché et al . , 2014 ) and the chloroplast 2-Cys peroxiredoxin ( 2-CP ) ( Heiber et al . , 2007; Hiltscher et al . , 2014 ) . Here we have addressed the role of RCD1 in the integration of ROS signals emitted by both mitochondria and chloroplasts . Abundance , redox status and oligomerization state of the nuclear-localized RCD1 protein changed in response to ROS generated in the chloroplasts . Furthermore , RCD1 directly interacted in vivo with ANAC013 and ANAC017 and appeared to function as a negative regulator of both transcription factors . The RST domain , mediating RCD1 interaction with ANAC transcription factors , was required for plant sensitivity to chloroplastic ROS . We demonstrate that RCD1 is a molecular component that integrates organellar signal input from both chloroplasts and mitochondria to exert its influence on nuclear gene expression . Methyl viologen ( MV ) enhances ROS generation in illuminated chloroplasts by catalyzing the transfer of electrons from Photosystem I ( PSI ) to molecular oxygen . This triggers a chain of reactions that ultimately inhibit Photosystem II ( PSII ) ( Farrington et al . , 1973; Nishiyama et al . , 2011 ) . To reveal the significance of nuclear protein RCD1 in these reactions , rosettes of Arabidopsis were pre-treated with MV in darkness . Without exposure to light , the plants displayed unchanged PSII photochemical yield ( Fv/Fm ) . Illumination resulted in a decrease of Fv/Fm in wild type ( Col-0 ) , but not in the rcd1 mutant ( Figure 1A ) , suggesting increased tolerance of rcd1 to chloroplastic ROS production . Analysis of several independent rcd1 complementation lines expressing different levels of HA-tagged RCD1 revealed that tolerance to MV inversely correlated with the amount of expressed RCD1 ( Figure 1—figure supplements 1 and 2 ) . This suggests that RCD1 protein quantitatively lowered the resistance of the photosynthetic apparatus to ROS . Treatment with MV leads to formation of superoxide that is enzymatically dismutated to the more long-lived H2O2 . Chloroplastic production of H2O2 in the presence of MV was assessed by staining plants with 3 , 3′-diaminobenzidine ( DAB ) in light . Higher production rate of H2O2 was evident in MV pre-treated rosettes of both Col-0 and rcd1 . Longer illumination led to a time-dependent increase in the DAB staining intensity in Col-0 , but not in rcd1 ( Figure 1—figure supplement 3 ) . In several MV-tolerant mutants , the resistance is based on restricted access of MV to chloroplasts ( Hawkes , 2014 ) . However , in rcd1 MV pre-treatment led to an initial increase in H2O2 production rate similar to that in the wild type ( Figure 1—figure supplement 3 ) , suggesting that resistance of rcd1 was not due to lowered delivery of MV to PSI . To test this directly , the kinetics of PSI oxidation was assessed by in vivo spectroscopy using DUAL-PAM . As expected , pre-treatment of leaves with MV led to accelerated oxidation of PSI . This effect was identical in Col-0 and rcd1 , indicating unrestricted access of MV to PSI in the rcd1 mutant ( Figure 1B ) . The MV toxicity was not associated with the changed stoichiometry of photosystems ( Figure 1—figure supplement 4A ) . However , in Col-0 it coincided with progressive destabilization of PSII complex with its light-harvesting antennae ( LHCII ) and accumulation of PSII monomer ( Figure 1—figure supplement 4B ) . No signs of PSI inhibition were evident either in DUAL-PAM ( Figure 1B ) or in PSI immunoblotting assays ( Figure 1—figure supplement 4B ) in either genotype . The fact that production of ROS affected PSII , but not PSI where these ROS are formed , suggests that PSII inhibition results from a regulated mechanism rather than uncontrolled oxidation by ROS , and that this mechanism requires the activity of RCD1 . Previous studies have described rcd1 as a mutant with altered ROS metabolism and redox status of the chloroplasts , although the underlying mechanisms are unknown ( Fujibe et al . , 2004; Heiber et al . , 2007; Hiltscher et al . , 2014; Cui et al . , 2019 ) . No significant changes were detected in rcd1 in transcript levels of chloroplast-related genes ( Brosché et al . , 2014 ) . Analyses of the low molecular weight antioxidant compounds ascorbate and glutathione did not explain the tolerance of rcd1 to chloroplastic ROS either ( Heiber et al . , 2007; Hiltscher et al . , 2014 ) . To understand the molecular basis of the RCD1-dependent redox alterations , the levels of chloroplast proteins related to photosynthesis and ROS scavenging were analyzed by immunoblotting . None of these showed significantly altered abundance in rcd1 compared to Col-0 ( Figure 1—figure supplement 5A ) . Furthermore , no difference was detected between the genotypes in abundance and subcellular distribution of the nucleotide redox couples NAD+/NADH and NADP+/NADPH ( Figure 1—figure supplement 5B , C ) . Finally , the redox status of chloroplast thiol redox enzymes was addressed . The chloroplast stroma-localized 2-Cys peroxiredoxin ( 2-CP ) is an abundant enzyme ( König et al . , 2002; Peltier et al . , 2006; Liebthal et al . , 2018 ) that was recently found to link chloroplast thiol redox system to ROS ( Ojeda et al . , 2018; Vaseghi et al . , 2018; Yoshida et al . , 2018 ) . The level of the 2-CP protein was unchanged in rcd1 ( Figure 1—figure supplement 5A ) . However , when protein extracts were subjected to thiol bond-specific labeling ( Nikkanen et al . , 2016 ) as described in Figure 1C , most 2-CP was reduced in rcd1 both in darkness and in light , while in Col-0 the larger fraction of 2-CP was present as oxidized forms . Thus , RCD1 is likely involved in the regulation of the redox status of chloroplastic thiol enzymes . Taken together , the results hinted that the mechanisms by which RCD1 regulates chloroplastic redox status are independent of the photosynthetic ETC , or steady-state levels and distribution of nucleotide electron carriers . However , they appear to be associated with changed thiol redox state of chloroplast enzymes . It was next tested whether the nuclear RCD1 protein could itself be sensitive to ROS , thus accounting for the observed alterations . For that , an RCD1-HA complementation line was used ( line ‘a’ in Figure 1—figure supplement 1 ) . No changes were detected in RCD1-HA abundance during 5 hr amid the standard growth light period , or during 5 hr high light treatment . On the other hand , both MV and H2O2 treatments led to a gradual decrease in RCD1 abundance ( Figure 2A ) . When plant extracts from these experiments were separated in non-reducing SDS-PAGE , the RCD1-HA signal resolved into species of different molecular weights ( Figure 2B ) . Under standard growth conditions or high light , most RCD1-HA formed a reduced monomer . In contrast , treatment with MV under light or H2O2 resulted in fast conversion of RCD1-HA monomers into high-molecular-weight aggregates ( Figure 2B ) . Importantly , MV-induced redox changes in RCD1-HA only occurred in light , but not in darkness , suggesting that the changes were mediated by increased chloroplastic ROS production ( Figure 2B and Figure 4—figure supplement 2B ) . To test whether oligomerization of RCD1 was thiol-regulated , a variant of RCD1-HA was generated where seven cysteines in the linkers between the RCD1 domains were substituted by alanines ( RCD1Δ7Cys; Figure 2—figure supplement 1A ) . The treatments of rcd1: RCD1Δ7Cys-HA plants with MV or H2O2 led to significantly less aggregation of RCD1Δ7Cys-HA compared to RCD1-HA . In addition , the levels of RCD1Δ7Cys-HA were insensitive to MV or H2O2 ( Figure 2—figure supplement 1B ) . In three independent complementation lines the RCD1Δ7Cys-HA variant accumulated to higher levels compared to RCD1-HA ( Figure 2—figure supplement 1C ) . This suggests the involvement of the tested RCD1 cysteine residues in the regulation of the protein oligomerization and stability in vivo . However , the tolerance of the RCD1Δ7Cys-HA lines to chloroplastic ROS and the expression of the selected RCD1-regulated genes in response to MV treatment were comparable to that of the RCD1-HA lines or Col-0 ( Figure 2—figure supplement 1C , D ) . These results suggest that the RCD1 protein is sensitive to chloroplastic ROS . However , the changes in RCD1 abundance and redox state did not explain the RCD1-dependent redox alterations observed in the chloroplasts . In further search for the mechanisms of RCD1-dependent redox alterations in the chloroplast ( Figure 1 ) , analysis of cell energy metabolism was performed by feeding uniformly labeled [U-14C] glucose to leaf discs from light- and dark-adapted Col-0 and rcd1 plants . Distribution of radioactive label between emitted 14CO2 and fractionated plant material was analyzed . This revealed significantly more active carbohydrate metabolism in rcd1 ( Figure 3—source data 1 ) . The redistribution of radiolabel to sucrose , starch and cell wall was elevated in rcd1 as were the corresponding deduced fluxes ( Figure 3 ) , suggesting that rcd1 displayed a higher respiration rate indicative of mitochondrial defects . Indeed , earlier transcriptomic studies in rcd1 have revealed increased expression of genes encoding mitochondrial functions , including mitochondrial alternative oxidases ( AOXs ) ( Jaspers et al . , 2009; Brosché et al . , 2014 ) . Immunoblotting of protein extracts from isolated mitochondria with an antibody recognizing all five isoforms of Arabidopsis AOX confirmed the increased abundance of AOX in rcd1 ( Figure 4A ) . The most abundant AOX isoform in Arabidopsis is AOX1a . Accordingly , only a weak signal was detected in the aox1a mutant . However , in the rcd1 aox1a double mutant AOXs other than AOX1a were evident , thus the absence of RCD1 led to an increased abundance of several AOX isoforms . To test whether the high abundance of AOXs in rcd1 correlated with their increased activity , seedling respiration was assayed in vivo . Mitochondrial AOXs form an alternative respiratory pathway to the KCN-sensitive electron transfer through complex III and cytochrome c ( Figure 4B ) . Thus , after recording the initial rate of O2 uptake , KCN was added to inhibit cytochrome-dependent respiration . In Col-0 seedlings KCN led to approximately 80% decrease in O2 uptake , versus only about 20% in rcd1 , revealing elevated AOX capacity of the mutant ( Figure 4C ) . The elevated AOX capacity of rcd1 was similar to that of an AOX1a-OE overexpressor line ( Umbach et al . , 2005 ) . In the rcd1 aox1a double mutant the AOX capacity was comparable to Col-0 or aox1a ( Figure 4C ) . Thus , elevated AOX respiration of rcd1 seedlings was dependent on the AOX1a isoform . Importantly , however , metabolism of rcd1 aox1a was only slightly different from rcd1 under light and indistinguishable from rcd1 in the darkness ( Figure 3—source data 1 ) . This again indicated that the studied phenotypes of rcd1 are associated with the induction of more than one AOX isoform . Taken together , the results suggested that inactivation of RCD1 led to increased expression and activity of AOX isoforms , which could contribute to the observed changes in energy metabolism of rcd1 ( Figure 3 ) . Inhibition of complex III by antimycin A ( AA ) or myxothiazol ( myx ) activates mitochondrial retrograde signaling ( Figure 4B ) . It leads to nuclear transcriptional reprogramming including induction of AOX genes ( Clifton et al . , 2006 ) . Accordingly , overnight treatment with either of these chemicals significantly increased the abundance of AOXs in Col-0 , rcd1 and rcd1 aox1a ( Figure 4—figure supplement 1 ) . Thus , sensitivity of rcd1 to the complex III retrograde signal was not compromised , rather continuously augmented . In addition , no major effect was observed on RCD1-HA protein level or redox state in the RCD1-HA line treated with AA or myx , suggesting that RCD1 acts as a modulator , not as a mediator , of the mitochondrial retrograde signal ( Figure 4—figure supplement 2 ) . To assess whether increased AOX abundance affected chloroplast functions , PSII inhibition was assayed in the presence of MV in AA- or myx-pre-treated leaf discs . Pre-treatment of Col-0 with either AA or myx increased the resistance of PSII to inhibition by chloroplastic ROS ( Figure 4D ) , thus mimicking the rcd1 phenotype . In addition to complex III , AA has been reported to inhibit plastid cyclic electron flow dependent on PGR5 ( PROTON GRADIENT REGULATION 5 ) . Thus , pgr5 mutant was tested for its tolerance to chloroplastic ROS after AA pre-treatment . AA made pgr5 more MV-tolerant similarly to the wild type , indicating that PGR5 is not involved in the observed gain in ROS tolerance ( Figure 4—figure supplement 3A ) . Mitochondrial complex III signaling induces expression of several genes other than AOX . To test whether accumulation of AOXs contributed to PSII protection from chloroplastic ROS or merely correlated with it , the AOX inhibitor salicylhydroxamic acid ( SHAM ) was used . Treatment of plants with SHAM alone resulted in very mild PSII inhibition , which was similar in rcd1 and Col-0 ( Figure 4—figure supplement 3B ) . However , pre-treatment with SHAM made both rcd1 and Col-0 plants significantly more sensitive to chloroplastic ROS generated by MV ( Figure 4E ) , thereby partially abolishing MV tolerance of the rcd1 mutant . Involvement of the plastid terminal oxidase PTOX ( Fu et al . , 2012 ) in this effect was excluded by using the ptox mutant ( Figure 4—figure supplement 3C ) . Noteworthy , analyses of AOX1a-OE , aox1a and rcd1 aox1a lines demonstrated that AOX1a isoform was neither sufficient nor necessary for chloroplast ROS tolerance ( Figure 4—figure supplement 4 ) . Taken together , these results indicated that mitochondrial AOXs contributed to resistance of PSII to chloroplastic ROS . We hypothesize that AOX isoforms other than AOX1a are implicated in this process . The pathway linking mitochondrial AOXs with chloroplastic ROS processing is likely to involve electron transfer between the two organelles . Chlorophyll fluorescence under light ( Fs; Figure 1—figure supplement 2 ) inversely correlates with the rate of electron transfer from PSII to plastoquinone and thus can be used as a proxy of the reduction state of the chloroplast ETC . After combined treatment with SHAM and MV ( as in Figure 4E ) , Fs increased in rcd1 , but not in Col-0 ( Figure 5A ) . This hinted that a pathway in rcd1 linked the chloroplast ETC to the activity of mitochondrial AOXs , with the latter functioning as an electron sink . When the AOX activity was inhibited by SHAM , electron flow along this pathway was blocked . This led to accumulation of electrons in the chloroplast ETC and hence to the observed rise in Fs . As a parallel approach , dynamics of PSII photochemical quenching was evaluated in MV-pre-treated Col-0 and rcd1 . In both lines , this parameter dropped within the first 20 min upon exposure to light and then started to recover . Recovery was more pronounced and more suppressed by SHAM in rcd1 ( Figure 5—figure supplement 1 ) . These experiments suggest that exposure of MV-pretreated plants to light triggered an adjustment of electron flows , which was compromised by SHAM . This was in line with the involvement of AOXs in photosynthetic electron transfer and chloroplast ROS maintenance . One of the mediators of electron transfer between the organelles is the malate shuttle ( Scheibe , 2004; Zhao et al . , 2018 ) . Thus , malate concentrations were measured in total extracts from Col-0 and rcd1 seedlings . Illumination of seedlings pre-treated with MV led to dramatic decrease in malate concentration in Col-0 , but not in rcd1 ( Figure 5B ) . Noteworthy , under standard light-adapted growth conditions , the concentration and the subcellular distribution of malate was unchanged in rcd1 ( Figure 5—figure supplement 2 ) . These observations suggest that exposure to light of MV-pre-treated plants resulted in rearrangements of electron flows that were different in Col-0 and rcd1 . Next , the activity of another component of the malate shuttle , the NADPH-dependent malate dehydrogenase ( NADPH-MDH ) , was measured . Chloroplast NADPH-MDH is a redox-regulated enzyme activated by reduction of thiol bridges . Thus , the initial NADPH-MDH activity may reflect the in vivo thiol redox state of the cellular compartment from which it has been isolated . After measuring this parameter , thiol reductant was added to the extracts to reveal the total NADPH-MDH activity . Both values were higher in rcd1 than in Col-0 ( Figure 5C ) . To determine the contribution of in vivo thiol redox state , the initial NADPH-MDH activity was divided by the total activity . This value , the activation state , was also increased in rcd1 ( Figure 5C ) . Taken together , our results suggested that mitochondria contributed to ROS processing in the chloroplasts via a mechanism involving mitochondrial AOXs and possibly the malate shuttle . These processes appeared to be dynamically regulated in response to chloroplastic ROS production , and RCD1 was involved in this regulation . Our results demonstrated that absence of RCD1 caused physiological alterations in both chloroplasts and mitochondria . As RCD1 is a nuclear-localized transcriptional co-regulator ( Jaspers et al . , 2009; Jaspers et al . , 2010a ) , its involvement in retrograde signaling pathways from both organelles was assessed . Transcriptional changes observed in rcd1 ( Jaspers et al . , 2009; Brosché et al . , 2014 ) were compared to gene expression datasets obtained after perturbations in energy organelles . This revealed a striking similarity of genes differentially regulated in rcd1 to those affected by disturbed organellar function ( Figure 6—figure supplement 1 ) . Analyzed perturbations included disruptions of mitochondrial genome stability ( msh1 recA3 ) , organelle translation ( mterf6 , prors1 ) , activity of mitochondrial complex I ( ndufs4 , rotenone ) , complex III ( AA ) , and ATP synthase function ( oligomycin ) , as well as treatments and mutants related to chloroplastic ROS production ( high light , MV , H2O2 , alx8/fry1 , norflurazon ) . In particular , a significant overlap was observed between genes mis-regulated in rcd1 and the mitochondrial dysfunction stimulon ( MDS ) genes ( De Clercq et al . , 2013 ) ( Figure 6A ) . Consistently , AOX1a was among the genes induced by the majority of the treatments . To address the role of RCD1 protein in the induction of other MDS genes , mRNA steady state levels for some of them were assayed 3 hr after AA treatment ( Figure 6—figure supplement 2 ) . As expected , expression of all these genes was elevated in rcd1 under control conditions . Treatment with AA induced accumulation of MDS transcripts to similar levels in Col-0 , rcd1 , and in rcd1: RCD1-HA lines that expressed low levels of RCD1 . For one marker gene , UPOX ( UP-REGULATED BY OXIDATIVE STRESS ) , AA induction was impaired in the lines expressing high levels of RCD1-HA or RCD1Δ7Cys-HA ( Figure 6—figure supplement 2 ) . In addition to MDS , the list of genes mis-regulated in rcd1 overlapped with those affected by 3'-phosphoadenosine 5'-phosphate ( PAP ) signaling ( Estavillo et al . , 2011; Van Aken and Pogson , 2017 ) ( Figure 6A ) . Given that PAP signaling is suppressed by the activity of SAL1 , expression of PAP-regulated genes was increased in the mutants deficient in SAL1 ( alx8 and fry1 , Figure 6A and Figure 6—figure supplement 1 ) . One of the MDS genes with increased expression in rcd1 encoded the sulfotransferase SOT12 , an enzyme generating PAP . Accordingly , immunoblotting of total protein extracts with αSOT12 antibody demonstrated elevated SOT12 protein abundance in rcd1 ( Figure 6B ) . To address the functional interaction of RCD1 with PAP signaling , rcd1-4 was crossed with alx8 ( also known as sal1-8 ) . The resulting rcd1 sal1 mutant was severely affected in development ( Figure 6C ) . The effect of PAP signaling on the tolerance of PSII to chloroplastic ROS production was tested . The single sal1 mutant was more tolerant to MV than Col-0 , while under high MV concentration rcd1 sal1 was even more MV-tolerant than rcd1 ( Figure 6—figure supplement 3 ) . Together with transcriptomic similarities between rcd1 and sal1 mutants , these results further supported an overlap and/or synergy of PAP and RCD1 signaling pathways . Expression of the MDS genes is regulated by the transcription factors ANAC013 and ANAC017 ( De Clercq et al . , 2013 ) . The ANAC-responsive cis-element ( De Clercq et al . , 2013 ) was significantly enriched in promoter regions of rcd1 mis-regulated genes ( Figure 6—figure supplement 1 ) . This suggested a functional connection between RCD1 and transcriptional regulation of the MDS genes by ANAC013/ANAC017 . In an earlier study , ANAC013 was identified among many transcription factors interacting with RCD1 in the yeast two-hybrid system ( Jaspers et al . , 2009 ) . This prompted us to investigate further the connection between RCD1 and ANAC013 and the in vivo relevance of this interaction . Association of RCD1 with ANAC transcription factors in vivo was tested in two independent pull-down experiments . To identify interaction partners of ANAC013 , an Arabidopsis line expressing ANAC013-GFP ( De Clercq et al . , 2013 ) was used . ANAC013-GFP was purified with αGFP beads , and associated proteins were identified by mass spectrometry in three replicates . RCD1 and its closest homolog SRO1 , as well as ANAC017 , were identified as ANAC013 interacting proteins ( see Table 1 for a list of selected nuclear-localized interaction partners of ANAC013 , and Figure 7—source data 1 for the full list of identified proteins and mapped peptides ) . These data confirmed that ANAC013 , RCD1 and ANAC017 are components of the same protein complex in vivo . In a reciprocal pull-down assay using transgenic Arabidopsis line expressing RCD1 tagged with triple Venus YFP under the control of UBIQUITIN10 promoter , RCD1-3xVenus and interacting proteins were immunoprecipitated using αGFP ( Table 1; Figure 7—source data 2 ) . ANAC017 was found among RCD1 interactors . To test whether RCD1 directly interacts with ANAC013/ANAC017 in vivo , the complex was reconstituted in the human embryonic kidney cell ( HEK293T ) heterologous expression system ( details in Figure 7—figure supplement 1 ) . Together with the results of in vivo pull-down assays , these experiments strongly supported the formation of a complex between RCD1 and ANAC013/ANAC017 transcription factors . RCD1 interacts with many transcription factors belonging to different families ( Jaspers et al . , 2009; Jaspers et al . , 2010a; Vainonen et al . , 2012; Bugge et al . , 2018 ) via its RST domain . The strikingly diverse set of RCD1 interacting partners may be partially explained by disordered flexible regions present in the transcription factors ( Kragelund et al . , 2012; O'Shea et al . , 2017; Bugge et al . , 2018 ) . To address structural details of this interaction , the C-terminal domain of RCD1 ( residues 468–589 ) including the RST domain ( RSTRCD1; 510–568 ) was purified and labeled with 13C and 15N for NMR spectroscopic study ( Tossavainen et al . , 2017 ) ( details in Figure 7—figure supplement 2 and Figure 7—source data 3 ) . ANAC013 was shown to interact with RCD1 in yeast two-hybrid assays ( Jaspers et al . , 2009; O'Shea et al . , 2017 ) . Thus , ANAC013235-284 peptide was selected to address the specificity of the interaction of the RST domain with ANAC transcription factors using NMR ( details in Figure 7—figure supplement 3A , B ) . Binding of RCD1468-589 to ANAC013235-284 caused profound changes in the HSQC spectrum of RCD1468-589 ( Figure 7A and Figure 7—figure supplement 3C ) . These data supported a strong and specific binary interaction between the RCD1 RST domain and the ANAC013 transcription factor . To evaluate the physiological significance of this interaction , stable rcd1 complementation lines expressing an HA-tagged RCD1 variant lacking the C-terminus ( amino acids 462–589 ) were generated . The rcd1: RCD1ΔRST-HA lines were characterized by increased accumulation of AOXs in comparison with the rcd1: RCD1-HA lines ( Figure 7B ) . They also had rcd1-like tolerance of PSII to chloroplastic ROS ( Figure 7C ) . Physiological outcomes of the interaction between RCD1 and ANAC transcription factors were further tested by reverse genetics . ANAC017 regulates the expression of ANAC013 in the mitochondrial retrograde signaling cascade ( Van Aken et al . , 2016a ) . Since ANAC017 precedes ANAC013 in the regulatory pathway and because no anac013 knockout mutant is available , only the rcd1-1 anac017 double mutant was generated . In the double mutant curly leaf habitus of rcd1 was partially suppressed ( Figure 8A ) . The rcd1-1 anac017 mutant was more sensitive to chloroplastic ROS than the parental rcd1 line ( Figure 8B ) . The double mutant was characterized by lower abundance of AOX isoforms ( Figure 8C ) , dramatically decreased expression of MDS genes ( Figure 8—figure supplement 1 ) and lower AOX respiration capacity ( Figure 8D ) compared to rcd1 . Thus , gene expression , developmental , chloroplast- and mitochondria-related phenotypes of rcd1 were partially mediated by ANAC017 . These observations suggested that the in vivo interaction of RCD1 with ANAC transcription factors , mediated by the RCD1 C-terminal RST domain , is necessary for regulation of mitochondrial respiration and chloroplast ROS processing . Plant chloroplasts and mitochondria work together to supply the cell with energy and metabolites . In these organelles , ROS are formed as by-products of the electron transfer chains ( photosynthetic in chloroplasts and respiratory in mitochondria ) . ROS serve as versatile signaling molecules regulating many aspects of plant physiology such as development , stress signaling , systemic responses , and programmed cell death ( PCD ) ( Dietz et al . , 2016; Noctor et al . , 2018; Waszczak et al . , 2018 ) . This communication network also affects gene expression in the nucleus where numerous signals are perceived and integrated . However , the molecular mechanisms of the coordinated action of the two energy organelles in response to environmental cues are only poorly understood . Evidence accumulated in this and earlier studies revealed the nuclear protein RCD1 as a regulator of energy organelle communication with the nuclear gene expression apparatus . The rcd1 mutant displays alterations in both chloroplasts and mitochondria ( Fujibe et al . , 2004; Heiber et al . , 2007; Jaspers et al . , 2009; Brosché et al . , 2014; Hiltscher et al . , 2014 ) , and transcriptomic outcomes of RCD1 inactivation share similarities with those triggered by disrupted functions of both organelles ( Figure 6 ) . The results here suggest that RCD1 forms inhibitory complexes with components of mitochondrial retrograde signaling in vivo . Chloroplastic ROS appear to exhibit a direct influence on redox state and stability of RCD1 in the nucleus . These properties position RCD1 within a regulatory system encompassing mitochondrial complex III signaling through ANAC013/ANAC017 transcription factors and chloroplastic signaling by H2O2 . The existence of such an inter-organellar regulatory system , integrating mitochondrial ANAC013 and ANAC017-mediated signaling ( De Clercq et al . , 2013; Ng et al . , 2013b ) with the PAP-mediated chloroplastic signaling ( Estavillo et al . , 2011; Chan et al . , 2016; Crisp et al . , 2018 ) has been previously proposed on the basis of transcriptomic analyses ( Van Aken and Pogson , 2017 ) . However , the underlying molecular mechanisms remained unknown . Based on our results we propose that RCD1 may function at the intersection of mitochondrial and chloroplast signaling pathways and act as a nuclear integrator of both PAP and ANAC013 and ANAC017-mediated retrograde signals . RCD1 has been proposed to act as a transcriptional co-regulator because of its interaction with many transcription factors in yeast two-hybrid analyses ( Jaspers et al . , 2009 ) . The in vivo interaction of RCD1 with ANAC013 and ANAC017 revealed in this study ( Table 1 , Figures 7 and 8 ) suggests that RCD1 modulates expression of the MDS , a set of ANAC013/ANAC017 activated nuclear genes mostly encoding mitochondrial components ( De Clercq et al . , 2013 ) . ANAC013 itself is an MDS gene , thus mitochondrial signaling through ANAC013/ANAC017 establishes a self-amplifying loop . Transcriptomic and physiological data support the role of RCD1 as a negative regulator of these transcription factors ( Figures 6–8 ) . Thus , RCD1 is likely involved in the negative regulation of the ANAC013/ANAC017 self-amplifying loop and in downregulating the expression of MDS genes after their induction . Induction of genes in response to stress is commonly associated with rapid inactivation of a negative co-regulator . Accordingly , the RCD1 protein was sensitive to treatments triggering or mimicking chloroplastic ROS production . MV and H2O2 treatment of plants resulted in rapid oligomerization of RCD1 ( Figure 2 ) . Involvement of chloroplasts is indicated by the fact that MV treatment led to redox changes of RCD1-HA only in light ( Figure 2B and Figure 4—figure supplement 2B ) . In addition , little change was observed with the mitochondrial complex III inhibitors AA or myx ( Figure 4—figure supplement 2A , B ) . Together with the fact that MDS induction was not compromised in the rcd1 mutant ( Figure 4—figure supplement 1 and Figure 6—figure supplement 2 ) , this suggests that RCD1 may primarily function as a redox sensor of chloroplastic , rather than mitochondrial , ROS/redox signaling . In addition to fast redox changes , the overall level of RCD1 gradually decreased during prolonged ( 5 hr ) stress treatments . This suggests several independent modes of RCD1 regulation at the protein level . The complicated post-translational regulation of RCD1 is reminiscent of another prominent transcriptional co-regulator protein NONEXPRESSER OF PR GENES 1 ( NPR1 ) . NPR1 exists as a high molecular weight oligomer stabilized by intermolecular disulfide bonds between conserved cysteine residues . Accumulation of salicylic acid and cellular redox changes lead to the reduction of cysteines and release of NPR1 monomers that translocate to the nucleus and activate expression of defense genes ( Kinkema et al . , 2000; Mou et al . , 2003; Withers and Dong , 2016 ) . Similar to NPR1 , RCD1 has a bipartite nuclear localization signal and , in addition , a putative nuclear export signal between the WWE and PARP-like domains . Like NPR1 , RCD1 has several conserved cysteine residues . Interestingly , mutation of seven interdomain cysteines in RCD1 largely eliminated the fast in vivo effect of chloroplastic ROS on redox state and stability of RCD1; however , it did not significantly alter the plant response to MV ( Figure 2 and Figure 2—figure supplement 1C , D ) . This suggests that redox-dependent oligomerization of RCD1 may serve to fine-tune its activity . How the RCD1-dependent induction of MDS genes contributes to the energetic and signaling landscape of the plant cell remains to be investigated . Our results suggest that one component of this adaptation is the activity of mitochondrial alternative oxidases , which are part of the MDS regulon . Consequently , AOX proteins accumulate at higher amounts in rcd1 ( Figure 4 ) . Pretreatment of wild type plants with complex III inhibitors AA or myx led to elevated AOX abundance coinciding with increased tolerance to chloroplastic ROS . Moreover , the AOX inhibitor SHAM made plants more sensitive to MV , indicating the direct involvement of AOX activity in the chloroplastic ROS processing . It thus appears that AOXs in the mitochondria form an electron sink that indirectly contributes to the oxidization of the electron acceptor side of PSI . In the rcd1 mutant , this mechanism may be continuously active . The described inter-organellar electron transfer may decrease production of ROS by PSI ( asterisk in Figure 9 ) . Furthermore , chloroplastic ROS are considered the main electron sink for oxidation of chloroplast thiol enzymes ( Ojeda et al . , 2018; Vaseghi et al . , 2018; Yoshida et al . , 2018 ) . Thus , the redox status of these enzymes could depend on the proposed inter-organellar pathway . This is in line with higher reduction of the chloroplast enzymes 2-CP and NADPH-MDH observed in rcd1 ( Figure 1C and Figure 5C ) . The malate shuttle was recently shown to mediate a chloroplast-to-mitochondria electron transfer pathway that caused ROS production by complex III and evoked mitochondrial retrograde signaling ( Wu et al . , 2015; Zhao et al . , 2018 ) . Altered levels of malate and increased activity of NADPH-dependent malate dehydrogenase in rcd1 ( Figure 5 ) suggest that in this mutant the malate shuttle could act as an inter-organellar electron carrier . Another MDS gene with more abundant mRNA levels in the rcd1 mutant encodes sulfotransferase SOT12 , an enzyme involved in PAP metabolism ( Klein and Papenbrock , 2004 ) . Accordingly , SOT12 protein level was significantly increased in the rcd1 mutant ( Figure 6B ) . Accumulation of SOT12 and similarities between transcript profiles of RCD1- and PAP-regulated genes suggest that PAP signaling is likely to be constitutively active in the rcd1 mutant . Unbalancing this signaling by elimination of SAL1 leads to severe developmental defects , as evidenced by the stunted phenotype of the rcd1 sal1 double mutant . Thus , the RCD1 and the PAP signaling pathways appear to be overlapping and somewhat complementary , but the exact molecular mechanisms remain to be explored . The MDS genes represent only a fraction of genes showing differential regulation in rcd1 ( Figure 6—figure supplement 1 ) . This likely reflects the fact that RCD1 interacts with many other protein partners in addition to ANACs . The C-terminal RST domain of RCD1 was shown to interact with transcription factors belonging to DREB , PIF , ANAC , Rap2 . 4 and other families ( Jaspers et al . , 2009; Vainonen et al . , 2012; Hiltscher et al . , 2014; Bugge et al . , 2018 ) . Analyses of various transcription factors interacting with RCD1 revealed little structural similarity between their RCD1-interacting sequences ( O'Shea et al . , 2017 ) . The flexible structure of the C-terminal domain of RCD1 probably determines the specificity and ability of RCD1 to interact with those different transcription factors . This makes RCD1 a hub in the crosstalk of organellar signaling with hormonal , photoreceptor , immune and other pathways and a likely mechanism by which these pathways are integrated and co-regulated . The changing environment requires plants to readjust continuously their energy metabolism and ROS processing . On the one hand , this happens because of abiotic stress factors such as changing light intensity or temperature . For example , a sunlight fleck on a shade-adapted leaf can instantly alter excitation pressure on photosystems by two orders of magnitude ( Allahverdiyeva et al . , 2015 ) . On the other hand , chloroplasts and mitochondria are implicated in plant immune reactions to pathogens , contributing to decisive checkpoints including PCD ( Shapiguzov et al . , 2012; Petrov et al . , 2015; Wu et al . , 2015; Van Aken and Pogson , 2017; Zhao et al . , 2018 ) . In both scenarios , perturbations of organellar ETCs may be associated with increased production of ROS . However , the physiological outcomes of the two situations can be opposite: acclimation in one case and cell death in the other . The existence of molecular mechanisms that unambiguously differentiate one type of response from the other has been previously suggested ( Trotta et al . , 2014; Sowden et al . , 2018; Van Aken and Pogson , 2017 ) . The ANAC017 transcription factor and MDS genes , as well as PAP signaling , were proposed as organelle-related components counteracting PCD during abiotic stress ( Van Aken and Pogson , 2017 ) . This suggests that RCD1 is involved in the regulation of the cell fate checkpoint . Accordingly , the rcd1 mutant is resistant to a number of abiotic stress treatments ( Ahlfors et al . , 2004; Fujibe et al . , 2004; Jaspers et al . , 2009 ) . Interestingly , in contrast to its resistance to abiotic stress , rcd1 is more sensitive to treatments related to biotic stress . The rcd1 mutant was originally identified in a forward genetic screen for sensitivity to ozone ( Overmyer et al . , 2000 ) . Ozone decomposes in the plant cell wall to ROS mimicking formation of ROS by respiratory burst oxidases ( RBOHs ) in the course of plant immune reactions ( Joo et al . , 2005; Vainonen and Kangasjärvi , 2015 ) . The opposing roles of RCD1 in the cell fate may be related to its interaction with diverse transcription factor partners and/or different regulation of its stability and abundance . For example , transcriptomic analyses showed that under standard growth conditions , a cluster of genes associated with defense against pathogens had decreased expression in rcd1 ( Brosché et al . , 2014 ) , and no ANAC013/ANAC017 cis-element motif is associated with these genes ( Figure 6—figure supplement 1 ) . In agreement with its role in biotic stress , RCD1 is a target for a fungal effector protein that prevents the activation of plant immunity ( Wirthmueller et al . , 2018 ) . Another possible factor determining varying roles of RCD1 in the cell fate is differential regulation of RCD1 protein function by ROS/redox signals emitted by different subcellular compartments . The sensitivity of RCD1 to chloroplastic ROS ( Figure 2 ) can be interpreted as negative regulation of the pro-PCD component . We hypothesize that this inactivation can occur in environmental situations that require physiological adaptation rather than PCD . For example , an abrupt increase in light intensity can cause excessive electron flow in photosynthetic ETC and overproduction of reducing power . The resulting deficiency of PSI electron acceptors can lead to changes in chloroplastic ROS production , which via retrograde signaling might influence RCD1 stability and/or redox status , inhibiting its activity and thus affecting adjustments in nuclear gene expression ( Figure 9 ) . Among other processes , RCD1-mediated suppression of ANAC013/ANAC017 transcription factors is released , allowing the induction of the MDS regulon . The consequent expression of AOXs together with increased chloroplast-to-mitochondrial electron transfer is likely to provide electron sink for photosynthesis , which could suppress chloroplast ROS production and contribute to the plant’s survival under a changing environment ( Figure 9 ) . Arabidopsis thaliana adult plants were grown on soil ( peat: vermiculite = 1:1 ) in white luminescent light ( 220–250 µmol m−2 s−1 ) at a 12 hr photoperiod . Seedlings were grown for 14 days on 1 x MS basal medium ( Sigma-Aldrich ) with 0 . 5% Phytagel ( Sigma-Aldrich ) without added sucrose in white luminescent light ( 150–180 µmol m−2 s−1 ) at a 12 hr photoperiod . Arabidopsis rcd1-4 mutant ( GK-229D11 ) , rcd1-1 ( Overmyer et al . , 2000 ) , aox1a ( SAIL_030_D08 ) , AOX1a-OE ( Umbach et al . , 2005 ) , ptox ( Wetzel et al . , 1994 ) , anac017 ( SALK_022174 ) , and sal1-8 ( Wilson et al . , 2009 ) mutants are of Col-0 background; pgr5 mutant is of gl1 background ( Munekage et al . , 2002 ) . ANAC013-GFP line is described in De Clercq et al . ( 2013 ) , RCD1-HA line labeled ‘a’ in Figure 1—figure supplement 1 is described in Jaspers et al . ( 2009 ) , rcd1 aox1a double mutant – in Brosché et al . ( 2014 ) . RCD1-3xVenus , RCD1∆7Cys-HA , RCD1∆RST-HA lines are described in Cloning . The rcd1 complementation line expressing RCD1 tagged with triple HA epitope on the C-terminus was described previously ( Jaspers et al . , 2009 ) . In this line the genomic sequence of RCD1 was expressed under the control of the RCD1 native promotor ( 3505 bp upstream the start codon ) . The RCD1∆7Cys-HA construct was generated in the same way as RCD1-HA . The cysteine residues were mutated to alanines by sequential PCR-based mutagenesis of the genomic sequence of RCD1 in the pDONR/Zeo vector followed by end-joining with In-Fusion ( Clontech ) . The RCD1∆RST-HA variant was generated in the same vector by removal with a PCR reaction of the region corresponding to amino acid residues 462–589 . The resulting construct was transferred to the pGWB13 binary vector by a Gateway reaction . To generate the RCD1-3xVenus construct , RCD1 cDNA was fused to the UBIQUITIN10 promoter region and to the C-terminal triple Venus YFP tag in a MultiSite Gateway reaction as described in Siligato et al . ( 2016 ) . The vectors were introduced in the rcd1-4 mutant by floral dipping . Homozygous single insertion Arabidopsis lines were obtained . They were defined as the lines demonstrating 1:3 segregation of marker antibiotic resistance in T2 generation and 100% resistance to the marker antibiotic in T3 generation . For HEK293T cell experiments codon-optimized N-terminal 3xHA-fusion of RCD1 and C-terminal 3xmyc-fusion of ANAC013 were cloned into pcDNA3 . 1 ( + ) . Full-length ANAC017 was cloned into pcDNA3 . 1 ( - ) in the Xho I/Hind III sites , the double myc tag was introduced in the reverse primer sequence . The primer sequences used for the study are presented in Supplementary file 1 . αRCD1 specific antibody was raised in rabbit using denatured RCD1-6His protein as the antigen for immunization ( Storkbio , Estonia ) . The final serum was purified using denatured RCD1-6His immobilized on nitrocellulose membrane , aliquoted and stored at −80°C . For immunoblotting , 200 μg of total protein were loaded per well , the antibody was used in dilution 1: 500 . For PSII inhibition studies , leaf discs were let floating on Milli-Q water solution supplemented with 0 . 05% Tween 20 ( Sigma-Aldrich ) . Final concentration of AA and myx was 2 . 5 μM each , of SHAM – 2 mM . For transcriptomic experiments , plant rosettes were sprayed with water solution of 50 μM AA complemented with 0 . 01% Silwet Gold ( Nordisk Alkali ) . Stock solutions of these chemicals were prepared in DMSO , equal volumes of DMSO were added to control samples . Pre-treatment with chemicals was carried out in the darkness , overnight for MV , AA and myx , 1 hr for SHAM . After spraying plants with 50 μM AA they were incubated in growth light for 3 hr . For chemical treatment in seedlings grown on MS plates , 5 mL of Milli-Q water with or without 50 µM MV were poured in 9 cm plates at the end of the light period . The seedlings were kept in the darkness overnight , and light treatment was performed on the following morning . For H2O2 treatment , the seedlings were incubated in 5 mL of Milli-Q water with or without 100 mM H2O2 in light . Plant rosettes were stained with 3 , 3′-diaminobenzidine ( DAB ) essentially as described in Daudi et al . ( 2012 ) . After vacuum infiltration of DAB-staining solution in the darkness , rosettes were exposed to light ( 180 µmol m−2 s−1 ) for 20 min to induce production of chloroplastic ROS and then immediately transferred to the bleaching solution . Chlorophyll fluorescence was measured by MAXI Imaging PAM ( Walz , Germany ) . PSII inhibition protocol consisted of repetitive 1 hr periods of blue actinic light ( 450 nm , 80 µmol m−2 s−1 ) each followed by a 20 min dark adaptation , then Fo and Fm measurement . PSII photochemical yield was calculated as Fv/Fm = ( Fm-Fo ) /Fm ( Figure 1—figure supplement 2 ) . To plot raw chlorophyll fluorescence kinetics under light ( Fs ) against time , the reads were normalized to dark-adapted Fo . For the measurements of photochemical quenching , Fm’ was measured with saturating pulses triggered against the background of actinic light ( 450 nm , 80 µmol m−2 s−1 ) , and the following formulae were used: qP = ( Fm' - Fs ) / ( Fm'-Fo' ) , where Fo' ≈ Fo / ( ( ( Fm – Fo ) /Fm ) + ( Fo/Fm' ) ) ( Oxborough and Baker , 1997 ) . The assays were performed in 96-well plates . In each assay , leaf discs from at least four individual plants were analyzed . Each assay was reproduced at least three times . PSI ( P700 ) oxidation was measured by DUAL-PAM-100 ( Walz , Germany ) as described ( Tiwari et al . , 2016 ) . Leaves were pre-treated in 1 µM MV for 4 hr , then shifted to light ( 160 µmol m−2 s−1 ) for indicated time . Oxidation of P700 was induced by PSI-specific far red light ( FR , 720 nm ) . To determine fully oxidized P700 ( Pm ) , a saturating pulse of actinic light was applied under continuous background of FR , followed by switching off both the actinic and FR light . The kinetics of P700+ reduction by intersystem electron transfer pool and re-oxidation by FR was determined by using a multiple turnover saturating flash of PSII light ( 635 nm ) in the background of continuous FR . Thylakoids were isolated as described in Järvi et al . ( 2016 ) . Chlorophyll content was determined according to Porra et al . ( 1989 ) and protein content according to Lowry et al . ( 1951 ) . For immunoblotting of total plant extracts , the plant material was frozen immediately after treatments in liquid nitrogen and ground . Total proteins were extracted in SDS extraction buffer [50 mM Tris-HCl ( pH 7 . 8 ) , 2% SDS , 1 x protease inhibitor cocktail ( Sigma-Aldrich ) , 2 mg/mL NaF] for 20 min at 37°C and centrifuged at 18 000 x g for 10 min . Supernatants were normalized for protein concentration and resolved by SDS-PAGE . For separation of proteins , SDS-PAGE ( 10–12% polyacrylamide ) was used ( Laemmli , 1970 ) . For thylakoid proteins , the gel was complemented with 6 M urea . To separate thylakoid membrane protein complexes , isolated thylakoids were solubilized with n-dodecyl β-D-maltoside ( Sigma-Aldrich ) and separated in BN-PAGE ( 5–12 . 5% polyacrylamide ) as described by Järvi et al . ( 2016 ) . After electrophoresis , proteins were electroblotted to PVDF membrane and immunoblotted with specific antibodies . αSOT12 antibodies have Agrisera reference number AS16 3943 . For quantification of immunoblotting signal , ImageJ software was used ( https://imagej . nih . gov/ij/ ) . Thiol redox state of 2-CPs in detached Col-0 and rcd1 leaves adapted to darkness or light ( 3 hr of 160 µmol m−2 s−1 ) , was determined by alkylating free thiols in TCA-precipitated proteins with 50 mM N-ethylmaleimide in the buffer containing 8 M urea , 100 mM Tris‐HCl ( pH 7 . 5 ) , 1 mM EDTA , 2% SDS , and 1/10 of protease inhibitor cocktail ( Thermo Scientific ) , reducing in vivo disulfides with 100 mM DTT and then alkylating the newly reduced thiols with 10 mM methoxypolyethylene glycol maleimide of molecular weight 5 kDa ( Sigma-Aldrich ) , as described in Nikkanen et al . ( 2016 ) . Proteins were then separated by SDS-PAGE and immunoblotted with a 2-CP-specific antibody . Leaves of Arabidopsis plants were harvested in the middle of the light period and snap-frozen in liquid nitrogen . Four grams of fresh weight of frozen plant material was ground to a fine powder using a mixer mill ( Retsch ) , transferred to Falcon tubes and freeze-dried at 0 . 02 bar for 5 days in a lyophilizer , which had been pre-cooled to −40°C . The NAF-fractionation procedure was performed as described in Krueger et al . ( 2011 ) , Arrivault et al . ( 2014 ) , and Krueger et al . ( 2014 ) , except that the gradient volume , composed of the solvents tetrachloroethylene ( C2Cl4 ) /heptane ( C7H16 ) , was reduced from 30 mL to 25 mL but with the same linear density . Leaf powder was resuspended in 20 mL C2Cl4/C7H16 mixture 66:34 ( v/v; density ρ = 1 . 3 g cm−3 ) , and sonicated for 2 min , with 6 × 10 cycles at 65% power . The sonicated suspension was filtered through a nylon net ( 20 μm pore size ) . The net was washed with 30 mL of heptane . The suspension was centrifuged for 10 min at 3 200 x g at 4°C and the pellet was resuspended in 5 mL C2Cl4/C7H16 mixture 66:34 . The gradient was formed in 38 mL polyallomer centrifugation tube using a peristaltic gradient pump ( BioRad ) generating a linear gradient from 70% solvent A ( C2Cl4/C7H16 mixture 66:34 ) to 100% solvent B ( 100% C2Cl4 ) with a flow rate of 1 . 15 mL min−1 , resulting in a density gradient from 1 . 43 g cm−3 to 1 . 62 g cm−3 . Five mL suspension containing the sample was loaded on top of the gradient and centrifuged for 55 min at 5 000 x g at 4°C using a swing-out rotor with acceleration and deceleration of 3:3 ( brakes off ) . Each of the compartment-enriched fractions ( F1 to F8 ) were transferred carefully from the top of the gradient into a 50 mL Falcon tube , filled up with heptane to a volume of 20 mL and centrifuged at 3 200 x g for 10 min . The pellet was resuspended in 6 mL of heptane and subsequently divided into 6 aliquots of equal volume ( 950 μL ) . The pellets had been dried in a vacuum concentrator without heating and stored at −80°C until further use . Subcellular compartmentation of markers or the metabolites of our interest was calculated by BestFit method as described in Krueger et al . ( 2011 ) and Krueger et al . ( 2014 ) . Percentage values ( % of the total found in all fractions ) of markers and metabolites have been used to make the linear regressions for subcellular compartments using BestFit . Before enzyme and metabolite measurements , dried pellets were homogenized in the corresponding extraction buffer by the addition of one steel ball ( 2 mm diameter ) to each sample and shaking at 25 Hz for 1 min in a mixer mill . Enzyme extracts were prepared as described in Gibon et al . ( 2004 ) with some modifications . The extraction buffer contained 50 mM HEPES-KOH ( pH 7 . 5 ) , 10 mM MgCl2 , 1 mM EDTA , 1 mM EGTA , 1 mM benzamidine , 1 mM ε-aminocaproic acid , 0 . 25% ( w/v ) BSA , 20 μM leupeptin , 0 . 5 mM DTT , 1 mM phenylmethylsulfonyl fluoride ( PMSF ) , 1% ( v/v ) Triton X-100 , 20% glycerol . The extract was centrifuged ( 14 000 rpm at 4°C for 10 min ) and the supernatant was used directly for the enzymatic assays . The activities of adenosine diphosphate glucose pyrophosphorylase ( AGPase ) and phosphoenolpyruvate carboxylase ( PEPC ) were determined as described in Gibon et al . ( 2004 ) but without using the robot-based platform . Chlorophyll was extracted twice with 80% ( v/v ) and once with 50% ( v/v ) hot ethanol/10 mM HEPES ( pH 7 . 0 ) followed by 30 min incubation at 80°C and determined as described in Cross et al . ( 2006 ) . Nitrate was measured by the enzymatic reaction as described in Cross et al . ( 2006 ) . For the light experiment , leaf discs were incubated in light in 5 mL 10 mM MES-KOH ( pH 6 . 5 ) , containing 1 . 85 MBq/mmol [U-14C] glucose ( Hartmann Analytic ) in a final concentration of 2 mM . In the dark experiment , leaf discs were incubated under green light for 150 min . Leaf discs were placed in a sieve , washed several times in double-distilled water , frozen in liquid nitrogen , and stored at −80°C until further analysis . All incubations were performed in sealed flasks under green light and shaken at 100 rpm . The evolved 14CO2 was collected in 0 . 5 mL of 10% ( w/v ) KOH . Extraction and fractionation were performed according to Obata et al . ( 2017 ) . Frozen leaf discs were extracted with 80% ( v/v ) ethanol at 80°C ( 1 mL per sample ) and re-extracted in two subsequent steps with 50% ( v/v ) ethanol ( 1 mL per sample for each step ) , and the combined supernatants were dried under an air stream at 35°C and resuspended in 1 mL of water ( Fernie et al . , 2001 ) . The soluble fraction was subsequently separated into neutral , anionic , and basic fractions by ion-exchange chromatography; the neutral fraction ( 2 . 5 mL ) was freeze-dried , resuspended in 100 μL of water , and further analyzed by enzymatic digestion followed by a second ion-exchange chromatography step ( Carrari et al . , 2006 ) . To measure phosphate esters , samples ( 250 μL ) of the soluble fraction were incubated in 50 μL of 10 mM MES-KOH ( pH 6 . 0 ) , with or without 1 unit of potato acid phosphatase ( grade II; Boehringer Mannheim ) for 3 hr at 37°C , boiled for 2 min , and analyzed by ion-exchange chromatography ( Fernie et al . , 2001 ) . The insoluble material left after ethanol extraction was homogenized , resuspended in 1 mL of water , and counted for starch ( Fernie et al . , 2001 ) . Fluxes were calculated as described following the assumptions detailed by Geigenberger et al . ( 1997 ) and Geigenberger et al . ( 2000 ) . Unfortunately , the discontinued commercial availability of the required positionally radiolabeled glucoses prevented us from analyzing fermentative fluxes more directly . Crude mitochondria were isolated from Arabidopsis rosette leaves as described in Keech et al . ( 2005 ) . Seedling respiration and AOX capacity were assessed by measuring O2 consumption in the darkness using a Clark electrode as described in Schwarzländer et al . ( 2009 ) . Primary metabolites were analyzed with GC-MS according to Roessner et al . ( 2000 ) . GC-MS analysis was executed from the plant extracts of eight biological replicates ( pooled samples ) . Plant material was homogenized in a Qiagen Tissuelyser II bead mill ( Qiagen , Germany ) with 1–1 . 5 mm Retsch glass beads . Soluble metabolites were extracted from plant material in two steps , first with 1 mL of 100% methanol ( Merck ) and second with 1 mL of 80% ( v/v ) aqueous methanol . During the first extraction step , 5 µL of internal standard solution ( 0 . 2 mg mL−1 of benzoic-d5 acid , 0 . 1 mg mL−1 of glycerol-d8 , 0 . 2 mg mL−1 of 4-methylumbelliferone in methanol ) was added to each sample . During both extraction steps , the samples were vortexed for 30 min and centrifuged for 5 min at 13 000 rpm ( 13 500 × g ) at 4°C . The supernatants were then combined for metabolite analysis . The extracts ( 2 mL ) were dried in a vacuum concentrator ( MiVac Duo , Genevac Ltd , Ipswich , UK ) , the vials were degassed with nitrogen and stored at −80°C prior to derivatization and GC-MS analysis . Dried extracts were re-suspended in 500 µL of methanol . Aliquot of 200 µL was transferred to a vial and dried in a vacuum . The samples were derivatized with 40 µL of methoxyamine hydrochloride ( MAHC , Sigma-Aldrich ) ( 20 mg mL−1 ) in pyridine ( Sigma-Aldrich ) for 90 min at 30°C at 150 rpm , and with 80 µL N-methyl-N- ( trimethylsilyl ) trifluoroacetamide with 1% trimethylchlorosilane ( MSTFA with 1% TMCS , Thermo Scientific ) for 120 min at 37°C at 150 rpm . Alkane series ( 10 µL , C10–C40 , Supelco ) in hexane ( Sigma-Aldrich ) and 100 µL of hexane was added to each sample before GC-MS analysis . The GC-MS system consisted of Agilent 7890A gas chromatograph with 7000 Triple quadrupole mass spectrometer and GC PAL autosampler and injector ( CTC Analytics ) . Splitless injection ( 1 µL ) was employed using a deactivated single tapered splitless liner with glass wool ( Topaz , 4 mm ID , Restek ) . Helium flow in the column ( Agilent HP-5MS Ultra Inert , length 30 m , 0 . 25 mm ID , 0 . 25 μm film thickness combined with Agilent Ultimate Plus deactivated fused silica , length 5 m , 0 . 25 mm ID ) was 1 . 2 mL min−1 and purge flow at 0 . 60 min was 50 mL min−1 . The injection temperature was set to 270°C , MS interface 180°C , source 230°C and quadrupole 150°C . The oven temperature program was as follows: 2 min at 50°C , followed by a 7 °C min−1 ramp to 260°C , 15 °C min−1 ramp to 325°C , 4 min at 325°C and post-run at 50°C for 4 . 5 min . Mass spectra were collected with a scan range of 55–550 m/z . Metabolite Detector ( versions 2 . 06 beta and 2 . 2N ) ( Hiller et al . , 2009 ) and AMDIS ( version 2 . 68 , NIST ) were used for deconvolution , component detection and quantification . Malate levels were calculated as the peak area of the metabolite normalized with the peak area of the internal standard , glycerol-d8 , and the fresh weight of the sample . From light-adapted plants grown for 5 weeks ( 100–120 µmol m−2 s−1 at an 8 hour day photoperiod ) , total extracts were prepared as for non-aqueous fractionation in the extraction buffer supplemented with 250 µM DTT . In microplates , 5 µL of the extract ( diluted x 500 ) were mixed with 20 µL of activation buffer , 0 . 1 M Tricine-KOH ( pH 8 . 0 ) , 180 mM KCl , 0 . 5% Triton X-100 ) . Initial activity was measured immediately after , while total activity was measured after incubation for 2 hr at room temperature in presence of additional 150 mM DTT . Then assay mix was added consisting of 20 µL of assay buffer [0 . 5 M Tricine-KOH ( pH 8 . 0 ) , 0 . 25% Triton X-100 , 0 . 5 mM EDTA] , 9 µL of water , and 1 µL of 50 mM NADPH ( prepared in 50 mM NaOH ) , after which 45 µL of 2 . 5 mM oxaloacetate or water control was added . The reaction was mixed , and light absorbance at 340 nm wavelength was measured at 25°C . Genes with misregulated expression in rcd1 were selected from our previous microarray datasets ( Brosché et al . , 2014 ) with the cutoff , absolute value of logFC <0 . 5 . These genes were subsequently clustered with the rcd1 gene expression dataset together with various Affymetrix datasets related to chloroplast or mitochondrial dysfunction from the public domain using bootstrapped Bayesian hierarchical clustering as described in Wrzaczek et al . ( 2010 ) . Affymetrix raw data ( . cel files ) were normalized with Robust Multi-array Average normalization , and manually annotated to control and treatment conditions , or mutant versus wild type . Affymetrix ATH1-121501 data were from the following sources: Gene Expression Omnibus https://www . ncbi . nlm . nih . gov/geo/ , AA 3 hr ( in figures labelled as experiment 1 ) , GSE57140 ( Ivanova et al . , 2014 ) ; AA and H2O2 , 3 hr treatments ( in figures labelled as experiment 2 ) , GSE41136 ( Ng et al . , 2013b ) ; MV 3 hr , GSE41963 ( Sharma et al . , 2013 ) ; mterf6-1 , GSE75824 ( Leister and Kleine , 2016 ) ; prors1-2 , GSE54573 ( Leister et al . , 2014 ) ; H2O2 30 min , GSE43551 ( Gutiérrez et al . , 2014 ) ; high light 1 hr ( in figures labelled as experiment 1 ) , GSE46107 ( Van Aken et al . , 2013 ) ; high light 30 min in cell culture , GSE22671 ( González-Pérez et al . , 2011 ) ; high light 3 hr ( in figures labelled as experiment 2 ) , GSE7743 ( Kleine et al . , 2007 ) ; oligomycin 1 and 4 hr , GSE38965 ( Geisler et al . , 2012 ) ; norflurazon – 5 day-old seedlings grown on plates with norflurazon , GSE12887 ( Koussevitzky et al . , 2007 ) ; msh1 recA3 double mutant , GSE19603 ( Shedge et al . , 2010 ) . AtGenExpress oxidative time series , MV 12 and 24 hr , http://www . arabidopsis . org/servlets/TairObject ? type=expression_set&id=1007966941 . ArrayExpress , https://www . ebi . ac . uk/arrayexpress/: rotenone , 3 and 12 hr , E-MEXP-1797 ( Garmier et al . , 2008 ) ; alx8 and fry1 , E-MEXP-1495 ( Wilson et al . , 2009 ) ; ndufs4 , E-MEXP-1967 ( Meyer et al . , 2009 ) . Quantitative PCR was performed essentially as described in Brosché et al . ( 2014 ) . The data were normalized with three reference genes , PP2AA3 , TIP41 and YLS8 . Relative expression of the genes RCD1 , AOX1a , UPOX , ANAC013 , At5G24640 and ZAT12 was calculated in qBase +3 . 2 ( Biogazelle , https://www . qbaseplus . com/ ) . The primer sequences and primer efficiencies are presented in Supplementary file 1 . Immunoprecipitation experiments were performed in three biological replicates as described previously ( De Rybel et al . , 2013 ) , using 3 g of rosette leaves from p35S: ANAC013-GFP and 2 . 5 g of rosette leaves from pUBI10: RCD1-3xVenus transgenic lines . Interacting proteins were isolated by applying total protein extracts to αGFP-coupled magnetic beads ( Milteny Biotech ) . Three replicates of p35S: ANAC013-GFP or pUBI10: RCD1-3xVenus were compared to three replicates of Col-0 controls . Tandem mass spectrometry ( MS ) and statistical analysis using MaxQuant and Perseus software was performed as described previously ( Wendrich et al . , 2017 ) . HEK293T cells were maintained at 37°C and 5% CO2 in Dulbecco’s Modified Eagle’s Medium F12-HAM , supplemented with 10% fetal bovine serum , 15 mM HEPES , and 1% penicillin/streptomycin . Cells were transiently transfected using GeneJuice ( Novagen ) according to the manufacturer’s instructions . For co-immunoprecipitation experiments , HEK293T cells were co-transfected with plasmids encoding HA-RCD1 and ANAC013-myc or ANAC017-myc . Forty hours after transfection , cells were lysed in TNE buffer [50 mM Tris-HCl ( pH 7 . 4 ) , 150 mM NaCl , 5 mM EDTA , 1% Triton X-100 , 1 x protease inhibitor cocktail , 50 µM proteasome inhibitor MG132 ( Sigma-Aldrich ) ] . After incubation for 2 hr at 4°C , lysates were cleared by centrifugation at 18 000 x g for 10 min at 4°C . For co-immunoprecipitation , cleared cell lysates were incubated with either αHA or αmyc antibody immobilized on agarose beads overnight at 4°C . Beads were washed six times with the lysis buffer . The bound proteins were dissolved in SDS sample buffer , resolved by SDS-PAGE , and immunoblotted with the specified antibodies . The C-terminal domain of RCD1 for NMR study was expressed as GST-fusion protein in E . coli BL21 ( DE3 ) Codon Plus strain and purified using GSH-Sepharose beads ( GE Healthcare ) according to the manufacturer’s instruction . Cleavage of GST tag was performed with thrombin ( GE Healthcare , 80 units per mL of beads ) for 4 hr at room temperature and the C-terminal domain of RCD1 was eluted from the beads with PBS buffer ( 137 mM NaCl , 2 . 7 mM KCl , 10 mM Na2HPO4 , 1 . 8 mM KH2PO4 , pH 7 . 4 ) . The protein was further purified by gel filtration with HiLoad 16/600 Superdex 75 column ( GE Healthcare ) equilibrated with 20 mM sodium phosphate buffer ( pH 6 . 4 ) , 50 mM NaCl at 4°C . ANAC013 peptides of >98% purity for surface plasmon resonance and NMR analysis were purchased from Genecust , dissolved in water to 5 mM final concentration and stored at −80°C before analyses . The C-terminal domain of RCD1 was covalently coupled to a Biacore CM5 sensor chip via amino-groups . 500 nM of ANAC013 peptides were then profiled at a flow rate of 30 µL min−1 for 300 s , followed by 600 s flow of running buffer . Analysis was performed at 25°C in the running buffer containing 10 mM HEPES ( pH 7 . 4 ) , 150 mM NaCl , 3 mM EDTA , 0 . 05% surfactant P20 ( Tween-20 ) . After analysis in BIAevaluation ( Biacore ) software , the normalized resonance units were plotted over time with the assumption of one-to-one binding . NMR sample production and chemical shift assignment have been described in Tossavainen et al . ( 2017 ) . A Bruker Avance III HD 800 MHz spectrometer equipped with a TCI 1H/ 13C/ 15N cryoprobe was used to acquire spectra for structure determination of RCD1468-589 . Peaks were manually picked from three NOE spectra , a 1H , 15N NOESY-HSQC and 1H , 13C NOESY-HSQC spectra for the aliphatic and aromatic 13C regions . CYANA 2 . 1 ( López-Méndez and Güntert , 2006 ) automatic NOE peak assignment – structure calculation routine was used to generate 300 structures from which 30 were further refined in explicit water with AMBER 16 ( Case et al . , 2005 ) . Assignments of three NOE peaks were kept fixed using the KEEP subroutine in CYANA . These NOE peaks restrained distances between the side chains of W507 and M508 and adjacent helices 1 and 4 , respectively . Fifteen lowest AMBER energy structures were chosen to represent of RCD1468-589 structure in solution . Peptide binding experiment was carried out by preparing a sample containing RCD1468-589 and ANAC013235-284 peptides in an approximately 1:2 concentration ratio , and recording a 1H , 15N HSQC spectrum . Amide peak positions were compared with those of the free RCD1468-589 .	Most plant cells contain two types of compartments , the mitochondria and the chloroplasts , which work together to supply the chemical energy required by life processes . Genes located in another part of the cell , the nucleus , encode for the majority of the proteins found in these compartments . At any given time , the mitochondria and the chloroplasts send specific , ‘retrograde’ signals to the nucleus to turn on or off the genes they need . For example , mitochondria produce molecules known as reactive oxygen species ( ROS ) if they are having problems generating energy . These molecules activate several regulatory proteins that move into the nucleus and switch on MDS genes , a set of genes which helps to repair the mitochondria . Chloroplasts also produce ROS that can act as retrograde signals . It is still unclear how the nucleus integrates signals from both chloroplasts and mitochondria to ‘decide’ which genes to switch on , but a protein called RCD1 may play a role in this process . Indeed , previous studies have found that Arabidopsis plants that lack RCD1 have defects in both their mitochondria and chloroplasts . In these mutant plants , the MDS genes are constantly active and the chloroplasts have problems making ROS . To investigate this further , Shapiguzov , Vainonen et al . use biochemical and genetic approaches to study RCD1 in Arabidopsis . The experiments confirm that this protein allows a dialog to take place between the retrograde signals of both mitochondria and chloroplasts . On one hand , RCD1 binds to and inhibits the regulatory proteins that usually activate the MDS genes under the control of mitochondria . This explains why , in the absence of RCD1 , the MDS genes are always active , which is ultimately disturbing how these compartments work . On the other hand , RCD1 is also found to be sensitive to the ROS that chloroplasts produce . This means that chloroplasts may be able to affect when mitochondria generate energy by regulating the protein . Finally , further experiments show that MDS genes can affect both mitochondria and chloroplasts: by influencing how these genes are regulated , RCD1 therefore acts on the two types of compartments . Overall , the work by Shapiguzov , Vainonen et al . describes a new way Arabidopsis coordinates its mitochondria and chloroplasts . Further studies will improve our understanding of how plants regulate these compartments in different environments to produce the energy they need . In practice , this may also help plant breeders create new varieties of crops that produce energy more efficiently and which better resist to stress .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology" ]	2019	Arabidopsis RCD1 coordinates chloroplast and mitochondrial functions through interaction with ANAC transcription factors
Although it is well known that long-term synaptic plasticity can be expressed both pre- and postsynaptically , the functional consequences of this arrangement have remained elusive . We show that spike-timing-dependent plasticity with both pre- and postsynaptic expression develops receptive fields with reduced variability and improved discriminability compared to postsynaptic plasticity alone . These long-term modifications in receptive field statistics match recent sensory perception experiments . Moreover , learning with this form of plasticity leaves a hidden postsynaptic memory trace that enables fast relearning of previously stored information , providing a cellular substrate for memory savings . Our results reveal essential roles for presynaptic plasticity that are missed when only postsynaptic expression of long-term plasticity is considered , and suggest an experience-dependent distribution of pre- and postsynaptic strength changes . In layer-5 pyramidal to pyramidal cell synapses , timing-dependent LTD is presynaptically expressed . It is mediated by the coincidence between a postsynaptic signal ( eCB release ) and a presynaptic signal ( presynaptic NMDA receptor activation ) ( Sjöström et al . , 2003 , 2004; Bender and Feldman , 2006; Yang and Calakos , 2013 ) . LTP is driven by postsynaptic coincidence detection of the combined binding of glutamate and postsynaptic depolarization ( Bender and Feldman , 2006; Sjöström et al . , 2007; Shouval et al . , 2010 ) , promoting an increase in the number and/or properties of postsynaptic AMPA receptors ( Malinow and Malenka , 2002 ) . However , timing-dependent LTP also has a presynaptic component , mediated by postsynaptic diffusion of NO ( Hardingham and Fox , 2006; Sjöström et al . , 2007; Hardingham et al . , 2013; Yang and Calakos , 2013 ) . Our phenomenological triplet model of long-term modification of pre- and postsynaptic components has three synaptic traces , two postsynaptic ( y+ and y− ) and one presynaptic ( x+ ) , which increase upon a post- or presynaptic spike , respectively ( see Appendix 1 for a more detailed comparison with the triplet model ( Pfister and Gerstner , 2006 ) ) . The traces are obtained by filtering the spike trains with a first-order low-pass filter . We defined the postsynaptic depression trace ( 3 ) dy− ( t ) dt=−y− ( t ) τy−+Y ( t ) , the postsynaptic potentiation trace ( 4 ) dy+ ( t ) dt=−y+ ( t ) τy++Y ( t ) , and the presynaptic potentiation trace ( 5 ) dx+ ( t ) dt=−x+ ( t ) τx++X ( t ) . The long-term modification in the weight is achieved by modifying the postsynaptic factor q and the presynaptic factor P . The postsynaptic factor is modified with every postsynaptic spike Y according to ( 6 ) Δq=c+x+ ( t ) y_ ( t−ϵ ) Y ( t ) ⏟TripletpostLTP , where c+ is a constant that sets the amount of postsynaptic LTP . The y− trace is evaluated at ( t − ϵ ) , so that the value of the respective synaptic trace is readout before being updated . The triplet character of this rule is expressed by the fact that it contains the presynaptic component once , but the postsynaptic activity twice ( Y and filtered version y− ) . This ensures that LTP only takes place when the postsynaptic spike follows both a presynaptic spike and a preceding postsynaptic spike ( Pfister and Gerstner , 2006 ) . As a result , low pairing frequencies do not lead to LTP , as y− will have decayed , consistent with data ( Sjöström et al . , 2001 ) . Similarly , the presynaptic factor is modified whenever the presynaptic cell is active according to ( 7 ) ΔP=−d_y_ ( t ) y+ ( t ) X ( t ) ⏟TripletpreLTD+d+x+ ( t−ϵ ) y+ ( t ) X ( t ) ⏟TripletpreLTP . For plasticity in P to occur , the presynaptic spikes X readout the postsynaptic traces ( presynaptic coincidence detection ) , y−y+ for presynaptic LTD and x+y+ for presynaptic LTP . d− and d+ are constants that set the amount of presynaptic LTD and LTP , respectively . While presynaptic LTD has a triplet form , it contains two postsynaptic traces and the raw presynaptic spike train . Therefore it does not vanish at low frequencies . Equivalently , this term could be written as a doublet rule with a double exponential as the presynaptic trace . The total synaptic strength is a product of both pre- and postsynaptic factors ( 8 ) w ( t ) =qp ( t ) r ( t ) . For a synapse that has not been stimulated recently this simplifies to w = Pq . Being a probability we hard-bounded P = [0 , 1] . The postsynaptic factor q had a lower bound of 0 , and an upper bound of 2 . Alternatively a soft-bounded rule could be used ( van Rossum et al . , 2012 ) . In the data used to fit the model ( see below ) , postsynaptic homosynaptic LTD was not apparent on the timescale of the experiment . Because it seems unrealistic that the postsynaptic factor q never decreases , slow homeostasic scaling of the postsynaptic factor was included for network simulations ( Turrigiano et al . , 1998 ) . This prevents weakly active synapses from potentiating the postsynaptic factor q . It was modelled as a postsynaptic subtractive normalization , so that the total change in q of synapse i was equal to Δqi−α1N∑j=1NΔqj ( Miller and MacKay , 1994 ) . The only condition on the speed α for it to be consistent with the data , is that it should not lead to noticable homeostasis on the timescale of the experiments . For computational efficiency we used α = 0 . 075 , which is still orders of magnitude faster than what has been observed in homeostasis experiments . The exact form of slow normalization ( α → 0 ) does not affect the qualitative behavior of the model . Note that the timescale of the slow normalization determines how long the memory savings effects are present . To speed up the numerical implementations , we integrated the synaptic traces between the pre- and postsynaptic spikes . In the following equations , we label the presynaptic spikes with k and the postsynaptic ones with l . ( 9 ) y−l+1=y− l exp ( −Δtpostτy− ) +1 , ( 10 ) y+l+1=y+ l exp ( −Δtpostτy+ ) +1 , ( 11 ) x+k+1=x+k exp ( −Δtpreτx+ ) +1 . We subsequently integrated the model between pre- and postsynaptic spikes ( 12 ) ql+1=ql+c+x+ k exp ( −Δtpost−preτx+ ) y− l exp ( −Δtpostτy− ) , ( 13 ) Pk+1=Pk−d−y−l exp ( −Δtpre−postτy− ) y+ l exp ( −Δtpre−postτy+ ) +d+y+ l exp ( −Δtpre−postτy+ ) x+ k exp ( −Δtpreτx+ ) , where Δtpost−pre is the time between the current postsynaptic spike and the last presynaptic spike , Δtpost is the time between the current postsynaptic spike and the last one , and similarly for Δtpre−post and Δtpre . Finally , we also integrated the STP ( Equations 1 , 2 ) between presynaptic spikes k and k + 1 , a time Δtpre apart , yielding ( 14 ) rk+1=1−[1−rk ( 1−pk ) ]exp ( −ΔtpreD ) , ( 15 ) pk+1=P+pk[1−P]exp ( −ΔtpreF ) . with initial conditions r0 = 1 and p0 = P . We fitted the free parameters of the long-term plasticity model θ = {d− , τy− , d+ , τy+ , c+ , τx+} to the frequency- and timing-dependent slice STDP data of layer-5 pyramidal cells ( Sjöström et al . , 2001 ) . Parameters are shown in Table 1 . Rather than fitting to changes in the weight w , we fitted directly to modifications in P and q ( see Equations 21 , 22 for our estimators of P and q ) . This was done by minimizing the mean squared error between the data and the experiments for both P and q ( as shown in Figure 1 ) ( 16 ) θ=argminθ1N∑jN[ ( PmodelafterPmodelbefore−PdataafterPdatabefore ) 2+ ( qmodelafterqmodelbefore−qdataafterqdatabefore ) 2] , where N denotes the number of protocols fitted , 10 in total ( 5 different pairing frequencies with −10 ms or +10 ms relative timing , see below ) . For induction protocols at high frequencies ( ≥10 Hz ) , pre- and postsynaptic spike trains consisted of five spikes that were paired 15 times at 0 . 1 Hz . Low-frequency pairings ( 0 . 1 Hz ) were done with a single pre- and postsynaptic spike ( as in Sjöström et al . , 2001 ) . Before plasticity induction , P and q were set to 0 . 5 and 1 , respectively . For the interaction of STP and STDP simulations ( Figure 1F , G ) , we used a standard passive neuron model with a membrane time constant of 25 ms . 10 . 7554/eLife . 09457 . 012Table 1 . Unified pre- and postsynaptic spike-timing-dependent plasticity ( STDP ) model parametersDOI: http://dx . doi . org/10 . 7554/eLife . 09457 . 012Parameterd−τy− ( ms ) d+τy+ ( ms ) c+τx+ ( ms ) Young rat visual cortex0 . 177132 . 70 . 1548230 . 20 . 061866 . 6The model was fitted to data from young rat visual cortex ( Sjöström et al . , 2001 ) . 10 . 7554/eLife . 09457 . 013Table 2 . Comparison between unified pre- and postsynaptic STDP model and different versions of the triplet model ( for simplicity we removed the function arguments ) ( Pfister and Gerstner , 2006 ) DOI: http://dx . doi . org/10 . 7554/eLife . 09457 . 013LTDLTP1LTP2pre-post STDPX d−y−y+X d+y+x+Y c+x+y−minimal HC TripletX A2−y1Y A2+x1Y A3+x1y2minimal VC TripletX A2−y1–Y A3+x1y2 Without further fitting this model also captured pharmacological blockade of the plasticity traces . In the model , we simulated the experimental effects of pharmacological blockade by setting the relevant parameter or variable to 0 . Specifically , we simulated the effects of blocking two different retrograde messenger systems shown to be involved in STDP in layer-5 pyramidal cell pairs , eCB signaling ( Sjöström et al . , 2003 ) and NO signaling ( Sjöström et al . , 2007 ) . To reproduce pharmacological blockade experiments , we used high-frequency pairing ( 50 Hz ) with +10 ms delay , which is comparable with our frequency-dependent results and approximates the long depolarizing currents used in Sjöström et al . ( 2007 ) . Blocking eCB receptors prevents presynaptic LTD ( Sjöström et al . , 2003 ) . By setting d− = 0 presynaptic LTD was disabled . This reveals presynaptic LTP and enhances short-term depression ( Figure 1—figure supplement 3 ) , consistent with experimental evidence ( Sjöström et al . , 2007 ) , as the drugs used are likely to block presynaptic eCB receptors . In contrast , blocking NO decreases LTP but does not affect short-term synaptic dynamics ( Sjöström et al . , 2007 ) ( Figure 1—figure supplement 3A ) . We simulated this by setting y+ = 0 , so that both presynaptic components were absent . The release of neurotransmitter was assumed to follow a standard binomial model ( Del Castillo and Katz , 1954 ) ( 17 ) Psyn ( X=k ) = ( Nk ) Pk ( 1−P ) N−k , which defines the probability of having k successful events ( neurotransmitter release ) given N trials ( release sites ) with equal probability P . The mean synaptic response is scaled by a postsynaptic factor q , which can be related to the quantal amplitude so that ( 18 ) μsyn=PqN , and the variance is ( 19 ) σsyn2=q2NP ( 1−P ) . Following the binomial release model ( Equation 18 ) , μsyn ( Equation 19 ) and σsyn2 ( Equation 20 ) , ( 20 ) P=μsynNq , and ( 21 ) q=σsyn2μsyn+μsynN . The number of release sites N is believed to change only after a few hours ( Bolshakov et al . , 1997; Saez and Friedlander , 2009 ) . As the slice synaptic plasticity experiments analysed here lasted only up to 1 . 5 hr ( Sjöström et al . , 2001 ) and we were interested in the relative changes we assumed constant N = 5 . 5 in our analysis below , as estimated in Markram et al . ( 1997 ) using data from the same connection type we used to fit our model . Equations 21 , 22 were used to estimate P and q from in vitro plasticity data ( see above ) , respectively ( dataset deposited at Dryad data repository at http://dx . doi . org/10 . 5061/dryad . p286g [Costa et al . , 2015] ) . Note that because the data had to be reanalized in full there are minor differences in the mean weights previously published ( Sjöström et al . , 2001 ) . We verified our P and q extraction method by analysing short-term plasticity experiments with pharmacological manipulation of presynaptic release or of postsynaptic gain ( Figure 1—figure supplement 2A , Sjöström et al . , 2003 ) , and experiments with pharmacological blockade of pre- or postsynaptic long-term plasticity ( Figure 1—figure supplement 2B , Sjöström et al . , 2007 ) ( Figure 1—figure supplement 2A , B ) . In addition , long-term changes in P but not in q were inversely correlated with changes in paired-pulse ratio , as expected ( Figure 1—figure supplement 2C , D ) . Taken together , these results lend experimental support to our binomial-distribution-based approach for extracting P and q to tune changes in the pre- and postsynaptic modifications of our unified STDP model ( Figure 1D , E ) . We extracted the effective P and q from the in vivo data obtained by Froemke et al . ( 2013 ) . Again using a binomial model , we obtained estimators for their variability measure given by v = q ( 1 − P ) and the mean by μ = PqN . To ease comparison with our simulations we set the initial P to the same initial condition used in our simulations P = 0 . 5 ( Costa et al . , 2013 ) . We then obtained the initial N=\|μ\|qP and the initial q=v ( 1−P ) . For the after pairing data we allowed both pre- and postsynaptic factors P and q to change , while N was fixed to the values extracted before pairing ( Bolshakov et al . , 1997 ) . The estimations after learning were obtained as q=v+\|μ\|N and P=\|μ\|Nq . We used these estimators to extract q and P from measurements for both the depression experienced for the unpaired ( best before pairing ) receptive field position and the potentiated paired position ( Froemke et al . , 2013 ) . After pairing , the effective q of the potentiated ( ‘on’ ) response increased from qbeforeon=23 . 3 pA to qafteron=27 . 1 pA ( +16 . 3% ) , while P increased from Pbeforeon=0 . 5 to Pafteron=0 . 73 ( +46% ) . Responses that were depressed ( ‘off’ ) , typically the original best frequency , yielded no statistically significant change in qbeforeoff , while Pbeforeoff=0 . 5 and Pafteroff=0 . 40 ( −20% ) ( Figures 2 , Figure 2—figure supplement 1 and Figure 2—figure supplement 3 ) . To ease comparison with the postsynaptic factor in the simulations we scaled the experimentally obtained q such that before plasticity it was 1 . We compared models where we allowed both P and q to change or only one of them , the lower variability estimation error was obtained by the one where both factors change ( Figure 2—figure supplement 3E ) . The estimation error was calculated as 1N∑iN ( vreali−vestimatedi ) 2 , where N is the number of data points . We calculated the SNR of a synaptic response defined here by a random variable s , amidst additive background noise defined by the random variable n as follows ( 22 ) SNRsyn=2 ( ⟨s⟩−⟨n⟩ ) 2σs2+σn2 , It is assumed that n∼𝒩 ( 0 , σn2 ) and we also used the Gaussian approximation to the binomial release model specified above , s∼𝒩 ( PqN , q2NP ( 1−P ) +σn2 ) , from which follows the SNR of the first postsynaptic response ( 23 ) SNRsyn=2 ( PqN ) 2q2NP ( 1−P ) +2σn2 . In Figure 2 , we used σn2=0 . 5 . Variance of the k-th postsynaptic response is given by σsynk2=q2Nrkpk ( 1−rkpk ) ( Figure 2—figure supplement 2A ) . The SNR of the k-th postsynaptic response is ( 24 ) SNRsynk=2 ( rkpkqN ) 2q2Nrkpk ( 1−rkpk ) +2σn2 , where pk and rk are given by Equations 15 , 16 , respectively . The SNR of the sum of the first K responses , evoked at a given presynaptic firing rate ρ therefore equals ( 25 ) SNRsynρ=2 ( ∑k=0K−1rkpkqN ) 2∑k=0K−1q2Nrkpk ( 1−rkpk ) +2∑k=0K−1σn2 . After unified STDP the first response has a higher amplitude and the second one a much lower amplitude due to synaptic depression . Combined with the background noise , the SNR can drop when the second or further responses are included . However , the SNR of the summed response will always be larger than when only postsynaptic modifications are made ( see Figure 2—figure supplement 2B ) . This holds for any frequency , Figure 2—figure supplement 2C and carries over to an information theoretic analysis of the response , Figure 2—figure supplement 2D . Next , we used ROC analysis to compute the false alarm and detection probability of the first postsynaptic response ( 26 ) pfalse alarm=∫T+∞Pn ( r ) dr=12erfc ( T2σn2 ) , ( 27 ) pdetection=∫T+∞Ps ( r ) dr=12erfc ( T−PqN2q2NP ( 1−P ) +σn2 ) . where T is the discrimination threshold , and erfc is the complementary error function defined as erfc ( x ) =2π∫x∞e−t2dt . To assess the overall discriminability , we used pdiscrimination , which is the area under the ROC curve ( AUC ) . The AUC was computed by integrating over the ROC curve using the trapezoid method ( see Figure 2D ) . Given that N is a simple constant we set it to 1 , unless otherwise stated ( see data inference above ) . For the receptive field development simulations , we used a feedforward network with 100 presynaptic neurons j with Poisson statistics and a single integrate-and-fire postsynaptic neuron . The postsynaptic neuron was modelled as an adaptive exponential integrate-and-fire neuron model ( Brette and Gerstner , 2005 ) . Model parameters were as reported in Brette and Gerstner ( 2005 ) ; Badel et al . ( 2008 ) and synapses were modelled as input currents . The firing rate of the presynaptic Poisson neurons was modelled using a Gaussian profile , defined as ( 28 ) ρ ( j;p , σ ) =ρmin+ ( ρmax−ρmin ) e− ( j−p ) 22σ2 . where ρ is the rate in the Poisson neuron model j , p the input position for which the rate is maximal , and σ = 5 Hz the distribution spread . ρmax and ρmin are the maximum and minimum rates , and were set to ρmax = 50 Hz and ρmin = 3 Hz . We scaled d− , d+ and c+ by a factor 0 . 15 to yield a smoother receptive field development . q was bounded between 0 nA and 20 nA , so that the synaptic input is appropriately scaled for the neuron model used . The network was simulated for 100 s to achieve convergence . For the memory savings experiment , we interleaved two receptive field positions . Results for receptive development and memory savings were averaged over 10 runs . The response of the postsynaptic neuron ( Figure 3C ) was assessed by presenting each stimulus alone with long-term synaptic plasticity inactive . Receptive field simulations were implemented in simulator Brian 1 . 41 ( Goodman and Brette , 2008 ) . Code for running and plotting the savings experiment is available online ( http://modeldb . yale . edu/184487 ) . Results are reported as mean ± SEM . Statistical comparisons were made with Student's t-test for equal means , if data was normally distributed as assessed using Kolmogorov–Smirnov test , Mann–Whitney U non-parametric test was used otherwise . For multiple comparisons we applied ANOVA or Kruskal–Wallis test for normally or non-normally distributed data , respectively . For correlation analysis the Spearman's coefficient was used together with one-tailed Student's t-test . Significance levels are p < 0 . 05 , p < 0 . 01 , and **p < 0 . 001 .	Throughout life , animals must learn how to respond accurately to new sensations and environments , while retaining knowledge of previous experiences . Learning is widely believed to modify the connections ( called synapses ) between neurons of the cerebral cortex and other brain areas . This process is known as synaptic plasticity . Experimentally , presynaptic and postsynaptic changes have been identified , but it is not known what advantages there are to changing both components when , in principle , changing either might suffice . To investigate this , Costa et al . developed a mathematical model of synaptic plasticity that captured both pre- and postsynaptic changes , based on a number of experiments over the last decade from recordings in the rat sensory cortices . There were two major findings from this model . First , if both presynaptic and postsynaptic changes occur , the modeling results indicated that sensory perception could become more precise , as has been recently found in the rat auditory system . Second , because the details of presynaptic and postsynaptic changes are different , previously triggered changes leave behind a type of memory trace that allows apparently forgotten information to be rapidly relearned . Interestingly , similar asymmetries have been reported in other brain regions . One future challenge is to understand whether these findings constitute a general principle of plasticity in the brain .	[ "Abstract", "Materials", "and", "methods" ]	[ "short", "report", "computational", "and", "systems", "biology", "neuroscience" ]	2015	Unified pre- and postsynaptic long-term plasticity enables reliable and flexible learning
Ribonucleotide reductases ( RNRs ) are key enzymes in DNA metabolism , with allosteric mechanisms controlling substrate specificity and overall activity . In RNRs , the activity master-switch , the ATP-cone , has been found exclusively in the catalytic subunit . In two class I RNR subclasses whose catalytic subunit lacks the ATP-cone , we discovered ATP-cones in the radical-generating subunit . The ATP-cone in the Leeuwenhoekiella blandensis radical-generating subunit regulates activity via quaternary structure induced by binding of nucleotides . ATP induces enzymatically competent dimers , whereas dATP induces non-productive tetramers , resulting in different holoenzymes . The tetramer forms by interactions between ATP-cones , shown by a 2 . 45 Å crystal structure . We also present evidence for an MnIIIMnIV metal center . In summary , lack of an ATP-cone domain in the catalytic subunit was compensated by transfer of the domain to the radical-generating subunit . To our knowledge , this represents the first observation of transfer of an allosteric domain between components of the same enzyme complex . Allosteric regulation of an enzyme is defined as regulation of activity by binding of an effector molecule to a different location of the enzyme than the active site . The effector influences the distribution of tertiary or quaternary conformations of an enzyme , alone or in combination , modulating its activity ( Swain and Gierasch , 2006 ) . Allostery is an intrinsic property of many , if not all , dynamic proteins ( Gunasekaran et al . , 2004 ) and an important factor in disease ( Nussinov and Tsai , 2013 ) , and has hence attracted considerable scientific interest . A substantial part of this interest has been focused on ribonucleotide reductase ( RNR ) , which has been termed a ‘paradigm for allosteric regulation of enzymes’ ( Aravind et al . , 2000 ) . RNRs are essential enzymes in all free-living cells , providing the only known de novo pathway for the biosynthesis of deoxyribonucleotides ( dNTPs ) , the immediate precursors for DNA synthesis and repair ( Hofer et al . , 2012; Nordlund and Reichard , 2006 ) . To avoid imbalanced levels of dNTPs and the increased mutation rates that are the inevitable consequences of this ( Kumar et al . , 2011; Mathews , 2006; Watt et al . , 2016 ) , RNRs are tightly controlled through transcriptional and allosteric regulation , subcellular compartmentalization and small protein inhibitors ( Pai and Kearsey , 2017; Torrents , 2014 ) . Allosteric regulation of RNRs affects both substrate specificity and overall activity . The specificity regulation has been intensively studied and described for all three classes of RNRs ( Andersson et al . , 2000; Hofer et al . , 2012; Larsson et al . , 2004; Reichard , 2010; Torrents et al . , 2000; Zimanyi et al . , 2016 ) . The s-site binds dNTPs and determines which nucleotide will be reduced at the active site to ensure balanced levels of the four dNTPs in the cell . Additionally , many RNRs possess an overall activity regulation site ( a-site ) ( Brown and Reichard , 1969; Thelander and Reichard , 1979 ) positioned in an N-terminal domain of ~85–100 amino acid residues called the ATP-cone ( Aravind et al . , 2000; Eriksson et al . , 1997 ) . Acting as a regulatory master switch , the a-site senses intracellular nucleotide concentrations by competitive binding of ATP and dATP . In the presence of ATP the enzyme is active , and when concentrations of dNTPs rise , binding of dATP switches the enzyme off . This mechanism ensures sufficient but not excessive amounts of nucleotides that may also cause increased mutation rates ( Mathews , 2006 ) . The ATP-cone is an example of allosteric regulation controlled by a domain that acts relatively independent of the catalytic core of proteins . This type of allosteric regulation has been shown to provide an evolutionarily dynamic process by which allosteric regulation can be lost or gained both in RNRs ( Lundin et al . , 2015 ) and other enzymes ( Aravind and Koonin , 1999; Lang et al . , 2014 ) . Although regulation has been lost and gained repeatedly in RNRs through evolutionary time , to date , the ATP-cone domain has been observed exclusively at the N-terminus of the catalytic subunit NrdA ( class I ) , NrdJ ( class II ) and NrdD ( class III ) . Class I RNRs consist of a large , catalytic subunit ( α or NrdA ) , and a smaller , radical-generating subunit ( β or NrdB ) ( Huang et al . , 2014; Nordlund and Reichard , 2006 ) . NrdA and NrdB interact to form the active complex , in which the two proteins need to be precisely positioned such that the radical can be transferred from NrdB , where it is generated and stored , to NrdA , where it starts the substrate reduction . In class I , it has long been noted that ATP-cones are absent from subclass Ib ( NrdE ) but present in several , but not all , NrdAs . A recent phylogenetic subclassification of RNRs reveals that three phylogenetically well-supported subclasses of class I never have ATP-cones ( Jonna et al . , 2015 ) ( http://rnrdb . pfitmap . org ) : NrdE , NrdAi and NrdAk . In two of these subclasses we instead discovered ATP-cones attached to their corresponding radical-generating subunit: NrdF ( the Ib subclass ) and NrdBi . It hence appears as if the lack of activity regulation through an ATP-cone in the catalytic subunit is compensated by the presence of this domain in the non-homologous radical-generating subunit of some RNRs . Three distinct types of class I complexes have been mechanistically characterized and found to operate via nucleotide-induced regulation of quaternary structure ( Johansson et al . , 2016; Jonna et al . , 2015; Kashlan et al . , 2002; Rofougaran et al . , 2008; Rofougaran et al . , 2006; Torrents et al . , 2006 ) . Crystal structures , cryo-electron microscopy reconstructions and small-angle X-ray scattering studies of inhibited complexes have revealed that when dATP is bound at the a-site , high molecular mass oligomers are formed , in which the radical transfer pathway is distorted ( Ando et al . , 2011; Ando et al . , 2016; Fairman et al . , 2011; Johansson et al . , 2016 ) . Conversely , when ATP is bound , an active enzyme complex is formed . Interestingly , the structure and organization of subunits in active and inactive complexes varies considerably between species ( Ahmad and Dealwis , 2013; Hofer et al . , 2012 ) . In Escherichia coli RNR , the active NrdAB complex is α2β2 , whereas the inactive form is an α4β4 ring-shaped octamer where the ATP-cones in the α subunits sequester the β subunits in a non-productive conformation ( Ando et al . , 2011 ) . In the eukaryotic class I RNR , the inactive complex differs from the one in E . coli in that it has an α6 stoichiometry . This hexamer can only bind one β2 subunit in an unproductive manner without a properly aligned electron transport chain ( Fairman et al . , 2011 ) . Activation by ATP creates a different type of α6 complex that binds one or more β2 complexes ( Ando et al . , 2016; Aye and Stubbe , 2011; Crona et al . , 2013; Fairman et al . , 2011; Rofougaran et al . , 2006 ) . The different complexes are formed by subtle changes at the a-site induced by binding of the different nucleotides ( Fairman et al . , 2011; Xu et al . , 2006 ) . Another oligomerization mechanism has been recently found in Pseudomonas aeruginosa class I RNR , which possesses two consecutive ATP-cones , of which only the N-terminal one binds nucleotides . The active complex is once again α2β2 , but the inactive P . aeruginosa RNR complex is a dATP-induced α4 complex consisting of a ring of four α subunits interacting via their outer ATP-cones ( Johansson et al . , 2016; Jonna et al . , 2015 ) . A single β2 can bind to this ring , but the complex is inactive . Oligomerization of RNRs may be a useful character to explore biomedically . RNRs have long attracted interest as potential targets for novel antibiotics as well as for cancer therapy ( Aye et al . , 2015; Julián et al . , 2015; Tholander and Sjöberg , 2012 ) . Several RNR drugs are directed towards the radical-containing subunit and the active site of the catalytic subunit . Recently , some nucleoside analogs used in cancer treatments and known to inhibit RNRs in vitro in their phosphorylated forms were shown to induce hexameric complexes in vivo ( Aye et al . , 2012; Aye and Stubbe , 2011; Wisitpitthaya et al . , 2016 ) . The unexpected finding of an ATP-cone fused to the radical-generating subunits poses questions regarding how it might regulate activity . Here we describe the mechanism of activity regulation by the ATP-cone N-terminally fused to the radical-generating NrdBi from Leeuwenhoekiella blandensis sp . nov . strain MED217 . L . blandensis was isolated from Mediterranean surface seawater and belongs to Flavobacteriaceae , the major family of marine Bacteroidetes , with important roles in carbon flow and nutrient turnover in the sea during and following algal blooms ( Fernández-Gómez et al . , 2013; Pinhassi et al . , 2006 ) . L . blandensis possesses two RNRs: a class II without ATP-cone , and the class I NrdAi/NrdBi , which lacks an ATP-cone in NrdA and instead contains an ATP-cone positioned at the N-terminus of NrdB . Superficially , the allosteric mechanism of L . blandensis NrdAi/NrdBi holoenzyme is similar to when the ATP-cone is contained in the catalytic subunit . At high dATP concentrations , inhibited higher oligomeric complexes of the holoenzyme are favoured . However , in the L . blandensis class I RNR , the oligomerization is driven by a shift towards tetramers of the radical-generating subunit . This illustrates how allosteric regulation controlled by ATP-cones can evolve in a highly dynamic way , requiring few adaptations to the core of the enzyme . The relative ease by which ATP-cone-driven activity regulation appears to evolve , provides a potential route to regulate engineered enzymes by dATP-inhibition for enzymes in which activity is affected by oligomerization . Addition of an ATP-cone to the protein could be used to induce higher oligomers controlled by dATP addition . We detected ATP-cones in the radical-generating subunits of RNRs from two distinct phylogenetic RNR subclasses: NrdBi and NrdF ( Figure 1a ) . In the NrdBi sequences , the ATP-cone was found at the N-terminus of the protein , whereas it was found at the C-terminus of the NrdF proteins ( Figure 1b ) . Interestingly , the corresponding catalytic subunit subclasses – NrdAi and NrdE respectively – have been found to lack ATP-cones . Ninety-three sequences in NCBI’s RefSeq database are NrdBi proteins with N-terminally positioned ATP-cones . They are encoded by viruses and bacteria from several phyla , although the Flavobacteriales order in the Bacteroidetes phylum predominate ( 70 sequences , http://rnrdb . pfitmap . org ) . NrdF proteins with a C-terminally positioned ATP-cone are only encoded by a few species of the Meiothermus genus of the Deinococcus-Thermus phylum ( http://rnrdb . pfitmap . org ) . All species encoding NrdB proteins with ATP-cones in their genomes also encode other RNRs . Initially we cloned , expressed and purified the L . blandensis NrdBi and NrdAi proteins . Using a four-substrate activity assay in the presence of saturating concentrations of the s-site effectors dTTP , dGTP or ATP , we found that L . blandensis RNR has a similar specificity regulation pattern to most characterized RNRs ( Hofer et al . , 2012 ) . ATP stimulated the reduction of CDP and UDP , whereas dTTP stimulated the reduction of GDP , and dGTP stimulated the reduction of ADP and GDP ( Figure 2 ) . The enzyme was completely inactive in the absence of allosteric effectors . Using mixtures of allosteric effectors , we observed that dTTP-induced GDP reduction dramatically increased in the presence of ATP ( Figure 2 ) . We performed a series of activity assays with CDP as substrate to elucidate the potential roles of ATP and dATP in activating and inhibiting the enzyme ( Figure 3 ) . The presence of ATP clearly activated the enzyme ( Figure 3a ) , while dATP activated the enzyme when used at low concentrations and was inhibitory at 30 µM and higher ( Figure 3b ) . An ATP-cone deletion mutant NrdB∆99 , lacking the N-terminal 99 residues , reached a lower maximum activity compared to the wild type enzyme , suggesting that it was not activated by ATP beyond saturation of the s-site in the NrdA , nor was it inhibited by dATP ( Figure 3a–b ) . From the NrdB∆99 effector titrations , we could calculate KL values – the concentrations of allosteric effectors that give half maximal enzyme activity – for binding of ATP and dATP to the s-site in NrdA to 30 and 1 . 4 μM respectively . Titration of ATP into wild type NrdB in the presence of an excess of NrdA saturated with dTTP showed that it activates the enzyme through the a-site with a KL of 96 μM ( Figure 3c ) . For the corresponding inhibition by dATP binding to the a-site , we calculated the Ki value - the binding constant of a non-competitive inhibitor - to be 20 µM in the presence of an s-site saturating dTTP concentration ( Figure 3d ) . We also tested if only the triphosphate form of ( deoxy ) adenosine nucleotides would interact with the a-site in presence of s-site saturating dTTP concentrations . In addition to ATP and dATP , dADP was also found to interact , whereas ADP , AMP and dAMP had no effect ( Figure 3e ) . Titration with dADP inhibited the enzyme activity , although less strongly than dATP ( Figure 3f ) . To elucidate the mechanism of allosteric overall activity regulation governed by the NrdB-linked ATP-cone , activity-independent oligomer-distribution experiments were performed by gas-phase electrophoretic macromolecule analysis ( GEMMA ) . In the absence of allosteric effectors , NrdB ( β ) is mainly monomeric ( theoretically 51 . 8 kDa ) and , in contrast to most studied NrdB proteins , it does not readily form dimers at the low protein concentration range suitable for GEMMA analyses ( Figure 4a ) . Addition of dATP ( 50 μM ) dramatically shifted the equilibrium towards tetramers β4 , which became the major form under these conditions ( Figure 4a ) . Titration of increasing concentrations of dATP to NrdB showed that the tetramers reached their half-maximum mass concentration at around 10 µM dATP ( Figure 4—figure supplement 1 ) . Addition of 50 μM dADP also induced formation of NrdB tetramers , although less efficiently than with dATP ( Figure 4a ) . In contrast , the NrdB∆99 mutant , lacking the ATP-cone ( Figure 4b ) , was mainly monomeric regardless if dATP was present or not ( Figure 4b ) , demonstrating that the NrdB-linked ATP-cone is required for dATP-induced tetramer formation . The protein had a tendency to aggregate , which can possibly explain the ladder of monomers , dimers , trimers and tetramers . NrdA was a monomer ( theoretically 70 . 6 kDa ) in the absence of allosteric effectors , while addition of dATP prompted formation of dimers ( Figure 4c ) To assess the oligomeric state of the complete enzymatic complex of the inactive L . blandensis RNR , a mixture of NrdA ( α ) and NrdB ( β ) in the presence of 100 µM dATP was analyzed with GEMMA ( Figure 4c ) . The two subunits formed a large complex of 340 kDa with the expected mass of an α2β4 complex ( theoretically 348 kDa ) , and at higher protein concentration a complex of 465 kDa with the expected mass of an α4β4 complex ( theoretically 488 kDa ) started to appear . To complement the GEMMA analyses of oligomer formation , we performed analytical size exclusion chromatography ( SEC ) using higher protein concentrations and physiologically reasonable concentrations of effectors ( 3 mM ATP and 0 . 1 to 0 . 5 mM dATP ) ( Bochner and Ames , 1982; Buckstein et al . , 2008 ) on an analytical Superdex 200 PC 3 . 2/30 column at 7°C ( Figure 5 ) . These experiments were complemented with SEC experiments at room temperature with a semi-preparative column ( Superdex 200 10/300 GL ) that gave better resolution but was less practical for the bulk experiments due to its much higher mobile phase consumption . The SEC experiments showed that NrdB was in a monomer-dimer-tetramer equilibrium ( Figure 5—figure supplement 1a ) with an average size comparable to dimers ( Figure 5a ) . Similar to the results with GEMMA , it was mainly a monomer at lower protein concentrations ( Figure 5—figure supplement 1a ) . ATP and dATP promoted the equilibrium to shift to dimers and tetramers , respectively ( Figure 5a , Figure 5—figure supplement 1b ) . Without effectors , NrdA was mainly in a monomeric state , while it was dominated by dimers in the presence of either of the two effectors ( Figure 5b ) . However , in agreement with the GEMMA results the formation of the nucleotide-induced dimer was much less efficient at lower protein concentration ( Figure 5—figure supplement 1c ) . When NrdA and NrdB were mixed in the absence of allosteric effectors , no complex was visible . After addition of ATP , a new complex of ~200 kDa appeared ( Figure 5c ) . To verify its composition and stoichiometry , we analyzed fractions eluted from SEC on SDS PAGE . Both NrdA and NrdB were visible on the gel in a 1:1 molar ratio ( Figure 5c insert ) . This complex , formed in conditions promoting active RNR , is conceivably an α2β2 complex . After addition of the inhibiting effector dATP , a complex with a molecular mass consistent with α4β4 eluted . Typical for complexes that equilibrate between each other faster than the time of analysis , a titration using increasing concentrations of NrdA-NrdB mixtures in the presence of 100 μM dATP showed a gradual movement of the NrdB tetramer peak to the left up to the position of an α4β4 complex rather than showing distinct β4 , α2β4 and α4β4 peaks ( Figure 5—figure supplement 1d ) . The crystal structure of the dATP-inhibited complex of L . blandensis NrdB at 2 . 45 Å resolution ( PDB 5OLK ) revealed a novel tetrameric arrangement hitherto not observed in the RNR family , with approximate 222 point symmetry ( Figure 6a , Table 1 ) . Each monomer consists of an ATP-cone domain ( residues 1–103 ) joined by a short linker ( 104-106 ) to a metal-binding α-helical core domain ( residues 107–398 ) and a disordered C-terminus ( 399-427 ) . The latter two features are typical of the NrdB/F family . This domain arrangement gives the NrdB monomer and dimer extended conformations that are presumably flexible in solution ( Figure 6b ) . The dimer buries about 1100 Å2 of solvent accessible area on each monomer . The tetrameric arrangement is completely dependent on interactions between the ATP cone domains , as no contacts are made between the core domains in the two dimers that make up the tetramer ( Figure 6a ) . The ATP-cone domain in L . blandensis NrdB is structurally very similar to the one recently identified in the NrdA protein of P . aeruginosa ( Johansson et al . , 2016 ) . The root-mean-square deviation for 92 equivalent Cα atoms is 1 . 2 Å . The electron density unambiguously confirms the L . blandensis ATP-cone’s ability to bind two molecules of dATP , which it shares with P . aeruginosa NrdA ( Figure 6c ) . Despite a local sequence identity of only 31% to P . aeruginosa NrdA , all amino acids involved in binding both dATP molecules are conserved ( Figure 6c ) . The two dATP molecules bind in a ‘tail-to-tail’ arrangement that orients the base of the ‘non-canonical’ dATP towards the fourth , most C-terminal helix , an arrangement made possible by the binding of a Mg2+ ion between the triphosphate moieties . Remarkably , the interactions between the ATP-cones in L . blandensis NrdB are also very similar to those seen in P . aeruginosa NrdA ( Johansson et al . , 2016 ) , despite the fact that the ATP-cone is attached to a structurally completely different core domain . The main interactions occur between the last two helices α3 and α4 in respective ATP-cones ( Figure 6d ) . A hydrophobic core in L . blandensis NrdB involving residues Met80 , Ile92 and Ile93 in both monomers is reinforced by salt bridges between residues Asp73 , Asp76 in one monomer and Lys89 in the other . The two domains bury ~510 Å2 of solvent accessible area . This is slightly less than the ~640 Å2 buried in the equivalent interaction involving the ATP-cones of P . aeruginosa NrdA . Within each ATP cone , helices α3 and α4 have the same relative orientation , partly determined by an internal salt bridge between the conserved Asp73 and Arg95 , but the helix pair in L . blandensis NrdB is rotated relative to its counterpart in the other ATP cone by about 15° compared to P . aeruginosa NrdA , which reduces the number of possible interactions between them . Interestingly , the dATP-induced tetramer leaves free the surfaces of both dimers of L . blandensis NrdB that are thought to interact with the NrdA subunit in productive RNR complexes , which implies that one or two dimers of L . blandensis NrdA could attach to these surfaces in a near-productive fashion in α2β4 or α4β4 complexes ( Figure 6a ) . Two metal ions are found to bind to each of the monomers of L . blandensis NrdB . Comparison of their B-factors with those of surrounding atoms suggest that they are fourth row elements with close to full occupancy , but does not enable us to distinguish between Mn , Fe or Ca . No metal ions were added to the protein preparation , but crystals were obtained in 0 . 2 M CaCl2 . The distance between metal ions varies between 3 . 4–3 . 8 Å in the four monomers ( Figure 6—figure supplement 1 ) , but the metal coordination is very similar . The coordination distances are long for an RNR metal center , with the shortest distances being 2 . 3–2 . 4 Å , in contrast to the more typical 1 . 9–2 . 1 Å seen in other NrdB proteins containing Mn or Fe . Anomalous diffraction at three wavelengths was used to determine the nature of the metal . Figure 6—source data 1 shows that there is little difference in the peak heights at the metal ion positions in anomalous difference maps irrespective of wavelength , which strongly suggests that the metal site is occupied by Ca2+ from the crystallization medium . Furthermore , the peaks were only about twice the height of those from well-ordered sulfur atoms in the structure . At these wavelengths , S has an anomalous f’ component of about 0 . 5 electrons . For Mn or Fe one would expect an anomalous signal around eight times that of S ( Figure 6—source data 2 ) . The presence of Ca2+ is also consistent with the long coordination distances . Furthermore , an X-ray fluorescence spectrum of the crystal ( not shown ) revealed only traces of Mn or Fe . Interestingly , Tyr207 is found near the metal site at the position expected for a radical-carrying Tyr , but it is not hydrogen-bonded to the metal site , its hydroxyl group being at around 6 Å from the side chain of Glu170 . Tyr207 is H-bonded to the side chain of Thr294 , but the latter is not H-bonded to a metal center ligand . On the other side of the metal site , Trp177 is H-bonded to the side chain of Glu263 . This tryptophan is completely conserved in the NrdBi subclass ( Figure 6—figure supplement 2 ) . It makes the same interaction as Trp111 from E . coli NrdB , despite the fact that it is projected from the first helix of the 4-helix bundle containing the metal center ligands , rather than the second helix . The first helix of the bundle has a very unusual distortion in the middle ( Figure 6—figure supplement 3 ) . Normally this is an undistorted α-helix , but in L . blandensis NrdB , residues 171–175 form a loop that bulges away from the metal site , with the exception of Lys174 , whose side chain is projected towards the metal site and is H-bonded to Glu170 ( Figure 6e ) . The significance of this distortion is at present not clear . However we are certain that the distortion is not caused by the coordination of Ca2+ instead of the native metal , as the structure is the same in crystals of the N-terminally truncated form grown in the presence of Mn2+ , which will be presented elsewhere . The nature of the native metallo-cofactor was addressed by X-band EPR spectroscopy and catalytically active samples were analyzed at 5–32 K . The spectra revealed a mixture of signals from low and high-valent manganese species ( Figure 7a ) . In particular , at 5 K a 6-line signal attributable to low valent Mn ( MnII ) was clearly visible , overlaid with a complex multiline signal with a width of approximately 1250G . Increasing the temperature to 32 K resulted in a significant decrease of the latter signal , while the 6-line feature remained relatively intense ( Figure 7a , top and middle ) . The shape , width and temperature dependence of the multiline signal are all highly reminiscent of the signal reported for super-oxidized manganese catalase as well as the 16-line signal observed during the assembly of the dimanganese/tyrosyl radical cofactor in NrdF RNR , and are attributable to a strongly coupled MnIIIMnIV dimer ( STotal = ½ ) ( Figure 7a , bottom ) ( Cotruvo et al . , 2013; Zheng et al . , 1994 ) . The low valent species appeared to be weakly bound to the protein , while the MnIIIMnIV dimer was retained in the protein following desalting ( Figure 7—figure supplement 1 ) . The observation of a high-valent Mn cofactor is consistent with the Mn-dependent increase in catalytic activity of the enzyme and its inhibition by the addition of Fe2+ ( Figure 7b–c ) , and is suggestive of a novel high-valent homodimeric Mn-cofactor . Indeed , earlier calculations have shown that high-valent Mn-dimers have reduction potentials similar to that of the tyrosyl radical in standard class I RNRs , but are hitherto not observed in RNRs ( Roos and Siegbahn , 2011 ) . A more detailed biophysical characterization of this cofactor is currently ongoing . The finding that two dATP molecules were bound to the ATP-cone in the 3D structure prompted us to investigate nucleotide binding using isothermal titration calorimetry ( ITC ) . Binding curves for dATP , ATP and dADP to NrdB , including a reverse titration of NrdB to dATP , were all consistent with a single set of binding sites except for the ATP-cone deletion mutant ( NrdBΔ99 ) , which did not bind nucleotides at all ( Figure 8 ) . In all other titrations the fitted apparent N value was significantly above one , but to be meaningful the N value needs to be an integer . Using information available from the 3D structure , a fixed stoichiometry of 2 was used to fit dATP binding to NrdB . The fit indicated 59% active ( i . e . nucleotide binding ) protein . A fixed stoichiometry of 1 would result in higher than 100% active protein concentration , which is impossible . Fitting dADP and ATP with a 59% proportion of active protein taken from the dATP experiment , we could calculate stoichiometries of 2 . 2 and 2 . 1 for ATP and dADP respectively , indicating that each NrdB monomer can bind two molecules of adenosine nucleotides ( Figure 8f ) . Kd for the three different nucleotides ( Figure 8f ) indicated a 38-fold and 15-fold lower affinity for ATP and dADP compared to dATP . Thermodynamic parameters ( Figure 8f ) indicated that the interactions are enthalpy-driven , with negative ΔH values of −79 , –44 and −103 for dATP , ATP and dADP respectively . Although allosteric regulation is built into many enzymes ( Gunasekaran et al . , 2004 ) , it can evolve in a surprisingly dynamic way ( Aravind and Koonin , 1999; Lang et al . , 2014; Lundin et al . , 2015 ) . This has also been shown experimentally by grafting an allosteric domain to an enzyme that was previously not allosterically regulated ( Cross et al . , 2013 ) . The presence of an ATP-cone in the radical-generating subunit of L . blandensis RNR , provides , to our knowledge , the first example of another degree of evolutionary dynamics by the transfer of a domain conferring allosteric regulation to a non-homologous component of an enzyme complex . It is critically important for an organism to control the supply of dNTPs to allow fidelity in DNA replication and repair ( Mathews , 2006 ) . Specificity regulation of RNR makes sure relative concentrations of dNTPs fit the organism’s DNA composition . On the other hand , activity regulation assures that absolute concentrations of dNTPs follow the different requirements through the cell cycle ( Hofer et al . , 2012 ) . Specificity regulation of RNRs is ubiquitous , integrated in the catalytic subunit of the enzyme , and works via the classical homooligomeric model of allosteric regulation ( Andersson et al . , 2000; Hofer et al . , 2012; Larsson et al . , 2004; Reichard , 2010; Swain and Gierasch , 2006; Torrents et al . , 2000; Zimanyi et al . , 2016 ) . In contrast , the activity regulation is controlled by an accessory domain , the ATP-cone , and works by affecting the distribution of the holoenzyme heteromeric complexes . Moreover , the ATP-cone is only found in some RNRs , and appears to be gained by domain shuffling when evolutionary selection favours it and lost when selection decreases ( Lundin et al . , 2015 ) . These dynamics are further evidenced by the differences in mechanisms recently discovered in class I RNRs ( Ando et al . , 2011 , 2016; Fairman et al . , 2011; Johansson et al . , 2016; Jonna et al . , 2015 ) . The current study was prompted by the interesting observation that several radical-generating subunits of the NrdBi subclass possess an N-terminal ATP-cone and that a few radical-generating subunits of the NrdF subclass possess a C-terminal ATP-cone . Our discovery evokes several pertinent questions: does the ATP-cone fused to a radical-generating subunit function as an allosteric on/off switch; how does it affect the distribution of holoenzyme complexes under active and inhibited conditions; is its structural mode of action similar to that of ATP-cones fused to the catalytic subunit ? We have delineated the function of the ATP-cone that is N-terminally fused to the L . blandensis NrdBi protein . In the presence of the positive effector ATP , L . blandensis NrdBi was a dimer , which by interaction with the L . blandensis catalytic subunit NrdAi formed the common active α2β2 complex , also found in e . g . E . coli , P . aeruginosa and eukaryotic class I RNRs . Binding of dATP to the ATP-cone instead promoted oligomerization of L . blandensis NrdBi to an inactive β4 complex , with a novel tetrameric structure revealed by crystallography . This oligomerization is reminiscent of the ‘ring-shaped’ α4 and α6 complexes formed by dATP binding to the NrdA-linked ATP-cones in P . aeruginosa and eukaryotic RNRs . When L . blandensis NrdAi was added to the dATP-loaded NrdBi tetramer , higher molecular mass complexes of α2β4 and α4β4 appeared . The NrdA dimers seem to bind to the NrdB tetramers in a ‘nonproductive’ orientation , and in other studied RNRs it usually means that the structure of the complexes does not allow efficient electron transfer between NrdA and NrdB . The crystal structure shows that the tetramerization of L . blandensis NrdBi leaves the putative interaction surface for NrdAi free , which is consistent with the possibility to form both α2β4 and α4β4 oligomers . However the structure does not suggest the structural basis for a disruption of the cysteinyl radical generation pathway in these non-productive complexes . The structure of the dATP-loaded L . blandensis NrdB shows that it binds two dATP molecules per ATP-cone . Both molecules bind to the same site and interact with each other through a Mg2+ ion . Most allosterically regulated RNRs characterized so far bind only one dATP molecule per ATP-cone . However , a novel class of ATP-cones that binds two dATP molecules was recently characterized in P . aeruginosa NrdA ( Johansson et al . , 2016; Jonna et al . , 2015 ) . The NrdB-linked ATP-cone of L . blandensis has sequence motifs characteristic of this kind of ATP-cone . The structure confirms that both dATP molecules bind essentially identically as in P . aeruginosa NrdA and that the ATP-cones make similar interactions to each other , distinct from those made in the eukaryotic α6 complexes . It has also been shown in the RNR transcriptional regulator NrdR , that ATP and dATP bind with positive cooperativity to its ATP-cone ( McKethan and Spiro , 2013 ) , implying binding of more than one nucleotide molecule . The ATP-cone in L . blandensis NrdBi offers a unique possibility to measure its binding capacity for ( deoxy ) adenosine nucleotides . This has not been possible in earlier studied RNRs , where the ATP-cone is bound to the α subunit that also possesses a binding site for allosteric regulation of substrate specificity . ITC ligand binding studies confirmed that the L . blandensis ATP-cone bound two dATP molecules , and showed that it also can bind two molecules of ATP or two molecules of dADP . Structurally , binding of dADP is not surprising , since the γ-phosphates of the dATP molecules make only one interaction with the protein , through Arg102 , and they only contribute 2 of the 6 coordinating atoms of the intervening Mg2+ ion . Nonetheless , this is the first observation of dADP binding to the ATP-cone of an RNR enzyme . dADP inhibits enzyme activity in a similar manner to dATP , but higher concentrations are required . In vivo , deoxyribonucleoside diphosphates are rapidly converted to triphosphates and cellular dADP concentrations are very low ( Mathews , 2014; Traut , 1994 ) . Nevertheless , dADP is one of the products of L . blandensis RNR , and perhaps local concentrations are higher . The ability of dADP to regulate the activity of L . blandensis RNR may enable it to react more rapidly to changes in ( deoxy ) adenosine nucleotide concentrations and provide the cell with a fitness advantage . It can therefore not be excluded that dADP has a physiological role , although we think that the 15 times stronger affinity for dATP suggests that it is the major inhibitory nucleotide effector . Binding of deoxyadenosine di- and monophosphates has been described for the ATP-cone in NrdR ( Grinberg et al . , 2006; McKethan and Spiro , 2013 ) . Based on the variable occurrence of the ATP-cone in RNR catalytic subunits , we have earlier hypothesized that its presence in RNRs is part of a dynamic process of gains and losses on a relatively short evolutionary time scale ( Lundin et al . , 2015 ) . This suggests that the ATP-cone , contrary to what might be expected for a domain involved in allostery , does not require a long evolutionary period of integration with a protein to contribute to regulation and that it hence lends itself to a highly dynamic evolution of allosteric activity control ( Lundin et al . , 2015 ) . This hypothesis is nicely supported by the N-terminally positioned ATP-cone , in the radical-generating subunit , NrdB , of L . blandensis RNR described here . No RNR in the phylogenetic subclass NrdAi/Bi contains an ATP-cone in the catalytic subunit , and only a minority – mostly encoded by Flavobacteriia , a marine class of Bacteroidetes – have an NrdB with an ATP-cone , suggesting the relatively recent acquisition of the ATP-cone to an RNR subclass that is otherwise not activity regulated ( Figure 1 ) . Moreover , in the Ib subclass , which also lacks ATP-cones in the catalytic subunit , we detected a C-terminally positioned ATP-cone in the radical-generating subunit . This was found in only two sequences from closely related organisms and might hence be an example of an even more recent evolutionary event . The evidence we have presented here for an ATP/dATP-sensing master switch of the L . blandensis NrdAi/NrdBi class I RNR , suggests a potential for the use of ATP-cones to control the activity of engineered proteins . Multimeric enzymes could be inactivated through sequestration of one member of an active complex , by control of dATP concentrations in the reaction mixture . The surprises did not end with the discovery of a functional master switch in the L . blandensis NrdAi/NrdBi RNR . The active center of the radical generating subunit of class I RNRs have earlier been found to consist either of an FeIIIFeIII or MnIIIMnIII pair coupled to a tyrosine residue acting as long-term storage for the catalytically essential radical ( Berggren et al . , 2017; Cotruvo and Stubbe , 2012 ) , or a FeIIIMnIV center not coupled to an amino acid radical ( Griese et al . , 2014 ) . Although further analyses are required to fully characterize the L . blandensis NrdBi active center , it appears to present a fourth type , a high-valent MnIIIMnIV center that lacks a suitably positioned radical storage amino acid . The evolutionary flexibility displayed by the ATP-cone appears hence all but equaled by the evolutionary tuning possibilities of the metal centers in the radical generating subunit of class I RNR . DNA fragments encoding NrdAi ( WP_009781766 ) and NrdBi ( EAQ51288 ) were amplified by PCR from Leeuwenhoekiella blandensis sp . nov . strain MED217 genomic DNA , isolated as described previously ( Pinhassi et al . , 2006 ) , using specific primers: NrdA: LBR1_For 5'- cgagCATATGAGAGAAAACACTACCAAAC-3' and LBR1_Rev 5'- gcaaGGATCCTTAAGCTTCACAGCTTACA-3' . NrdB: LBR2_For 5'-cgagCATATGAGTTCACAAGAGATCAAA-3' , LBR2_REV 5'- gcaaGGATCCTTAAAATAAGTCGTCGCTG-3' , The PCR products were purified , cleaved with NdeI and BamHI restriction enzymes and inserted into a pET-28a ( + ) expression vector ( Novagen , Madison , Wisconsin , USA ) . The obtained constructs pET-nrdA and pET-nrdB contained an N-terminal hexahistidine ( His ) tag and thrombin cleavage site . To construct a truncated NrdB mutant , lacking the entire ATP-cone domain , new forward primer LBR2∆99_For 5'-cgatCATATGCTGGAGCGTAAAACAAAT-3' was used with LBR2_REV to yield a pET-nrdB∆99 . The cloning process and the resulting construct was similar to that of the wild type protein , except that it lacked sequence coding for the N-terminal 99 amino acids . Overnight cultures of E . coli BL21 ( DE3 ) /pET28a ( + ) bearing pET-nrdA , pET-nrdB and pET-nrdB∆99 were diluted to an absorbance at 600 nm of 0 . 1 in LB ( Luria-Bertani ) liquid medium , containing kanamycin ( 50 μg/ml ) and shaken vigorously at 37°C . At an absorbance at 600 nm of 0 . 8 , isopropyl-β-D-thiogalactopyranoside ( Sigma ) was added to a final concentration of 0 . 01 mM for NrdA expression and 0 . 5 mM for NrdB and NrdB∆99 expression . For particular experiments , 0 . 5 mM MnSO4 or 0 . 5 mM FeNH4 ( SO4 ) 2 or the combination of both metals ( 0 . 4 mM and 0 . 25 mM respectively ) were added to NrdB∆99 cultures . The cells were grown overnight at 14°C for NrdA expression and 20°C for NrdB and NrdB∆99 expression and harvested by centrifugation . The cell pellet was resuspended in lysis buffer: 50 mM Tris-HCl pH 7 . 6 containing 300 mM NaCl , 20% glycerol , 10 mM imidazole , 1 mM PMSF . Cells were disrupted by high pressure homogenization and the lysate was centrifuged at 18 , 000 × g for 45 min at 4°C . The recombinant His-tagged protein was first isolated by metal-chelate affinity chromatography using ÄKTA prime system ( GE Healthcare ) : the supernatant was loaded on a HisTrap FF Ni Sepharose column ( GE Healthcare ) , equilibrated with lysis buffer ( w/o PMSF ) , washed thoroughly with buffer and eluted with buffer containing 500 mM imidazole . Further purification was accomplished by fast protein liquid chromatography ( FPLC ) on a 125 ml column packed with Superose 12 Prep Grade or HiLoad 16/600 Superdex 200 pg column ( GE Healthcare ) using ÄKTA prime system , equilibrated with buffer containing 50 mM Tris-HCl pH 7 . 6 , 300 mM NaCl , 10–20% glycerol . Eluted protein was collected . In the case of NrdA , all purification steps were performed in the presence of 2 mM DTT . Protein concentration was determined by measuring the UV absorbance at 280 nm based on protein theoretical extinction coefficients 91 , 135 , 46 , 870 and 39 , 420 M−1 cm−1 for NrdA , NrdB and NrdB∆99 respectively . Proteins were concentrated using Amicon Ultra-15 centrifugal filter units ( Millipore ) , frozen in liquid nitrogen and stored at −80°C until used . For crystallization , NrdB was subsequently cleaved by thrombin ( Novagen ) to remove the hexahistidine tag . 41 mg NrdB was incubated with 25 U thrombin at 6°C for 3 hr , in a buffer containing 40 mM Tris-HCl pH 8 . 4 , 150 mM NaCl , and 2 . 5 mM CaCl2 in a total volume of 50 ml . Subsequently , imidazole to a final concentration of 20 mM was added and the reaction mixture was applied to a HisTrap FF Ni Sepharose column in a buffer containing 20 mM imidazole . Unbound protein was collected , concentrated and further purified by FPLC ( see above ) in a buffer containing Tris-HCl pH 7 . 6 , 300 mM NaCl , and 10% glycerol . The thrombin-cleaved NrdB contained three additional residues ( GlySerHis ) that originated from the cleavage site of the enzyme at its N-terminus . Protein purity was evaluated by SDS–PAGE ( 12% ) stained with Coomassie Brilliant Blue . For EPR measurements , NrdB∆99 was frozen in liquid nitrogen in EPR tubes directly after imidazole elution from the HisTrap column . Additional EPR samples were frozen after desalting the protein using PD-10 desalting columns ( GE Healthcare ) . Enzyme assays were performed at room temperature in 50 mM Tris-HCl at pH 8 in volumes of 50 μl . Reaction conditions , giving maximal activity were determined experimentally . In a standard reaction the constituents were; 10 mM DTT , 10 mM MgAc , 10 mM KCl , 0 . 8 mM ( or when indicated 3 mM ) CDP , and various concentrations of allosteric effectors ATP , dATP or dADP . Mixtures of 0 . 5 μM NrdA and 2 μM wild type NrdB or NrdB∆99 ( for determination of s-site KL ) ( Figure 3a–b ) or 0 . 5 μM of NrdB and 2 uM NrdA ( for determination of a-site KL ) , ( Figure 3c–f ) were used . Some components were explicitly varied in specific experiments . Generally , 0 . 8 mM CDP was used as substrate . High CDP concentration ( 3 mM ) was used for dADP titrations , to exclude potential product inhibition by binding of dADP to the active site . When dTTP was used as an s-site effector , 0 . 5 mM or 0 . 8 mM GDP was used as substrate . In four substrate assays , the four substrates CDP , ADP , GDP and UDP were simultaneously present in the mixture at concentrations of 0 . 5 mM each . The substrate mixture was added last to start the reactions . Certain assays were performed in the presence of Mn ( CH3COO ) 2 or FeNH4 ( SO4 ) : 20 μM of the indicated metal was added to 10 μM NrdB or NrdB∆99 protein , incubated for 10 min , mixed with NrdA and added to the reaction mixture . Enzyme reactions were incubated for 10–30 min and then stopped by the addition of methanol . The chosen incubation time gave a maximum substrate turnover of 30% . Substrate conversion was analyzed by HPLC using a Waters Symmetry C18 column ( 150 × 4 . 6 mm , 3 . 5 μm pore size ) equilibrated with buffer A . 25 μl samples were injected and eluted at 1 ml/min with a linear gradient of 0–100% buffer B ( buffer A: 10% methanol in 50 mM potassium phosphate buffer , pH 7 . 0 , supplemented with 10 mM tributylammonium hydroxide; buffer B: 30% methanol in 50 mM potassium phosphate buffer , pH 7 . 0 , supplemented with 10 mM tributylammonium hydroxide ) . Compound identification was achieved by comparison with injected standards . Relative quantification was obtained by peak height measurements in the chromatogram ( UV absorbance at 271 nm ) in relation to standards . Specific activities of either NrdA ( Figure 2a–b ) or NrdB ( Figure 2c–f ) were determined . Specific activities varied between protein preparations . In some cases the data was standardized to activity percent , where 100% was determined as maximum enzyme activity in a specific condition . From a direct plot of activity versus concentration of effector , the KL values for binding of effectors to the s-site and the a-site , were calculated in SigmaPlot using the equation:V=Vmax×[dNTP]/ ( KL+[dNTP] ) and Ki for non-competitive dATP inhibition at NrdB was calculated in Sigmaplot using the equation:V=Vmax/ ( 1+[dNTP]/Ki ) In GEMMA , biomolecules are electrosprayed into gas phase , neutralized to singly charged particles , and the gas phase electrophoretic mobility is measured with a differential mobility analyzer . The mobility of an analyzed particle is proportional to its diameter , which therefore allows for quantitative analysis of the different particle sizes contained in a sample ( Kaufman et al . , 1996 ) . The GEMMA instrumental setup and general procedures were as described previously ( Rofougaran et al . , 2008 ) . NrdA , NrdB and NrdB∆99 proteins were equilibrated by Sephadex G-25 chromatography into a buffer containing 100 mM ammonium acetate , pH 7 . 8 . In addition , 2 mM DTT was added to the NrdA protein solutions to increase protein stability . Prior to GEMMA analysis , the protein samples were diluted to a concentration of 0 . 025–0 . 1 mg/ml in a buffer containing 20 mM ammonium acetate , pH 7 . 5 , 0 . 005% ( v/v ) Tween 20 , nucleotides ( when indicated ) , and magnesium acetate ( equimolar to the total nucleotide concentration ) , incubated for 5 min at room temperature , centrifuged and applied to the GEMMA instrument . Protein concentrations higher than normally recommended for GEMMA were needed to see the larger oligomeric complexes and the experiments to measure NrdA-NrdB interactions were run at as low flow rate as possible ( driven by 1 . 4 Psi pressure ) to minimize false interactions that may appear with elevated protein concentration if the flow-rate recommended by the manufacturer is used ( 3 . 7 Psi ) . Most of our experiments were performed with a flow rate driven by 2 Psi . Fast protein liquid chromatography on a Superdex 200 PC 3 . 2/30 column ( with a total volume of 2 . 4 ml ) and ÄKTA prime system ( GE Healthcare ) was performed . The column was equilibrated with SEC buffer containing 50 mM Tris-HCl pH 8 , 50 mM KCl , 10% glycerol , 10 mM magnesium acetate , 2 mM DTT and when applicable either 3 mM ATP or 0 . 1–0 . 5 mM dATP . 50 µL samples containing NrdA , NrdB or both subunits in the presence or the absence of indicated amounts of nucleotides , were pre-incubated for 10 min in room temperature , centrifuged and applied to the column at a temperature of 7°C with a flow rate of 0 . 07 ml/min . When nucleotides were added to proteins , they were also included in the buffer at the same concentration to avoid dissociation of nucleotide-induced protein complexes during the run . Varying concentrations of proteins were used in the range of 10–113 µM and 5–150 µM for NrdA and NrdB respectively . For complex formation , 10–20 µM NrdA and 10–40 µM NrdB were used in ratios of 1:1 or 1:2 . Representative SEC chromatograms in which 20 µM NrdA , 20 µM NrdB or a mixture of 25 µM and 50 µM NrdA and NrdB respectively are shown in Figure 5 . Molecular weight was estimated based on a calibration curve , derived from globular protein standards using high molecular weight SEC marker kit ( GE Healthcare ) . Standard deviations were calculated from at least 3 s experiments . The experiments in Figure 5—figure supplement 1 were run in a similar way but using a Superdex 200 10/300 GL column ( loop size 100 µl and flow rate 0 . 5 ml/min ) with a SEC buffer containing 50 mM Tris-HCl pH 7 . 6 , 150 mM KCl , and 10 mM magnesium chloride . Some experiments with ATP or dATP in the mobile phase were monitored at 290 nm to reduce the absorbance from the nucleotides . Measurements were performed on a Bruker ELEXYS E500 spectrometer using an ER049X SuperX microwave bridge in a Bruker SHQ0601 cavity equipped with an Oxford Instruments continuous flow cryostat and using an ITC 503 temperature controller ( Oxford Instruments , Oxford , United Kingdom ) . Measurement temperatures ranged from 5 to 32 K , using liquid helium as coolant . The spectrometer was controlled by the Xepr software package ( Bruker ) . RNR protein sequences were collected and scored using HMMER ( Eddy , 2011 ) HMM profiles in the RNRdb ( http://rnrdb . pfitmap . org ) . ATP-cones were identified with HMMER and the Pfam ATP-cone HMM profile: PF03477 ( Finn et al . , 2010 ) . Sequences representing the diversity of NrdB were selected by clustering all NrdB sequences from RefSeq at an identity threshold of 70% using VSEARCH ( Rognes et al . , 2016 ) . Sequences were aligned with ProbCons ( Do et al . , 2005 ) and reliable positions for phylogenetic reconstruction were manually selected . A maximum likelihood phylogeny was estimated with FastTree 2 ( Price et al . , 2010 ) . Isothermal titration calorimetry ( ITC ) experiments were carried out on a MicroCal PEAQ-ITC system ( Malvern Instruments Ltd ) in a buffer containing 25 mM HEPES ( pH 7 . 65 ) , 150 mM NaCl , and 10% glycerol , 2 mM tris ( 2-carboxyethyl ) phosphine , and 5 mM MgCl2 . Measurements were done at 10°C . The initial injection volume was 0 . 4 μl over a duration of 0 . 8 s . All subsequent injection volumes were 2–2 . 5 μl over 4–5 s with a spacing of 150 s between the injections . Data for the initial injection were not considered . For dATP binding analysis , the concentration of NrdB in the cell was 12 µM and dATP in syringe 120 or 140 µM . Reverse titrations were performed with 113 µM NrdB in the syringe and 12 or 30 µM dATP in the cell . For titration of dATP into NrdB∆99 , protein concentration in the cell and dATP concentration in the syringe were 50 µM and 900 µM respectively . For dADP binding analysis , the NrdB concentration in the cell was 20–50 µM and ligand concentrations in the syringe were 500–750 µM . For titration of ATP into NrdB , cell and syringe concentrations were 50 and 1600 µM respectively . The data were analyzed using the built-in one set of sites model of the MicroCal PEAQ-ITC Analysis Software ( Malvern Instruments Ltd ) . A fixed ligand/protein stoichiometry of 2 was used for dATP to NrdB titrations . Standard deviations in thermodynamic parameters , N and Kd were estimated from the fits of three different titrations . The purified NrdB , digested by thrombin to remove the hexahistidine tag ( see above ) , was used for crystallization . The protein at a concentration of 9 . 6 mg/ml was mixed with 20 mM MgCl2 , 2 mM TCEP and 5 mM dATP , incubated for 30 min and used for setting up drops using commercially available screens . An initial crystal hit was obtained by the sitting drop vapor diffusion method with a protein to reservoir volume ratio of 200:200 nL and incubated with a 45 μl reservoir at 20 °C in a Triple Drop UV Polymer Plate ( Molecular Dimensions , UK ) . A Mosquito nanoliter pipetting robot ( TTP Labtech , UK ) was used to set up drops , which were imaged by the Minstrel HT UV imaging system ( Rigaku Corporation , USA ) available at the Lund Protein Production Platform ( LP3 ) . Crystals were obtained with a reservoir containing 0 . 2 M CaCl2 , 0 . 1 M Tris pH 8 . 0% and 20% w/v PEG 600 ( condition #57 of the PACT screen ) . The crystals were then further optimized using an additive screen ( Hampton Research , USA ) and diffraction quality crystals were obtained within 1 week from a crystallization solution containing an additional 3% 6-aminohexanoic acid . Crystals were picked up directly from the drop without cryoprotectant and data were collected at 100 K using the ID23-1 beamline at the ESRF , Grenoble , France . The crystals are in space group P1 with unit cell dimensions a = 74 . 0 , b = 90 . 5 , c = 90 . 9 , α = 110 . 8 , β = 99 . 0 , γ = 114 . 1 , containing one NrdB tetramer in the unit cell . The solvent content is 49 . 5% . The diffraction images were integrated using the program XDS ( Kabsch , 2010 ) and scaled using the program Aimless ( Evans and Murshudov , 2013 ) from the CCP4 package ( Winn et al . , 2011 ) . The structure was solved by molecular replacement ( MR ) in two steps , using Phaser ( McCoy et al . , 2007 ) . First the structure of NrdB∆99 was solved to 1 . 7 Å resolution using the most homologous structure in the PDB , that of NrdF from Chlamydia trachomatis ( PDB: 1SYY ) ( Högbom et al . , 2004 ) . This structure was rebuilt manually in Coot ( Emsley et al . , 2010 ) and using Buccaneer ( Cowtan , 2006 ) then refined to convergence using Refmac5 ( Murshudov et al . , 1997 ) and Buster ( Bricogne et al . , 2016 ) . Full details of this structure will be presented elsewhere . In the second step , a multi-body molecular replacement search was carried out using four copies of NrdB∆99 and four copies of the ATP-cone domain from P . aeruginosa NrdA ( PDB: 5IM3 ) ( Johansson et al . , 2016 ) prepared by side chain truncation using Chainsaw ( Stein , 2008 ) . A single solution was found in which all 8 bodies were placed , with a translation function Z score ( TFZ ) of 12 . 2 . After rearrangement of the ATP-cones from the MR solution to the N-termini of their respective core domains , the structure was refined using Refmac5 . Some automatic building was performed using Buccaneer and manual rebuilding in Coot . Final refinement was done using Buster ( Bricogne et al . , 2016 ) . Automatically-generated non-crystallographic symmetry restraints were used . Geometry was validated using the MolProbity server ( Chen et al . , 2010 ) . To establish the nature of the metal ions in the crystal , an anomalous diffraction experiment was conducted at three wavelengths , one on the near-high-energy side of the Fe K edge ( 1 . 72 Å ) , where both Mn and Fe should have a strong anomalous signal , one near the Mn K edge ( 1 . 87 Å ) , where only Mn should have a strong signal , and one on the low energy side of the Mn edge ( 1 . 92 Å ) , where neither Mn nor Fe should have a strong signal but Ca2+ will have a small residual signal ( Figure 6—source data 1 ) . These data were collected at EMBL station P14 of the PETRA-III synchrotron in Hamburg , Germany . The data at different wavelengths were collected in non-overlapping stripes on the crystal to avoid radiation damage that might affect the relative anomalous signals , which were found to be weak at all wavelengths . The structure was refined against all three datasets to a common resolution of 2 . 5 Å and the relative heights of peaks in the anomalous difference maps at were used to identify the metal . These are presented in Figure 6—source data 2 . The refinement was done and anomalous difference maps calculated and analyzed using Buster ( Bricogne et al . , 2016 ) .	When a cell copies its DNA , it uses four different building blocks called deoxyribonucleotides ( dNTPs ) . These consist of one of the four ‘bases’ ( A , T , C and G ) , which pair up to link the two strands of DNA in the double helix , bound to a sugar and a phosphate group . If the cell contains too little or too much of one of these building blocks , an incorrect base may be inserted into the DNA . This results in a mutation , which in bacteria can cause death , and in animals may lead to cancer . The enzyme that fabricates and carefully controls the amount of each dNTP building block inside a cell is called ribonucleotide reductase . Once there are enough building blocks in a cell the enzyme is turned off . A part of the enzyme called the ATP-cone acts as an on/off switch to control this activity . The ribonucleotide reductase consists of a large component and a small component . Until now , studies of the ATP-cone have found it only in the large component of the enzyme . However , when looking through a public database of sequence data , Rozman Grinberg et al . noticed that ribonucleotide reductases in some bacteria have their ATP-cone joined to the small component . Does this ATP-cone also control the amounts of dNTP building blocks inside cells and , if so , how ? Rozman Grinberg et al . studied one such ATP-cone in a ribonucleotide reductase from a bacterium ( named Leeuwenhoekiella blandensis ) found in the Mediterranean Sea . This revealed that when the amount of dNTP building blocks reaches a certain limit , the ATP-cone turns off the enzyme . Examining the three-dimensional structure of the enzyme using a technique called X-ray crystallography revealed that when turned off , the enzyme’s small components are glued together in pairs . This prevents them from working . Rozman Grinberg et al . also discovered that this enzyme contains a new type of metal center with two manganese ions suggesting that a new reaction mechanism may operate in this class of ribonucleotide reductase . These findings support a theory that biological on/off switches can evolve rapidly . In addition to its evolutionary and biomedical interest , understanding how the ATP-cone works might help to improve the enzymes used in industrial processes .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2018	Novel ATP-cone-driven allosteric regulation of ribonucleotide reductase via the radical-generating subunit
Plant leaves constitute a huge microbial habitat of global importance . How microorganisms survive the dry daytime on leaves and avoid desiccation is not well understood . There is evidence that microscopic surface wetness in the form of thin films and micrometer-sized droplets , invisible to the naked eye , persists on leaves during daytime due to deliquescence – the absorption of water until dissolution – of hygroscopic aerosols . Here , we study how such microscopic wetness affects cell survival . We show that , on surfaces drying under moderate humidity , stable microdroplets form around bacterial aggregates due to capillary pinning and deliquescence . Notably , droplet-size increases with aggregate-size , and cell survival is higher the larger the droplet . This phenomenon was observed for 13 bacterial species , two of which – Pseudomonas fluorescens and P . putida – were studied in depth . Microdroplet formation around aggregates is likely key to bacterial survival in a variety of unsaturated microbial habitats , including leaf surfaces . The phyllosphere – the aerial parts of plants – is a vast microbial habitat that is home to diverse microbial communities ( Lindow and Brandl , 2003; Lindow and Leveau , 2002; Vorholt , 2012; Vacher et al . , 2016; Leveau , 2015; Bringel and CouÃce , 2015 ) . These communities , dominated by bacteria , play a major role in the function and health of their host plant , and take part in global biogeochemical cycles . Hydration conditions on plant leaf surfaces vary considerably over the diurnal cycle , typically with wet nights and dry days ( Beattie , 2011; Brewer and Smith , 1997; Magarey et al . , 2005; Klemm et al . , 2002 ) . An open question is how bacteria survive the dry daytime on leaves and avoid desiccation . While leaf surfaces may appear to be completely dry during the day , there is increasing evidence that they are frequently covered by thin liquid films or micrometer-sized droplets that are invisible to the naked eye ( Burkhardt and Hunsche , 2013; Burkhardt and Eiden , 1994; Burkhardt et al . , 2001 ) ( Figure 1A ) . This microscopic wetness results , in large part , from the deliquescence of hygroscopic particles that absorb moisture until they dissolve in the absorbed water and form a solution . One ubiquitous source of deliquescent compounds on plant leaf surfaces is aerosols ( Pöschl , 2005; Tang and Munkelwitz , 1993; Tang , 1979 ) . Notably , during the day , the relative humidity ( RH ) in the boundary layer close to the leaf surface is typically higher than that in the surrounding air , due to transpiration through open stomata . Thus , in many cases , the RH is above the deliquescent point , leading to the formation of highly concentrated solutions in the form of thin films ( <a few µm ) and microscopic droplets ( Burkhardt and Hunsche , 2013 ) . The phenomenon of deliquescence-associated microscopic surface wetness is under-studied , and little is known about its impact on microbial ecology of the phyllosphere and on its contribution to desiccation avoidance and cell survival during the dry daytime . The microscopic hydration conditions around bacterial cells are expected to significantly affect cell survival in the largest terrestrial microbial habitats – soil , root , and leaf surfaces – that experience recurring wet-dry cycles . Only a few studies have attempted to characterize the microscopic hydration conditions surrounding cells on a drying surface under moderate RH and the involvement of deliquescent substrates in this process . Bacterial survival in deliquescent wetness has mainly been studied in extremely dry deserts ( Davila et al . , 2008; Davila et al . , 2013 ) and on Mars analog environments ( Nuding et al . , 2017; Stevens et al . , 2019 ) . Soft liquid-like substances wrapped around cells , whose formation was suggested to be due to deliquescence of solute components , were reported ( Méndez-Vilas et al . , 2011 ) . Yet , the interplay between droplet formation , bacterial surface colonization , and survival , has not been studied systematically . Bacterial cells on leaf surfaces are observed in solitary and aggregated forms . The majority of cells are typically found within surface-attached aggregates , that is biofilms ( Monier and Lindow , 2004; Morris et al . , 1997 ) . This is consistent with the reported increased survival rate in aggregates under dry conditions on leaves , and poor survival of solitary cells ( Monier and Lindow , 2003; Rigano et al . , 2007; Yu et al . , 1999 ) . The conventional explanation for the increased survival in aggregates is the protective role of the extracellular polymeric substances ( EPS ) , a matrix that acts as a hydrogel ( Chang et al . , 2007; Or et al . , 2007; Roberson and Firestone , 1992; Ophir and Gutnick , 1994 ) . Here , we ask if aggregation plays additional roles in protection from desiccation . We hypothesize that the resulting microscale hydration conditions around cells on a drying surface depend on cellular organization ( i . e . solitary/aggregated cells and aggregate size ) and that the microscale hydration conditions ( i . e . droplet size ) affect cell survival . To this end , we designed an experimental system that creates deliquescent microscopic wetness on artificial surfaces . This system conserves some basic important features of natural leaf microscopic wetness while eliminating some of the complexities of studying leaf surfaces directly . The system enabled us to perform a systematic microscopic analysis of the interplay between bacteria’s cellular organization on a surface , microscopic wetness , and cell survival on surfaces drying under moderate humidity . We observed that bacterial cells – aggregates in particular – retained a hydrated micro-environment in the form of stable microscopic droplets ( of tens of µm in diameter ) while the surface was macroscopically dry . We then quantitatively analyzed the distribution of droplet size , its correlation with aggregate size , and the fraction of live and dead cells in each droplet . The significance of our results is discussed in the context of survival strategies on drying surfaces , microbial ecology of the phyllosphere , and possible relevance to other habitats . Studying bacteria in microscopic surface wetness directly on leaves poses a significant technological challenge due to strong auto-fluorescence , surface roughness , and transparency of films and microdroplets . We therefore constructed a simple experimental system , accessible to microscopy , that enables studying the interplay between bacterial surface colonization , cell survival , and microscopic wetness on artificial surfaces . This system enables capturing microscopic leaf wetness central properties , including contribution of deliquescent substrates , droplet persistence , thickness , and patchiness ( Figure 1B - see Materials and methods ) . We studied in depth two model bacterial strains – Pseudomonas fluorescens A506 ( a leaf surface dweller strain; Wilson and Lindow , 1993; Hagen et al . , 2009 ) and P . putida KT2440 ( a soil and root bacterial strain extensively studied under unsaturated hydration conditions; Nelson et al . , 2002; Molina , 2000; van de Mortel and Halverson , 2004; Espinosa-Urgel et al . , 2002 ) . Qualitatively similar results were observed for 16 additional strains ( 13 bacterial species in total - see Materials and methods ) . Briefly , bacterial cells were inoculated in diluted M9 minimal media onto hollowed stickers applied to the glass substrate of multi-well plates and placed inside an environmental chamber under constant temperature and RH ( 28°C; 70% or 85% RH ) ( Figure 1B - Materials and methods ) . Results shown here are from 85% RH though 70% RH yielded qualitatively similar results . At 85% RH , it took about 14 ± 1 hr for the bulk water to evaporate . During this time , for both studied strains , some of the cells attached to the surface and , over time , grew and formed aggregates . Other cells formed cell clusters at the liquid-air interface ( pellicles ) . The rest of the cells remained solitary: either surface-attached , or planktonic . The glass substrate appeared dry to the naked eye after 14 ± 1 hr of incubation . We then examined the surface of the wells under the microscope ( see Materials and methods ) . Remarkably , the surface was covered by stable microscopic droplets , mainly around bacterial aggregates ( Figure 2A–B ) . Notably , while solitary cells were surrounded by miniscule droplets ( possibly similar to those reported by Méndez-Vilas et al . , 2011 ) , larger aggregates ( of ~100 cells ) were surrounded by large droplets measuring tens of µm in diameter . Microscopic wetness was retained around bacterial cells for more than 24 hr , while uncolonized surface areas appeared completely dry . In order to assess the distribution of droplet size and the correlation between droplet size and aggregate size , we scanned a large area of the surface ( ~10 mm2 ) to collect and analyze information on thousands of microdroplets ( Materials and methods ) . We found that droplet size ( measured by droplet area ) follows a power law distribution with similar exponents for the two studied strains ( Figure 2C ) . When droplet size was plotted as a function of area covered by cells within each droplet ( as a proxy for cell number - see Materials and methods ) , a clear positive correlation between cell abundance and droplet size emerged ( Figure 2D ) . Experiments using hydrophobic polystyrene substrate rather than glass also yielded qualitatively similar results ( Figure 2—figure supplement 1 ) . To understand how these microdroplets form , we tested what components of the system were essential to this process . First , we repeated the experiments with fluorescent beads ( 2 µm in diameter ) instead of bacteria . Interestingly , we found that microdroplets formed even around beads ( Figure 2—figure supplement 2 ) with a similar droplet-size distribution , as in experiments with bacteria; and a surprisingly similar correlation between the size of the droplet and the number of beads therein ( Figure 2C–D ) . In a control experiment without any particulates – bacterial cells or beads – a much smaller number of droplets formed ( <1 droplets of >10 µm2 area per mm2 , as opposed to >100 droplets of that size in experiments with bacteria ) . These results indicate that the presence of particles is necessary for droplet formation , whereas biological activity is not . Last , we repeated the beads experiment with pure water instead of M9 medium . This time we did not observe any droplets ( Figure 2—figure supplement 2 ) , indicating that the solutes control droplet formation and retention through their deliquescent properties . To observe the surface’s final drying phase , we used time-lapse imaging , enabling us to capture the receding front of the remaining thin liquid layer and the formation of microdroplets . Retention of droplets around aggregates as well as solitary cells , through pinning of the liquid-air interface , is clearly evident ( Figure 2E , Videos 1–3 ) . The cause of this pinning is the strong capillary forces acting on the rough surfaces produced by the presence of particulates ( Bonn et al . , 2009; Herminghaus et al . , 2008 ) . This phenomenon supports the notion that aggregate sizes ( but possibly also other properties ) determine droplet size . We note that under our experimental conditions , the droplets were not formed through the wetting ‘direction’ of a deliquescence process , by which solid salts absorb water until dissolution . Rather , the deliquescent properties of the solutes prevented complete evaporation at RH above the point of deliquescence of the salts mixture . In summary , both particulates and deliquescent solutes are essential for the differential formation and retention of microscopic wetness around cells and aggregates . Direct measurements of the solute concentrations within microdroplets constitute a technical challenge . To overcome that challenge , we added a fluorescent dye ( Alexa Fluor 647 ) to the initial M9 medium , as a reporter for the solute concentration induced by evaporation . We compared the fluorescent intensity of dye-labeled microdroplets to a calibration curve built by measuring the intensities of known concentrations of the standard M9 supplemented with Alexa 647 ( see Materials and methods , Figure 2—figure supplement 3 ) . We found that the microdroplet solution is highly concentrated – as can be expected from deliquescent wetness – and is estimated to be 23 . 3 ± 3 . 5 ( mean ± SD ) more concentrated than a standard M9 ( ~50 times more concentrated than the diluted 0 . 5x M9 used in our experiments ) ( See Materials and methods , Figure 2—figure supplement 3 ) . The high estimated mean osmolarity within the droplets ( ~6 . 7 Osm/L , see Supplementary file 1 ) likely imposes severe osmotic stress on cells within them . Indeed , growth curves of the two strains ( P . fluorescens and P . putida ) in liquid cultures of equivalent concentrated M9 and M9+NaCl media showed delayed or complete growth inhibition ( Figure 3—figure supplement 5 , Supplementary file 2 ) . This result accords with the observation that cell divisions within droplets was rarely seen in our experiments ( at 85% RH ) . As cells inhabit a heterogeneous landscape of droplets of various sizes , we next asked whether droplet size affects cell survival . We applied a standard bacterial viability assay by adding propidium iodide ( PI ) to the medium ( see Materials and methods ) . Thus , live cells emit green-yellow fluorescence , while dead cells exhibit red emission ( Figure 3A , B ) . The assay’s validity was further confirmed by the observation that following further incubation at 95% RH , YFP-expressing cells were dividing ( some were even motile ) , while red cells lacked signs of physiological activity ( Figure 3—figure supplement 1 , Videos 4 and 5 ) . Notably , although the overall population distribution along droplet size was strain specific , survival of cells was nearly exclusively restricted to large droplets for both strains ( >103 µm2 area; Figure 3C , D , Materials and methods ) . P . putida showed higher overall survival than did P . fluorescens ( 16% vs . 7% , 24 hr after drying ) . We note that the overall survival often varied between experiments , and in some cases P . fluorescens had higher survival than did P . putida . Importantly , regardless of this stochasticity , common to all experiments was a clear trend for both strains: The fraction of live cells within droplets increases with droplet size ( Figure 3E ) . Accordingly , survival probabilities in small droplets ( <102 µm2 area ) were poor ( <5% ) , in contrast to >50% survival of both strains in the largest droplets ( >104 µm2 ) . Next , we sought to understand what the net contribution of droplet size is to cell survival . Analysis of cell survival rates as a function of both aggregate size ( which by itself affects survival; Monier and Lindow , 2003 , cf . Figure 3—figure supplement 2 ) and the size of the droplet they inhabit , shows that for both strains , droplet size strongly affects survival , whereas aggregate size has only a marginal ( P . fluorescens ) or moderate ( P . putida ) effect on survival ( Figure 3F , G ) . The relative contribution of each of these two variables was also assessed by a multinomial logistic regression model , giving significantly higher weight to droplet size in comparison to aggregate size , for both strains ( Figure 3—figure supplement 3 ) . To further study droplet size effect on survival , we repeated the drying experiment , but inoculated the cells into the drying medium only at a later stage – closer to the macroscopic drying stage – so that the cells did not have time to grow and form aggregates , and were thus mostly solitary . Notably , live cells were observed nearly exclusively in large droplets ( >103 µm2 area , cf . Figure 3—figure supplement 4 ) , and survival increased with droplet size . These results indicate that large droplets promote cell survival even when aggregates are absent . Experiments with 16 additional strains , including Gram-negative and Gram-positive bacteria from a variety of microbial habitats , yielded qualitatively similar results to those described in the preceding paragraphs ( Table 1 ) . Although not all the strains formed aggregates under our experimental conditions , the general picture was same for all strains: Larger droplets were observed around aggregates or surface areas more densely populated by cells ( for strains that did not form aggregates ) , and higher survival was observed in larger droplets . Lastly , we repeated our experiments using solutes and microbiota extracted from the surface of a natural leaf . We found that stable microdroplets also formed around natural microbiota cells , in some cases only at higher RH ( >85% ) or at lower temperatures , suggesting that condensation is involved in microdroplet formation . Furthermore , the microscopic wetness from natural leaf wash was visibly similar to those in our experiments with inoculated bacteria and a synthetic medium ( Figure 4 , Figure 4—figure supplement 1 ) . Our study demonstrates that stable microdroplets of concentrated liquid solutions form around cells and aggregates on bacterial-colonized surfaces that are drying under moderate to high RH . We show that bacterial cell organization on a surface strongly affects the microscopic hydration conditions around cells , and that droplet size strongly affects cell survival . We reveal an additional function of bacterial aggregation: improving hydration by retaining large stable droplets ( >tens of µm in diameter ) around aggregates . Why survival is enhanced in larger droplets remains an open question . We hypothesize that larger droplets provide favorable conditions due to higher water potential; further research is required to test this hypothesis . We note that the evaporation dynamics of a drop of a liquid solution – even without bacteria – is a surprisingly rich and complex physical process and a subject of intensive research ( de Gennes et al . , 2013; Bonn et al . , 2009 ) . Our results point to two central mechanisms promoting the formation and stability of microdroplets around bacterial aggregates: The first is pinning of the liquid-air interface due to the large interfacial tension force associated with the rough surfaces of particulate aggregates ( Herminghaus et al . , 2008; Bonn et al . , 2009 ) , as observed in Videos 1–3 . The second is the deliquescent property of solutes that prohibits complete evaporation of the pinned droplets at RH that is higher than the point of deliquescence of the solutes , such that the droplets are in equilibrium with the surrounding humid air . We suggest that bacterial self-organization on surfaces can improve survival in environments with recurrent drying that lead to microscopic wetness . A simple conceptual model that captures the system’s main components and their interactions is depicted in Figure 5A . Aggregation is an important feature that can affect self-organization , and in turn , the resulting waterscape , by increasing the fraction of the population that ends up in large droplets . Preliminary evidence for this is provided by the comparison of the fraction of the population residing in droplets above a given size , using beads , ‘solitary’ and ‘aggregated’ cells as particles ( Figure 5B , Figure 5—figure supplement 1 ) . The interplay between self-organization , waterscape , and survival is an intriguing open question that merits further research . Interestingly , the ecological origin of the strains ( Table 1 ) did not always predict their survival rates . Some phyllospheric bacteria ( mostly plant pathogens ) exhibited low survival , soil bacteria exhibited variable survival rates , E . coli exhibited a surprising medium survival , and the aquatic strain P . veronii exhibited high survival . Survival in microscopic surface wetness is likely a complex trait that combines physiological adaptation of the individual cell and collective protection that results from self-organization and cooperation ( i . e . aggregation ) . In nature , bacteria live in complex communities comprised of many bacterial species and are exposed to various chemical and physical environmental cues . Thus , our single-strain experiments , with M9 medium on glass-bottom wells , may not capture survival strategies that might be triggered by environmental cues and that rely on other members of the community . For example , joining existing aggregates of other species can be a beneficial strategy in environments with recurrent drying events ( Grinberg et al . , 2019; Steinberg et al . , 2019 ) . Obviously , there are more differences between natural leaf surfaces and our simplified experimental system . Firstly , leaf surfaces have heterogeneous 3D topography due to leaf microscale anatomy such as the cavities between epidermal cells , stomata openings , and trichomes ( Koch et al . , 2008 ) . This microscale topography affects drying and wetting of the leaf surface , and hence can impact droplet formation both by its effect on interfacial forces and pinning as well as imposing stronger flow upon topological sinks . Secondly , leaf surfaces tend to be hydrophobic to a degree that varies among plant species ( Koch et al . , 2008 ) . The impact of both microscale topography and surface hydrophobicity on drying and droplet formation can be studied using artificial leaves ( Doan and Leveau , 2015; Soffe et al . , 2019 ) or leaf cuticle peels ( Schönherr and Riederer , 1986; Remus-Emsermann et al . , 2011 ) . Finally , the chemical composition of leaf surface wetness varies considerably with multiple factors including plant species , soil characteristics ( e . g . salinity ) , geography , and environmental variables that affect atmospheric aerosol composition , deposition , and retention , such as wind and rain ( Pöschl , 2005; Tang and Munkelwitz , 1993; Tang , 1979 ) . All of these factors are likely to affect the formation and retention of microscopic leaf wetness . Our results suggest that microscopic surface wetness , predicted to occur globally on plant leaves ( Burkhardt and Hunsche , 2013 ) , can explain how microorganisms survive on leaf surfaces during daytime by avoiding complete desiccation . Yet , they also imply that phyllospheric bacteria have evolved mechanisms to cope with the highly concentrated solutions associated with deliquescent wetness . The ability to tolerate periods of such high salinities could thus be a ubiquitous and necessary trait for phyllospheric bacteria . Better understanding of bacterial survival in microscopic deliquescent surface wetness , and how it is affected by agricultural practices and anthropogenic aerosol emissions , is thus of great importance to microbial ecology of the phyllosphere and to plant pathology . Finally , as deliquescent substances are prevalent in many other microbial habitats , it is safe to assume that deliquescent microscopic wetness occurs in many microbial habitats , including soil and rock surfaces ( Davila et al . , 2008; Davila et al . , 2013 ) , the built environment , human and animal skin , and even extraterrestrial systems ( e . g . Mars; Nuding et al . , 2017; Stevens et al . , 2019 ) . Moreover , microscopic surface wetness is likely to have a significant impact not only on survival , but also on additional key aspects of bacterial life , including motility , communication , competition , interactions , and exchange of genetic material , as demonstrated for soil and other porous media ( Tecon et al . , 2018; Or et al . , 2007 ) . Microbial life in deliquescent microscopic surface wetness remains to be further explored . A simple experimental system , accessible to microscopy , that enables studying the interplay between bacterial surface colonization , cell survival , and microscopic wetness on artificial surfaces was built ( see Figure 1B and section Drying surface experiments ) . Fluorescently tagged bacterial cells are inoculated in liquid media onto hollowed stickers adhered to the glass substrate of multi-well plates and placed inside an environmental chamber under constant temperature and RH ( Figure 1B , and Drying surface experiments and Strains and culture condition ) . After macroscopic drying is achieved , plates are examined under the microscope ( see Microscopy ) and microscopic wetness , bacterial surface colonization , and cell survival are analyzed ( see Image analysis , Statistical analysis , and Estimation of solution concentrations within droplets ) . Similar experiments with natural leaf washes are described in the section Natural leaf washes . Imaging spacers ( 20 mm SecureSeal SS1 × 20 , Grace Bio-Labs ) were used to confine the inoculum on the surface of six-well glass bottom plates ( CellVis ) ( Figure 1B ) . The spacer was used to reduce flow dynamics effect that result in transfer of biomass to the edge of an evaporating body of liquid drops on flat surfaces ( e . g . coffee ring effect; Deegan et al . , 1997; Larson , 2017 ) . Reduction of flow was achieved through a more spatially uniform evaporation rate . The corners of the spacer were cut to fit the well , adhesive liner was removed from one side of the spacer , and the exposed adhesive was applied to the center of the well by applying gentle pressure against the glass using a sterile disposable cell spreader . The upper liner was removed and the hollow of the spacer was loaded with 340 µl of diluted suspended cells ( ~2×103 cell/ml ) at half-strength M9 medium ( with 2 mM glucose conc . ) . For survival assay , propidium iodide ( component B , LIVE/DEAD Bac-Light Bacterial Viability Kit , L-7012 , Molecular Probes ) was added to the starting inoculum to obtain a final concentration of 20 nM . The typical ‘live’ SYTO dye was not used; instead , we used the constitutive YFP expression of live cells ( see below ) as indication of living cells . In the experiments with fluorescent beads , rhodamine-tagged micro particles ( 2 µm ) based on melamine resin were used ( melamine-formaldehyde resin , FLUKA ) . The plates were placed , with the plastic lid open , on the uppermost shelf of a temperature- and humidity-controlled growth chamber ( FitoClima 600 PLH , Aralab ) . Temperature was set to 28°C , RH to 70% or 85% , and fan speed to 100% . Prior to the microscopy imaging acquisition , diH2O was added to the empty spaces between the wells of the plate , plates were covered with the plastic lid , and the plate’s perimeter was sealed with a stretchable sealing tape to maintain a humid environment ( >95% RH ) . Pseudomonas fluorescens A506 ( Wilson and Lindow , 1993; Hagen et al . , 2009 ) and Pseudomonas putida KT2440 ( Nelson et al . , 2002 ) ( ATCC 47054 ) were chromosomally tagged with YFP using the mini-Tn7 system ( Choi and Schweizer , 2006 ) ( Plasmid pUC18T-mini-Tn7T-Gm-eyfp and pTNS1 , Addgene plasmid # 65031 , and # 64967 respectively ( Choi et al . , 2005 ) . Prior to the gradual drying experiments , strains were grown in LB Lennox broth ( Conda ) supplemented with gentamicin 30 µg/ml for 12 hr ( agitation set at 220 rpm; at 28°C ) . 50 µl of the 12 hr batch culture was transferred into 3 ml of fresh LB medium , and incubated for an additional 3–6 hr ( until OD reached a value of ~0 . 5–0 . 7 ) . Suspended cells were transferred to a half-strength M9 medium supplemented with glucose by a two-step washing protocol ( centrifuge at 6000 rcf for 2 min . , and resuspension of the pellet in 500 µls medium ) . The half-strength M9 medium consisted of 5 . 64 g M9 Minimal Salts Base 5x ( Formedium ) , 60 mgs of MgSO4 , and 5 . 5 mgs of CaCl2 per liter of de-ionized water supplemented with 360 mgs glucose as a carbon source ( final glucose concentration of 2 mM ) . The full list of strains used in this study is given in Table 1 . Microscopic inspection and image acquisition were performed using an Eclipse Ti-E inverted microscope ( Nikon ) equipped with 40x/ ( 0 . 95 N . A . ) air objective . A LED light source ( SOLA SE II , Lumencor ) was used for fluorescence excitation . YFP fluorescence was excited with a 470/40 filter , and emission was collected with a T495lpxr dichroic mirror and a 525/50 filter . Propidium iodide fluorescence was excited with a 560/40 filter , and emission was collected with a T585lpxr dichroic mirror and a 630/75 filter ( filters and dichroic mirror from Chroma ) . A motorized encoded scanning stage ( Märzhäuser Wetzlar GmbH ) was used to collect multiple stage positions . In each well , five xy positions were randomly chosen , and 5 × 5 adjacent fields of view ( with a 5% overlap ) were scanned . Images were acquired with an SCMOS camera ( ZYLA 4 . 2PLUS , Andor ) . NIS Elements 5 . 02 software was used for acquisition and basic image processing . The images were exported from NIS Elements as four separate 16-bit grayscale images per image: bright field ( BF ) , YFP fluorescence ( green ) , propidium fluorescence ( red ) , and a shorter wavelength fluorescence that highlights the droplets ( blue ) . Image analysis was performed in MATLAB . The droplets were segmented by processing the blue fluorescence channel . Droplets were segmented by setting thresholds on the image intensity and gradient following Gaussian filtering ( the centers of the droplets are brighter than their periphery and background , and the gradient is more pronounced at the periphery ) . The two resulting masks were combined , and holes in the connected components were removed . Live and dead cells within each droplet were segmented by the histogram-based threshold of the green and red fluorescent channels respective intensities , producing binary segmentation and live/dead classification of the cells . The segmented droplet image was then used to assign cells and aggregates to their ‘host’ droplet , and to quantify the live/dead surface coverage within each droplet and aggregate . Our analysis relies on the projected 2D features of 3D objects: droplets and bacterial cells and aggregates . Although some information is lost in the projection , it was deemed a necessary tradeoff for the analysis of the large scanned area and the quantity of data involved . We assume that the relationship between droplet area and volume is monotonous , and that the great majority of cellular aggregates are single layered . To affirm these assumptions , we performed 3D analysis using z-stacking and 3D deconvolution on a small surface area . This analysis verified that our droplet identification and segmentation does not capture flat discolorations as droplets , and that indeed the cells within the droplets are generally arranged in a single layer on the surface , or suspended in the liquid at densities low enough to maintain the validity of 2D projections . Data analyses and statistics for experiments with bacterial cells were based on microscopy images of five different surface sections ( each of an area of 2 . 5 mm2 ) per well . Data analyses and statistics for experiments with beads were based on microscopy images of surface sections of areas of 10 mm2 . For statistical analysis of mean values and standard errors , droplets and aggregates were binned by their size on a logarithmic scale . In Figure 2C , standard errors are based on the five surface sections ( of 2 . 5 mm2 ) per strain ( n = 5 ) and nine different surface sections ( of 1 . 1 mm2 ) for the beads experiment ( n = 9 ) . In Figure 2D and Figure 3E , standard errors are calculated for all droplets within each bin ( size range of droplets ) of the combined data of the five surface sections for experiments with bacteria . In Figure 3B , C and Figure 4B , data is combined for all five surface sections . In Figure 3F , G standard errors are calculated for all aggregates within each bin of the combined data of the five surface sections . We employed two different methods for the extraction and drying of leaf washes .	A single plant leaf can be home to about 10 million bacteria and other microbes . These microscopic organisms are part of a larger community of microbes – the microbiome – that plays an important role in the life and health of their plant host . Like all other organisms , bacteria need water to survive , but the surfaces of leaves experience daily changes in moisture , tending to be much wetter at night than during the day . While the surfaces of leaves often appear dry during the day , previous studies suggest they may actually be covered by thin films or tiny droplets of fluid that are invisible to the naked eye . This microscopic wetness forms because hygroscopic particles such as aerosols , which tend to absorb moisture from the air , are common on the leaf surface . These molecules absorb water until they become dissolved in it , leaving behind a concentrated solution ( a process known as deliquescence ) . However , it is not clear if this microscopic wetness can protect bacteria from drying out . Here , Grinberg , Orevi et al . investigated how bacteria , including several species that are commonly found on plants , survived episodes of drying on an artificial surface that produces microscopic wetness . The experiments revealed that as the surfaces dried out , stable microscopic droplets formed around the bacterial cells . The droplets that formed around aggregates of bacterial cells were larger than those that formed around solitary cells . Bacteria inside these droplets can survive longer than 24 hours , and survival rates were much higher in larger droplets . Further experiments found that 11 other species of bacteria could also survive an episode of drying for over 24 hours if microscopic droplets formed around them . Together , these findings suggest that by organizing themselves into aggregates , bacteria can improve their chance of surviving on the surface of leaves and other environments that are frequently exposed to drying . These results help explain how microbes avoid drying and survive during the daytime on leaf surfaces . Understanding how microscopic leaf wetness protects the plant microbiome is important because it helps explain how it can be disrupted by agricultural practices and human-made aerosols , information that can be used to better protect plants . Microscopic surface wetness is likely to occur in many other situations including in the soil , on human and animal skin , and in homes and workplaces . These findings may have broad implications on the way we understand bacterial life on these seemingly dry surfaces , potentially leading to future benefits for human health , agriculture , and nature conservation .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "microbiology", "and", "infectious", "disease" ]	2019	Bacterial survival in microscopic surface wetness
During embryonic development signalling pathways act repeatedly in different contexts to pattern the emerging germ layers . Understanding how these different responses are regulated is a central question for developmental biology . In this study , we used mouse embryonic stem cell ( mESC ) differentiation to uncover a new mechanism for PI3K signalling that is required for endoderm specification . We found that PI3K signalling promotes the transition from naïve endoderm precursors into committed anterior endoderm . PI3K promoted commitment via an atypical activity that delimited epithelial-to-mesenchymal transition ( EMT ) . Akt1 transduced this activity via modifications to the extracellular matrix ( ECM ) and appropriate ECM could itself induce anterior endodermal identity in the absence of PI3K signalling . PI3K/Akt1-modified ECM contained low levels of Fibronectin ( Fn1 ) and we found that Fn1 dose was key to specifying anterior endodermal identity in vivo and in vitro . Thus , localized PI3K activity affects ECM composition and ECM in turn patterns the endoderm . Understanding the mechanisms regulating axis formation in vertebrate development has been one of the central questions in modern developmental biology . It involves the activity of a number of conserved signal transduction pathways . However , the complex set of morphological rearrangements that occur during gastrulation make it difficult to uncouple the role of specific signals in mediating cell migration from those acting directly on cell specification ( Villegas et al . , 2010 ) . Embryonic stem cell ( ESC ) differentiation offers a tractable in vitro model , which can be used to complement the analysis of embryos . ESCs are karyotypically normal , self-renewing and pluripotent cell lines derived from the mammalian blastocyst that can be driven to differentiate toward all the three germ layers . Adherent ESC differentiation models , therefore , allow direct access to mechanisms regulating cell fate decisions in a stable defined environment . The inductive activity of specific signalling pathways can be tested on isolated progenitor populations , allowing the deciphering of specific target cells and paracrine interactions ( Murry and Keller , 2008 ) . The specification of visceral organs is intricately linked to the establishment of positional identity and begins with the formation of the endoderm germ layer . Endoderm progenitors will give rise to the entire digestive track in addition to thymus , thyroid , liver , pancreas , lungs , gallbladder as well as the extra-embryonic component of the visceral yolk sac . Therefore , understanding the basis for endoderm induction is the first step in attempting directed differentiation to organ cell types . In mammals , endoderm induction occurs in two waves . The first occurs at implantation and leads to the formation of the extra-embryonic visceral endoderm and the second occurs during gastrulation . During gastrulation , the precursors of the definitive endoderm emerge from the anterior region of a transient embryonic processing centre known as the primitive streak ( PS ) ( Zorn and Wells , 2009 ) . These precursors will either push away , or intercalate into , the visceral endoderm that surrounds the embryo proper ( Kwon et al . , 2008; Burtscher and Lickert , 2009 ) . The first endoderm to emerge from the PS region is the anterior definitive endoderm ( ADE ) or prospective foregut . These cells move forward along the midline of the embryo to convey anterior information to the emerging neural axis and eventually give rise to the liver , pancreas and other derivatives of the foregut ( Zorn and Wells , 2009 ) . ADE induction is dependent on a network of specific transcription factors downstream of Wnt and high levels of Nodal related TGF-beta signalling . However , as this emerging endodermal signalling centre moves forward during gastrulation , it starts to express antagonists of both these signals , protecting the anterior neural ectoderm from the mesoderm inducing properties of these pathways ( Tam and Loebel , 2007 ) . In addition to Nodal and Wnt signalling , we have recently uncovered an additional requirement for Fibroblast Growth Factor ( FGF ) signalling in endoderm and ADE specification ( Morrison et al . , 2008 ) . FGF signalling is also required to induce the morphological rearrangements that are normally associated with gastrulation and endoderm migration ( Yamaguchi et al . , 1994 ) , including the induction of an epithelial to mesenchymal transition ( EMT ) that enables the beginning of cell migration . EMT and cell migration are also regulated by the substrate upon which cells adhere , the extra-cellular matrix ( ECM ) ( Hynes , 2009 ) . For example , PS formation can be inhibited on the anterior side of the embryo as a consequence of the visceral endoderm depositing ECM components that block cell migration ( Egea et al . , 2008 ) and FGF signalling has been linked to visceral endoderm-dependent deposition of basement membrane components in ESC differentiation ( Villegas et al . , 2010 ) . However , while these studies link FGF signalling to ECM composition and cell migration , they do not address how ECM could be involved in patterning the embryonic axis or induction of specific cell types during ESC differentiation . In this study , we resolve the ability of FGF signalling to induce both endoderm and morphogenetic movements . We uncoupled the signalling downstream of the FGF receptor using an in vitro mESC based model for endoderm induction . We found that endodermal patterning by FGF was dependent on a PI3K/Akt1 activity that temporally restricts EMT . This unexpected PI3K/Akt1 activity , which we show is required both in vivo and in vitro , is transduced via a modification to ECM composition , which in itself , promotes cell type specification . Using liquid chromatography and mass spectrometry ( LC-MS ) , we further identified ECM components and found that Fibronectin ( Fn1 ) levels were an essential determinant of the ECM-dependent endoderm patterning activity . Taken together , our findings link anterior-posterior ( A–P ) patterning of the endoderm to EMT and uncover a novel ECM-based mechanism for the activity of PI3K/Akt1 signalling during germ layer specification . To address the basis of the requirement for FGF during endoderm differentiation and anterior specification , we applied several small molecule inhibitors to a defined adherent mESC endoderm differentiation model ( Figure 1A; Livigni et al . , 2009 ) . Differentiation conditions are shown in Figure 1A and the medium used is defined in ‘Materials and methods’ section . We used BMP4 together with modest doses of the Nodal-like TGF-beta Activin A , to promote the differentiation of mESCs to epiblast-like cells , followed by the further differentiation of epiblast towards PS and endoderm in the presence of FGF and Activin A . Through the use of this staged differentiation protocol we have been able to generate committed ADE from ESCs more efficiently ( Morrison et al . , 2008 ) . We monitored the formation of ADE , anterior primitive streak ( APS ) and endoderm in general , using either single or combinatorial fluorescent reporter ESC lines generated by gene targeting . These included the fluorescent reporter gene RedStar under the transcriptional control of the early anterior endoderm marker Hhex ( HRS ) ( Figure 1—figure supplement 1A ) and a GFP under the control of the Goosecoid ( Gsc ) locus , a marker of PS and APS ( Figure 1—figure supplement 1A , B ) . For additional resolution of ESC differentiation to endoderm we used the cell surface marker , Cxcr4 ( Morrison et al . , 2008 ) . ADE was identified as either double positive for Hhex and Cxcr4 ( H+C+ ) or Hhex and Gsc ( H+Gsc+ ) . Differentiation of these reporter lines was assessed by quantitative RT-PCR ( q-RT-PCR ) ( Figure 1—figure supplement 1C ) , which confirmed that these cells pass through the in vitro equivalents of specific developmental stages , including PS , APS ( or mesendoderm ) , and ADE , thereby recapitulating in vivo development . Based on these data , we designated the specification phase leading up to PS specification as phase 1 and the stage of subsequent mesoderm and endoderm segregation as phase 2 ( Figure 1A ) . The transition between phase 1 and 2 is marked by the appearance of Gsc-GFP cells during day 3 . 0–3 . 5 of ESC differentiation . 10 . 7554/eLife . 00806 . 003Figure 1 . Inhibition of PI3K signalling disrupts Hhex positive ADE specification , but does not interfere with mesendoderm induction . ( A ) Schematic representation of ESC differentiation towards ADE . PrE: primitive ectoderm , PS: primitive streak , APS: anterior primitive streak , ADE: anterior definitive endoderm . ( B ) Fluorescence and brightfield images of HRS ADE cultures generated in the absence or presence of LY . LY was present throughout phase 2 . ( C ) ADE ( H+C+ ) , but not Cxcr4+ mesendoderm ( H−C+ ) was impaired by LY treatment . Gates were set using parental E14 Tg2A ( E14 ) cells without fluorescent reporters . Hereinafter , LY treatment refers 10 μM at d3 . 5 and 20 μM at d4 . 5 onwards . For some of the experiments described in this paper the base media SFO3 was substituted by ADEM with identical results . DOI: http://dx . doi . org/10 . 7554/eLife . 00806 . 00310 . 7554/eLife . 00806 . 004Figure 1—figure supplement 1 . A model for monitoring endoderm specification from ESCs . ( A and B ) Schematic showing the strategy used in the construction of the HRS/Gsc-GFP dual reporter cell line . ( A ) is adapted from Figure 1 in Morrison et al . , 2008 . ( C ) Response of key lineage markers by q-RT-PCR in our defined culture system confirms that ADE differentiation in vitro recapitulates the in vivo developmental program of gene expression . The expression of the pluripotency markers Oct4 and Nanog decreased as differentiation proceeded . Mesendoderm markers T and Mixl1 peaked between phase 1 and 2 of differentiation and their expression was down-regulated as markers of anterior endoderm ( ADE ) , Hhex , Lefty1 and Fzd5 were up-regulated . Transcript levels were normalised to the Tbp value obtained for each sample . Normalised values are related to the level obtained for ESC . DOI: http://dx . doi . org/10 . 7554/eLife . 00806 . 00410 . 7554/eLife . 00806 . 005Figure 1—figure supplement 2 . MAPK kinase signalling is required for endoderm induction . ( A ) Flow cytometry on differentiating HRS/Gsc-GFP cells showing the effect of specific MAPK inhibitors on both mesendoderm differentiation and ADE emergence ( d6 ) . Inhibitors were added during phase 2 of differentiation . ( B ) Q-RT-PCR showing the response of mesoderm and endoderm markers to MAPK signalling inhibition in endoderm differentiation . Transcript levels were normalised to the Tbp value obtained for each sample . Normalised values are related to the level obtained for ESC . ( C ) Fluorescence microscopy on differentiating HRS cells ( d6 ) showing a failure in ADE specification in the presence of MAPK inhibitors . ( D ) Morphology and gene expression in response to inhibition of p38 with the SP inhibitor . Hhex-IRES-Venus differentiating cells , reporting low levels of Hhex ( Canham et al . , 2010 ) , showed broad Hhex/Venus expression when cultures were exposed to SP during ADE differentiation . Cell morphology was different from that observed in normal conditions and cells also co-expressed the pluripotency marker Oct4 . ( E ) A blockade to ERK and p38 MAPK signalling during the phase 1 of differentiation disrupted the formation of mesendodermal intermediates . Treatment with PD03 led to the generation of tightly packed ESC-like colonies surrounded by large flat cells , whereas SB treatment produced highly homogeneous non-mesendodermal cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00806 . 005 Application of inhibitors of MEK- ( PD0325901–PD032− ) , JNK ( SP600125–SP− ) , p38 ( SB239063–SB− ) , and PI3K ( LY249002–LY− ) during phase 2 of differentiation all resulted in an inhibition of ADE specification ( Figure 1C , Figure 1—figure supplement 2A–C ) . However , while inhibition of different MAPKs ( with PD032 , SP and SB ) also resulted in a dramatic reduction in mesendodermal and pan-endodermal , Hhex−Cxcr4+ ( H−C+ ) populations , only PI3K inhibition with LY had a specific effect on induction of ADE ( Figure 1B , C ) . While each of these kinases were required for ADE specification at a certain level , some Hhex+ cells were observed in SP treated cultures , although endodermal gene expression was reduced ( Figure 1—figure supplement 2B ) and these cells co-express the ESC marker Oct4 ( Figure 1—figure supplement 2D ) . Thus , all these kinases were required broadly for ESC differentiation towards mesoderm and endoderm , but only PI3K appeared specific to the transition between mesendoderm and committed ADE . To confirm that these signalling requirements were specific to phase 2 , we also examined the effects of these inhibitors in phase 1 . Inhibition of either JNK or PI3K was highly toxic , leading to extensive cell death , even at low concentrations ( Supplementary file 1 ) . Inhibition of MEK resulted in ESC-like colonies that maintained Nanog expression ( Figure 1—figure supplement 2B , E ) consistent with a requirement for MEK signalling during early ESC differentiation ( Kunath et al . , 2007; Stavridis et al . , 2007; Ying et al . , 2008 ) . Suppression of p38 signalling with SB also blocked differentiation toward APS derivatives , although SB was not able to support ESC-like phenotypes ( Figure 1—figure supplement 2E ) . Gene expression analysed by q-RT-PCR also indicated that PI3K signalling was essential for anterior endoderm specification . We found that the expression of pan-endodermal markers Sox17 , Foxa2 and Gsc were enhanced by PI3K inhibition , while induction of all ADE specific gene expression ( Hhex , Cer1 , Lefty1 , Sfrp5 and Fzd5 ) was repressed ( Figure 2A ) . The loss of Hhex expression in the absence of PI3K signalling within the Sox17+/Foxa2+ progenitor population was confirmed by immunocytochemistry ( IC ) ( Figure 2B and data not shown ) . Taken together these findings suggest that PI3K is required after PS stages for the acquisition of anterior positional identity in the endoderm and , when it is suppressed , cells acquire an earlier more naïve or potentially posterior endodermal state . 10 . 7554/eLife . 00806 . 006Figure 2 . Inhibition of PI3K signalling inhibits ADE but not DE differentiation . ( A ) Differential regulation of ADE , pan-endoderm and pluripotency markers as a result of LY treatment . Q-RT-PCR showing relative gene expression for ESC , endodermal and ADE markers in ESC differentiation towards endoderm . Transcript levels were normalised to the Tbp value obtained for each sample . Normalised values are shown relative to ESC expression level . ESCs were differentiated to ADE using the protocol described in Figure 1 in the presence or absence of LY . ( B ) Treatment with LY impaired HRS but not Sox17 expression . Fluorescence images showing HRS expression and Sox17 antibody staining at day 6 of differentiation . ( C ) Phosphorylation of Akt1 in ESC differentiation . Western blot showing phosphorylation of Akt1 during a time course for endoderm differentiation toward ADE ( upper panel ) and its dephosphorylation after the addition of 20 µM LY at different time points ( middle panel , h; hours , top and bottom panel , d; days ) . Total ERK was used as loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 00806 . 00610 . 7554/eLife . 00806 . 007Figure 2—figure supplement 1 . PI3K is not specifically required for ADE survival . ( A ) Flow cytometry showing Annexin V staining of Gsc- and Gsc+ differentiating ESCs . At the time of LY application ( Day 3 ) some apoptosis is apparent in the Gsc- population . Caspase inhibitor Z-VAD-FMK ( 10 µM ) was used to inhibit apoptosis . ( B ) Flow cytometry at day 6 showing ADE induction and that cell death inhibition does not rescue the LY mediated block to ADE induction . High dose of LY ( 50 µM ) led to extensive cell death during differentiation and therefore no data were available at this time point . ( C ) Quantification of flow cytometry for each day of differentiation in both Gsc- and ( D ) Gsc+ differentiating cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00806 . 007 We confirmed that PI3K was active during normal endoderm differentiation based on the phosphorylation of its main target , Akt1 ( pAkt1 ) . pAkt1 was down-regulated just as cells enter phase 1 , but levels were rapidly restored at day 2 of differentiation , and maintained throughout the endodermal differentiation ( Figure 2C , upper panel ) . LY rapidly suppressed Akt1 phosphorylation and while pAkt1 levels exhibited a slight recovery 6 hr after LY addition , they remained low throughout differentiation ( Figure 2C , middle and bottom panel ) . As phosphorylation of Akt1 is required for cell survival , we asked if the phenotype of LY was driven by selective death of ADE precursor cells . Doses of LY that completely suppressed pAkt1 ( 50 μM ) resulted in extensive cell death ( Figure 2 , Figure 2—figure supplement 1B , Supplementary file 1 ) , while medium doses ( 20 μM ) lead to a transient increase in apoptosis in Gsc− , but not Gsc+ populations of day 3 cultures as judged by Annexin V staining ( Figure 2 , Figure 2—figure supplement 1A , C , D ) . However , by day 4 , LY had no effect on survival of either Gsc− or Gsc+ cells ( Figure 2—figure supplement 1C , D ) . To further confirm that LY was not blocking ADE induction through the selective cell death of a precursor population , we inhibited apoptosis with Z-VAD-FMK , a specific caspase inhibitor . The addition of this inhibitor to LY containing cultures eliminated the transient increase in apoptosis observed in the Gsc− population , but produced no rescue of the ADE induction ( Figure 2 , Figure 2—figure supplement 1B ) . We found that PI3K was required early during endoderm specification for the eventual induction of anterior cell types . Figure 3A shows the induction of ADE at day 7 , in cultures exposed to LY for different periods of time , and demonstrates that PI3K was essential between days 3 and 5 of differentiation . Thus , the induction of ADE was dependent on the action of PI3K early during phase 2 . During this phase of differentiation two populations of Gsc-GFP expressing cells ( Gsclow and Gschigh ) become apparent . We found that Hhex+ ADE cells were always Gschigh suggesting they arose specifically from the earlier GschighHhex− population . We also found that this Gschigh population was diminished in the presence of LY while the Gsclow population was considerably expanded ( Figure 3—figure supplement 1A , B ) . Thus , PI3K was required in early PS precursors to generate Gschigh cells that can later give rise to Hhex+ ADE . When LY-treated cells were released from the PI3K block and cultured for an additional 3 days in phase 2 differentiation conditions , we observed a complete recovery of Hhex induction , suggesting that in the absence of PI3K cells fail to progress in differentiation , but they were maintained in a less differentiated endodermal naïve state ( Figure 3B ) . To better characterize this LY-dependent naïve state we assessed a panel of markers by q-RT-PCR and IC . Alongside the high levels of endodermal markers previously observed ( Sox17 , Foxa2 ) in LY-treated cultures , they also showed expression of early primitive streak markers such as Mixl1 and Cdh2 , but not mesodermal specific gene expression ( Meox1 , Meox2 , Mesp1 , Mesp2 , Isl1 , Gata1 ) ( Figure 3—figure supplement 1C and data not shown ) . While these cultures expressed high level of PS gene expression , this appears to be predominantly endoderm as 90% of the Gsc+ cells in these LY-treated cultures co-expressed Sox17 ( Figure 3—figure supplement 1D , E ) . These cells also do not up-regulate the expression of posterior ( Cdx2 ) , or visceral ( Hnf4a , Dab2 ) endoderm markers . Taken together , these data indicate that blocking PI3K signalling retains endodermal cells in a naïve state . 10 . 7554/eLife . 00806 . 008Figure 3 . PI3K promotes exit from a naïve endoderm state . ( A ) PI3K is essential during endoderm segregation . Time course for endoderm differentiation analysed by flow cytometry on the HRS/Gsc-GFP cells showing a requirement for PI3K signalling during days 3–5 of ADE differentiation . Cells were differentiated under normal conditions or treated with LY for different periods of time . The period of LY treatment is stated for each graph and ADE induction was assessed at day 7 of differentiation . Gates were set with parental E14s as in Figure 1 . ( B ) Cells exposed to LY during d3–6 could be returned to normal differentiation to generate ADE . Flow cytometry showing the emergence of the Hhex+/Gsc+ ADE after LY-treated naïve endoderm was returned to normal differentiation for a further 3 days . Gates were set as in A . ( C ) Hepatic differentiation of ADE and naïve endoderm . Immunostaining for AFP and Hnf4a on differentiated cells showing the formation of hepatocyte progenitors . ADE ( control-upper panel ) , cells differentiated in LY ( middle panel ) and ADE generated from LY-treated naïve cells ( lower panel ) were subjected to hepatocyte differentiation . LY-treated cells showed reduced differentiation efficiency , whereas both ADE and recovered LY-treated ADE efficiently generated hepatocytes . DOI: http://dx . doi . org/10 . 7554/eLife . 00806 . 00810 . 7554/eLife . 00806 . 009Figure 3—figure supplement 1 . PI3K inhibition supports a naïve endoderm state . ( A ) Histograms based on flow cytometry showing that while the Gsc-GFP cell number is increased by LY treatment , the population is shifted to a lower level of Gsc expression ( Gsclow ) . ( B ) Fluorescence imaging of HRS/GscGFP differentiating cultures showing clusters of cells with Gschigh expression ( pre-ADE ) ( upper panel ) that are eliminated when PI3K is inhibited . Cultures treated with LY failed to form ordered epithelial junctions and showed uniform Gsclow expression ( lower panel ) . ( C ) Q-RT-PCR showing markers of posterior , visceral endoderm and mesoderm/PS in LY-treated cultures . Meox1&2 ( mesoderm ) transcripts were not detected . ( D ) Fluorescence images of the Gsclow population stained for Sox17 in cultures differentiated in the presence of LY . The majority of these cultures co-expressed the endodermal marker Sox17 alongside low levels of Gsc-GFP . ( E ) Quantification of cell types found in LY differentiating cultures similar to those depicted in C and D , at day 6 of differentiation . ( F ) Pdx-1 immunostaining ( pancreatic progenitor marker ) showing expression in further differentiation of ADE cultures . ADE ( upper panel ) and LY-treated cells ( lower panel ) were subjected to further pancreatic differentiation and stained for Pdx-1 . LY-cells showed significantly reduced differentiation efficiency compared to ADE . DOI: http://dx . doi . org/10 . 7554/eLife . 00806 . 009 To further confirm that the LY-mediated block to differentiation maintained a true functional naïve state , we tested the capacity of these cells for further differentiation towards hepatocyte or Pdx1 positive pancreatic progenitors ( Morrison et al . , 2008 ) . While ADE cells progressed efficiently into both lineages , LY-treated cells failed to differentiate efficiently ( Figure 3C , Figure 3—figure supplement 1F ) . However , when LY-treated naïve cells were returned to normal differentiation for 3 days prior to subjecting them to hepatocyte differentiation , they performed as normal Hhex positive ADE cultures , efficiently generating Hnf4a/AFP positive hepatocyte progenitors ( Figure 3C ) . Endodermal cultures were also subjected to intestinal differentiation ( Spence et al . , 2011 ) , but we were unable to generate intestinal spheres from either ADE or LY-treated cells ( data not shown ) . This may indicate that specific posterior pre-patterning of naïve endoderm is required . Taken together , these experiments suggest that the cells produced by PI3K/Akt1 inhibition are non-regionalized endoderm ( naïve ) that retains the ability to give rise to specific endodermal domains when returned to normal differentiation . We found that the Gsclow population generated in the presence of LY failed to make cell–cell contacts and no longer maintained the epithelial morphology normally observed at this early stage of endoderm differentiation ( Figure 3—figure supplement 1B , D , bright field ) . Importantly , normal up-regulation of E-cadherin was not observed and residual E-cadherin was not localized to cell–cell junctions , while the pro-EMT transcription factor Snai1 was up-regulated ( Figure 4A , B ) . These observations indicated that in the absence of PI3K signalling , naïve endoderm precursor cells remained in an extended EMT-like process , instead of forming an organized epithelium . 10 . 7554/eLife . 00806 . 010Figure 4 . PI3K/Akt1 regulates EMT and is localized to the pre-ADE population . ( A ) PI3K activity is necessary to maintain epithelial integrity in endoderm differentiation . Immunofluorescense images showing strong expression of the epithelial marker E-cadherin during ADE specification . Addition of LY during phase 2 resulted in a reduction of E-cadherin levels and disappearance of E-cadherin from cell–cell junctions . ( B ) Q-RT-PCR showing a failure in E-cadherin transcriptional up-regulation as well as the up-regulation of its repressor and EMT inducer , Snai1 , in LY-treated cultures . Transcript levels were normalised as described in Figure 2A . Control: ADE differentiation protocol . ( C and D ) Normal differentiating HRS/Gsc-GFP cells were sorted by flow cytometry based on Gsc levels at day 4 . 5 of ADE differentiation . ( C ) Western blots showing Akt1 phosphorylation in different fractions of differentiating ESC . pAkt1 was localized to the Gschigh population . Total ERK was used as loading control . ( D ) Q-RT-PCR showing lineage markers in sorted cells . The Gschigh/Akt1high population coincides with the Sox17 positive emergent endoderm . ( E ) Inhibition of PI3K during phase 2 alters the identity of the Gsclow population . Q-RT-PCR showing the mesoderm and endoderm markers T and Sox17 . Transcript levels for D and E were normalised as described in Figure 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 00806 . 010 Since differentiating populations of ESC showed heterogeneous expression of Gsc , we investigated whether Akt1 was preferentially activated in a particular population . Figure 4C shows that pAkt1 levels were significantly higher in the Gschigh cells sorted from differentiating ESCs at the time at which Gschigh and Gsclow states could first be distinguished ( day 4 . 5 ) . Q-RT-PCR analysis on the different fractions also indicated that the GschighpAkt1high population was endodermal ( Sox17+ ) , whereas the GsclowpAkt1low population formed under control conditions was still mesendodermal ( Brachyury+ ) ( T+ ) ( Figure 4D ) . As the GschighpAkt1high population is lost upon LY addition , this indicates that anterior endoderm generation would appear to require the activation of Akt1 . The Gsclow population formed in the presence of LY was also distinct from the Gsclow population formed under control conditions . Gsclow-LY treated cells expressed high levels of Sox17 whereas T expression was reduced ( Figure 4E ) . Thus , while LY may promote an EMT-like state , it is not promoting mesodermal , but rather naïve mesenchyme-like endodermal state . We uncoupled Akt1 activation from PI3K signalling by using an Akt1 fusion to the oestrogen receptor , Myr-Akt1-mER ( Kohn et al . , 1998 ) , that placed activated and myristoylated Akt1 under the control of the oestrogen analogue 4-hydroxy-tamoxifen ( Tam ) ( Figure 5—figure supplement 1A , B ) . The Myr-Akt1-mER fusion protein was constitutively induced in HRS cells and its expression was visualized based on GFP expression from an internal ribosomal entry site ( IRES ) ( Myr-Akt1-mER-IRES-GFP/HRS ) ( Akt1-GFP-HRS ) ( Figure 5—figure supplement 1A–C ) . In this cell line , we found that Tam stimulated Akt1 activation rescued ADE generation in the presence of LY . This ability of pAkt1 to support ADE specification was observed both by flow cytometry ( Figure 5A ) and by q-RT-PCR ( Figure 5C ) . ADE markers ( Hhex , Sfrp5 , Fzd5 ) were induced robustly in response to Tam , alongside the rescue of ADE epithelial morphology , even in the presence of LY ( Figure 5B ) . 10 . 7554/eLife . 00806 . 011Figure 5 . Akt1 activation rescues the block to PI3K in ADE differentiation . ( A ) ADE induction is rescued by Akt1 activation . Flow cytometry on Akt1-GFP-HRS differentiating cells showing that induction of Akt1 with Tamoxifen ( Tam ) rescues HRS expression . ( B ) Fluorescence microscopy showing Akt1-GFP-HRS differentiating cultures in presence of LY and Tam . HRS expression and ADE epithelial morphology were both rescued . ( C ) Q-RT-PCR showing rescue of ADE marker expression in LY-treated cultures by Tam-induced Akt1 expression . Transcript levels were normalised as described in Figure 2A . ( D ) Flow cytometry showing cell non-autonomous rescue of mixed Akt1-GFP-HRS and HRS differentiating co-cultures treated with Tam and/or LY . Akt1-GFP-HRS cells can be distinguished based on GFP expression ( X-axis ) . HRS expression ( Y-axis ) was rescued in both Akt1-GFP-HRS expressing and non-expressing cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00806 . 01110 . 7554/eLife . 00806 . 012Figure 5—figure supplement 1 . Activation of Akt1 supports Hhex induction in the presence of a block to PI3K . ( A ) Schematic showing Akt1-GFP-HRS cell line . HRS cells were modified to express a myristoylated fusion protein that includes the activated form of Akt1 fused to a mutant version of the ligand binding domain of the murine oestrogen receptor ( mER ) . This mutant ligand binding domain recognizes 4-Hydroxy-tamoxifen ( Tam ) in place of oestrogen and places Akt1 under the control of Tam . ( B ) Western blot showing endogenous and conditional transgenic Akt1 expression in response to Tam and LY treatment . Differentiating cells were harvested on d4 , 1 hr after Tam/LY treatment . ( C ) Flow cytometry showing uniform expression of the Akt1 containing transgene in Akt1-GFP-HRS cells during endoderm differentiation . Akt1-GFP-HRS cells were compared to HRS on day 6 of differentiation . ( D ) Western Blot confirming the Tam-induced expression of the Akt1 trangene in the GFP+ fraction of mixed HRS and Akt1-GFP-HRS co-cultures following cell sorting by flow cytometry based on GFP . Endogenous pAKT1 expression was observed in control conditions and was blocked by LY treatment . Cells were sorted based on GFP expression 6 hr after LY and Tam+LY ( T/LY ) treatments . ( E ) Quantification of the rescue of ADE induction as assessed by flow cytometry in Figure 5D . The experiments were performed by mixing HRS cells with Akt1-GFP-HRS cells ( 1:1 ) , differentiating in the presence of LY and Tam , and assessing HRS expression in the different GFP populations at day 6 . Blue columns represent the percentage of HRS expressing cells in the GFP− fraction ( non-Akt1 expressing cells ) . Red columns represent the GFP+ fraction ( Akt1 expressing cells ) . Treatments are stated below each set of columns . For normalization , HRS expression in control mixed cultures was taken as 100% for each fraction . Quantification of the non-cell autonomous effect of Tam was obtained by the ratio between HRS+GFP−/GFP− , while similar quantification of the cell-autonomous effect was obtained by the ratio between HRS+GFP+/GFP+ , and compared to the control . LY decreased HRS expression in both GFP− ( 14 . 76% +/− 8 . 12 ) and GFP+ ( 21 . 68% +/− 7 . 26 ) fractions , whereas in Tam/LY treated cultures rescue of HRS expression was observed in both GFP− ( 65 . 95% +/− 21 . 87 ) , and GFP+ ( 81 . 66% +/− 14 . 03 ) fractions , ( n = 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00806 . 012 As the Akt1-GFP-HRS cells contained a constitutive GFP reporter , we mixed Akt1-GFP-HRS expressing cells with unmodified HRS ESCs , and assessed the induction of endoderm in the presence of LY , in both the Akt1 expressing and non-expressing populations ( Figure 5—figure supplement 1D ) . We found that the stimulation of Akt1 by Tam administration rescued the block to ADE induction in both Akt1 expressing and non-expressing cells ( Figure 5D , Figure 5—figure supplement 1E ) , indicating cell autonomous and non-autonomous rescue of the LY phenotype . However , based on a time course for induction of gene expression in sorted Akt1-GFP+ and Akt1-GFP− populations ( +/− Ly , +/− Tam ) , it appeared that Akt1 also had cell autonomous effects on Snai1 and Sox17 transcription ( Figure 6—figure supplement 1A ) . To identify the factors downstream of pAkt1 that mediate non-autonomous ADE induction , we tested the ability of supernatants produced during normal differentiation to rescue ADE generation in LY-treated cultures , but failed to observe any effect ( data not shown ) , suggesting that these factors might not readily diffuse . We therefore tested the hypothesis that the ADE inducing activity downstream of Akt1 might be the result of specific Akt1-dependent ECM proteins . We reasoned that if the role of Akt1 in anterior patterning is conducted by the production of a specific ECM , then if we expose differentiating cells to the action of an ECM generated under normal conditions , pAkt1 would no longer be necessary for ADE induction . To do so , we prepared ECM from untreated differentiating ESC cultures ( ECM1 ) or from cultures treated with LY ( ECM2 ) . These ECM preparations were then tested for their ability to induce or rescue endoderm differentiation in the presence of LY ( Figure 6B ) . HRS-Gsc-GFP cells were differentiated to the APS GFP+ stage ( Figure 1A ) , collected , re-plated onto the different matrices or gelatine ( Figure 6A ) and differentiated in the presence or absence of LY . Figure 6B shows that ECM1 , but not ECM2 , could support ADE differentiation in the presence of LY . Moreover , ECM1 not only restored anterior induction , but the combination of LY and ECM1 enhanced anterior endoderm specification , such that these cultures were almost 60% ADE ( Figure 6B ) . These data suggest that the capacity of LY to enhance Sox17+Foxa2+ expression was harnessed by ECM1 , which was able to convert the Sox17+Foxa2+ naïve population into Hhex+ prospective foregut ( Figure 6C ) . Cells plated on ECM1 in the presence of LY displayed epithelial morphology , enhanced E-cadherin and reduced Snai1 expression ( Figure 6D , Figure 6—figure supplement 1B ) . While treatment of differentiating cultures with LY resulted in a loss of expression of the tight junction and polarity markers , ZO-1 and aPKC , organized expression of these was restored within 12 hr of plating onto ECM1 ( Figure 6—figure supplement 1C ) . Cells plated on ECM2 in the presence of LY did not generate ADE or form an organized epithelium ( Figure 6—figure supplement 1B ) . To exclude the possibility that Akt1 activation was somehow stimulated by re-plating on these ECMs , despite the block to PI3K signalling , we assessed Akt1 activation after re-plating and observed no rescue of the LY-mediated block to Akt1 phosphorylation ( Figure 6—figure supplement 1D ) . Thus , ECM1 has the capacity to redirect the LY-mediated enhancement of Sox17/Foxa2 expression through an Akt1 independent pathway and convert the majority of the culture to properly patterned anterior endoderm . These data suggest that ECM1 together with LY could be used to enhance physiological ESC differentiation toward foregut ( Figure 6B , middle panel ) . 10 . 7554/eLife . 00806 . 013Figure 6 . PI3K/Akt1 dependent-ECM generated during endoderm differentiation can support ADE specification . ( A ) Schematic showing the experimental strategy designed to test the role of ECM in supporting ADE specification . ECM1 and ECM2 were produced by differentiating cells in the absence or presence of LY respectively . Matrices were obtained by removing differentiated cells from day 6 cultures . ( B ) ADE generation assessed by flow cytometry on differentiating cells exposed to LY . ADE induction was rescued and enhanced when cells were re-plated onto ECM1 . APS-like cells re-plated on either ECM2 or gelatine failed to counteract LY-induced phenotypes . ( C ) Q-RT-PCR analysis showing rescue of ADE markers in differentiating cells re-plated onto ECM1 and simultaneously treated with LY . Transcript levels were normalised to the Tbp value obtained for each sample . Normalised values are related to the level obtained for ESC . ( D ) Q-RT-PCR showing the regulation of EMT markers by ECM1 . Transcript levels were normalised to the Tbp value obtained for each sample . Normalised values are related to the level obtained in control conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 00806 . 01310 . 7554/eLife . 00806 . 014Figure 6—figure supplement 1 . PI3K/Akt1 signalling modulates ECM activity to induce ADE specification . ( A ) A time course for induction of Snai1 and Sox17 by Akt1 . Q-RT-PCR on mixed HRS and Akt1-GFP-HRS co-cultures ( +/− Ly , +/− Tam ) that were sorted based on GFP and analysed at the indicated times . Transcript levels were normalised to the Tbp value obtained for each sample . Normalised values are related to the level obtained for ESC . ( B ) Phase contrast image showing APS cells re-plated onto ECM1 and 2 . On ECM1 cells formed epithelial clusters in the presence of LY , whereas cells plated onto ECM2 retained a mesenchymal-like morphology in the presence of LY . ( C ) Cell–cell contacts and localization of the tight junction marker , ZO-1 , and the polarity marker , aPKC , were severely affected by LY treatment , but were unaffected in APS cells exposed to LY and plated onto ECM1 . IC showing expression of aPKC and ZO-1 12 hr after plating on indicated ECMs . ( D ) Western blots showing activation of Akt1 in cells plated on different substrates with or without LY . Re-plating onto ECM1 did not prevent Akt1 dephosphorylation by LY . DOI: http://dx . doi . org/10 . 7554/eLife . 00806 . 014 Gene expression profiling data ( Morrison et al . , 2008 ) suggested that components of the matrisome were up-regulated in ADE compared to ESC ( Figure 7—source data 1 ) , including a number of ECM proteases that proved to be PI3K-dependent ( Figure 7—figure supplement 1A ) . To provide a better picture of PI3K/Akt1-dependent changes in the ECM that might be induced by post translational modification , we used mass spectrometry coupled to liquid chromatography ( LC-MS ) to define PI3K-dependent differences between ECM1 and ECM2 . Details of ECM extraction and LC-MS methodologies can be found in ‘Material and methods . ’ Specific proteins enriched in ECM1 include two heparan sulphates proteoglycans ( HSPG ) , Perlecan and Col18a1; two HSPG regulators Sulf1 and Cyr61; the metalloprotease Adamts15 , the anterior endoderm associated diffusible factors Nodal and Lefty1; and the endoderm/epithelial associated cytokeratines Krt8 , Krt18 and Krt19 ( Figure 7A , Figure 7—source data 2 , 3 ) . While a number of these proteins are not canonical matrix proteins , they are generally associated with integrin signalling and tight junction formation . Interestingly , Fibronectin ( Fn1 ) , a promoter of EMT and migration , was highly enriched in ECM2 as was the PS associated diffusible factor Fgf8 ( Figure 7A , Figure 7—source data 2 , 3 ) . Analysis by q-RT-PCR confirmed that Fn1 expression was transcriptionally regulated by PI3K/Akt1 signalling ( Figure 7B ) , and that the induction of Fn1 expression was a cell autonomous response to Akt1 activation ( Figure 7—figure supplement 1B ) . Moreover , the addition of Fn1 to differentiating ESC cultures , either through re-plating of PS stage cultures onto Fn1-coated dishes ( data not shown ) or through direct addition to phase 2 differentiation ( Figure 7C ) abolished the induction of ADE specific transcription . As Fn1 is known to signal through Akt1 ( Khwaja et al . , 1997 ) , and Akt1 is involved in ADE specification , we asked whether exogenous Fn1 resulted in Akt1 activation during normal endoderm differentiation . Figure 7D shows that Fn1 did not stimulate Akt1 phosphorylation , but it actually inhibited it ( Figure 7D ) . 10 . 7554/eLife . 00806 . 015Figure 7 . Composition of ECM determines cell fate choices within endoderm . ( A ) Different compositions of ECM1 and ECM2 as determined by mass spectrophotometry . Significantly over-represented peptides ( blue dots ) or evenly represented peptides ( red dots ) in the ECM1 ( left side ) , and ECM2 ( right side ) are shown in the volcano plot . Relevant peptides are indicated . ( B ) Q-RT-PCR showing the regulation of Fn1 expression as a result of PI3K inhibition ( d6 ) . ( C ) Q-RT-PCR analysis showing inhibition of ADE gene expression when Fn1 was added to the culture media . Transcript levels were normalised to the Tbp value obtained for each sample . Normalised values are related to the level obtained in ESC ( B ) and in control conditions with no exogenous Fn1 ( C ) . ( D ) Fn1 inhibits Akt1 activation . Western blots showing inhibition of Akt1 phosphorylation by Fn1 during ESC differentiation . Tubulin was used as loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 00806 . 01510 . 7554/eLife . 00806 . 016Figure 7—source data 1 . ECM global gene analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 00806 . 01610 . 7554/eLife . 00806 . 017Figure 7—source data 2 . LC-MS analysis I . DOI: http://dx . doi . org/10 . 7554/eLife . 00806 . 01710 . 7554/eLife . 00806 . 018Figure 7—source data 3 . LC-MS analysis II . DOI: http://dx . doi . org/10 . 7554/eLife . 00806 . 01810 . 7554/eLife . 00806 . 019Figure 7—figure supplement 1 . Expression of ECM components during endoderm differentiation and in response to Akt1 activation . ( A ) Q-RT-PCR showing a time course for metalloprotease transcription during endoderm differentiation in the presence and absence of LY . Metalloproteases belonging to the matrisome identified based on transcriptomics of the different differentiating endoderm populations ( Morrison et al . , 2008 ) . Transcript levels were normalised to the Tbp value obtained for each sample . Normalised values are related to the level obtained for ESC . ( B ) Q-RT-PCR sorted mixed population ( HRS and Akt1-GFP-HRS ) in a time course experiment ( +/− Ly , +/− Tam ) showing that Fn1 expression is Akt1 dependent . Transcript levels were normalised to the Tbp value obtained for each sample . Normalised values are related to the level obtained for ESC . DOI: http://dx . doi . org/10 . 7554/eLife . 00806 . 019 A number of ECM components identified by LC-MS were also screened by q-RT-PCR in different dissected regions of E7 . 5 embryos ( Figure 8—figure supplement 1A ) . ECM1 components showed expression in the endoderm ( Figure 8—figure supplement 1B and data not shown ) , as well as anterior enrichment ( Figure 8—figure supplement 1C and data not shown ) . Importantly , we found that Fn1 was both highly expressed in the endoderm ( Figure 8—figure supplement 1B ) and robustly enriched in the posterior region of these embryos ( Figure 8A , C–E ) , consistent with a role in promoting naïve endoderm . To address whether endoderm regionalization and Fn1 expression were regulated in vivo by PI3K/Akt1 , E6 . 5 embryos ( initiation of gastrulation ) were dissected and cultured ex-utero in the presence or absence of LY . Gene expression analysis in individual cultured embryos was assessed in response to inhibition of PI3K by LY . We observed enhanced expression of Fn1 and Snai1 , slight increases in Sox17 and Foxa2 , and a reciprocal reduction in Cer1 and Lefty1 ( Figure 8B ) . Consistent with these observations , immunohistochemistry ( IHC ) showed that embryos cultured in the presence of LY exhibit enhanced Fn1 expression ( Figure 8C , D ) and a complete loss of Cer1+ ADE ( Figure 8—figure supplement 1D ) . 10 . 7554/eLife . 00806 . 020Figure 8 . Anterior specification in vivo requires PI3K-dependent ECM . ( A ) Q-RT-PCR showing differential expression of Fn1 in vivo . Individual E7 . 5 mouse embryos were bi-dissected into anterior and posterior halves , and Lefty1 and T expression were used as a control for bi-dissection efficiency . Transcript levels were normalised to the Tbp value obtained for each sample . Normalised values are related to the level obtained in the anterior region . ( B ) Fn1 is regulated by PI3K in vivo . Q-RT-PCR on single embryos dissected at E6 . 5 and cultured ex-vivo for 24 hr in the presence or absence of LY . Wild-type ( WT ) embryos were dissected at E7 . 5 and used as control . Embryos treated with LY exhibited a failure in the expression of anterior markers , and up-regulation of Snai1 and Fn1 , similar to that observed in vitro . Error bars represent the standard deviation between embryos ( >0 . 05; >0 . 01; >0 . 001 ) . Both in A and B the transcript levels were normalised to the Tbp value obtained for each sample . Normalised values are related to the level obtained in the anterior region ( A ) and in E7 . 5 wild-type embryos ( B ) . ( C ) Immunohistochemistry ( IHC ) on sectioned embryos showing higher levels Fn1 on the posterior side of E7 . 5 wild-type embryos ( left ) and increased expression of Fn1 in embryos dissected at E6 . 5 and cultured ex-vivo during 24 hr in the presence of LY ( right ) . ( D ) Images analysis on the areas highlighted in ( C ) indicating differential expression of Fn1 along the A–P axis in WT embryos ( I , II ) , and increased expression of Fn1 in embryos cultured ex vivo in the presence of LY ( III , IV ) ( n = 4 , >0 . 001 ) . ( E ) IHC on E7 . 5 embryos showing differential expression of Fn1 along the anterior–posterior axis and anterior expression of Cer1 . In Fn1−/− embryos the Cer1 domain is significantly expanded . Embryos are shown in frontal and in lateral view ( anterior to the left ) . ( F ) IHC on WT and mutant embryos ( dorsal view ) showed an expanded Foxa2 expression domain at E7 . 75 , and defects in gut folding and in the formation of the anterior intestinal portal ( AIP ) at E8 . 0 . Arrowhead points to the node and arrows the foregut . DOI: http://dx . doi . org/10 . 7554/eLife . 00806 . 02010 . 7554/eLife . 00806 . 021Figure 8—figure supplement 1 . Composition of ECM in the gastrulation stage mouse embryo . ( A ) To analyse the in vivo regional expression of the ECM components detected by LC-MS , the endoderm layer was dissected away from the epiblast in E7 . 5 embryos . Schematic showing the details of the dissections used in Figure 8A and Figure 8—figure supplement 1B , C . ( B and C ) Q-RT-PCR showing differential expression levels of ECM components in the epiblast vs endoderm ( B ) , and in the anterior vs posterior embryonic region ( C ) . Hhex , Lefty1 and T expression were used to validate the efficacy of individual germ layer and A–P dissections . ( D ) IHC in ex-vivo gastrulating cultured embryos . Embryos collected at E6 . 5 exposed for 24 hr to LY showed an impairment in Cer1 expression . DOI: http://dx . doi . org/10 . 7554/eLife . 00806 . 02110 . 7554/eLife . 00806 . 022Figure 8—figure supplement 2 . Reduction in Fn1 activity resulted in expansion of the anterior endodermal domain in mouse gastrula . ( A ) Image analysis measuring the intensity of Cer1 expression across A–P axis in WT and Fn1−/− E7 . 5 embryos . ( B ) IHC on Fn1−/− E7 . 5 embryos showed Cer1 expression is expanded alongside enhanced posterior Foxa2 expression . Embryos are shown in a lateral view . ( C ) Image analysis measuring the intensity of Foxa2 expression across A-P axis in WT and Fn1−/− E7 . 5 embryos . ( D ) IHC on WT and Itga5−/− E7 . 5 embryos showing a similar expansion of the Cer1 domain . Embryos are shown in a lateral view . ( E ) In situ hybridization on E7 . 5 Itga5−/− mutants showing Cer1 RNA expression is also expanded . DOI: http://dx . doi . org/10 . 7554/eLife . 00806 . 02210 . 7554/eLife . 00806 . 023Figure 8—figure supplement 3 . Fibronectin mutants exhibit defects in foregut and naïve endoderm . ( A ) Lateral view of WT and Fn1−/− E7 . 5 embryos after in situ hybridization showing expanded domain of Foxa2 RNA expression . ( B ) Expanded Foxa2 expression becomes more apparent as the foregut begins to form at E7 . 75 in the Itga5−/− mutants . In situ hybridisation showing Foxa2 expression at E7 . 75 and E8 . 0 in WT and mutant embryos . ( C ) IHC on WT and Fn1−/− E7 . 5 embryos showing the relationship between Cer1 , Sox17 and Fn1 . Sox17 expression is reduced in the prospective foregut domain in WT embryos , similar to the graded expression observed for Fn1 . Sox17 expression in the Fn1−/− embryos showed decreased intensity and more homogeneous distribution alongside the expansion of Cer1 . Embryos are shown in a lateral view . DOI: http://dx . doi . org/10 . 7554/eLife . 00806 . 023 Analysis of wild-type ( WT ) embryos at E7 . 5 indicated that Fn1 protein was non-uniformly distributed along the A-P axis , with relatively low expression in the prospective foregut region where Cer1 was expressed ( Figure 8A , C–E ) . We therefore assessed the requirement of Fn1 during endoderm specification by examining the endoderm of embryos homozygous for a null mutation in the Fn1 locus ( Fn1−/− ) ( George et al . , 1993 ) . In Fn1−/− embryos , the domain of Cer1 expression was expanded toward the posterior-proximal side of the embryo within the domain of Foxa2 expression ( Figure 8E , Figure 8—figure supplement 2A–C ) . We also assessed the protein expression of the pan-endoderm marker Sox17 and found that like Fn1 , Sox17 was expressed at lower levels in the Cer1 positive foregut region . In Fn1 mutants , this region of Sox17low was expanded across the embryo alongside Cer1 ( Figure 8—figure supplement 3C ) and consistent with the increase in Sox17 transcript that we have observed in LY-treated embryos . Moreover , while there are a number of integrins that bind to Fn1 , integrin Itga5 is thought to mediate Fn1 functions during early embryogenesis and we found that Itga5−/− mutant embryos ( Yang et al . , 1993 ) also exhibited an expanded domain of Cer1 expression ( Figure 8—figure supplement 2D , E ) . As the Fn1 and Itga5 phenotypes are known to be background dependent , we confirmed that we could observe expansion of the anterior endoderm domain in both C57BL/6J ( Figure 8E , Figure 8—figure supplement 2B , D ) and 129S4 mice strains ( Figure 8—figure supplement 2E ) . Consistent with the expansion of Cer1 in these mutants , further foregut morphogenesis was also defective and there was evidence of an expansion in the prospective foregut . In Fn1 and Itga5 null embryos: the foregut did not fold into a tube and remained as a flat field of Foxa2+ cells ( Figure 8F , Figure 8—figure supplement 3B ) . This domain of Foxa2+ cells remained enlarged compared to control embryos and this was particularly apparent early in foregut morphogenesis ( E7 . 75 ) ( Figure 8F , Figure 8—figure supplement 3A , B ) . We have defined a novel ECM-dependent mechanism that can explain the diverse developmental outcomes induced by a single signalling pathway . Thus , a requirement for FGF signalling in endoderm induction is based on a non-canonical activity downstream of PI3K/Akt1 signalling . PI3K/Akt1 acts during positional specification in the endoderm , by regulating the deposition of regionalized ECM with patterning activity ( Figure 9 ) . One essential component of this activity is Fn1 . This observation led to the identification of PI3K/Akt1-dependent graded Fn1 expression within the developing embryo with the capacity to differentially regulate the timing of endoderm differentiation and establish positional identity . 10 . 7554/eLife . 00806 . 024Figure 9 . A role for ECM in positional specification in the endoderm . Schematic illustration of the proposed role played by PI3K/Akt1 signalling in positional identity within the endoderm . Cells in PS region are in the process of generating mesoderm and endoderm ( mesendoderm ) . At the anterior end of the PS , high levels of pAkt1 regulate proper epithelialization and anterior endoderm specification through the production of a unique ECM low in Fn1 . Low levels of PI3K/Akt1 led to the production of an Fn1-enriched ECM that favours naïve endoderm generation . DOI: http://dx . doi . org/10 . 7554/eLife . 00806 . 024 Mesendoderm induction is associated with specific levels of Nodal , Wnt and FGF signalling , but is also accompanied by morphogenetic rearrangements that result in changes in a cell’s immediate environment . These cellular movements , the ECM that cells move on and the signals they are exposed to during their migration all contribute to the patterning of the embryonic endoderm ( Arnold and Robertson , 2009 ) . The non-canonical ECM-dependent pathway downstream of PI3K described here provides an explanation for the diversity of cellular responses that can be achieved through the activation of pAkt1 . PI3K/Akt1 signalling has been extensively associated with the induction of EMT ( Larue and Bellacosa , 2005 ) rather than its suppression . However , in the context described here , Akt1 modifies the ECM to drive cell–cell contact and either induce or preserve epithelial identity . Thus , high levels of PI3K activity ensure that Fn1 levels remain low and that EMT is suppressed . The involvement of an ECM intermediate also provides context for an endoderm specific role for FGF signalling . FGF receptor activation is essential for ESC differentiation towards multiple lineages and leads to Grb2 and SHP2 recruitment , activating both PI3K and MAPK signalling pathways ( Kouhara et al . , 1997; Xu et al . , 1998; Villegas et al . , 2010 ) . These pathways have been associated with mesoderm induction , EMT and cell migration during gastrulation ( Casey et al . , 1998; Rodaway et al . , 1999; Ciruna and Rossant , 2001; Mizoguchi et al . , 2006; Poulain et al . , 2006 ) . In endoderm , where FGF signalling is required , but does not induce EMT ( Morrison et al . , 2008; Hansson et al . , 2009 ) , the signalling is contextualized and rendered distinct as a result of the activity of this pathway in remodelling the ECM . PI3K activity in the visceral endoderm of embryoid body differentiation ( Li et al . , 2001 ) has also been associated with the production of ECM required to support the epithelialization of the epiblast ( Jeanes et al . , 2009 ) . While the activity of PI3K/Akt1 is generally associated with induction of EMT , activation of this pathway has also been associated with a block to EMT in human ESCs ( Singh et al . , 2012 ) . In this context , Akt1 activation supports human ESC self-renewal by inhibiting Raf/Mek/Erk and canonical Wnt signalling , which in turn may also promote EMT , although it is not clear whether this route takes place in vivo during early endoderm differentiation or how comparable the mouse and human ESC differentiation models are . Accordingly , the transient reduction in pAkt1 we observed at the beginning of differentiation may be coupled to early activation of the ERK and Wnt pathways as cells become prepared for mesoderm and endoderm induction . However , prior to the appearance of endoderm gene expression , differentiating ESCs contained high levels of pAkt1 and this appeared essential for the deposition of the correct , Fn1low regionalized ECM . That cells rapidly down-regulate Snai1 upon re-plating suggests that the ECM-dependent pathway described here allows rapid modifications to differentiation as cells become exposed to a new environment . The basement membrane produced during endoderm differentiation is complex and proteomic analysis supports the association of the ECM with functional cell membrane components in a matrisome ( Hynes and Naba , 2011 ) . How do these elements impact on the key signalling events regulating lineage specification ? Several elements of this matrisome are directly involved in Fn1 signalling and exogenous Fn1 can alter ECM activity . Genetic studies indicate that Fn1-integrin interactions are essential for axial extension of the mesendoderm ( Yang et al . , 1999; Davidson et al . , 2006; Marsden and DeSimone , 2003 ) . However , Fn1 is not required for gastrulation and specific endodermal patterning defects have not been previously reported . The Fn1 and Itga5 null mutants are both embryonic lethal and die by E10 . 5 due to cardiovascular problems , posterior truncations , and kinking of the neural tube among others defects ( George et al . , 1993; Yang et al . , 1993 ) . We observed higher levels of Fn1 in the posterior region of gastrulation stage embryos and the region where the nascent foregut moves into was uniquely low in Fn1 . In Fn1−/− embryos , the region of Cer1 positive ADE is expanded , suggesting that the deposition of Fn1 may help to restrict the prospective foregut domain as it migrates up from the distal end of the embryo . Consistent with this observation , the foregut did not fold into a tube and remained as a flat field of expanded Foxa2+ cells in Fn1−/− embryos , indicating that the regulation of Fn1 deposition helps delineate and form the prospective foregut . We also observed a similar inverse correlation between the expression of Sox17 and Cer1 , indicating that Sox17 is expressed at higher levels in Fn1 rich regions of the endoderm . This is consistent with our in vivo and in vitro data , which show significantly higher levels of Sox17 expression in naïve endoderm that was generated in response to inhibition of PI3K . In addition to Fn1 , we identified additional potential determinants that could interact with Fn1 in modulating ECM activity . Two out of the three main Heparan Sulphate-bearing species , Perlecan and Col18a1 ( Halfter et al . , 1998 ) were expressed at high levels in ECM1 . Knock down of these two HSPG’s have been shown to impair endoderm differentiation in vitro ( Higuchi et al . , 2010 ) . Potentially , this HSPG-bearing ECM could be acting in conjunction with low levels of Fn1 to modulate growth factor bioavailability , and operate as a platform for cytokine signalling . As the cytokines we added exogenously to the cultures were not found in either ECMs , we think it is unlikely that this explains the full activity of these ECMs in vitro . However , we did find that ECM1 contained Nodal and Lefty1 , whereas ECM2 contained Fgf8 ( Figure 7A , Figure 7—source data 2 , 3 ) . It has been suggested that specific modifications in the ECM are required for Nodal transport within the embryo ( Oki et al . , 2007; Marjoram and Wright , 2011 ) . Moreover , the inclusion of Nodal , Lefty1 and Fgf8 within these matrices fits with the locus of their extracellular activity in vivo . Our observation that these cytokines appear specifically embedded in the ECM suggests that the mechanism by which these proteins function in vivo involves association with ECM proteins ( Hynes , 2009 ) . Despite widespread attempts to use ESC differentiation towards endodermal derivatives , the production of fully functional differentiated cell types has not been particularly successful . Part of the explanation for this lack of success may lie in the inability to pattern the correct intermediates . The observation that inhibition of PI3K leads to enhanced Sox17 and Foxa2 expression ( McLean et al . , 2007 ) has been used in ESC differentiation protocols to generate endoderm derivatives . Our data suggest that these cells are less efficient in further differentiation to derivatives of the ventral foregut such as pancreatic or hepatic progenitors . Based on both molecular and functional analysis of these endoderm cells they appear to be naïve endoderm , similar to the mesenchymal cell type that emerges from the primitive streak and intercalates into the visceral endoderm layer ( Kwon et al . , 2008; Burtscher and Lickert , 2009 ) to form the hindgut . These Sox17high mesenchyme-like cells that arise from the primitive streak region of the embryo exist only transiently in vivo , and while we show that prolonged and enhanced induction of this state in vitro may be reversed by reintroducing cells back into normal differentiation , the direct onward differentiation of these cells may have deleterious effects on their capacity to form functional cell types in vitro . Interestingly , while these cells become epithelialized as they intercalate in the visceral endoderm , their levels of Sox17 remain high throughout gastrulation , suggesting that a naïve endoderm state could extend beyond the initial EMT event . As ECM1 is able to pattern differentiating cells to anterior fate , it suggests that it may be possible to use it in combination with the LY-dependent enhancement of naïve endoderm to efficiently generate high levels of foregut in vitro . Our finding that ECM is an essential downstream signalling response required for generation of positional identity in vitro indicates that the presence of appropriate ECM may be as important as the dose and identity of the cytokines used in directing context dependent signalling responses during ESC differentiation . Mouse ESC cultures and ADE differentiation in adherent monolayer were performed according to Livigni et al . ( 2009 ) except that in some cases SFO3 media was replaced by ADEM ( see below ) . Re-plating of APS-like cells was done by differentiating the HRS/Gsc-GFP ESC until Gsc-GFP expression became apparent ( to day 3–3 . 5 ) . The media was collected , centrifuged at 3000 rpm for 8 min and mixed with freshly prepared media in ratio 1:1 ( new:old ) . The APS-like cells were detached with Accutase ( Sigma ) , 5 min at 37°C , washed with the collected media , resuspended in 1:1 media , and re-plated onto different substrates . For the induction of pancreatic progenitor cells , ADE and LY-cells in N2B27 media supplemented with 10 ng/ml FGF4 , 2 mM all-trans retinoic acid , and 0 . 25 mM KAAD cyclopamine ( Toronto Research Chemicals ) . Hepatocyte differentiation was as described in Gouon-Evans et al . ( 2006 ) , except N2B27 that was used as the base media at all stages and Activin A was not included , whereas BMP4 , FGF2 , and VEGF were included . ADEM ( Anterior definitive endoderm media ) preparation: 50% DMEM/F12 no Glutamine ( Invitrogen , UK ) , 50% RPMI1640 no Glutamine ( Invitrogen , UK ) , 2 mM L-Glutamine ( Invitrogen , UK ) , 10 mM HEPES ( Invitrogen ) , 0 . 5 μg/ml Cholesterol ( Sigma-Aldrich , Denmark ) , 80 μM Ethanolamine ( Sigma-Aldrich , Denmark ) , 10 μg/ml Cytidine ( Sigma-Aldrich , Denmark ) . Filter sterilize . 2 µg/ml Apo-transferrin ( Sigma-Aldrich , Denmark ) , 2 ng/ml Na Selenite ( Sigma-Aldrich , Denmark ) . Just before use add: 0 . 1 μM 2-ME ( Sigma-Aldrich , Denmark ) , 0 . 1% BSA ( Invitrogen , UK ) , 0 . 5 µg/ml human recombinant insulin ( Sigma-Aldrich , Denmark ) . Media must be kept at 4°C and used within 1 month . Differentiated ADE cells were detached with 2 mM EDTA , 37°C , 8–12 min . The remaining ECM was washed with PBS and either used immediately or preserved in 70% EtOH , −20°C . Doses and timing of inhibitor usage are described in Supplementary file 1 . To determine the window of PI3K activity during phase 2 of ADE differentiation LY ( LY294002; Promega , UK ) was added for periods of 24 hr ( d3–4 , d4–5; d5–6 , d6–7 ) , 48 hr ( d3–5 , d4–6 , d5–7 ) , 72 hr ( d3–6 , d4–7 ) , or 96 hr ( d3–7 ) . ADE generation was assessed by flow cytometry based on HRS/Gsc-GFP expression . LY was used at a concentration of 10 µM during d3–4 and at 20 µM from d4 . 20–50 µM LY was used for experiments assessing apoptosis from d3-d6 . For co-culture experiments , HRS and Akt1-GFP-HRS ESC were mixed in a 1:1 ratio . 0 . 5 µM Tam was added 15 min before LY addition at day 3 . For ex-vivo gastrula cultures , embryos were dissected at E6 . 5 and cultured in 25 , 50 , or 75 µM LY for 24 hr . 25 µM and showed no morphological or differences in housekeeping gene transcription compared to untreated embryos . While 50 µM LY showed a significant difference compared to untreated embryos , 50 and 75 µM LY gave similar results ( Data not shown ) , and 50 µM LY was therefore used for further experiments . Apoptosis inhibitor ( Z-VAD-FMK; Santa Cruz , USA; 10 µM ) was added at d3–6 . The effects of LY or Z-VAD-FMK addition were analysed by immunostaining with the apoptotic marker Annexin-V Alexa647 ( Invitrogen , UK ) followed by flow cytometry analysis ( as described below ) . Cells were trypsinized and stained with Topro3-iodide or DAPI ( Invitrogen , UK ) to exclude dead cells from the analysis . FITC-conjugated rat anti-CD184 ( Cxcr4 ) was purchased from BD PharMingen ( UK ) , cells were analysed using a Becton Dickinson FACS Aria II or III cell sorter and a Becton Dickinson LSR Fortessa . Total RNA was prepared from a minimum of 1 × 104 cells using Trizol reagent ( Invitrogen ) and 1 µg of RNA was used as a template for cDNA synthesis using Superscript III ( Invitrogen , UK ) . Real-time RT-PCR was performed using a LightCycler 480 ( Roche ) and LightCycler 480 SYBR Green 1 Master or UPL Assay ( Roche ) . Primers and PCR conditions are listed in Supplementary file 3 . Monolayer differentiated cells were washed in PBS; fixed in 4% paraformaldehyde ( PFA ) for 10 min; 1M Glycine , pH 7 , 4 was used for 10 min to block residual PFA , and cells were permeabilized by washing in PBS supplemented with 0 . 1% Triton X ( PBST ) . The fixed cells were blocked in 1% Bovine serum albumin ( Sigma-Aldrich , Denmark ) and 3% appropriate serum in PBST for 1 hr at room temperature ( RT ) . For whole mount IHC , embryos were dissected in ice-cold M2 media and transferred to cold 4% PFA , overnight ( O/N ) . The embryos were dehydrated in methanol series ( 25 , 50 , 75 , 100% , in PBS ) , bleached with 5% H2O2 for 1 hr , rehydrated , blocked with 3% of the appropriate serum in PBST , incubated with primary antibody ( in PBST at 4°C , O/N ) , washed in PBST 3 × 15 min , 5 × 1 hr , incubated with secondary antibody for 2 hr , RT , washed in PBST 3 × 15 min , 5 × 1 hr . For confocal imaging embryos were placed in microscope slides mounted with Vectastain/DAPI ( Vector Labs ) and gently covered with glass cover slips with vaseline drops in the corners to avoid embryo squashing . For cryo-sectioned embryos and IHC , whole-mount embryos were dissected and fixed as above and allowed to sink in 15% sucrose in PBS , 2 hr at 4°C , then incubated in 15% sucrose/7% gelatine/PBS at 37°C , assembled in an aluminium mould , frozen in liquid nitrogen , and cut into 10 μm sections on a Cryostat ( Leica ) . Both transverse and sagittal sections were the same thickness and were counterstained with DAPI . Sections were collected on Poly-Lysine microscope slides ( VWR International ) , air-dried for 30 min to 1 hr , and stored at −20°C until use . Immunocytochemistry was performed essentially as described above for cells . A list of antibodies and conditions used is provided in Supplementary file 2 . Cells were lysed in lysis buffer ( 1% Triton X-100 , 150 mM NaCl , 10 mM Tris_HCl , at pH 7 . 4 , 1 mM EDTA , 1 mM EGTA , containing 2 mM NaF , 1 mM sodium orthovanadate , 10 μg/ml leupeptin , 10 μg/ml pepstatin , 10 μg/ml aprotinin , and 1 mM Pefabloc ) . Equal amounts of protein lysates were loaded and separated by SDS-PAGE and followed by western blotting . A list of antibodies used is provided in Supplementary file 2 . Whole-mount embryos ( E7 . 5 ) were photographed on an AZ100 Multizoom microscope ( Nikon ) with DIC optics and standard epifluorescence using an EXI-BLU-R-F-M-14 camera ( QImaging , Canada ) and Volocity software ( Improvision ) . For higher resolution , DMIRE2 inverted confocal ( Leica ) microscope was used . For confocal imaging , a series of 5 μm optical sections were taken and deconvolved with Volocity or equivalent software according to the instructions of the manufacturer . For A–P expression analysis , E7 . 5 embryos were dissected out of decidua in ice-cold M2 media ( Sigma ) , and bisected into anterior and posterior halves with glass needles after removal of the extraembryonic tissues . Individual halves were prepared for RNA isolation with Trizol reagent ( Invitrogen ) . For endoderm dissection , E7 . 5 embryos were dissected as stated before and the embryonic fragments were incubated for 10–15 min at 4°C in a solution of 0 . 5% trypsin and 2 . 5% pancreatin in PBS . The embryos were drawn into a hand-pulled glass pipette with a diameter slightly smaller than the fragment to peel away the endoderm and separate it from the epiblast . Both fragments were recovered and prepared for RNA isolation . For embryo culture , E6 . 5 mouse embryos were dissected as stated above , but leaving the ectoplacental cone intact and removing only Reichart’s membrane . The embryos were cultured in 1:1 rat serum/GMEM media ( 1 ml per embryo ) supplemented with 200 mM L-glutamine/100 mM sodium pyruvate ( Millipore , UK ) , 25 U/Ml Penicillin/25 µg/Ml Streptomycin , treated or not with 50 μM LY , in rotating wheel incubator at 37°C , 5% CO2 . After 24 hr , the embryos were collected and fixed with 4% PFA for IHC or prepared for RNA isolation . The embryos from the same litter ( WT ) were dissected at E7 . 5 and used as wild-type controls . T-test was used to analyse the relationship between groups . WT , n = 12; C , n = 10; LY , n = 9; p values: p<0 . 05 , p<0 . 01 , or *p<0 . 001 . For ECM analysis samples were obtained and prepared according to an adaptation from Turoverova et al . ( 2009 ) . Briefly , cells were removed by 2 mM EDTA ( 500 μl per 35 mm well ) , 37°C , 8–12 min , without shaking , and the cell detachment was controlled by eye . The solution was gently aspirated and residual cells and debris removed by 1×PBS wash/aspiration . Then the wells were covered with 800 ml 5% acetic acid , 4°C , ON . The wells were scrapped , the solution collected , and the wells covered with extraction buffer: 125 mM tris-HCL , pH 6 . 8 , 1% SDS , 20 mM DDT , 10% glycerol , 0 . 05 mM PSMF , protease inhibitor cocktail ( Santa Cruz Biotech , USA ) , and placed at 37°C , on a shaking platform incubator , 2 × 1 hr . Proteins were scrapped , collected and combined with the acetic acid fraction , and precipitated using methanol:chloroform . Briefly , equal volumes of protein extract and methanol were mixed together with one quarter of volume of chloroform . The mixture was centrifuged at 9000×g for 1 min and the upper phase was removed . At least 3 volumes of methanol was added to the lower phase and interphase with precipitated protein , mixed and centrifuged at 9000×g for 2 min . The pellets were allowed to air dry before solubilisation . Protein concentration was measured using the Amido Black Protein Assay ( Dieckmann-Schuppert and Schnittler , 1997 ) . Proteins were dissolved in 200 μl of HEPES buffer ( 200 mM HEPES , pH 8 . 0 containing 0 . 1% of RapiGest SF ) ( Waters Corp . , UK ) . Protein sample was reduced in 5 mM THP for 30 min at 37°C , and cysteines were alkylated in 10 mM iodoacetamide at RT in the dark for 30 min . Trypsin ( modified , sequencing grade , Roche ) was added to the sample at a ratio of 1:50 enzyme/protein , and allowed to digest overnight at 37°C . After tryptic digestion , formic acid was added to 2% final concentration to stop the reaction , and samples were then incubated at 37°C for an additional 4 hr . The samples were centrifuged for 30 min at 10 , 000×g to remove insoluble material and filtered with a 0 . 2 μm cartridge ( Varian , UK ) . Filtered peptides were dried by a Speed Vac , and stored at −20°C . Capillary-HPLC-MSMS analysis was performed on an on-line micro-pump ( 1200 binary HPLC system , Agilent , UK ) coupled to a hybrid LTQ-Orbitrap XL instrument ( Thermo-Fisher , UK ) . The LTQ was controlled through Xcalibur 2 . 0 . 7 and LTQ Orbitrap XL MS2 . 4SPI . HPLC-MS methods have been described previously ( Le Bihan et al . , 2010 ) . Samples were reconstituted in 10 µl loading buffer before injection , and analysed on a 2 hr gradient for data dependant analysis . As a control , pure gelatine was treated and analysed using the same procedure . Figure 7—source data 2 also shows the results for gelatine MS-LC analysis . LC-MS Label-free quantification was performed using Progenesis 2 . 6 ( Nonlinear Dynamics , UK ) . The number of Features was reduced to only MSMS peaks with a charge of 2 , 3 , or 4+ and the five most intense MSMS spectra per ‘Feature’ were kept . The generated MGF files were searched using MASCOT Versions 2 . 3 ( Matrix Science Ltd , UK ) against a mouse plus contaminant proteins IPI database with 55413 sequences downloaded from www . ebi . ac . uk ( version v3 . 42 ) . Variable methionine oxidation , STY phosphorylation , protein N-terminal acetylation and fixed cysteine carbamidomethylation were used in all searches . Precursor mass tolerance was set to 7 ppm and MSMS tolerance to 0 . 4 amu . The significance threshold ( p ) was set below 0 . 05 ( MudPIT scoring in Mascot ) . The list of proteins and quantitation are reported in Supplemental Information . Q values for false discovery rate control based on p value generated by Progenesis software were calculated from a R package ( Storey and Tibshirani , 2003 ) . A similar approach to Hartmann et al . ( 2009 ) in terms of threshold choice was used in this study . Hierarchical clustering was performed using Genesis ( Sturn et al . , 2002 ) with an average linkage clustering .	From conception to birth , a single fertilised egg will multiply into trillions of cells , with each cell becoming one of the 200 or so different types of cell that are found in the human body . The development of an embryo is complex and dynamic , with cells giving up their ability to become any cell type and committing to becoming a specific cell type within a given tissue . At the same time , different groups of cells migrate to the appropriate locations within the developing embryo . Although it is challenging to decipher the roles of the individual signalling pathways that control an embryo’s development , several important components have been found . Fibroblast growth factor ( FGF ) is a protein that regulates the formation of the endoderm: this is the innermost of the three layers of cells that form in the early embryo , and it gives rise to internal organs such as the gut , liver and pancreas . As well as ‘telling’ cells to become the front part , or anterior , of the endoderm , FGF also controls the migration of these cells within the embryo . However , uncoupling these two roles has been a major challenge , and the molecular mechanisms behind them are unclear . Now , Villegas et al . have discovered that FGF activates a signalling cascade involving two enzymes called PI3K and Akt1 . In lab-grown embryonic stem cells—cells that can be coaxed to become any of the cell types formed during development—this signalling cascade is essential for FGF to trigger differentiation of the cell types found in the anterior endoderm . The PI3K/Akt1 signalling cascade achieves this by reducing the level of a protein called fibronectin in the ‘extracellular matrix’ that surrounds the cells . This low level of fibronectin will in turn induce cells to stick together in an organized layer; and this rearrangement of cell-cell and cell-matrix interactions appears linked to triggering the differentiation of anterior endoderm cell types . Villegas et al . showed that the PI3K/Akt1 pathway was also essential for endoderm formation in living mouse embryos . As a normal embryo develops , the anterior endoderm cells move into a ‘groove’ at the front the embryo , where the level of fibronectin is lower than it is at the posterior end of the embryo . These findings highlight the importance of the extracellular matrix in the regulation of embryonic development , and should assist in the effort to turn lab-grown stem cells into the useful cell types found in internal organs .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2013	PI3K/Akt1 signalling specifies foregut precursors by generating regionalized extra-cellular matrix
In response to an osmotic challenge , the synthesis of the antidiuretic hormone arginine vasopressin ( AVP ) increases in the hypothalamus , and this is accompanied by extension of the 3′ poly ( A ) tail of the AVP mRNA , and the up-regulation of the expression of RNA binding protein Caprin-2 . Here we show that Caprin-2 binds to AVP mRNAs , and that lentiviral mediated shRNA knockdown of Caprin-2 in the osmotically stimulated hypothalamus shortens the AVP mRNA poly ( A ) tail at the same time as reducing transcript abundance . In a recapitulated in vitro system , we confirm that Caprin-2 over-expression enhances AVP mRNA abundance and poly ( A ) tail length . Importantly , we show that Caprin-2 knockdown in the hypothalamus decreases urine output and fluid intake , and increases urine osmolality , urine sodium concentration , and plasma AVP levels . Thus Caprin-2 controls physiological mechanisms that are essential for the body's response to osmotic stress . The maintenance of salt and water homeostasis is a sine qua non condition for the survival of all living organisms ( Antunes-Rodrigues et al . , 2004 ) . In mammals , this process is centrally regulated by the hypothalamo-neurohypophyseal system ( HNS ) , the source of the antidiuretic neuropeptide hormone arginine vasopressin ( AVP ) . The HNS consists of the large peptidergic magnocellular neurones ( MCNs ) of the hypothalamic supraoptic nuclei ( SON ) and paraventricular nuclei ( PVN ) , the axons of which course though the internal zone of the median eminence and terminate on blood capillaries of the posterior pituitary ( PP ) gland ( Bargmann , 1966; Burbach et al . , 2001; Antunes-Rodrigues et al . , 2004 ) , a neuro–vascular interface through which the brain regulates peripheral organs in order to maintain homeostasis . The biosynthesis of the AVP ( Brownstein et al . , 1980; Burbach et al . , 2001 ) starts with the translation of AVP-encoding messenger ribonucleic acids ( mRNAs ) into a large precursor preprohormone that comprises a signal peptide ( SP ) , the nine amino acid AVP moeity itself , the neurophysin II carrier molecule , and a C-terminal glycopeptide ( copeptin ) of unknown function . Delivery of AVP to the pituitary starts with the insertion of the preprohormone into the lumen of the endoplasmic reticulum with the removal of the SP . The resulting prohormone is folded then routed unidirectionally to the trans-Golgi network , where it is targeted specifically into the large dense core vesicle of the regulated secretory pathway . Whilst these granules are anterogradely transported some considerable distance towards the PP , the prohormone is processed to generate the biologically active , mature AVP , which is stored in PP axon terminals until mobilised for secretion into the systemic circulation . The rise in plasma osmolality that follows dehydration is detected by intrinsic MCN mechanisms ( Bourque , 2008 ) and by specialised osmoreceptive neurons in the circumventricular organs such as the subfornical organ , which provide excitatory inputs to shape the firing activity of MCNs for hormone secretion ( McKinley et al . , 2004 ) . Upon release , AVP travels through the blood stream to specific receptor targets located in the kidney where it promotes water reabsorption in the collecting duct ( Breyer and Ando , 1994 ) , and sodium reabsorption in the thick ascending limb of the loop of Henle ( Ares et al . , 2011 ) . As a consequence of the depletion of pituitary stores that accompanies a chronic osmotic stimulation , there is a need to synthesize more AVP . This starts with an increase in transcription ( Murphy and Carter , 1990 ) , which results in an increase in the abundance of both the precursor heteronuclear RNA ( Kondo et al . , 2004 ) and the mature mRNA ( Sherman et al . , 1986 ) . In addition , following the onset of an osmotic challenge , the AVP mRNA is subject to a curious post-transcriptional modification in the form of an increase in the length of the 3′ poly ( A ) tail ( Carrazana et al . , 1988; Zingg et al . , 1988; Carter and Murphy , 1989; Murphy and Carter , 1990 ) . Until this report , the regulation and physiologiocal function of this poly ( A ) tail length increase were not understood . In order to better understand the network of plastic events in the osmotically challenged SON and PVN , and in an attempt to identify novel regulatory players , we ( Hindmarch et al . , 2006 ) and others ( Yue et al . , 2006 ) have used microarrays to ask how chronic osmotic stress evokes changes in the transcriptome . One of the genes found to be differentially up-regulated in the SON and PVN was cytoplasmic activation/proliferation-associated protein-2 ( Caprin2 ) ( Mutsuga et al . , 2004; Hindmarch et al . , 2006; Yue et al . , 2006 ) , otherwise known as C1q domain containing 1 ( C1qdc1 ) , EEG1 and RNA granule protein 140 ( RNG140 ) . In these various guises , Caprin2 has been implicated in the inhibition of cell growth ( Aerbajinai et al . , 2004 ) , differentiation ( Aerbajinai et al . , 2004; Lorén et al . , 2009 ) , the enhancement of canonical Wnt signaling ( Ding et al . , 2008; Miao et al . , 2014; Flores and Zhong , 2015 ) , and , as an RNA-binding protein , the maintenance of the dendritic structure in the adult vertebrate brain ( Shiina and Tokunaga , 2010 ) . We sought to understand the physiological role of Caprin2 in the function related plasticity exhibited following an osmotic challenge . Specifically , given that Caprin2 is an RNA binding protein , we tested the hypothesis that it may associate with AVP transcripts and hence mediate changes in poly ( A ) tail length and/or mRNA stability . Quantitative RT-PCR analysis ( qRT-PCR ) was used to demonstrate robust up-regulation of Caprin-2 mRNA expression in the rat PVN ( Figure 1A ) and SON ( Figure 1B ) following two different types of chronic osmotic stress—7 days of salt-loading ( obligate consumption of 2% NaCl wt/vol in tap water ) or 72 hr of dehydration ( complete fluid deprivation ) . In parallel , AVP mRNA levels are significantly increased in both PVN and SON ( Figure 1A , B ) . There was no significant change in Rpl19 levels ( Figure 1A , B ) . Fluorescent immunostaining of brain slices revealed the presence of Caprin-2 in AVP MCNs in the PVN ( Figure 2A ) and SON ( Figure 2B ) . Salt-loading ( Figure 2A , B ) and dehydration ( not shown ) both elicit a apparent increase in the intensity of staining . Mean Caprin-2 fluorescence in identified AVP cells also significantly increased ( Figure 2D ) , whilst AVP signal in AVP cells significantly decreased ( Figure 2D ) . We note that some MCNs express Caprin-2 but not AVP . These are possibly MCNs that express the closely related oxytocin neuropeptide hormone . 10 . 7554/eLife . 09656 . 003Figure 1 . Caprin-2 messenger ribonucleic acid ( mRNA ) expression in the rat paraventricular nuclei ( PVN ) and supraoptic nuclei ( SON ) increases following a chronic osmotic stimulus . Quantitative RT-PCR analysis ( qRT-PCR ) analysis of Caprin-2 mRNA expression in the PVN ( A ) and SON ( B ) of euhydrated ( EU ) , salt-loaded ( SL ) and dehydrated ( DH ) rats . p ≤ 0 . 05 , p ≤ 0 . 01 , p ≤ 0 . 001 , p ≤ 0 . 0001 , n = 5 , One-way ANOVA with Sidak's post-hoc test . Compared with euhydrated ( EU ) rats , both SL and DH resulted in significant up-regulation of Caprin-2 mRNA in the PVN and SON . SL and DH significantly increased Caprin-2 mRNA levels in the PVN ( euhydrated PVN , 1 . 03 ± 0 . 04; salt-loaded PVN , 4 . 62 ± 0 . 34; dehydrated PVN , 3 . 84 ± 0 . 28; n = 5; p ≤ 0 . 0001 salt-loaded and dehydrated vs euhydrated ) . In the SON , SL further increased the Caprin-2 mRNA levels compared to DH ( p < 0 . 05 ) and SON ( euhydrated SON , 1 . 01 ± 0 . 07; salt-loaded SON , 6 . 13 ± 0 . 53; dehydrated SON , 4 . 9 ± 0 . 22 p ≤ 0 . 0001 salt-loaded and dehydrated vs euhydrated , salt-loaded vs dehydrated ) . In parallel , arginine vasopressin ( AVP ) mRNA levels are significantly increased in both PVN ( euhydrated PVN , 1 . 01 ± 0 . 07; salt-loaded PVN , 3 . 03 ± 0 . 18; dehydrated PVN , 1 . 79 ± 0 . 05; n = 6; p ≤ 0 . 001 salt-loaded and dehydrated vs euhydrated , salt-loaded vs dehydrated ) and SON ( euhydrated SON , 1 . 03 ± 0 . 11; salt-loaded , SON 2 . 10 ± 0 . 07; dehydrated SON , 1 . 45 ± 0 . 09 p ≤ 0 . 001 salt-loaded vs euhydrated , p ≤ 0 . 05 dehydrated vs euhydrated , and p ≤ 0 . 001 salt-loaded vs dehydrated ) . Rpl19 levels are unchanged in both the PVN ( euhydrated PVN , 1 . 03 ± 0 . 11; salt-loaded PVN , 0 . 93 ± 0 . 10; dehydrated PVN , 0 . 83 ± 0 . 09 , ns ) and SON ( euhydrated SON , 1 . 03 ± 0 . 10; salt-loaded SON , 0 . 95 ± 0 . 08; dehydrated SON 0 . 94 ± 0 . 09 , ns ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09656 . 00310 . 7554/eLife . 09656 . 004Figure 2 . Caprin-2 protein expression in the rat PVN and SON . Immunohistochemical analysis of Caprin-2 protein expression ( green ) in the rat PVN ( A ) and SON ( B ) in euhydrated and salt-loaded rats; co-localization with AVP-neurophysin II ( AVP; red ) . Scale bar 100 µm . ( C ) Quantification of fluorescence signals for Caprin-2 and ( D ) AVP in AVP cells . Mean fluorescence intensities were quantified in 10 SON wide field fluorescence microscope images acquired from 3 rats . AVP signal was selected ( above the same threshold for EU and SL samples ) and mean fluorescence intensity was measured for Caprin 2 and AVP signal . Each result was corrected by subtraction of mean background intensities for each channel . Salt-loading ( SL ) significantly increases Caprin-2 signal in AVP cells ( euhydrated , 13 . 68 ± 1 . 47; salt-loaded , 19 . 92 ± 0 . 86 , p = 0 . 0018 ) , but results in a significant decrease of AVP in AVP cells ( euhydrated , 56 . 94 ± 2 . 439; salt-loaded 48 . 3 ± 1 . 05 , p = 0 . 0044 ) . p ≤ 0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 09656 . 004 To study the physiological roles of Caprin-2 in the hypothalamus , we created a lentiviral vector ( LV ) expressing a specific Caprin-2 shRNA , which we used for Caprin-2 gene knockdown ( Cap2 KD ) in vivo . This shRNA was designed to target all known Caprin-2 transcript variants ( see Supplementary file 1 ) . As a control ( Ctrl ) , we used an LV expressing a scrambled shRNA . Viruses with eGFP tags expressing the hairpins were delivered bilaterally into both the PVN and SON using stereotaxic brain surgery . First we showed , at the RNA level , that Cap2 KD results in a significant reduction in Caprin-2 mRNA levels in both the salt-loaded SON ( Figure 3A ) and PVN ( Figure 3A′ ) . Caprin-2 knockdown had no effect on GAPDH mRNA levels , which were similar in the Ctrl and Cap2 KD SON and PVN . The affinity of each LV for MCNs was confirmed by co-expression of eGFP with Caprin-2 and AVP ( Figure 3B ) . We quantified the Caprin-2 fluorescent signal in eGFP-positive ( transduced ) compared to eGFP-negative ( non-transduced ) neurons in Ctrl and Cap2 KD rats . We found that in Cap2 KD rats , the Caprin-2 signal was significantly reduced in eGFP-positive neurons compared to eGFP-negative cells , whereas there was no significant difference in Caprin-2 signal between eGFP-negative and positive neurons in the Ctrl rats ( Figure 3C ) . 10 . 7554/eLife . 09656 . 005Figure 3 . Lentivirus-mediated Caprin-2 shRNA knockdown in the SON and PVN . qRT-PCR analysis of the effect of Caprin-2 knockdown in the SON ( A ) and PVN ( A′ ) on Caprin-2 mRNA levels ( 1 . 10 ± 0 . 11 vs 0 . 60 ± 0 . 04 in the Ctrl , n = 18 , and Cap2 KD , n = 11 , SON , p = 0 . 0023; 1 . 03 ± 0 . 07 vs 0 . 64 ± 0 . 06 in the Ctrl , n = 16 , and Cap2 KD , n = 7 , PVN , p = 0 . 0017 ) . GAPDH mRNA levels are unchanged by Cap2 KD ( 1 . 03 ± 0 . 07 vs 1 . 04 ± 0 . 1 , in the Ctrl , n = 18 , and Cap2 KD , n = 11 , SON , n . s . ; 1 . 02 ± 0 . 05 vs 1 . 01 ± 0 . 1 in the Ctrl , n = 16 , and Cap2 KD , n = 7 , PVN , n . s . ) . ( B , C ) Immunohistochemistry-based quantification of Caprin-2 gene knockdown in magnocellular neurones ( MCNs ) . ( B ) MCNs in the SON transduced with control scrambled shRNA and Caprin-2 shRNA lentiviruses visualized by immunostaining for eGFP ( GFP , green ) and co-localised with Caprin-2 ( red ) and AVP neurophysin II ( AVP; blue ) . Full arrows—GFP positive cells . Arrow heads—GFP negative cells . Whilst the scrambled shRNA has no effect on Caprin-2 levels , in cells expressing the specific Caprin-2 shRNA , Caprin-2 expression is much reduced . ( C ) There was no significant difference in Caprin-2 signal between eGFP-negative and positive neurons in the Ctrl rats ( respectively , 59 . 86 ± 2 . 89 and 66 . 25 ± 6 . 61; n = 18 and 23; n . s . ) In contrast , in Cap2 KD SON , Caprin-2 signal was significantly reduced in eGFP-positive neurons compared to eGFP-negative cells ( respectively , 20 . 16 ± 1 . 97 vs 76 . 19 ± 6 . 22; n = 15 and 26; p ≤ 0 . 0001 ) . p ≤ 0 . 01 , *p ≤ 0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 09656 . 005 In separate groups of animals , we then asked about the effects of Caprin-2 knockdown on fluid homeostasis in vivo . In euhydrated rats , Caprin-2 knockdown had no significant effect on any of the measured parameters . However , following salt-loading , Cap2 KD has profound effects . In Ctrl rats , as in naïve rats ( Greenwood et al . , 2015 ) , salt loading resulted in increases in urine output and fluid intake that were both attenuated by Cap KD ( Figure 4A , B; Table 1 ) . Urine osmolality dropped significantly after SL in both Ctrl and Cap2 KD rats , but this was less pronounced in the latter animals ( Figure 4C; Table 1 ) . During salt-loading , there was a significant increase of urine [Na+] in both groups , however in Cap2 KD rats , urine [Na+] was significantly higher than in controls ( Figure 4D; Table 1 ) . At the end of the experiment , we measured plasma osmolality and AVP content . Caprin-2 knockdown had no significant effect on plasma osmolality ( 308 . 7 ± 1 . 49 mOsmol/kg in Ctrl rats and 309 . 4 ± 2 . 16 mOsmol/kg in Cap2 KD rats , p = 0 . 78 ) . However , plasma AVP levels in the Cap2 KD rats were significantly higher than in the Ctrl rats ( 34 . 7 ± 5 . 5 pg/ml in Cap2 KD rats vs 21 . 6 ± 2 . 8 pg/ml in Ctrl rats; p ≤ 0 . 05 ) ( Figure 4E ) . We then used qRT-PCR to ask if the levels of AVP transcripts in the hypothalamus were changed following Cap2 KD . Paradoxically , in contrast to the increase in plasma AVP , we found that Caprin-2 mRNA knockdown ( Figure 3A , A′ ) was accompanied by a significant decrease in AVP mRNA levels in SON ( Figure 4F ) and PVN ( Figure 4F′ ) . Note that Caprin-2 knockdown had no significant effect on food intake or body weight ( not shown ) . 10 . 7554/eLife . 09656 . 006Figure 4 . Physiological effects of Caprin-2 gene knockdown in euhydrated and salt-loaded rats . Urine output ( A ) , fluid intake ( B ) , urine osmolality ( C ) and urine sodium concentration ( D ) were measured in control , scrambled shRNA ( Ctrl ) and Caprin-2 shRNA lentivirus-injected ( Cap2 KD ) euhydrated ( received water for 3 days , W1–3 ) and salt-loaded rats ( received 2% wt/vol NaCl ad libitum for 7 days , SL1-7 ) . Plasma AVP concentration ( E ) was measured after 7 days of SL , at the end of the experiment . p ≤ 0 . 05 , p ≤ 0 . 01 , p ≤ 0 . 001 , **p ≤ 0 . 0001 , n = 9 ( Ctrl ) and 5 ( Cap 2 KD ) . ( F , F′ ) qRT-PCR analysis of the effects of Caprin-2 knockdown in the SON ( F ) and PVN ( F′ ) on AVP mRNA levels ( 1 . 12 ± 0 . 13 vs 0 . 75 ± 0 . 08 for Ctrl , n = 18 , and Cap2 KD , n = 11 , SON , p = 0 . 047; 1 . 06 ± 0 . 08 vs 0 . 68 ± 0 . 13 for Ctrl , n = 16 , and Cap2 KD , n = 7 , PVN , p = 0 . 019 ) . p ≤ 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 09656 . 00610 . 7554/eLife . 09656 . 007Table 1 . Urine output ( A ) , fluid intake ( B ) , Urine osmolality ( C ) and urine sodium concentration ( D ) in rats injected into the SON and PVN with either , control , scrambled shRNA or Caprin-2 shRNA , in euhydrated ( water: W1–3 ) and salt-loading ( SL 1–7 ) conditionsDOI: http://dx . doi . org/10 . 7554/eLife . 09656 . 007Scrambled shRNACaprin 2 shRNAMeanSEMNMeanSEMNA . Urine output ( ml/100 g b . m . ) W14 . 6610 . 34493 . 9540 . 2225W24 . 7840 . 33994 . 0480 . 4235W34 . 8030 . 41394 . 2300 . 4615SL111 . 6491 . 590910 . 6001 . 1555SL214 . 4822 . 00599 . 4881 . 0625SL317 . 7902 . 294910 . 8141 . 7765SL420 . 8843 . 086911 . 8601 . 4855SL522 . 0463 . 564911 . 9822 . 2235SL622 . 4733 . 367913 . 5982 . 6665SL722 . 8333 . 828912 . 1942 . 1355B . Fluid intake ( ml/100 g b . m . ) W19 . 3530 . 66899 . 6740 . 4215W29 . 7520 . 60798 . 8320 . 2515W39 . 6840 . 54299 . 2030 . 4255SL114 . 1271 . 665913 . 8061 . 2245SL218 . 2702 . 145914 . 4831 . 1565SL322 . 0022 . 325915 . 7071 . 4605SL425 . 3513 . 211916 . 6801 . 8865SL526 . 8423 . 580916 . 8182 . 4375SL627 . 1663 . 912918 . 2093 . 1245SL728 . 9523 . 805916 . 5542 . 1085C . Urine osmolality ( mOsmol/kg ) W11943 . 333154 . 93791994 . 000175 . 5165W21986 . 667190 . 34392090 . 00065 . 6515W32105 . 556204 . 70291860 . 000176 . 4375SL11532 . 222108 . 18991362 . 00087 . 7725SL21390 . 00085 . 26191690 . 000120 . 8315SL31280 . 00080 . 03591606 . 000178 . 3425SL41181 . 11183 . 77391520 . 000149 . 7675SL51188 . 889115 . 54891620 . 000195 . 0905SL61121 . 11175 . 06491438 . 000152 . 9185SL71093 . 33379 . 86191474 . 000156 . 8315D . Urine sodium ( mM ) W1-3295 . 92915 . 17925321 . 30613 . 19615SL1-3726 . 2217 . 10227746 . 22114 . 04815SL2-4721 . 8506 . 85227761 . 95911 . 23315SL3-5713 . 6537 . 92527759 . 99213 . 38915SL4-6703 . 2707 . 88827758 . 02514 . 09315SL5-7701 . 0858 . 31927763 . 92615 . 83815 Previous studies have shown that Caprin-2 is an RNA binding protein ( Shiina and Tokunaga , 2010 ) . We thus tested the hypothesis that in the rat SON and PVN , Caprin-2 might bind to the AVP mRNA . We performed an RNA immunoprecipitation assay on extracts from the SON and PVN of EU and SL rats using anti-Caprin-2 antibodies , followed by qRT-PCR ( Figure 5A ) . The level of Caprin-2-AVP mRNA binding was quantified by comparing it to the signal detected with the non-specific binding control IgG performed simultaneously for each individual sample . In all samples incubated with Caprin-2 antibody , we found AVP mRNA at levels 1–2 orders of magnitude higher than in samples incubated with non-specific IgG . AVP mRNA levels in the Caprin-2-enriched extracts from EU and SL SON were respectively 49 . 44 ± 10 . 77 ( n = 5 , p = 0 . 002 ) and 23 . 77 ± 4 . 22 ( n = 5 , p = 0 . 0006 ) times higher , as compared to extracts incubated with non-specific IgG . In the EU and SL PVN , these values were respectively , 91 . 92 ± 24 . 25 ( n = 4 , p = 0 . 0037 ) and 108 ± 11 . 74 ( n = 5 , p ≤ 0 . 0001 ) . Binding to the control Rpl19 mRNA was negligible ( not shown ) . 10 . 7554/eLife . 09656 . 008Figure 5 . Caprin-2 binds to the AVP mRNA in the SON and PVN . Binding of AVP by Caprin-2 protein in the SON and PVN of euhydrated ( EU ) and salt-loaded ( SL ) rats determined by RNA immunoprecipitation assay . ( A ) In the RNA immunoprecipitation assay , SON or PVN tissue punches from EU or SL rats were first exposed to formaldehyde in order to covalently cross-link RNA with associated proteins . Cell extracts were then incubated with antibodies recognizing Caprin-2 . Following immunoprecipitation , and hence enrichment of specific complexes , cross-links were reversed and extracted RNA was subject to qRT-PCR to detect AVP mRNA sequences . ( B ) Effects of salt-loading on the amount of Caprin-2 binding to AVP mRNA in the SON ( 1 . 2 ± 0 . 36 EU vs 2 . 05 ± 0 . 20 SL , n = 5 , p = 0 . 072 ) and PVN ( 1 . 03 ± 0 . 13 EU vs 3 . 6 ± 0 . 92; n = 5 , p = 0 . 0243 ) . ( C ) Salt-loading has no effect on the amount of Caprin-2 binding to Rpl19 mRNA in the SON and PVN . p ≤ 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 09656 . 008 To examine the effect of salt loading on the amount of Caprin-2- bound AVP transcripts , we quantified Caprin-2-bound fractions in the SL rats relative to samples from the EU rats . The results showed significantly higher levels of AVP mRNA bound to Caprin-2 in the SL PVN and the similar trend was observed in the SON ( Figure 5B ) . There was no change in the already low level of Caprin-2 binding to Rpl19 mRNA ( Figure 5C ) , which does not change in abundance in the PVN or SON following an osmotic stimulus ( Figure 1A , B ) . We and others have shown that osmotic stress results in the extension of the poly ( A ) tails of the AVP mRNA ( Carrazana et al . , 1988; Zingg et al . , 1988; Carter and Murphy , 1989; Murphy and Carter , 1990 ) . We thus hypothesized that Caprin-2 binding to the AVP mRNA might be involved in this process . Quantification of transcript sizes by Northern Blot revealed that Caprin-2 knockdown prevented the salt-loading-induced increase in AVP mRNA length in SON ( Figure 6A ) and PVN ( Figure 6A′ ) ; the poly ( A ) returns to the length seen in euhydrated rats when Caprin-2 is knocked down . No changes were observed with the GAPDH mRNA ( Figure 6B , B′ ) . Removal of the poly ( A ) tails by hybridisation with oligo ( dT ) and incubation with RNase H reduced the size of the AVP mRNA , and removed differences between Ctrl and Cap2 KD animals ( Figure 6C , C′ ) . 10 . 7554/eLife . 09656 . 009Figure 6 . In vivo effects of Caprin-2 gene knockdown on the length of the AVP mRNA poly ( A ) tail . AVP mRNAs were analyzed by Northern Blotting . Quantification results of AVP ( A , A′ ) and GAPDH mRNA ( B , B′ ) in samples obtained from the SON ( A , B ) and PVN ( A′ , B′ ) of SL rats injected with lentivirus with either , scrambled shRNA ( Ctrl ) or Caprin-2 shRNA ( Cap2 KD ) . Insets on the plots with GAPDH mRNA represent typical images . Representative images of AVP mRNA before and after removing poly ( A ) tails ( RNaseH ) in the SON ( C ) and PVN ( C′ ) of euhydrated ( − ) and salt-loaded ( SL ) rats , as well as in salt-loaded rats injected with scrambled ( Scr shRNA ) or Caprin-2 ( Cap2 shRNA ) shRNA . Quantification of transcript sizes revealed that Caprin-2 knockdown prevented the SL-induced increase of the AVP mRNA length , reducing it from 822 . 7 ± 3 . 76 nt . ( n = 8 , Ctrl ) to 801 . 7 ± 4 . 47 nt . ( n = 7 , Cap2 KD ) ( p = 0 . 0031 ) in the SON , and from 877 . 7 ± 4 . 77 nt . ( n = 6 , Ctrl ) to 851 . 3 ± 9 . 23 nt . ( n = 6 , Cap2 KD ) in the PVN ( p = 0 . 0296 ) . No size changes were observed with the GAPDH mRNA ( Ctrl SON , n = 8 , 1393 ± 7 . 55 nt . vs Cap2 KD , n = 7 , 1392 ± 6 . 12 nt . , n . s . ; Ctrl PVN , n = 6 , 1388 ± 4 . 65 nt . vs Cap2 KD , n = 6 , 1368 ± 11 . 24 nt . , n = 6 , n . s . ) . p ≤ 0 . 05 , p ≤ 0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 09656 . 009 We then developed a recapitulated in vitro system to ask if the effect of Caprin-2 on AVP mRNA abundance and poly ( A ) tail length was direct . We overexpressed the rat AVP structural gene and full-length rat Caprin-2 ( Supplementary file 1 ) in HEK293T cells , both under the control of the heterologous CMV promoter ( Figure 7 ) . A vector expressing eGFP was used as a control . Western blotting showed robust expression of eGFP and Caprin-2 proteins ( Figure 7 ) . For knockdown studies , cells were additionally co-transfected with a Caprin-2 shRNA vector ( overexpression control ) or scrambled shRNA in place of the Caprin-2 shRNA ( knockdown control ) . 10 . 7554/eLife . 09656 . 010Figure 7 . A recapitulated in vitro system for examining effect of Caprin-2 on the metabolism of the AVP mRNA . In order to further examine to role of Caprin-2 in AVP mRNA metabolism , we developed a recapitulated in vitro system . We overexpressed the rat AVP structural gene and full-length rat Caprin-2 in HEK293T cells , both under the control of the heterologous CMV promoter . A vector expressing eGFP was used as a control . Western blotting of triplicate independent samples showing robust expression of eGFP and Caprin-2 proteins in transduced cells . DOI: http://dx . doi . org/10 . 7554/eLife . 09656 . 010 Firstly , we showed using qRT-PCR that Caprin-2 knockdown significantly reduced Caprin-2 mRNA level , compared to the respective control ( Figure 8A ) . Further , we found that the decrease in Caprin-2 mRNA abundance was accompanied by a 24% drop in AVP mRNA levels ( Figure 8B ) . In contrast , overexpression of Caprin-2 resulted in a significant increase in AVP mRNA abundance ( Figure 8C ) . 10 . 7554/eLife . 09656 . 011Figure 8 . Effects of Caprin-2 overexpression or knockdown on AVP mRNA levels in vitro . For Caprin-2 knockdown experiments , cells were co-transfected with the CMV-driven rat Caprin-2 overexpression construct , the CMV-driven rat AVP structural gene , and either scrambled shRNA ( Ctrl ) or Caprin-2 shRNA ( Cap2 KD ) constructs . Caprin-2 knockdown ( Cap2 KD ) reduces ( A ) Caprin-2 ( Ctrl , 1 . 00 ± 0 . 03 vs Cap2 KD , 0 . 13 ± 0 . 01 , n = 5 , p ≤ 0 . 0001 ) and ( B ) AVP ( Ctrl , 1 . 00 ± 0 . 02 vs Cap2 KD , 0 . 76 ± 0 . 02; n = 5 , p ≤ 0 . 0001 ) mRNA levels . In the Caprin-2 overexpression experiments , the CMV-driven rat AVP structural gene was co-transfected with either the CMV-driven rat-Caprin-2 overexpression construct ( Cap2 ) , or a control CMV-eGFP construct ( Ctrl ) . ( C ) Caprin-2 overexpression increases the level of AVP mRNAs ( Ctrl , 1 . 0 ± 0 . 05 vs Cap2 , 2 . 17 ± 0 . 14 , n = 5 , p ≤ 0 . 0001 ) . p ≤ 0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 09656 . 011 Northern blot analysis demonstrated that Caprin-2 knockdown significantly decreased the length of the AVP mRNA ( Figure 9A , E ) . Removal of poly ( A ) tails reduced the size of the AVP mRNAs from Ctrl and Cap2 KD cells to the same length ( Figure 9A , E ) . In contrast , Caprin-2 overexpression significantly increased the length of the AVP mRNA ( Figure 9B , F ) . Removal of the poly ( A ) tails by hybridisation with oligo ( dT ) and incubation with RNase H reduced the size of the AVP mRNAs from all samples to the same length ( Figure 9B , F ) . Neither , Caprin-2 overexpression nor Caprin-2 knockdown had an effect on the length of the GAPDH mRNA ( Figure 9C–F ) . 10 . 7554/eLife . 09656 . 012Figure 9 . Effects of Caprin-2 overexpression or knockdown on AVP mRNA poly ( A ) tail length in vitro . Effects of Caprin-2 knockdown ( Cap2 KD ) ( A ) and overexpression ( Cap2 ) ( B ) in HEK293 cells on the length of the poly ( A ) tail in AVP mRNA , analyzed by Northern Blotting . Knockdown reduces ( Ctrl , 833 . 3 ± 2 . 21 nt . vs Cap2 KD , 799 . 6 ± 5 . 07 nt . , n = 5 , p = 0 . 0003 ) whilst overexpression increases ( Ctrl , 804 . 4 ± 6 . 19 nt . vs Caprin-2 , 884 . 3 ± 5 . 51 nt . , n = 5; p ≤ 0 . 0001 ) the length of the AVP mRNA . This is due to modulation in the lenth of the poly ( A ) tail , when this is removed there are no differences between the groups ( Ctrl , 674 . 8 ± 3 . 75 nt . vs Cap2 KD , 674 ± 3 . 48 nt . , n = 5; n . s . ; Ctrl , 679 ± 7 . 62 nt . vs Cap2 , 678 . 9 ± 6 . 79 nt . , n = 5; n . s . ) . Neither Caprin-2 knockdown ( C ) ( Ctrl , 1351 ± 2 . 94 nt . , vs Cap2 KD , 1347 ± 4 . 98 nt . , n = 5 , n . s . ) nor Caprin-2 overexpression ( D ) ( Ctrl , 1377 ± 6 . 13 nt . vs Caprin-2 , 1380 ± 8 . 05 nt . , n = 5; n . s . ) had any effect on the size of the GAPDH mRNA . Typical images representing the effects of Caprin-2 knockdown ( E ) and overexpression ( F ) on AVP mRNA in HEK293T cells before and after removing poly ( A ) tails ( RNaseH ) are shown . p ≤ 0 . 0001 , *p ≤ 0 . 0003 . DOI: http://dx . doi . org/10 . 7554/eLife . 09656 . 012 We provide the first evidence that RNA binding protein Caprin-2 plays a critical role in mediating brain responses to osmotic stress . We show that Caprin-2 is expressed in AVP MCNs in the SON and PVN and that Caprin-2 expression in these neurons increases during osmotic stress . Importantly , we demonstrate that Caprin-2 knockdown in the SON and PVN disrupts physiological osmoregulatory mechanisms . Normally , chronic salt loading leads to a gradual increase of urine output and fluid intake . Loss of Caprin-2 significantly reduces the salt loading-induced urine output and fluid intake and increases urine osmolality , urine sodium concentration and plasma AVP levels . We then explored the molecular mechanisms of Caprin-2 action in homeostatic osmoregulatory hypothalamic neuronal circuits . Caprin-2 has recently been identified in vitro as an RNA binding protein RNG140 ( Shiina and Tokunaga , 2010 ) . However , up until now , no specific RNA binding partners of Caprin-2 have been identified . Here we show that Caprin-2 binds to the AVP mRNA in the SON and PVN of EU and SL rats in vivo , and , in doing so , modulates both its abundance and poly ( A ) tail length . By stimulating water retention at the level of the kidneys ( Breyer and Ando , 1994 ) , AVP plays a central role in the maintenance of cardiovascular homeostasis , particularly blood volume and osmolality ( Antunes-Rodrigues et al . , 2004 ) . The increase in AVP biosynthesis in MCNs that follows osmotic stimulation is accompanied by a plethora of changes in gene expression ( Mutsuga et al . , 2004; Hindmarch et al . , 2006; Yue et al . , 2006 ) . One of the novel genes identified as such by transcriptome analysis was Caprin-2 . Here we confirm the transcriptome data , and that Caprin-2 protein is up-regulated at the mRNA level by chronic hyperosmotic cues . Using double immunostaining we also provide evidence that Caprin-2 protein is present in the cytoplasm of VP MCNs , and that its expression increases during osmotic stress . Caprin-2 has been reported to play a role in several physiological processes , including differentiation , apoptosis in erythroblasts or lens fiber cells , and synaptic plasticity in the mouse cerebellum ( Aerbajinai et al . , 2004; Ding et al . , 2008; Lorén et al . , 2009; Shiina and Tokunaga , 2010; Miao et al . , 2014; Flores and Zhong , 2015 ) . Although expression of Caprin-2 in the HNS has been reported ( Mutsuga et al . , 2004 ) , its role in the maintenance of fluid homeostasis had not previously been addressed . We therefore investigated the physiological consequences of Caprin-2 gene knockdown using virally delivered specific shRNAs in the SON and PVN in vivo . The Caprin-2 shRNA had no effect on water homeostasis in euhydrated rats . In naïve rats ( Greenwood et al . , 2015 ) , and in rats transduced with a scrambled control shRNA , as shown here , salt loading results in production of large amounts of dilute urine and excessive fluid intake . However , compared to the control animals , salt loaded Caprin-2 knockdown rats only minimally increased their urine production and they drank less fluid . Urine osmolality and sodium levels were also higher in Caprin-2 knockdown rats . These observations may be explained by water retention resulting from the elevated plasma AVP levels found in the Caprin-2 shRNA-transduced rats . Our data suggest that Caprin-2 knockdown rats increase their ability to concentrate urine and so doing they prevent the decrease in sodium excretion ( i . e . , elevation of plasma sodium ) resulting from lower urine output by producing more concentrated urine . We then explored the molecular mechanisms of Caprin-2 action in the hypothalamus , based on our hypothesis that this RNA binding protein ( Shiina and Tokunaga , 2010 ) might associate with the AVP mRNA . Here we show that Caprin-2 does indeed bind to AVP mRNAs in the SON and PVN in vivo . We then asked if Caprin-2 binding has any role in the alterations in AVP mRNA metabolism that occur as a consequence of the chronic osmotic stimuli of dehydration or salt-loading , namely an increase in abundance ( Sherman et al . , 1986; Murphy and Carter , 1990 ) , and the unusual post-transcriptional modification of an increase in the length of the 3′ poly ( A ) tail ( Carrazana et al . , 1988; Zingg et al . , 1988; Carter and Murphy , 1989; Murphy and Carter , 1990 ) . Our in vivo studies show that knockdown of Caprin-2 in the rat SON and PVN prevented the salt-loading-induced increases in the abundance of the AVP mRNA , and the elongation of its poly ( A ) tail . In HEK293T cells transfected with a plasmid containing rat AVP genomic sequences under constitutive CMV promoter , concomitant overexpression of Caprin-2 increased AVP mRNA abundance , whilst addition of the Caprin-2 shRNA significantly reduced AVP mRNA levels . At the same time , Caprin-2 overexpression resulted in elongation of the poly ( A ) tail of the AVP mRNA , whilst Caprin-2 knockdown shortened the poly ( A ) tails of the AVP mRNA . The direct binding and positive regulation of AVP transcript levels suggest that Caprin-2 may be important for stabilization of AVP mRNAs during osmotic stress , in a similar way to another AVP mRNA binding protein , poly ( A ) -binding protein ( PABP ) ( Mohr et al . , 2002 ) . It is thus tempting to speculate that the two processes , increased poly ( A ) tail length and increased mRNA abundance , are consequential . An important role of poly ( A ) tail length in regulation of mRNA degradation and stability has been demonstrated for many transcripts ( Curinha et al . , 2014; Zhang et al . , 2014 ) . It is therefore possible that one of the functions of Caprin-2-mediated poly ( A ) tail length regulation is protection of AVP transcripts from degradation , which in turn would lead to the observed increase of AVP mRNA levels . Note that we are unable to distinguish between a process that promotes polyadenylation , or one that prevents deadenylation . Unexpectedly , whilst Caprin-2 knockdown in the PVN and SON resulted in a decreased AVP mRNA levels in those structures , plasma AVP ( peptide ) levels increased . Although the precise mechanisms underlying this paradoxical effect require separate investigation , examination of the existing literature leads us to propose two possible hypotheses . Firstly , Caprin-2 binding to AVP mRNA might inhibit translation , a phenomenon observed as a consequence of other association mRNA-RNA-binding protein interactions , such as the binding of PABP to the AVP mRNA ( Mohr et al . , 2002; Richter , 2008 ) . This is not necessarily in disagreement with the fact that Caprin-2 stimulates extension of poly ( A ) tail in AVP mRNAs—recent studies have revealed that a correlation between the length of poly ( A ) tail and translational activity is not always present ( Weill et al . , 2012; Subtelny et al . , 2014 ) . It has been shown previously that recombinant Caprin-2 inhibits the translation of a luciferase mRNA in a cell-free rabbit reticulocyte lysate system ( Shiina and Tokunaga , 2010 ) . This is consistent with our observation that Caprin-2 knockdown in the SON and PVN increases circulating plasma AVP levels . It is conceivable that this effect results from the increased translation of AVP mRNA under conditions when Caprin-2 protein level is too low to suppress this process . Moreover , knockdown of Caprin-2 , whilst increasing the translation of the AVP mRNA , might increase its turnover , and hence decreases steady state transcript levels and poly ( A ) tail length . We have previously examined the polysome distribution of the AVP mRNA in euhydrated and salt-loaded SON ( Murphy and Carter , 1990 ) , and showed no difference in the pattern association with heavy polysome fractions . However , the AVP mRNA is small , and subject to a high rate of translation , even in euhydrated animals . Thus , the rates of translation initiation , elongation and termination , none of which are directly measured by polysome analysis , especially when the size of the RNA limits ribosome number , could be influenced , positively or negatively , by poly ( A ) tail length . Quantification of Caprin-2 and AVP levels in AVP in the SON neurons ( Figure 2 ) revealed that salt-loading elicits a significant increase in the former protein , but a significant decrease in the latter , consistent with Caprin-2 being an inhibitor of the translation of the AVP mRNA . However , it is important to point out the steady-state level of AVP in hypothalamic MCNs is not only determined by translation rate . As we describe in some detail in the Introduction , the AVP precursor is subject to axonal transport from cell bodies in the hypothalamus to terminals in the PP , a process that is activated by the need to deliver mature peptide to the circulation following an osmotic stimulus . Thus , further studies are needed to examine the translation of the AVP mRNA under different physiological conditions . A second , but not mutually exclusive hypothesis is that the effects that we observe after Caprin-2 knockdown could be modulated by changes in dendritic release of AVP . Dendritic release and autocrine or paracrine action of AVP has been well documented ( Ludwig and Stern , 2015 ) . AVP released from dendrites inhibits the electrical activity of AVP neurons , thus suppressing axonal release to the blood stream ( Ludwig and Stern , 2015 ) . Caprin-2 protein is present in the RNA granules localized in the rat neuronal dendrites ( Shiina and Tokunaga , 2010 ) . Therefore , it is possible that Caprin-2 knockdown results in a decrease of the synthesis and dendritic release of AVP , which in turn decreases AVP-mediated auto-inhibition of AVP neurons and leads to an increase of AVP release in the neurohypophysis and increased plasma concentration , as we observe after Caprin-2 knockdown in the SON and PVN . In this report , we have shown that , during an osmotic stimulus , three things happen to the AVP mRNA in the SON and PVN due to the action of Caprin-2: the abundance of Caprin-2 protein/AVP mRNA complexes increases , the AVP mRNA poly ( A ) tail elongates , and there is an increase in steady state levels of the AVP mRNA . On the basis of these data , we propose a new mechanism mediating the regulation of AVP gene expression ( Figure 10 ) . Whilst it has previously been demonstrated that a chronic osmotic stimulus results in an increase in the transcription of the AVP gene ( Murphy and Carter , 1990 ) , the data presented here provides new evidence that post-transcriptional process , governed by Caprin-2 , are also operative . Both the transcriptional and the post-transcriptional mechanisms would contribute to overall steady-state AVP mRNA abundance . However , it would appear that the increase in AVP mRNA mediated by Caprin-2 does not result in an increase in AVP peptide production . Indeed , our results are consistent with previous reports that showed that Caprin-2 inhibits translation ( Shiina and Tokunaga , 2010 ) . Thus we suggest that association with Caprin-2 may inhibit the production of AVP peptide , perhaps by directly interfering with the translational apparatus , or through sequestration of AVP mRNA in translationally inert mRNP storage or transport granules ( Buchan , 2014 ) . Interestingly , polysome analysis revealed a heavy class of EDTA-resistant AVP mRNA containing fractions that might represent storage granules ( Murphy and Carter , 1990 ) . We also note that AVP transcripts are known to be transported to dendrites ( Mohr et al . , 2002 ) and to axons ( Murphy et al . , 1989 ) , although in the latter case , the AVP RNAs are characterised by a short poly ( A ) tail that does not change in length following osmotic stimulation ( Murphy et al . , 1989 ) . 10 . 7554/eLife . 09656 . 013Figure 10 . A model for the actions of Caprin-2 . Caprin-2 expression increases in the rat SON and PVN following an osmotic stimulus . Caprin-2 binds to the AVP mRNA , and mediates an increase in the length of the poly ( A ) tail , either by stimulating further synthesis , or by inhibiting deadenylation . Caprin-2 also mediates an increase in VP mRNA abundance , possibly by increasing transcript stability . Paradoxically , we propose that Caprin-2 inhibits translation of the AVP mRNA , perhaps by displacing ribosomes or by slowing initiation or elongation , or through mediating sequestration into translationally inert storage or transport granules . DOI: http://dx . doi . org/10 . 7554/eLife . 09656 . 013 The osmotic status-dependent modulation of AVP mRNA poly ( A ) tail length in the SON and PVN was first described over 20 years ago ( Carrazana et al . , 1988; Zingg et al . , 1988; Carter and Murphy , 1989 ) , but only now , with the discovery of Caprin-2 , do we have insight into the regulation and physiological function of this post-transcriptional process . To get a more comprehensive picture of Caprin-2 functioning in the HNS , it will be of interest to identify other mRNAs that bind to Caprin2 in the SON and PVN , in addition to those encoding AVP . Further , identification of the sequences within the AVP mRNA that are recognised by Caprin-2 , and the Caprin-2 protein modules that mediate binding to the AVP mRNA and poly ( A ) tail extension , might reveal interactions with other known Caprin-2 functions , such as the modulation of Wnt signaling ( Ding et al . , 2008; Miao et al . , 2014; Flores and Zhong , 2015 ) that may have regulatory and physiological relevance . Male Sprague–Dawley rats ( Harlan , UK ) weighing ∼300 g were housed at a constant temperature of 22°C and a relative humidity of 50–60% ( vol/vol ) under a 14:10 hr light/dark cycle . Rats had free access to food and tap water for at least 1 week prior to experimentation . To induce hyperosmotic stress , water was removed for 3 days ( dehydration , DH ) or replaced by 2% ( wt/vol ) NaCl in drinking water for 7 days ( salt loading , SL ) . The control group ( euhydrated , EU ) had access to food and water ad libitum throughout the experimental period . Rats were randomly allocated into groups . Food was available ad libitum . Tissue collections were performed between 10 am–2 pm . All experiments were carried out under the licensing arrangements of the UK Animals ( Scientific Procedures ) Act ( 1986 ) with local ethics committee approval . Rats were sacrificed by stunning and decapitation . Brains were removed and immediately frozen in powdered dry ice . SON and PVN samples were isolated from 60 µm frozen sections in a cryostat using 1 mm tissue punch tool ( Fine Scientific Tools , Germany ) . The accuracy of the tissue punch was controlled by staining each slice with 2% ( wt/vol ) toluidine blue , and visualizing it on a light microscope . Samples were stored in −80°C until further analysis . Total RNA was extracted as described ( Greenwood et al . , 2014 ) . 1 µg of RNA isolated from the rat forebrain was reverse transcribed using Super Script II RT kit ( Life Technologies , Carlsbad , CA , USA ) . 2 µl of cDNA was used as a template for PCR with Caprin-2- specific primers rCAP2_F 5′ GCTCGAGGCCACCATGAAGTCAGCCAAGTCC 3′ and rCAP2_R 5′ GGGGATCGATTCAATCTTGATAAAGAAGATAGCC 3′ , and Phusion High-Fidelity DNA Polymerase ( New England Biolabs , Ipswich , MA , USA ) under the following conditions: 98°C—2 min , ( 98°C—30 s , 53°C—30 s , 72°C—3 min ) ×40 , 72°C—10 min . The 3 kb PCR product was cloned using routine procedures , and sequenced with SP6 and T7 primers ( Source BioScience , UK ) as well as with Caprin-2-specific internal primers F 5′ GAAGGAACTTGTACAGCCAGA 3′ and R 5′ GATAAATGGCTGAGCAGGTC 3′ designed using OligoPerfect tool ( Life Technologies ) . To identify different isoforms of Caprin-2 several different clones were analysed by restriction with frequently cutting enzymes ( 4-cutters ) : Alu I ( Gibco , Life Technologies , USA ) , Dpn I ( Stratagene , San Diego , CA , USA ) , Hae I and Rsa I ( New England Biolabs ) , followed by agarose gel , and sequencing analysis ( Source BioScience ) . We found that , similar to mouse Caprin-2 ( ENSMUST00000111569 ) , the rat Caprin-2 gene consists of 18 coding exons ( Supplementary file 1; Genbank accession number—KT867373 ) . To maintain consistency with the mouse sequence , we labeled them 2–19 . Since several different isoforms of Caprin-2 have been found in human ( Aerbajinai et al . , 2004 ) , we also looked for different Caprin-2 isoforms in rat brain . We found that the longest isoform was 3090 bp long , which encodes 1029 aa . Exons were predicted on the basis of GT and AG flanking regions of introns in rat Caprin-2 gene ( NCBI reference sequence AC_000072 . 1 , 170273255–170328842 , complement strand ) and verified by comparison to the mouse Caprin-2 sequence ENSMUST00000111569 . We then identified four other isoforms , which are shorter than the first one due to the absence of several base pairs ( Cap2-2 ) , or to the removal of entire exons ( E13 and 14 in Cap2-3 , E17 in Cap2-4 , E13 in Cap2-5 ) ( Supplementary file 1 ) . Previous studies have shown that human Caprin-2 is homologous to Xenopus RNG105 , which is also known in rodents as Caprin 1 ( Shiina and Tokunaga , 2010 ) . The sequence of RNG105 has been thoroughly analyzed , therefore we used it as a reference to check if the rat Caprin-2 contains similar functional domains . Comparison of the rat 3090 bp Caprin-2 sequence to the Xenopus RNG105 revealed that they share several domains , including N-terminal coiled-coil domain thought to be involved in RNA binding , a nuclear localization signal , as well as a C-terminal RNA-binding motif , RGG ( Supplementary file 1 ) also reported to be present in human Caprin-2 . Steady state mRNA levels were assessed using quantitative real-time RT-PCR analysis . For cDNA synthesis , 100 ng of total RNA was treated with DNAse I ( Invitrogen , Life Technologies , UK ) and reverse transcribed using SuperScript III Reverse Transcriptase ( Invitrogen Life Technologies ) with random primers ( Invitrogen Life Technologies ) , in the presence of RNaseOUT Inhibitor ( Invitrogen Life Technologies ) . RNA isolated from cell cultures ( 100 ng ) was reverse-transcribed using QuantiTect Reverse Transcription Kit ( Qiagen , UK ) according to the manufacturer protocol . Quantitative PCRs were conducted in duplicate , in 25 μl reaction volumes containing 1 or 2 µl of cDNA template , 0 . 4 µM gene-specific primers and FastStart Universal SYBR Green Master mix ( ROX ) ( Roche , Switzerland ) , using ABI 7500 Real-Time PCR System ( Applied Biosystems , UK ) . The following primers were used: Caprin-2 ( F 5′ AGGTATCCAAGCCTGTGGTG 3′ , R 5′ AGGATCTGCTGCCACTCTGT 3′ ) , AVP ( F 5′ TACGCTCTCTGCTTGCTTCC 3′ , R 5′ ACTGTCTCAGCTCCATGTCG 3′ ) , Rpl19 ( ribosomal protein L19 , F 5′ GTCCTCCGCTGTGGTAAAAA 3′ , R 5′ GGCAGTACCCTTCCTCTTCC 3′ ) , Gapdh ( glyceraldehyde-3-phosphate dehydrogenase , F 5′ ATGATTCTACCCACGGCAAG 3′ , R 5′ CTGGAAGATGGTGATGGGTT 3′ [Colomer et al . , 2010] ) , eGFP ( F 5′ ACTTCTTCAAGTCCGCCATGCC 3′ , R 5′ TGAAGTCGATGCCCTTCAGCTC 3′ ) , β-actin ( F 5′ CACCCGCGAGTACAACCTTC 3′ , R 5′ CCCATACCCACCATCACACC 3′ ) . Unless specified otherwise , primers were designed using either OligoPerfect ( Invitrogen Life Technologies ) , MWG Operon ( Eurofins Scientific , Luxembourg ) or NCBI Primer Blast ( NCBI Resource Coordinators , 2015 ) tools . Primers efficiency was validated using serial dilutions of cDNA from the SON or PVN . All qRT-PCR reactions were followed by dissociation curve analysis . Relative quantification of gene expression was performed using the 2ΔΔCT method ( Livak and Schmittgen , 2001 ) and Rpl19 or GAPDH ( in HEK cells ) housekeeping genes . Rats were anesthetised with sodium pentobarbitone ( 100 mg/kg i . p . ) and transcardially perfused with 0 . 1 M phosphate buffered saline ( PBS , pH 7 . 4 ) followed by 4% ( wt/vol ) paraformaldehyde ( PFA ) in 0 . 1 M PBS . The brains were removed , post-fixed overnight in 4% ( wt/vol ) PFA , followed by 3 day-incubation in 30% ( wt/vol ) sucrose prepared in PBS , and frozen in liquid nitrogen . Coronal sections of the forebrain ( 30 µm ) were cut on a cryostat , washed in 0 . 1 M PBS ( pH 7 . 4 ) and subjected to antigen retrieval in 0 . 1 sodium citrate pH 6 . 0 at 100°C for 15 min . After three washes in PBS sections were blocked in 10% ( vol/vol ) donkey serum ( Sigma–Aldrich , UK ) in PBS supplemented with 0 . 3% Triton-X100 for 30 min . Floating sections were incubated with primary antibodies in 0 . 1 M PBS supplemented with 1% normal donkey serum , 0 . 3% ( vol/vol ) Triton X-100 for 1 hr at RT and then overnight at 4°C . The following primary antibodies were used: goat anti-Caprin-2 ( Santa Cruz Biotechnology , Dallas , TX , USA ) , rabbit anti-Caprin-2 ( kindly provided by Prof . Lin Li , Shanghai , China [Ding et al . , 2008] ) , rabbit anti-GFP ( Abcam , UK ) , and mouse anti-AVP-neurophysin ( NP-II , PS41; 1:100 ) ( kindly provided by Prof . Harold Gainer , Bathesda , MD , USA [Ben-Barak et al . , 1985] ) . Note that similar results were obtained with both of the antibodies recognising Caprin-2 . The data shown in Figures 2 , 3 were obtained using the goat anti-Caprin-2 from Santa Cruz Biotechnology . After three washes in PBS ( 15 min each ) sections were incubated for 1 hr at RT with a mixture of two or three fluorescent secondary antibodies: donkey anti-goat IgG—AF 488 or AF 594 , donkey anti-mouse IgG—AF 594 or AF 647 , donkey anti-rabbit IgG—AF 488 ( 1:500 , Life Technologies ) , diluted 1:500 in PBS containing 1% ( vol/vol ) normal donkey serum and 0 . 3% ( vol/vol ) Triton X-100 . At the end sections were incubated for 5 min at RT with DAPI ( 1 µg/ml in PBS , Sigma–Aldrich , St . Louis , MO , USA ) and washed three times for 15 min , before mounting on slides . Fluorescent imaging was performed using either , Leica DM IRB epifluorescent microscope with Volocity ( Improvision ) acquisition system and cooled CCD camera ( Hamamatsu ORCA ER Firewire ) or Leica SP2 confocal microscope in the Wolfson Bioimaging Facility , University of Bristol . Fiji software ( NIH , USA [Schindelin et al . , 2012] ) was used to quantify fluorescence intensities and co-localization of Caprin-2 , AVP , GFP and DAPI in multichannel confocal image stacks . The Caprin-2 shRNA was designed using BLOCK-iT RNAi Designer ( https://rnaidesigner . lifetechnologies . com/rnaiexpress/; Life Technologies ) . A scrambled control shRNA for this oligonucleotide was generated using siRNA Wizard v3 . 1 ( www . sirnawizard . com/scrambled . php ) . We designed 4 different shRNAs , then tested them in HEK293 cells co-transfected with a Caprin-2 overexpression plasmid . Only one shRNA delivered >55% knockdown ( c2784 , 40 . 51% ± 7 . 81 of control values ) and was used in all further studies . Oligonucleotides ( c2784 Caprin-2 shRNA , top strand: 5′ GGGAGAGACCTTTGATCTTCATTCAAGAGATGAAGATCAAAGGTCTCTCCCTTTTTT 3′ , bottom strand: 5′ AATTAAAAAAGGGAGAGACCTTTGATCTTCATCTCTTGAATGAAGATCAAAGGTCTCTCCCGGCC 3′; scrambled Ctrl shRNA , top strand: 5′ GCGTTAAGCGAGCATGTTCTATTCAAGAGATAGAACATGCTCGCTTAACGCTTTTTT 3′ , bottom strand: 5′ AATTAAAAAAGCGTTAAGCGAGCATGTTCTATCTCTTGAATAGAACATGCTCGCTTAACGCGGCC 3′ ) containing the loop sequence TTCAAGAGA were purchased from Eurofins MWG Operon , Germany . Double-stranded oligonucleotides were cloned into the pSilencer 1 . 0—U6 vector ( Ambion , Life Technologies , Carlsbad , CA , USA ) . The efficiency of the specific Caprin-2 shRNA was tested in HEK293T/17 cells ( human embryonic kidney cell line , CRL-11268 , ATCC , Manassas , VA , USA ) transiently overexpressing rat Caprin-2 . The U6-shRNA sequences were then amplified from pSilencer 1 . 0-U6 ( F 5′ ATAGATTTAATTAACACTATAGGGCGAATTGGGTA 3′ , R 5′ AGTCTTTCTCGAGCCCGGGCTGCAGGAATTA 3′ ) then cloned into LV pRRL . sin . U6 . shRNA . cppt . CMV . GFP . wpre . High titer LVs were propagated as previously reported ( Greenwood et al . , 2014 ) . Preliminary experiments were performed to assess the efficiency of Caprin-2 shRNA lentivirus transduction in vivo . Rats ( 300–340 g ) were anaesthetized by i . m . administration of Domitor/Vetalar ( Pfizer , New York City , NY , USA ) and placed in a stereotaxic frame in the flat skull position . A 2 cm rostral-caudal incision was made to expose the surface of the skull . One 0 . 8 mm hole were drilled at co-ordinates 1 . 3 mm posterior to bregma and 1 . 95 mm left to the midline for SON injection . An additional one 1 mm hole was drilled at co-ordinates 1 . 8 mm posterior to bregma , and ±0 . 4 mm lateral to midline for PVN injection . A 5 µl pulled glass pipette ( Sigma–Aldrich ) was positioned −8 . 8 mm ( SON ) or −7 . 5 mm ( PVN ) ventral to the surface of the brain and 1 µl of LV at the concentration of 2 × 109 particles/ml , was delivered separately into each nucleus , over 5 min/nucleus . At the end of surgery the rats received an agonist Antisedan ( Norden Laboratories , Lincoln , NE , USA ) and analgesic Rimadyl ( Zoetis , Florham Park , NJ , USA ) . After 3 weeks the animals were deeply anaesthetized ( Lethobarb , Fort Dodge , IA , USA ) and perfused with PBS and PFA , as described before . Brain samples were further fixed in 4% ( wt/vol ) PFA for 24 hr , cryoprotected in 30% ( wt/vol ) sucrose in PBS and frozen in liquid nitrogen . Coronal hypothalamic sections were stained with anti-GFP ( Abcam ) , AVP NP-II ( PS-41 , kindly provided by Prof . Harold Gainer , NIH , Bethesda , USA [Ben-Barak et al . , 1985] ) and Caprin-2 antibodies ( Santa-Cruz Biotechnology ) and the imaging was performed using fluorescent confocal microscope ( Leica , Germany ) as described earlier . Quantification analysis has been performed on the Z-stack and single-plane confocal microscope images obtained from the SON and PVN of 5 rats injected with scrambled ( Ctrl ) or Caprin-2 ( Cap2 KD ) shRNA , using Fiji software ( Schindelin et al . , 2012 ) . Similar surgical procedure was performed for determination of physiological results of Caprin-2 knockdown , except that it involved bilateral injections into the SON and PVN . To this end two 0 . 8 mm holes were drilled at co-ordinates 1 . 3 mm posterior to bregma and 1 . 95 mm lateral to midline for SON injection and one 1 mm hole was drilled at co-ordinates 1 . 8 mm posterior to bregma , and ±0 . 4 mm lateral to midline for PVN injection , and 1 µl of LV at the concentration of 2 × 109 particles/ml , was delivered separately into four nuclei , over 5 min/nucleus . Like previously , at the end of surgery the rats received Antisedan ( Norden Laboratories ) and analgesic Rimadyl ( Zoetis ) . After the surgery the animals were individually housed in standard laboratory cages for two and a half week before being transferred to metabolic cages ( Techniplast , Italy ) , for precise daily measurements of fluid intake and urine output . Animals were weighed and allowed to acclimatize to the cage for 48 hr . Fluid intake , urine output , urine osmolality and body weights were recorded for 10 days , between 10–11 am . The animals received ad libitum water for the first 3 days , and 2% ( wt/vol ) NaCl for the next 7 days . At the end of the experiment the animals were culled by stunning and decapitation and trunk blood samples were collected and processed immediately for plasma isolation . Brain tissue was frozen in powdered dry ice and stored in −80°C for RNA extraction and qRT-PCR and Northern blot analysis . Accuracy of injections and Caprin-2 knockdown were confirmed by eGFP and Caprin-2 mRNA expression analysis in each individual SON and PVN using qRT-PCR . Two separate groups of animals were used . In the first group , individual Caprin-2 shRNA-injected nuclei with 25% or higher Caprin-2 knockdown ( as compared to the average control , scrambled shRNA-injected PVNs or SONs ) were retrospectively chosen for gene expression analysis . In the second group , rats with more than 30% of Caprin-2 gene knockdown in both SON and PVN ( 5 out of the 10 animals injected ) were retrospectively chosen for analysis of physiological parameters . Urine osmolality was measured in 100 µl of 10×-diluted samples by freezing point depression using a Roebling micro-osmometer ( Camlab , UK ) . Urine sodium levels were determined in 2 ml of 10×-diluted samples using Cole–Parmer Sodium Combination Epoxy Body Electrode ( Vernon Hills , IL , USA ) and Jenway 3310 pH/mV meter ( UK ) . The ionic strength of standards and samples was adjusted with 4 M NH4Cl/4 M NH4OH added in 1:50 ratio . Sodium concentration in samples was calculated from the calibration curve representing potentials ( mV ) recorded for 0 . 1–100 mM NaCl standards . Blood plasma was isolated from trunk blood collected after decapitation into chilled EDTA-treated plastic tubes ( 5 µl 0 . 5 M EDTA pH 8 . 0/1 ml blood ) and centrifuged at 1600×g for 15 min at 4°C . 100 µl plasma samples were placed on ice and plasma osmolality was determined within 2 hr , as described before . The remaining plasma was aliquoted and stored in −80°C for AVP determination . AVP was extracted from 0 . 5 ml of plasma with acetone and petroleum ether and measured using arg8-Vasopressin EIA kit , according to the manufacturer protocol ( Enzo Life Sciences , Farmingdale , NY , USA ) . Tissue punches from the left and right SON or PVNs were cross-linked with 1% ( vol/vol ) formaldehyde ( Sigma–Aldrich ) in PBS , for 10 min at room temperature . Cross-linking was terminated by 5 min incubation with 0 . 125 M glycine ( pH 7 . 0 ) and samples were washed three times in PBS , using a centrifuge ( 5 min , 2000×g , 4°C ) . After homogenization in 100 μl of NT- RNA immunoprecipitation buffer ( 50 mM Tris , pH 7 . 4 , 150 mM NaCl , 1 mM MgCl2 , 0 . 5% Nonidet P40 , 1 mM EDTA pH 8 . 0 , 1 mM DTT ) , Complete protease inhibitor ( Roche ) , 200 U/ml RNase Out ( Invitrogen , USA ) samples were incubated for 10 min on ice and pre-cleared with 25 μl of Protein G-coated Dynabeads ( Life Technologies ) . Protein levels were determined using BCA method ( Thermo Fisher Scientific , Pierce , Waltham , MA , USA ) . First , we analyzed suitability of two different anti-Caprin-2 antibodies: rabbit ( kindly provided by Prof . Lin Li , Shanghai , China ) and goat ( Santa Cruz Biotechnology ) , against non-specific rabbit IgG ( Santa Cruz Biotechnology ) and goat IgG specific to an irrelevant antigen DKK ( Dickkopf-related protein 1 , Santa Cruz Biotechnology ) . Both antibodies recognising Caprin-2 gave similar results . The data presented here was derived using the antibody raised in goat from Santa Cruz Biotechnology . Tissue extracts containing 60 µg ( PVN samples ) or 30 µg ( SON samples ) of protein were incubated overnight on a rotating wheel at 4°C with either , goat anti-Caprin-2 antibody ( Santa Cruz Biotechnology ) or with non-specific goat IgG ( Santa Cruz Biotechnology ) at a concentration ratio 1:8 , in PBS buffer supplemented with RNase Out ( 200 U/ml ) and Complete Protease Inhibitor . Extracts equivalent to 20% of each sample were preserved in −80°C for input mRNA analysis . Next , the protein—antibody extracts were incubated with G-protein Dynabeads ( pre-washed with PBS and blocked with 10% BSA for 10 min at RT ) for 2 hr , at 4°C , on a rotating wheel . After 3 washes with PBS , the protein-G-adsorbed complexes were washed out with 50 µl of the crosslinking reversal buffer ( 50 mM Tris-HCl pH 7 . 0 , 5 mM EDTA , 10 mM DTT , 200 U/ml RNase Out ) at 70°C , for 45 min . All steps have been carried out in RNase-free environment , in a presence of RNase Out ( washing buffers—50 U/ml , reaction buffers—200 U/ml ) and Complete Protease Inhibitor ( Roche ) . Total RNA was isolated with 1 ml of Reagent ( Invitrogen , Life Technologies ) , cleaned with RNeasy MinElute Cleanup kit ( Qiagen ) and eluted with 14 µl of water , of which 8 µl was treated with DNase I ( Life Technologies ) and used for qRT-PCR analysis , as described above . RNA extracted from SONs and PVNs of rats injected with lentiviruses and from the in vitro experiment on HEK cells was subjected to Northern blot analysis using Ambion NorthernMax kit ( Life Technologies ) . Briefly , samples containing 300 ng of RNA were incubated for 15 min at 65°C with formaldehyde load dye containing ethidium bromide ( 10 µg/ml ) and separated ( 5 V/cm ) on 1% ( wt/vol ) agarose gel along with the BrightStar Biotinylated RNA Millenium marker ( Life Technologies ) . The gel was exposed under UV light and photographed , then downward-transferred overnight to the BrightStar-Plus membrane and crosslinked for 2 min using a standard UV transilluminator . Quality of the transfer was assessed by visualizing the gel and the membrane under the UV light . Membrane was placed in the hybridization bag ( Roche ) and subjected to prehybridization , blocking and overnight hybridization at 37°C ULTRAhyb buffer . We used 100 pM oligonucleotide probes double biotinylated at the 5′ and the 3′ ends ( Eurofins MWG Operon ) . The sequences of the anti-sense ( AS ) oligonucleotide probes used are as follows: rAVP-AS–5′ GTAGACCCGGGGCTTGGCAGAATCCACGGACTCTTGTGTCCCAGCCAG 3′ , rGAPDH-AS–5′ CCAGCCTTCTCCATGGTGGTGAAGACGCCAGTAGACTCCACGACA 3′ . After washing in low stringency solution ( 2 × SSC , 0 . 1% wt/vol SDS ) , the signal was developed using the chemiluminescent BrightStart BioDetect kit ( Ambion , Life Technologies , USA ) and captured on Amersham Hyperfilm ECL ( GE Healthcare , UK ) . At the end the probes were stripped by dipping membranes in boiling DEPC-treated water with 0 . 1% SDS until they reach RT , and re-probed for GAPDH . To determine if the observed changes in the size of AVP mRNA were associated with changes in the length of their poly ( A ) tails , in parallel we performed Northern blot analysis on samples subjected to poly ( A ) tail removal . 200 ng of RNA was incubated with oligo ( dT12–18 ) primers ( Life Technologies ) for 5 min at 85°C to denature RNA , followed by 10 min hybridization at 42°C and slow cooling ( 1°C/min ) to 32°C . Double-stranded poly ( A ) tails were then digested with 3 . 75 U of RNase H ( New England Biolabs ) at 37°C for 30 min . Samples were cleaned with 1 volume of phenol-chloroform , precipitated for 30 min at −20°C with 20 µg/ml of linear polyacrylamide ( Ambion , Life Technologies ) , 10% ( vol/vol ) of 5 M ammonium acetate and 2 vol of 100% ( vol/vol ) ethanol and centrifuged at 12 , 000×g , 4°C for 20 min . After cleaning with ethanol the RNA pellet was resuspended in 20 µl of formaldehyde load dye containing ethidium bromide ( 10 µg/ml , Sigma–Aldrich ) , incubated for 15 min at 65°C and separated in agarose gel alongside the non-digested samples as described before . RNase-free conditions were maintained throughout the whole protocol . Molecular sizes of the RNA bands were calculated based on the migration distances of the ladder bands in each individual gel . HEK 293T/17 cells ( ATCC , USA ) were plated into 6-well plates at a density of 300 , 000 cells/well in DMEM high glucose media ( Sigma–Aldrich ) supplemented with 10% ( vol/vol ) FBS , 2 mM L-glutamine and 1× NEAA ( Sigma–Aldrich ) . The following day media was changed and cells were co-transfected with the rat genomic AVP structural gene under the transcriptional control of the CMV promoter ( derived from pSP72-VP [Chooi et al . , 1994] ) in combination with either: ( a ) control pRRL . sin . cppt . CMV . eGFP . wpre vector ( ‘eGFP’ ) , ( b ) Caprin-2 overexpression vector pRRL . sin . cppt . CMV . rCaprin2 . ires . eGFP . wpre ( ‘Caprin-2’ ) , ( c ) Caprin-2 overexpression vector and control scrambled shRNA vector pRRL . sin . U6 . Scr_shRNA . cppt . cmv . GFP . wpre ( ‘Ctrl’ ) and ( d ) Caprin-2 overexpression vector and Caprin-2 shRNA vector pRRL . sin . U6 . Caprin-2_shRNA . cppt . CMV . GFP . wpre ( ‘Cap2 shRNA’ ) . Transfection with 2 µg of each plasmid and Lipofectamine LTX reagent ( Life Technologies ) was performed according to the manufacturer protocol . 48 hr after transfection cells were lysed with TRIzol reagent ( Life Technologies ) and subjected to RNA extraction , as described above . HEK293T/17 cells were cultured in 6 well-tissue culture plate for 24 hr ( 800 , 000 cells/wells ) . Transfections were performed in triplicate using standard calcium phosphate transfection method . Cells were either mock transfected , or were tranfected with control pRRL . sin . cppt . CMV . eGFP . wpre vector ( ‘eGFP’ ) or Caprin-2 overexpression vector pRRL . sin . cppt . CMV . rCaprin2 . ires . eGFP . wpre ( ‘Caprin-2’ ) . Both vectors express eGFP . After 8 hr of transfection the culture medium was replaced with fresh media . At 2 days after transfection , total protein extraction was performed using RIPA buffer ( 1% vol/vol Nonidet P-40 , 0 . 5% wt/vol sodium deoxycholate , 0 . 1% wt/vol SDS in phosphate buffer saline ) containing protease inhibitors ( P8340; Sigma ) . The lysate was incubated on ice for 30 min , followed by centrifugation at 10 , 000×g for 10 min . Supernatant was collected and stored at −80°C . Protein concentration was determined using Bradford assay ( Bio-Rad , Hercules , CA , USA ) . For immunoblot , proteins were separated by SDS-PAGE , then transferred to PVDF membranes ( Millipore , Billerica , MA , USA ) . The membranes were blocked with 5% ( vol/vol ) ECL prime blocking reagent ( Amersham , UK ) /0 . 1% vol/vol tween20/TBS for 2 hr at room temperature and incubated with primary antibody diluted in 2% ( vol/vol ) ECL prime blocking reagent/0 . 1% vol/vol tween20/TBS at room temperature for at least 2 hr at room temperature or overnight at 4°C . Incubations with secondary antibodies conjugated with HRP were performed at room temperature for 1 hr . The signal was visualised using Westar EtaC HRP Detection Substrate ( Cyanagen , Italy ) . Primary antibodies used were: goat polyclonal anti-Caprin2 ( 1:500 , sc-107473; Santa Cruz Biotechnology ) , rabbit polyclonal anti-GFP ( 1:10 , 000 , ab290; Abcam ) , mouse monoclonal anti-GAPDH ( 1:10 , 000 , sc32233; Santa Cruz Biotechnology ) . Statistical analysis was performed using Graphpad Prism ( Graphpad Software , La Jolla , CA , USA ) . Statistical differences between two groups in qRT-PCR , immunohistochemistry , Northern blot , RNA immunoprecipitation experiments and in plasma osmolality and AVP measurements were evaluated using unpaired Student's t-test . Results of qRT-PCR and immunofluorescence with three experimental groups were evaluated using one-way ANOVA with respectively , Holm-Sidak's and Bonferroni post-hoc tests . Two-way ANOVA with uncorrected Fisher's LSD test was used to determine the differences between more than two groups ( i . e . , urine output , osmolality and sodium concentration , fluid intake ) . p < 0 . 05 was considered significant . All data are expressed as the mean ± s . e . m .	Cells are only able to work properly if they maintain a more or less constant balance of water and salts . In mammals , a hormone called arginine vasopressin regulates water and salt levels in the whole body . This hormone is made by cells in a region of the brain called the hypothalamus , and is then transported to the pituitary gland . When the level of water relative to the level of salts in the blood starts to drop ( i . e . , during dehydration ) , arginine vasopressin is released into the blood and travels to the kidneys where it acts as a signal to retain more water in the body . However , if water levels continue to remain low , the stores of arginine vasopressin in the pituitary gland may run out and so more protein needs to be made in the hypothalamus . Like all proteins , arginine vasopressin is made by first copying a template encoded in a particular gene into a molecule called messenger ribonucleic acid ( mRNA ) . During dehydration , the cells in the hypothalamus produce more of these corresponding mRNA molecules . Also , the mRNAs are slightly larger than normal because they have longer ‘polyA tails’ ( structures added to the ends of all newly-made mRNAs ) . However , it was not clear how or why this happens . Here , Konopacka et al . studied the production of arginine vasopressin in rats . The experiments show that a protein called Caprin-2 accumulates in hypothalamic neurons when rats are dehydrated . Furthermore , Caprin-2 is able to directly bind to the mRNA that encodes arginine vasopressin and is responsible for increasing the length of the polyA tail . To test whether this interaction is important for regulating the balance of water and salts , Konopacka et al . decreased the levels of Caprin-2 protein in the hypothalamus of live rats . When these rats became dehydrated , they had lower levels of the arginine vasopressin mRNA and these mRNAs had shorter polyA tails . Furthermore , the rats drank less water and urinated less than normal rats . Further experiments show that Caprin-2 helps to stabilize the structure of these mRNAs so that they accumulate in cells . Together , Konopacka et al . 's findings show that Caprin-2 regulates the production of arginine vasopressin by interacting with and modifying its corresponding mRNA in the rat hypothalamus . The next challenge is to find out which other mRNAs in the hypothalamus are regulated by Caprin-2 , and to determine their roles in the body .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "neuroscience" ]	2015	RNA binding protein Caprin-2 is a pivotal regulator of the central osmotic defense response
Witnessing another person’s suffering elicits vicarious brain activity in areas that are active when we ourselves are in pain . Whether this activity influences prosocial behavior remains the subject of debate . Here participants witnessed a confederate express pain through a reaction of the swatted hand or through a facial expression , and could decide to reduce that pain by donating money . Participants donate more money on trials in which the confederate expressed more pain . Electroencephalography shows that activity of the somatosensory cortex I ( SI ) hand region explains variance in donation . Transcranial magnetic stimulation ( TMS ) shows that altering this activity interferes with the pain–donation coupling only when pain is expressed by the hand . High-definition transcranial direct current stimulation ( HD-tDCS ) shows that altering SI activity also interferes with pain perception . These experiments show that vicarious somatosensory activations contribute to prosocial decision-making and suggest that they do so by helping to transform observed reactions of affected body-parts into accurate perceptions of pain that are necessary for decision-making . Prosocial behavior — actions intended to benefit others despite costs to self ( Batson , 1981 ) — is important in social animals but poorly understood . The role of empathy in motivating prosocial behavior is intuitive but also intensely debated ( Bloom , 2017; Zaki , 2017 ) . Some researchers have shown that people are more likely to engage in prosocial behaviour when they feel empathy for the person in distress ( Batson , 1981 ) and that self-reported emphatic concern is related to prosocial behaviour ( FeldmanHall et al . , 2015 ) . Others have shown that empathy is a poor predictor of prosociality ( Vachon et al . , 2014; Jordan et al . , 2016 ) , with prosocial decisions often driven by other motives ( e . g . status; Bloom , 2017 ) . Unfortunately , experiments that specifically manipulate brain activity in empathy-related regions and that measure prosociality are missing , limiting our neuroscientific understanding of whether and how empathy is mechanistically linked to prosociality ( Keysers and Gazzola , 2017; Zaki et al . , 2016 ) . Here , we use a combination of electroencephalography ( EEG ) , transcranial magnetic stimulation ( TMS ) and transcranial direct current stimulation ( tDCS ) to explore whether altering somatosensory activity that vicariously represents the pain of others would alter the decision to donate money to alleviate that pain , and whether alterations in pain perception mediate this effect . Witnessing somebody in pain activates two networks , depending on the nature of the stimulus ( Keysers et al . , 2010; Lamm et al . , 2011 ) . If the pain of the other is deduced from abstract symbols or facial expressions alone , a network involving the anterior insula ( AI ) and the anterior cingulate cortex ( ACC ) is activated . The AI and ACC activity correlates with personal distress ( Singer et al . , 2004 ) , perceived unpleasantness ( Rainville et al . , 1997 ) and perceived pain intensity ( Lamm et al . , 2011 ) , and is therefore thought to code for the unpleasantness of the pain of the other ( Lamm et al . , 2011 ) . This network is often referred to as the affective path . If the injured body part is visible and in the focus of attention , the somatosensory cortices ( SI and SII ) are also recruited ( Bufalari et al . , 2007; Cheng et al . , 2008; Keysers and Gazzola , 2009; Lamm et al . , 2011; Nummenmaa et al . , 2012; Morrison et al . , 2013; Ashar et al . , 2017; Christov-Moore and Iacoboni , 2016 ) and consequently included in the network of regions participating in human empathy ( Keysers et al . , 2010; de Waal and Preston , 2017 ) . The affective and somatosensory networks are also active when experiencing pain ( Melzack , 2001; Iannetti and Mouraux , 2010; Lockwood et al . , 2013 ) . Because of that , many interpret vicarious activation while witnessing the pain of others as the neural correlate of emotional contagion , feeling vicariously what we see someone else experience , a core component of empathy ( Koban et al . , 2013; Corradi-Dell'Acqua et al . , 2011; Jackson et al . , 2006; Lamm et al . , 2011; Cui et al . , 2015; Singer et al . , 2004 ) . In this perspective , the emotional states of others are understood through personal , embodied representations that allow empathy and accuracy in perceiving other emotions to increase on the basis of the observer’s past experiences ( Preston and de Waal , 2002; de Waal and Preston , 2017 ) . Pharmacological studies provide causal evidence for the involvement of our own experience of pain , via the insula and anterior cingulate cortex , in perceiving the pain of other individuals ( Rütgen et al . , 2015 , 2017 ) . This evidence pin-points to the opioidergic circuit , which might code representations of aversive outcomes that are derived through both direct and indirect experiences ( Haaker et al . , 2017 ) . Much like the original proposal of Adam Smith ( Smith , 1759 ) , the pain vicariously felt because of the recruitment of these pain-related regions while viewing the pain of others is then thought to motivate prosocial behavior , which then simply serves to reduce the vicariously felt pain ( Batson , 1981; Hein et al . , 2010 , 2016; Ma et al . , 2011; Tomova et al . , 2016 ) . Testing the causal contribution of mapping the pain of others onto our own pain representations in social decision-making has remained poorly explored in neuroscience because most scientists focus on the affective network ( AI , ACC ) , which lies too deep in the brain to allow its selective targeting with traditional non-invasive neuro-manipulation tools ( Keysers and Gazzola , 2017 ) . Here , we leverage the less investigated hand representation of SI , which is superficial and reachable with TMS , to address two questions: does activity in the somatosensory cortex ( measured using EEG ) explain prosocial behavior on a trial-by-trial basis and does disturbing ( with TMS ) this activity alter decision-making ? We explore prosocial behavior using a costly helping paradigm ( Figure 1A ) , in which participants make a moral decision between two conflicting motives: maximizing their financial gains and minimizing the pain of another ( see FeldmanHall et al . [2015] ) for use of a similar tradeoff scenario ) . We then measure and alter brain activity in the hand region of SI to explore the impact of this activity on the decision-making . If somatosensory activity contributes to our decision to help , does it do so by being necessary for an accurate perception of how much pain other people experience ? Whether somatosensory activation levels reflect the intensity of the pain experienced by others remains unclear ( Lamm et al . , 2011; Morrison et al . , 2013 ) , and we therefore use data from a third experiment in which high-definition tDCS ( HD-tDCS ) is used to alter SI activity to measure whether an SI perturbation also alters the accuracy of pain perception . As mentioned above , somatosensory cortices are mainly involved when the injured body parts are visible . In addition , ventral regions of SI have been reported to be involved in emotional facial perception ( e . g Adolphs et al . , 2000; van der Gaag et al . , 2007; Preston and de Waal , 2002; but see Rütgen et al . , 2015 for absence of SI activity in the perception of facial expression of pain ) . Given its somatotopic organization , SI activity could therefore be involved in pain perception in two ways . If we only see the painful reaction of a hand , the hand region of SI could reflect the intensity of the observed reaction by simulating the movements of the hand and/or the somatosensory consequences of the harm ( Keysers et al . , 2010 ) . If we know that the painful stimulation originates from the hand , but the intensity of the pain has to be inferred from the facial expressions , somatosensory activation in the hand region could still reflect the intensity of the stimulation , through an indirect route in which information derived from the facial expression is referred back onto the hand region through a process akin to somatosensory imagery , or activation could instead fail to reflect pain intensity . Either way , we could expect the more ventral representation of the face in SI to be involved in representing the intensity of the facial expression ( Adolphs et al . , 2000; van der Gaag et al . , 2007 ) , but less involved in representing the intensity of a painful hand movement when the face is not visible . To shed further light on the properties of the hand region of SI in the decision-making task presented in this study , we therefore designed two types of input stimuli that probe the above-mentioned scenarios , both showing different intensity of pain ( Videos 1 , 2 , 3 and 4 ) in order to look at quantitative relations between brain activity , perception and behavior ( Wager et al . , 2013 ) . Specifically , participants witnessed a confederate receiving a noxious stimulation of randomly selected intensity ( InputMovie ) delivered as ( a ) a swat with a belt on the right hand , with only the hand reaction visible ( Hand condition , Videos 1 and 2 ) or ( b ) an electroshock on the right hand with a visible facial expression and no hand movement ( Face condition , Videos 3 and 4 ) . In both kinds of videos , the confederate’s right hand receiving the painful stimulation was clearly visible on the screen , but in the case of the belt ( Hand condition ) the reaction of the hand itself was the only cue for the participants to deduce the painfulness of the stimulation . By contrast , when receiving the electroshock ( Face condition ) , the hand did not show any reaction and the painfulness was deduced only by the confederate’s facial expression . At each trial , participants then received an endowment of 6€ and could reduce the intensity of the next noxious stimulation ( OutputMovie ) by giving up some of that money , knowing that the remainder would be part of their compensation ( Figure 1A and Videos 1–6 ) . First , we investigated whether activation of the hand region of the left SI , as measured with EEG , explains prosocial behavior . The SI hand region was identified in an independent pool of participants by correlating fMRI BOLD responses within SI with subjective experience of pain elicited by electrical stimulations on the participant’s right hand . We hypothesized that activation of the hand region of SI would correlate with the donation given by the participants in the Hand condition , when the intensity of the stimulation had to be deduced from the hand movement . In the Face condition , we predicted that activity more ventrally in SI , where facial expressions are represented , would correlate with the donation . We were able to make this prediction because the relevance of facial mimicry has been highlighted in many studies ( Oberman et al . , 2007; Hess and Fischer , 2013; Fischer and Hess , 2017; Wood et al . , 2016b ) , because we know that the ventral somatosensory cortex causally contributes to emotion perception from facial expression ( Adolphs et al . , 2000; Paracampo et al . , 2017 ) and because we know that the emotions from visually presented facial expressions requires ventral somatosensory-related cortices ( Adolphs et al . , 2000 ) . As mentioned above , for the hand region of SI during the Face condition , we had less defined predictions: the presence or absence of correlation of SI hand region activity with the donation while perceiving facial expressions will inform whether facially deduced pain intensity is re-represented in the SI locations reflecting the inferred origin of that pain . In a second experiment , we then perturbed the SI activity of the hand region with repetitive TMS ( rTMS ) to test whether disturbing SI vicarious activity altered prosocial behavior . We expected TMS over the hand-region of SI to disrupt the accuracy with which participants can transform the observed kinematics of the belt and hand into an accurate perception of how painful this particular stimulation was for the other . We had this expectation because disrupting SI activity using TMS or neurological lesions has been shown to alter the accuracy with which participants perceive some emotions ( Adolphs et al . , 2000; Paracampo et al . , 2017 ) and hand actions ( Valchev et al . , 2017 ) , and because , in the nociceptive literature , SI has been associated more with perceptual than with motivational processes ( Keysers et al . , 2010; Lee and Tracey , 2010 ) . We thus expect decision-making to become noisier , and less attuned to the level of pain experienced by the other on a trial-by-trial basis , particularly when the reaction of the hand is the only source of information on which the decision is based ( Hand condition ) . This effect would be weaker or absent when information is derived from the Face , where alternative sources of information are available . Finally , we used data from a third experiment to explore whether a disruption of the perception of pain intensity does indeed mediate how disrupting SI activity alters decision-making . Brain activity in SI was altered using high-definition tDCS while participants had to rate how much pain the person in the Hand and Face movies experienced on a trial-by-trial basis . Because the specific montage used in this experiment was expected to facilitate brain activity under the anode placed over the SI hand region and to inhibit brain activity under the return cathodes , one of which was placed over the face region of SI , we expected the accuracy of the ratings to increase in the Hand stimuli and to decrease in the Face stimuli . Participants ( Table 1 ) donated on average the same amount in the Face and Hand conditions ( Face: M = 2 . 14€ , SD = 1 . 2; Hand: M = 2 . 16€ , SD = 1 . 2; dependent sample t-test t ( 28 ) =–0 . 2 , p=0 . 8 ) , but comparing the standard deviation in donation within each participant showed more variability in donation for the Face condition ( Face: MSD = 1 . 47 , SDSD = 0 . 44; Hand: MSD = 1 . 22 , SDSD = 0 . 40; dependent sample t-test t ( 28 ) =4 , p=0 . 0004 ) . To avoid this confound in further analysis , we Z-transformed the donation of each participant separately for the two conditions . To assess whether participants’ donation was driven by the intensity of the reaction shown in the InputMovie , for each participant , we performed a robust linear regression ( Holland and Welsch , 1977 ) between the intensity attributed to the movies by an independent pool of participants and the Z-donation . The analysis confirmed that participants’ Z-donation closely followed the pain intensity shown in the InputMovies ( Figure 2A ) . In the Face condition , all participants had regression slopes that were positive and significantly different from zero ( Face slope: M = 0 . 48 , SD = 0 . 6 , average t value for Face slope = 21 . 9 , SD = 17 , all p<0 . 05; group one-sample t-test on Face slopes t ( 28 ) = 37 . 9 p=0 . 0006E-21 ) . In the Hand condition , one participant had a negative slope but the regression was not significant ( t = −0 . 17 , p=0 . 8 ) , a second participant had a positive but not significant slope ( t = 1 . 89 , p=0 . 06 ) , whereas the remaining participants all showed positive and significant slopes ( all p<0 . 05 ) . We considered the former two as normal variation along the population spectrum and kept them in the analysis ( Hand slope: M = 0 . 45 , SD = 1 . 4; average t value for Hand slope = 7 . 4 , SD = 5 . 4; group single sample t-test on Hand slopes t ( 28 ) = 16 . 3 p=0 . 0001E–11 ) . Importantly , the average slope did not differ between the two conditions ( paired t-test t ( 28 ) =–1 p=0 . 3 ) . To interrogate the electrical activity originating from the primary somatosensory cortex , we used a linear constrained minimum variance beam-forming approach ( Van Veen et al . , 1997 ) , in which spatial filters were designed to isolate brain electrical activity from the specified locations of interest . To identify regions of SI that reflect perceived pain intensity while participants experience pain on their own bodies , we performed an independent fMRI experiment in which participants received electrical stimulation at different intensities on their right hand and reported how painful each stimulation was . We then identified voxels in the left SI where the BOLD signal correlated with reported painfulness ( see 'Supplementary information , Pain localizer experiment' ) , and used the following clusters as ROIs for the EEG beam-forming analysis: a dorsal ( dSI-L; peak at MNI ( x , y , z ) = −30 , –36 , 62 ) and a ventral ( vSI-L; peak at MNI ( x , y , z , ) = −54 , –19 , 32 ) somatosensory cluster ( Figure 2B; Maldjian et al . , 1999; Mancini et al . , 2012 ) . The ventral cluster has a dorso-ventral extent similar to the face representation in SI but seems unusually posterior ( Figure 2B; van der Gaag et al . , 2007 ) . This suggests that the ventral cluster could originate from the posterior parietal region PF or could represent the facial reaction to the pain in SI . We focused on the left hemisphere , because electrical stimulation was always delivered to the right hand of both the confederate shown in the movies , and of the participants in the pain localizer study . For each participant and ROI , beam-forming returned activity time-courses along three standard dipole directions . Given the low spatial resolution of EEG , the mixed orientation of cells in the somatosensory cortex ( encompassing gyri and sulci ) , and the fact that we use videos , for which the dipole capturing most of the variance could change over time , we included all three dipole directions in multivariate analyses . To assess whether EEG activity explained variance in participants’ donations , we used a random effect , summary statistics approach routinely used in fMRI analysis ( Holmes et al . , 1998 ) . First , at the single subject level , we modelled the relationship between SI ROI activity and donation by calculating a robust regression between brain activity at a given time-point and the Z-donation for all the trials of that participant ( Figure 3A , left ) . Repeating this analysis for each time point generated a time-course of the slope separately for each participant , condition and dipole ( Figure 3A , right ) . Second , at the group level , we analyzed the group distribution of these slopes: if EEG activity in an ROI does not carry systematic information about the donation , the slopes would be randomly distributed around zero . To test this null hypothesis , we used the Hotelling's t-squared statistic , a multivariate generalization of the Student's t-test that combines evidence from all three dipoles and all participants in a given condition and time-point . We controlled for multiple time point testing using a cluster-based randomization test implemented in Field-Trip ( Oostenveld et al . , 2011 ) , which compares the sum of the Hotelling statistics within the clusters in the real data against those in clusters obtained after switching the sign of the entire slope-time course of randomly selected participants . We repeated the procedure for each condition and ROI and accounted for those additional comparisons using a Bonferroni procedure . Only results surviving those corrections are reported in yellow in Figure 3B . Results show that variation in activity in dSI-L explained variation of Z-donation in the Hand condition in the time windows between 420 and 452 ms ( cluster-statistic = 142 . 2 , p=0 . 002 ) and between 458 and 476 ms ( cluster-statistic = 91 . 4 ms , p=0 . 006 ) after the belt hits the confederate’s hand . Activity in dSI-L also related to Z-donation in Face condition but later in time , between 516 and 602 ms ( cluster-statistic = 602 . 5 ms , p=0 . 0009 ) , between 608 and 754 ms ( cluster-statistic = 333 . 8 ms , p=0 . 0009 ) , and also between 808 and 880 ms ( cluster-statistic = 330 ms , p=0 . 0009 ) after the noxious stimulation was delivered . vSI-L brain activity explained Z-donation but only in the Face condition in the time window between 402 and 430 ms ( cluster-statistic = 102 ms , p=0 . 005; Figure 3B ) . To visualize this effect , we categorized the trials into low and high donation categories ( median split per participant ) and calculated grand-averages voltage time courses during the InputMovie ( Figure 3C ) for each dipole . The grand averages along Y and Z present a negative deflection after the InputMovie onset independently of condition and donation , which probably reflects general attentional processes . For the Hand condition , the dipoles further present a positive peak some milliseconds after the slap , which is sustained along Z and transient along X and Y . No clear peak is recognizable after the shock ( time 0 ) during the Face condition . Differences between low and high donation responses can be observed , in particular along Y , for all ROIs and conditions ( Supplementary file 1 and 2 ) . There is evidence for a bilateral receptive field in the Brodmann 1 and 2 sub-regions of SI ( Iwamura et al . , 2002 ) , and for the involvement of the right hemisphere in the perception of emotion ( not including pain ) from facial expressions ( Adolphs et al . , 2000 ) ; bilateral activation reported by Ashar et al . , 2017; Lamm et al . , 2011; Cui et al . , 2015; Carr et al . , 2003 ) , and from hand movements ( Christov-Moore and Iacoboni , 2016 ) . Figure 3D shows the signal originating from mirroring our left ROIs onto the right hemisphere , and in yellow , the time points significantly explaining the donation . For the hand region of SI ( d-SI ) , results for the two hemispheres are very similar , suggesting a lack of clear hemispheric specificity ( Figure 3D and 'Supplementary Information' ) . For the more ventral , putative face region of SI ( v-SI ) , responses appear stronger on the right hemisphere , in line with previous findings ( Adolphs et al . , 2000 ) . Our EEG findings therefore suggest that while witnessing the pain of another person , the magnitude of brain activity in the hand region of SI ( d-SI ) could inform decision-making . To examine the causal contribution of this region to decision-making , in a second experiment , we used TMS to disturb the activity of the SI hand region . We targeted the left hemisphere because it is contralateral to the hand that is stimulated in the confederate . A separate group of participants ( Table 1 ) performed the costly helping paradigm for the Face and Hand conditions under both active and sham rTMS over the left SI hand region . The within-subject ANOVA with two factors , Condition ( Face and Hand ) and TMS ( Active and Sham ) , on the standard deviation calculated across the donation of each participant reveals that participants varied their donations from trial-to-trial more in the Face than in the Hand condition ( main effect of condition F ( 1 , 14 ) = 98 , p=0 . 001E–4 ) , but there was no variance difference between the two TMS protocols ( main effect of TMS F ( 1 , 14 ) = 0 . 2 , p=0 . 65 , interaction between Condition and TMS F ( 1 , 14 ) = 0 . 2 , p=0 . 7 ) . Therefore , the same Z-transformation of the donation was used as in Experiment 1 , standardizing separately all of the donations of the Hand and Face conditions of each participant ( but without separating Active and Sham to preserve TMS effects ) . TMS had no effect on the average Z-donation in either task , as indicated by the repeated ANOVA with factor Condition ( Face and Hand ) and TMS ( Active and Sham ) : TMS F ( 1 , 14 ) = 2 . 3 , p=0 . 15 , TMS x Condition F ( 1 , 14 ) = 0 . 5 , p=0 . 56 . To test whether TMS interferes with the relationship between the intensity of the movie and the donation , for each participant we calculated the slope between the intensity of the movie and the Z-donation by means of a robust regression , separately for the Condition ( Face and Hand ) and the TMS protocol ( Sham and Active ) . We then performed a two-factor repeated ANOVA on these slopes , using the above-mentioned two-level summary statistic approach . The analysis showed that the slopes for the Face condition are steeper than those for the Hand ( Face: M = 0 . 56 , SD = 0 . 06; Hand: M = 0 . 49 , SD = 0 . 14; Main effect of Condition: F ( 1 , 14 ) = 7 , p=0 . 02 ) . TMS did not have a general effect common to the two conditions ( Main effect of TMS: F ( 1 , 14 ) = 1 . 9 , p=0 . 19 ) . Interestingly , the ANOVA showed a significant interaction effect ( F ( 1 , 14 ) = 4 . 8 , p=0 . 04 , partial η²=0 . 25; Figure 2C ) with a larger TMS effect in the Hand compared to that in the Face condition . A Newman-Keuls post-hoc test indicated that active rTMS on SI significantly flattened the relationship between intensity and donation only for the Hand condition ( Hand Sham M = 0 . 54 , DS = 0 . 1; Hand TMS M = 0 . 45 , DS = 0 . 2; p=0 . 02 ) . There is no evidence for such an effect in the Face condition ( Face Sham M = 0 . 55 , SD = 0 . 06; Face TMS M = 0 . 56 , SD = 0 . 06; p=0 . 6 ) . To test whether the lack of effect in the Face condition was due to limited statistical power or whether it provides evidence for the null hypothesis ( H0 ) of no ( sizable ) effect of TMS , we used Bayesian statistics . We calculated an index of TMS effect for each participant in the Face condition as follows: slope in the Active session – slope in the Sham session . We then performed a Bayesian one sample t-test using JASP with default priors ( https://jasp-stats . org ) that showed that the null hypothesis is 5 . 7 times more likely than the alternative H1 ( Bayes factor p[H0:index ≥0\|data]/p[H1:index <0\|data]=5 . 7 ) , providing positive evidence for the absence of a sizable effect in the Face condition ( Kass and Raftery , 1995 ) . The effect on the Hand condition did not correlate with any of the TMS side effects perceived by the participant while performing the experiment ( measured by questionnaire , all p>0 . 05 ) . This suggests that TMS on the SI hand representation interferes with the process that normally couples a person’s donation to the needs of the other person ( i . e . the observed pain intensity ) when this need is perceived through the movements of the affected body part ( Hand condition ) , but this is less the case when the need is perceived through facial expressions . To test whether the impact of TMS is mediated by an effect on pain perception , we used data from a third experiment . In this experiment , participants ( Table 1 ) had to rate how much pain they perceived while watching the Face and Hand videos under the effect of tDCS centered over left SI . The high density 4 × 1 electrodes tDCS montage that we used would be expected to have a facilitatory effect on the hand region of SI under the anode ( Figure 2D ) , and a weaker inhibitory effect on ventral SI , including the face representation , under one of the four return cathodes . To control for unspecific effects on intensity rating processes unrelated to pain in this experiment , we introduced a new type of video in which participants needed to rate color saturation intensity ( Color condition , Figure 1B ) . A 2 Stimulation ( tDCS and Sham ) x 3 Condition ( Face , Hand and Color ) repeated measures ANOVA on the standard deviation calculated for the participants’ rating reveals that participants did not use the rating scale differently in the three conditions ( main effect of condition F ( 1 , 24 ) = 0 . 4 , p=0 . 5 ) , nor between the two tDCS sessions ( main effect of Stimulation F ( 1 , 48 ) = 2 , p=0 . 1 , interaction between Condition and Stimulation F ( 1 , 48 ) = 0 . 6 , p=0 . 5 ) . To be consistent between our studies , we applied Z-transformation of the ratings separately for the three conditions but while pooling the two tDCS conditions . A Stimulation ( tDCS and Sham ) x Condition ( Face , Hand and Color ) ANOVA on the average rating revealed that average rating remained stable across conditions ( Stimulation F ( 1 , 24 ) = 0 . 5 , p=0 . 5; Condition F ( 1 , 48 ) = 0 . 8 , p=0 . 5; Interaction F ( 1 , 48 ) = 0 . 3 , p=0 . 7 ) . For each condition and session , we then correlated the intensity assigned to the movies during the validation process and the Z-rating given by the participant . The correlation values were normalized using the Fisher z-transformation . We then performed a two-factor repeated ANOVA on the correlation coefficients obtained . The analysis showed that the correlation coefficients differed between conditions ( Face: M = 1 . 3 , SD = 0 . 2; Hand: M = 1 , SD = 0 . 2; Color: M = 1 . 2 , SD = 0 . 3; Main effect of Condition: F ( 2 , 48 ) = 17 , p<0 . 002E–3 ) . A post-Hoc Newman-Keuls test revealed that the Hand condition had a lower correlation coefficient than both the Face and Color conditions , while the latter did not differ from each other . tDCS had no main effect on the correlation coefficients ( main effect of tDCS F ( 2 , 48 ) = 0 . 5 , p=0 . 5 ) . Interestingly , tDCS had a different effect depending on the conditions ( Interaction: F ( 2 , 48 ) = 3 . 4 , p=0 . 04 , η²=0 . 12 ) . Planned paired t-test comparison between sham and tDCS session for each condition indicated that the tDCS on SI significantly improved the relationship between intensity and rating for the Hand condition ( Hand Sham: M = 1 , DS = 0 . 2; Hand tDCS: M = 1 . 1 , DS = 0 . 2 , t ( 24 ) = –2 , p=0 . 04 ) , while it showed a trend for reduction in the Face condition ( Face Sham: M = 1 . 3 , DS = 0 . 2; Face tDCS: M = 1 . 2 , DS = 0 . 2; t ( 24 ) = 1 . 8 p=0 . 07 ) , and no appreciable change in the Color condition ( Color Sham: M = 1 . 3 , DS = 0 . 2; Color tDCS: M = 1 . 2 , DS = 0 . 3; t ( 24 ) = 1 , p=0 . 3 ) . Again , to test whether the lack of effect in the Color condition supports the null hypothesis , we calculated an index of the stimulation effect by subtracting the z-transformed correlation score calculated in sham from the one calculated in the real tDCS session . As Hand and Face conditions showed opposite effects , we performed a 2-tailed one sample Bayesian t-test on the Color condition . The Bayes factor indicated that the H0 was 3 . 1 times more likely than the H1 , confirming that the Color condition does not change after HD-tDCS . The Hand and Face effects did not correlate with any of the tDCS side effects perceived by participants while performing the experiment ( all p>0 . 05 ) . We used a helping task in which brain activity could be related to prosocial behavior on a trial-by-trial basis , and concentrated on measuring and altering activity in the hand region of SI , to shed light on the contribution of this region to prosocial decision-making . Specifically , we localized regions in the left SI that encode the intensity of pain experienced by a group of participants using fMRI . This evidenced two clusters in the left SI: a dorsal cluster corresponding to the hand representation of SI and a ventral cluster that had a dorso-ventral extent similar to the face representation of SI . These ROIs served as the ROIs for our EEG experiment , which showed that the magnitude of brain activity originating from the dorsal cluster had a significant relationship with helping . This was true whether victims expressed their reaction through their afflicted body part or through their face . Figure 3 shows how the timing of that activity followed the timing of the information in the movies: the hand immediately retracts at the moment the belt hits the hand , and SI activity had a sharp and sudden peak in explanatory power , whereas the facial expression develops more slowly after the shock is delivered , and SI activity showed a more progressive and sustained explanatory power . Figure 3 also shows that the ventral and dorsal sector of our functionally localized SI nociceptive representation behave differently in our task , with the dorsal ( hand ) ROI explaining donation for either source of information ( Face or Hand ) and the ventral ( face ) ROI being explanatory only for facial expressions . The choice to interrogate the signal during pain observation within regions coding the intensity of pain during self-pain experience was , as mentioned in the introduction , dictated by the theoretical framework of emotional contagion and vicarious activity . In this framework , the activation of cells involved in experiencing pain during the observation of other people’s pain would help someone to ‘feel’ what another person is experiencing by inducing a psychological state similar to that to which these neurons contribute during the experience of pain . FMRI overlaps between the experience and observation of pain have been widely documented , and taken as support for such a framework ( Keysers et al . , 2010; Lamm et al . , 2011 ) . This notion was recently in the focus of debate because of mixed results from multi-voxels pattern analyses ( Zaki et al . , 2016; Corradi-Dell’'Acqua et al . , 2016; Krishnan et al . , 2016 ) . The logic of these analyses is to identify a pattern across voxels that discriminates different intensities of experienced pain from fMRI signals . If pain observation triggers a neuronal representation of felt pain , the logic goes , the same pattern should discriminate different intensities of observed pain — so-called above-chance cross-modal classification . Some scientists find this to be true ( Corradi-Dell’'Acqua et al . , 2016 ) , others not ( Krishnan et al . , 2016 ) . It is important to realize , however , that a region may have neurons involved in experiencing and observing pain , as our framework predicts , without significant cross-modal classification of fMRI signals ( Zaki et al . , 2016 ) : decades of work on mirror neurons for actions show that only 10% of neurons involved in performing an action become recruited while observing that action ( Gallese et al . , 1996; Keysers et al . , 2003; Mukamel et al . , 2010 ) . If the same is true for pain , only 10% of neurons involved in pain experience may also be recruited during pain observation . This low percentage means that a pattern classifier trained on pain experience would be dominated by the signals originating from the 90% of neurons that are not involved in pain observation , and would then fail to interrogate the 10% involved in observation reliably when tested with pain-observation data . Until we have systematic single-cell data during the experience and observation of pain ( Hutchison et al . , 1999 ) , it will not become clear whether neurons represent felt and observed pain reliably ( Zaki et al . , 2016 ) . Accordingly , our finding that signals from a region of SI involved in actual pain experience explains variance in helping in our paradigm is compatible with the notion that this signal originates from neurons also involved in pain experience . Alternatively , the signal could originate from neurons not involved in pain experience that are simply spatially intertwined with those involved in pain experience ( Zaki et al . , 2016; Keysers and Gazzola , 2009 ) . In this manuscript , when we speak of vicarious pain activations , we therefore mean activations of regions involved in pain experience during the observation of the pain of others , without having the means to assess whether this activation originates from the neurons that are involved in actual pain experience . When pain is expressed by the reaction of the hand , the dorsal SI hand region activation correlates with decision-making ( EEG ) . In addition , altering this activity changes ( TMS ) decision-making , suggesting that the SI hand region activation feeds into the decision-making process . When pain is expressed by the face , the SI hand region activation correlates with decision-making ( EEG ) but altering this activity influences decision-making less . This latter finding is compatible with three interpretations . ( i ) Activity in the hand-region of SI is an epiphenomenon , representing an imagination of what the painful stimulation would have felt on the hand ( Fairhurst et al . , 2012 ) , which is not used in decision-making . ( ii ) The SI hand region activity is used for decision-making in the sham condition , but can be substituted by alternative sources of information derived from the face elsewhere in the brain in the active TMS condition . Both of these interpretations are reminiscent of the notion that pain information takes different paths depending on the stimulus it is derived from ( Keysers et al . , 2010; Lamm et al . , 2011 ) . ( iii ) Information in the SI hand region has a higher signal to noise ratio in the Face condition than in the Hand condition , making it less susceptible to TMS interference . Our results show that trial-by-trial amplitude of the EEG activity from ventral SI significantly explains changes in donation in the Face condition only . This effect could be driven by a covert internal simulation of the other’s facial expression or by overt facial mimicry . Interfering with facial mimicry has been shown to impair visual recognition of expressions ( Oberman et al . , 2007; Wood et al . , 2016a ) and interfering with activity in ventral somatosensory cortex alters the recognition of emotion from faces ( Adolphs et al . , 2000; Paracampo et al . , 2017 ) . Future research should neuro-modulate brain activity in ventral SI in addition to the hand representation we targeted here , while measuring the willingness to help in order to further investigate the dissociation suggested by our results . The emergence of focused ultrasounds as a focal neuro-modulation method ( Mueller et al . , 2014; Lee et al . , 2015 , 2016 ) could enable such studies without the muscle artefacts that are inevitable with TMS . HD-tDCS , as used in our third experiment , also has the advantage of not causing muscle twitches , but lacks the focality needed to allow a confident argument that one can disentangle the contributions of the face and hand regions that are located only 2 cm apart . In the Hand movie , movements are displayed both in the first half of the video by the agent wielding the belt and in the second half by the victim’s hand being compressed by and reacting to the swat . Activity in SI has been shown to have the potential to encode all of these ( Keysers et al . , 2010 ) . Interestingly , SI activity significantly predicted the donation only during the second half , in which the victim’s hand is compressed by the belt and reacts to it . This suggests that it is SI’s ability to represent the impact of the belt on the hand or the reaction of the victim to the stimulation on the hand that induced the prosocial decision-making . Furthermore , we addressed the issue of how SI contributes to decision-making , leveraging a third HD-tDCS experiment that allowed us to discriminate between perceptual or motivational contributions . Our results suggest that SI activation in the hand region contributes to prosocial decision-making by transforming the sight of hand-movements caused by a swat into a perception of pain-intensity , which then serves as an input to a decision-making process elsewhere . If this trial-by-trial perception is perturbed , our decision to help no longer optimally follows the trial-by-trial variance in pain experienced by others . This function is similar to the function that SI has during the observation of actions . For instance , disturbing the activity of the SI hand region with TMS makes ratings of the weight of an object seen lifted noisier than in a sham condition , suggesting that the region is necessary for transforming observed hand kinematics into an estimate of the forces that have been acting on the hand ( Valchev et al . , 2017 ) . Similar kinematic analysis may underpin the transformation of the observed hand kinematics following the swat into a painfulness estimate in our Hand condition . Affective social reactions , be they personal distress or empathic concern , would be informed by this kinematic analysis in SI , but require additional processes that the pain experience literature would ascribe to the anterior insula and cingulate ( Lee and Tracey , 2010 ) . In this interpretation , a neural network including SI informs the participant on how intense the swat was on a given trial , and determines the ability of the participant to adjust the donation to the circumstances of a specific trial . By contrast , the mean donation could reflect trait differences in empathic concern ( measured by the Interpersonal Reactive Index [Davis , 1983] ) and money attitude ( Yamauchi and Templer , 1982 ) ( 'Supplemental analysis — Correlation with self-reported questionnaires' ) . A more in-depth understanding of what emotional feelings ( pain-like personal distress vs . more positively valenced empathic concern ) accompany the motivational effect of SI activation on high-pain trials remains unclear from our data , and could be studied in future research by asking participants to provide specific ratings of their own affect on a trial-by-trial basis . In summary , we provide evidence that activity in the hand region of SI that occurs while witnessing the bodily reactions of a victim to a painful stimulation is not only correlated with the willingness to help but significantly influences prosocial decision-making . Our data further constrain the mechanisms through which SI influences prosocial decision-making by showing that altering SI activity also influences the perception of other people’s pain intensity . This suggests that the role of SI is to help us transform the kinematics of affected body-parts into a perception of pain , which is then a significant input to a decision-making process that occurs elsewhere in the brain . If pain is not expressed through the affected body part but communicated through facial expressions , SI activity in the somatotopic representation of the initially affected body-part no longer seems to be a necessary input to this decision-making . These neuromodulation findings support the notion derived from neuroimaging literature that multiple networks , depending on the nature of the stimulus , can be recruited during the perception of the pain of others ( Keysers et al . , 2010; Lamm et al . , 2011 ) . Future studies will be needed to isolate and characterize the causal contribution and interaction across the nodes of these networks , and to further characterize the conditions under which each network is necessary . An empirical foundation for the intuitively attractive and often suggested causal links between the ability to represent what other people feel and prosocial actions is provided by the evidence that SI vicarious activations directly influence prosociality . A total of 169 healthy , right-handed volunteers , with normal or corrected-to-normal vision , ( mean age = 25 +/– 5 SD ) were recruited for our studies ( Table 1 ) . Because previous studies reported that racial biases modulate empathy ( Xu et al . , 2009; Avenanti et al . , 2010; Cikara et al . , 2014 ) and our videos showed a Caucasian confederate , only Caucasian individuals were recruited . All participants received monetary compensation and gave their informed consent for participation in the study . None of the participants reported neurological , psychiatric , or other medical problems or any contraindication to fMRI , TMS or tDCS ( Rossini et al . , 2015; Rossi et al . , 2009 ) . No discomfort or adverse TMS effects were reported by the participants or noticed by the experimenter . Table 1 summarizes the number and characteristics of the participants for each study . All studies have been approved by the Ethics Committee of the University of Amsterdam , The Netherlands ( project identifiers: 2016-BC-7394 , 2016-BC-7130 , 2017-EXT-8467 , 2016-PSY-6485 , 2014-EXT-3476 , and 2014-EXT-3432 ) . All participants received monetary compensation and gave their informed consent for participation in the study . Consent authorization for the publication of images has been obtained . Central to our task was the aim of inducing an effective , naturalistic moral dilemma , in which the state induced by witnessing the distress of another individual is pitched against financial rewards . The other person’s distress was elicited by delivering electroshocks or slaps on the right-hand dorsum . To limit the total number of shocks or slaps delivered throughout the experiments and to avoid uncontrollable variance in the reactions of the victim , we developed a cover story . Each participant was made to believe that she/he would be paired to another participant , with whom she/he will draw lots to decide who would play the role of the observer and who of the pain-taker . The observer and the pain-taker will be allocated to separate adjacent rooms , connected through a video camera . While the pain-taker would receive the electroshocks and slaps , the observer would have an EEG recorded while they witnessed the reaction of the pain-taker to the stimulations ( Experiment 1 ) or while brain stimulation was delivered over SI ( Experiment 2 ) . In reality , the lots were manipulated in such a way that the confederate would always be selected as the pain-taker and the participant as the observer . In addition , participants were misled to think that the noxious stimulations were delivered to the confederate in real-time , and that what the participants saw on the monitor was a live feed from the pain-taker’s room . In reality , we presented prerecorded videos of face and hand reactions to noxious stimuli previously delivered to the confederate ( Videos 1 , 2 , 3 and 4 ) . The confederate’s appearance during the experiment was carefully matched to the pre-recorded videos . The exact setup shown in the videos was recreated at every session , including the belt and electric stimulator used , and shown to the participants . Face and Hand videos were shown in separate sessions ( order randomized across participants ) , with a long break in between . During the break , participants could move , leave the experimental room and , importantly , briefly interact with the confederate . This short break helped to maintain the cover story . The choice to use a cover story was dictated by ( i ) pilot data showing that relaxing the cover-story , for instance by acknowledging that the confederate was in fact an experimenter , led to a notable reduction in donation , and ( ii ) the effort to keep the variance introduced by having different victims at a minimum , facilitating group analyses . In addition , our initial piloting also revealed that testing participants over multiple days led to increased skepticism , which was why we decided to use different pools of participants for each experiment , and why we had to limit the number of trials in the TMS study to what could fit a within subject design . At the end of the costly helping paradigm , participants answered the question ‘Do you think the experimental setup was realistic enough to believe it’ on a scale from 1 ( strongly disagree ) to 7 ( strongly agree ) . Five was used as cut off to discriminate participants who believed in the cover story from those who did not , and participant’s who reported four or less were excluded from the analyses . The credibility values for the whole sample of participants are shown in Figure 1—figure Supplement 1 . Two types of 2 s long videos were generated ( Figure 1A ) . The Hand-videos depicted the confederate’s right hand reactions to a slap delivered by a brown leather belt ( procedure adapted from Meffert et al . [2013] ) . The hand , right arm and shoulder are the confederate’s only visible body parts . While the belt was visible , the hand holding it only entered the field of view marginally at times , and was covered by a black glove to blend in with the black background . The videos started with the belt laying on the hand dorsum . The slap occurred at the end of the first second , during which the belt would be lifted and prepared to hit . Videos ended one second after the slap , after showing the hand and shoulder reaction . A total of 200 Hand-videos were recorded by varying the intensity of the slap at every trial , with 30 s between each trial . The Face-videos showed the actor’s facial expressions in response to an electroshock delivered to the right-hand dorsum . The upper part of her body was clearly visible on a black background . Even though the stimulation was given to her hand and the hand was visible , the hand did not move in response to the shock making the face the main source of information about the stimulation intensity . The videos started with the face in a neutral expression . During the first second , the expression was kept neutral until the stimulation occurred . Both the hand and face movies were centered not on the moment in which the noxious stimulation was delivered but on the reaction of the actor to them . In the Hand videos , the central frame was the one in which the belt hit the hand with consequent immediate reaction of the hand . In the Face videos , the central frame was the one in which the facial expression began to change ( i . e . at +1 s from the beginning of the movie ) . A total of 392 Face videos were recorded . The electrical stimulation was a 100 Hz train of electrical pulses of 2 ms pulse duration ( square pulse waveform ) delivered via a bipolar concentric surface electrode ( stimulation areas: 16 mm2 ) . The electrodes were attached to the skin with tape , which was also left in place during the Hand-video recording . A black ‘X’ was drawn on the tape to clearly show the electrode's position . Each stimulation lasted 1000 ms and varied in current intensity , which ranged from 0 . 2 mA to 8 . 0 mA . Thirty seconds were left between stimulations . Current intensity was determined prior to video recording , by following a procedure well established in the literature ( de Vignemont and Singer , 2006; Cui et al . , 2015 ) : starting from 0 . 1 mA , the current was gradually increased until reaching a maximum of 8 . 0 mA in increments of 0 . 2 mA . The actor was instructed to evaluate how painful each stimulation was on a 10-point scale , and the current intensities to be used during video recording were chosen according to a maximum perceived intensity of 8 . The actor was the same Caucasian woman for all of the videos and experiments: author SG who played the confederate role . The videos were recorded using a Sony DSR-PDX10P Camcorder ( Sony , Minato , Tokyo , Japan ) , and were edited using Adobe Premiere Pro CS6 ( Adobe , San Jose , CA , USA ) . The 392 videos that were generated were validated by an independent group of 40 participants who did not participate in the other experiments ( Table 1 ) . Hand and Face videos were presented ( OkazoLab Ltd , 2012 ) in separate blocks , counterbalanced across participants . Participants were instructed to observe them and to report the perceived pain intensity that the person in the video felt , using the same 10-point scale employed in the pain-threshold assessment . Average ratings were computed and rounded off . The results yielded six different movie categories with average pain intensities perceived as 2 , 3 , 4 , 5 , 6 , and 7 out of 10 . Of those , we selected different subsamples for each experiment on the basis of the following two criteria: ( a ) a low standard deviation in rating across participants ensuring that a specific pain intensity was communicated reliably , and ( b ) maximized the statistical power of the regression ( i . e . privileging movies at the extremes of the intensity range ) . For the EEG task , we privileged the criterion ( b ) and included a larger number of trials to reduce neural habituation ( 95 trials in total ) . For the TMS experiment , we privileged criterion ( a ) to maximize the sensitivity to small changes in perceptual accuracy , and only used those movies in which ratings were most concordant ( Table 2 ) . In both the EEG and TMS experiments , the Face and Hand conditions did not differ in average perceived intensity ( EEG: MFace = 3 . 7 , SD = 1 . 7; MHand = 4 , SD = 1 . 4; t ( 93 ) = 0 . 88 , p=0 . 4 . TMS: MFace = 3 . 8 , SD = 1 . 6; MHand = 3 . 9 , SD = 1 . 2; t-test t ( 118 ) = −0 . 26 , p=0 . 9 ) or standard deviation ( EEG: F ( 42 , 51 ) = 1 . 49 , p=0 . 09; TMS: F ( 14 , 14 ) = 1 . 67 , p=1 . 17 ) . Because the intensity of the OutputMovie depended on the participant’s donation , it was impossible to precisely predict the number of videos needed for each intensity and participant . This means that in some cases , the number of recorded videos was lower than the number of actual presentations of a particular intensity , and few videos had to be shown more than once . Care was taken to maximize the distance between repetitions of the same stimulus . The number of repeated videos is low and it was never the case that a participant saw all of the movies twice . During debriefing , we asked our participants to indicate on a seven step scale ( from 1 = strongly disagree to 7 = strongly agree ) how well the following statement applied to them: ‘You think you saw the same movie twice . ’ On average , participants in the EEG experiment reported a value of 6 . 1 ( Figure 1—figure Supplement 1 ) , whereas those in the TMS experiment reported a value of of 6 . 3 , suggesting that movie repetition was not easily recognized by our participants . For the participant , each trial of the costly helping paradigm started with a 1 s red fixation cross , followed by a video showing the actor’s reaction to the first stimulation ( InputMovie , Figure 1A ) , the intensity of which was randomly determined by the computer . Participants then received an endowment of 6€ , and had to decide whether to donate all or part of this money to reduce the intensity of the following stimulation . Participants knew , that every 1€ donated reduced the intensity by one unit , and that money was not donated directly to the confederate , who did not have monetary benefit from the donation . Instead , 10% of the money not donated was added as a ‘bonus’ to participants’ compensation for taking part in the study . InputMovie intensity varied from trial to trial to cover the entire range from 2 to 7 , and was randomly chosen . Participants’ implicit task was to infer the intensity of the stimulation from the confederate’s reaction . Participants selected the donation amount by moving a rectangle along a bar with possible donations ( 0€ to 6€ ) . The starting position of the cursor was randomized to avoid motor preparation of the response . Participants were instructed to use a two-pedal controller with their right foot to select the donation ( USB Double Foot Switch II , Scythe Co . , Ltd . , Tokyo , Japan ) . We used a foot-pedal rather than a traditional button-box because SI activity in the hand region , critical for this experiment , may otherwise have been contaminated by the planning/performance of the button-presses . After three seconds without pressing any pedal , the software would select the current position of the rectangle as the chosen donation and move to the following black screen , which had a random duration from 1 . 5 to 3 s . Finally , depending on the donation , 2 s feedback video ( OutputMovie ) showed the confederate’s response to the second stimulation . OutputMovie would represent the end of a trial . A 5 s black screen separated consecutive trials . For the entire duration of the costly helping paradigm , electrophysiological brain signals were recorded from 64 active channels ( 10–20 positioning ) by an ActiCHamp Brain Vision system . The ground electrode was placed on Fz . Electrode impedances were kept below 5 kΩ , and all signals were digitized ( rate of 500 Hz ) and stored for off-line processing . All data analyses were performed using the FieldTrip Toolbox ( Oostenveld et al . , 2011 ) and customized MATLAB ( Mathworks Inc . , Natick , MA , USA ) scripts . The signals were low-pass filtered at 60 Hz and band-stop filtered within the range of 49 . 5–50 . 5 Hz , and harmonics were used to eliminate the electrical line noise . The data were re-referenced to a common average and segmented in epochs of 7 s , containing 5 s before InputMovie onset and lasting until InputMovie end . The segmented signals were visually inspected across all channels . Trials containing muscular and other non-ocular movement artifacts were discarded . The artifact rejection procedure resulted in 41 . 9 ± 1 . 7 artifact-free trials in the Hand task and 50 . 6 ± 1 . 9 in the Face task . Blinks and eye movements were corrected using Independent Component Analysis ( Jung et al . , 2000 ) . Each trial was then baseline corrected using the average of the signal from 400 to 200 ms before the appearance of the fixation cross . We used a linear constrained minimum variance beam-former approach ( Van Veen et al . , 1997 ) . We used a three-layer BEM volume conductance model of 1 cm³ resolution as the forward model ( Figure 2B , Oostenveld et al . , 2003 ) . For each participant , we used the BEM model together with the covariate matrix of the ERP ( entire trial length , obtained by averaging all of the videos belonging to each condition ) to create an adaptive spatial filter such that its inverse applied to the sensor level representation would reconstruct the source power with maximum strength at the location of the source of interest , and would suppress the output from the sources of no-interest . To derive the complex source estimates , the time-course of each trial was multiplied with the real-valued filters . The procedure resulted in three orthogonal dipoles within each ROI , oriented in the three spatial dimensions . We constructed our filters separately for the Face and Hand sessions to account for the possibility that the noise structure changed during the long pause ( ~20 min ) separating the two sessions . This maximized our ability to compare different trials from the same session to establish whether donation and brain activity are related within a condition , but reduced our ability to compare the two conditions directly . To test whether brain activity in the ROIs while watching the videos could explain the Z-donation , we first conducted a mass-univariate robust regression analysis ( within subjects ) with the brain activity as the predictor variable and Z-donation as the observed variable . This was done separately for each time point , dipole , task , subject and ROI . Robust regression was chosen to be less sensitive to outliers in the data ( Wager et al . , 2005 ) . The regression slopes of this subject level analysis were then subjected to Hotelling's T-Squared test ( Ht2 ) at the group level . This is a multivariate test that examines whether the average slopes for the three dipoles are all zero . This test was repeated for each time point , condition and ROI . The family-wise error rate arising from multiple comparison of time-points was dealt with by using a cluster-based non-parametric Monte-Carlo correction . Neighboring values exceeding the cluster-cutting threshold ( corresponding to Ht2 >4 . 675 and punc <0 . 01 ) were combined into a single cluster . Cluster-level statistics were computed by comparing the summed Ht2 values of each cluster against a permutation distribution . The permutation distribution was constructed by randomly flipping the sign of all of the slopes of randomly selected participants ( 1000 iterations ) and by calculating the maximum group cluster statistic for each iteration . The null distribution of the cluster-based test statistic was obtained by taking the most extreme value of the statistic in each permutation . The cluster-based test statistic in the observed data was then associated with a corrected pseudo-p value based on its percentile in the null distribution for each cluster . Furthermore , we corrected for four ROIs ( two left and two right mirror ROIs ) and two conditions tested using a Bonferroni correction of 6 ( Figure 3 ) , leading to a pseudo-p-value of 0 . 0063 as the cut-off for the cluster-based test statistic . Sample size for the TMS experiment was determined though a power analysis conducted using GPower 3 ( Faul et al . , 2007 ) , with power ( 1 – β ) set at 0 . 95 and α = 0 . 05 . The effect size was chosen on the basis of the work of Paracampo et al . ( 2017 ) ( Cohen’s d = 0 . 94 ) , because this work was conducted by the experimenter who was also responsible for the TMS part of the work described in the current manuscript ( RP ) , and because the Paracampo et al . ( 2017 ) study involved a task that made similar cognitive demands , used an equivalent rTMS protocol , and targeted the same brain region ( SI ) . A comparable effect size ( Cohen’s d = 0 . 89 ) was found when taking into account TMS studies in which participants are required to observe others and understand their behavior ( Paracampo et al . , 2017 , 2018; Valchev et al . , 2017; Tidoni et al . , 2013 ) . In our within subjects study , we hypothesed a perturbation of the activity in SI , and therefore a reduction in performance as a consequence of the perturbation , so we conducted a power analysis to compare performance in active stimulation with performance in sham stimulation using a matched paired one-tailed t-test at the second ( group ) level . This analysis yielded a required sample size of 15 participants . TMS was administered using a figure-of-eight coil ( diameter: 70 mm ) connected to a Magstim Rapid2 stimulator ( Magstim , Whitland , Dyfed , UK ) . To set rTMS intensity and to determine coil location , the resting motor threshold ( rMT ) was estimated for all participants in a preliminary phase of the experiment using standard procedures ( Rossi et al . , 2009 ) . Motor-evoked potentials ( MEPs ) induced by stimulation of the left motor cortex were recorded from the right first dorsal interosseous ( FDI ) by means of a Biopac MP-35 . EMG signals were band-pass filtered ( 30–500 Hz ) and digitized ( sampling rate: 5 kHz ) . Pairs of Ag-AgCl surface electrodes ( Ø35mm ) were placed in a belly-tendon montage with a ground electrode on the wrist . The intersection of the coil was placed tangentially to the scalp with the handle pointing backward and laterally at a 45° angle away from the midline . The rMT was defined as the minimal intensity of stimulator output that produces MEPs with an amplitude of at least 50 μV in the FDI with 50% probability ( Rossini et al . , 2015 ) . After the rMT procedure , the optimal scalp location for the hand representation in the left motor cortex was marked . A large body of evidence shows that the hand area in the somatosensory cortex can be successfully targeted by positioning the coil 1–4 cm posterior to the motor hotspot ( Harris et al . , 2002; Balslev et al . , 2004; Merabet et al . , 2004; Fiorio and Haggard , 2005; Tegenthoff et al . , 2005; Bufalari et al . , 2007; Azañón and Haggard , 2009; Jacquet and Avenanti , 2015; Valchev et al . , 2017 ) . This approach is based on the close correspondence between the motor and somatic homunculi ( Buccino et al . , 2001; Yang et al . , 1994; Schulz et al . , 2004; Nakamura et al . , 1998; Amunts and Zilles , 2015; Kuehn et al . , 2014 ) . In line with this , we identified our region of interest using a two-step procedure . First , we localized the hand region in the primary motor cortex , corresponding to the optimal scalp position ( OSP ) for evoking MEPs in the FDI muscle . After that , we moved the coil 2 cm backward following a parasagittal plane , assuming that this displacement would not produce effects on M1 . We tested this assumption directly by checking that TMS pulses applied at 105% rMT with the coil in the final target position did not elicit any detectable MEPs . To rule out any possible interference with the primary motor cortex , intensity for the rTMS was set at 90% of the resting motor threshold . Moreover , before the rTMS session , the position of the coil over the SI-L was verified by applying single pulses of TMS to ensure that no muscle activity was associated with our repetitive stimulations . While performing the costly helping paradigm , a time-locked single train of subthreshold 6 Hz rTMS ( 12 pulses , 2 s ) was delivered ( Tidoni et al . , 2013; Paracampo et al . , 2017 ) , starting at the onset of the movie and thus covering its entire duration . During active rTMS blocks , the intersection of the coil was placed tangentially to the scalp directly above the scalp location of the target region with the handle pointing backward and laterally at a 45° angle away from the midline . Sham rTMS blocks were performed by tilting the coil by 90° over the same target region , to provide some scalp sensations and TMS sounds comparable to active stimulation but without inducing a current in the brain . The general procedure of the experiment was the same as that for Experiment 1 , except for changes in the number of trials and in the video presented , which were necessary to adapt the task to the TMS ( sham vs active ) set-up . Participants underwent a total of 60 trials for Face Condition and 60 trials for Hand condition . The conditions were presented in four blocks ( two blocks for the Face and two for the Hand condition ) , separated by a long break in the middle in which participants further interacted with the confederate . Each block was equally divided into two parts , which were assigned to active and sham TMS in a pseudo-randomized order ( e . g . for one participant: ActiveTMSFaceBlock1part1 , ShamTMSFaceBlock1part2 , ShamHandBlock1part2 , ActiveHandBlock1part2 , long break , ActiveHandBlock2part1 , ShamHandBlock2part2 , ActiveFaceBlock2part2 , ShamFaceBlock2part1 ) . Although some videos might have been shown twice within block1 or block2 , videos in block1 were different from those in block2 . At the end of the experimental session , participants had to rate from 1 to 4 how much headache , neck stiffness , itching on the skin , pain on the skin below the stimulation site , sleepiness and mood-swing they experienced , and whether it was difficult to concentrate . Answers were then compared with the TMS effect in the two tasks by calculating the difference in slopes between the Active and the Sham Condition and by correlating this difference with the answer to these questions across participants . Results were corrected for multiple comparison and found to be non-significant . For this experiment , in addition to the Face and Hand movies , a new set of videos was created , in which no pain was depicted but the color saturation changed over time ( Figure 1B , and Videos 5–6 ) . To match the temporal dynamic of the Hand and Face videos , the saturation change started after 1 s from the beginning of the videos and reached its peak 0 . 5 s afterwards . Videos were created with three different levels of saturation changes . An independent group of 20 participants ( Table 1 ) watched Color , Face and Hand videos ( presented using EventIDE; OkazoLab Ltd . , 2012 ) , and rated from 1 to 10 how painful the stimulation was for the person in the Face and Hand videos , and how much the saturation changed in the Color videos . Using their left hand , they moved a rectangle along a bar with possible ratings , from 0 to 10 . The starting position of the cursor was randomized to avoid motor preparation of the response . The validation procedure resulted in a total of 32 videos per category matched for average rating ( F ( 2 , 93 ) =0 . 2 , p=0 . 8 ) and accuracy , calculated as the square of the difference from the expected value ( F ( 2 , 93 ) =0 . 4 , p=0 . 6 ) . Sample size for the tDCS experiment was determined though a power analysis conducted using GPower 3 ( Faul et al . , 2007 ) , with power ( 1 – β ) set at 0 . 95 and α = 0 . 05 . We expected a small effect size on the basis of recent transcranial electrical stimulation experiments ( Bolognini et al . , 2013; Avenanti et al . , 20172018 ) . In these studies , the somatosensory cortices were targeted , and similar design and task requirements were used . This analysis yielded a required sample size of 26 participants ( Table 1 ) . 1 . 5 mA was delivered to the left primary somatosensory cortex for 18 min through a 4 × 1 ring-electrode set-up consisting of a central active anode and four surrounding return electrodes ( Kuo et al . , 2013 ) ; Figure 2D ) , which were connected to a battery-operated tDCS MXN-9 High-Definition ( HD ) Stimulator ( Soterix Medical Inc . , USA ) . The HD-tDCS electrodes were fixed to a cap by means of HD-tDCS electrode holders , with the central anode placed over the primary somatosensory cortex , between the EEG electrode sites C3 and CP3 . The HD-tDCS electrodes’ impedance was kept below 10 kΩ . Throughout the stimulation , the participant was sitting comfortably in an arm chair . Participants received both real and sham stimulations on two different days , separated by an average of 7 . 8 days ( SD = 3 ) . After the sham and real stimulations , participants performed the rating task as in the validation procedure . Each trial started with the presentation of a white fixation cross ( 1 s ) , which was followed by the video clip ( 2 s ) , and finally the visual analogue scale . Using their left hand , participants moved a rectangle along the scale using two keys ( one for moving the rectangle to the right , one to the left ) . A third key was pressed to confirm the intensity selection . A variable interval of between 2800 and 3200 ms separated the trials . Videos were presented in six blocks ( two per task ) containing multiple intensities . Block presentation order was randomized among participants and among sessions . After both the real and the sham sessions , participants rated from 1 to 4 how much headache , neck stiffness , itching on the skin , pain on the skin below the stimulation site , sleepiness and mood-swing they experienced , and whether it was difficult to concentrate .	When we experience physical pain , certain areas in our brain that process bodily sensation and emotions switch on . If we see someone else in pain , many of the same regions also get activated . In contrast , convicted criminals with psychopathic traits have less activation in these areas of the brain when witnessing someone’s pain; they also show less empathy and disregard the needs of others . This suggests that a lack of this ‘shared activations’ may lead to problems in empathy . In fact , many scientists believe that shared activations are why we feel empathy for people in pain , and why we are driven to help them . Yet , there is little direct evidence about how the activity in the pain processing parts of the brain actually influences helpful behavior . As a result , some scientists now argue that empathy-related processes may actually contribute very little to helping behavior . Gallo et al . designed an experiment where participants watched videos of someone having their hand swatted with a belt , and showing different levels of pain as a result . The volunteers could decide to reduce the amount of pain the person received by donating money they could have taken home . The more pain the participants thought the victim was in , the more money they gave up to lessen it . During the study , the activity in the brain region that processes pain in the hand was also measured in the participants . The more active this region was , the more money people donated to help . Then , Gallo et al . used techniques that interfered with the activity of the brain area involved in perceiving sensations from the hand . This interference changed how accurately participants assessed the victim's pain . It also disrupted the link between donations and the victim's perceived pain: the amount of money people gave no longer matched the level of pain they had witnessed . This suggests that the brain areas that perceive sensations of pain in the self , which evolved primarily to experience our own sensations , also have a social function . They transform the sight of bodily harm into an accurate feeling for how much pain the victim experiences . The findings also show that we need this feeling so we can adapt our help to the needs of others . In the current debate about the role of empathy in helping behaviors , this study demonstrates that empathy-related brain activity indeed promotes helping by allowing us to detect those that need our assistance . Understanding the relationship between helping behavior and the activity of the brain may further lead to treatments for individuals with antisocial behavior and for children with callous and unemotional traits , a disorder that is associated with a lack of empathy and a general disregard for others .	[ "Abstract", "Introduction", "Results", "Discussion", "Methods", "and", "material" ]	[ "neuroscience" ]	2018	The causal role of the somatosensory cortex in prosocial behaviour
Urodynamic status must interact with arousal and attentional processes so that voiding occurs under appropriate conditions . To elucidate the central encoding of this visceral demand , multisite recordings were made within a putative pontine-cortical micturition circuit from the pontine micturition center ( PMC ) , locus coeruleus ( LC ) and medial prefrontal cortex ( mPFC ) during cystometry in unanesthetized rats . PMC neurons had homogeneous firing patterns , characterized by tonic activity and phasic bursts that were temporally associated with distinct phases of the micturition cycle . LC and cortical activation became synchronized 20-30 s prior to micturition . During this pre-micturition interval , a theta oscillation developed in the LC , the mPFC desynchronized and LC-mPFC coherence increased in the theta frequency range . The temporal offset between the shift in LC-mPFC network activity and micturition may allow time to disengage from ongoing behaviors unrelated to micturition and initiate specific voiding behaviors so that micturition occurs in environmentally and socially appropriate conditions . The physiological elimination of urine , micturition , must occur in socially appropriate and environmentally safe conditions . In some species micturition is integral to social communication , for example , with regard to sexual status and social rank ( Ralls , 1971; Desjardins et al . , 1973 ) . These neurobehavioral aspects of micturition demand precise temporal coordination between bladder activity and the initiation of motor patterns that characterize voiding behavior . For example , bladder filling should increase arousal and shift the focus of attention before pressure reaches the micturition threshold so that ongoing behavior is suspended and voiding behavior is initiated prior to the passage of urine . A lack of coordination between voiding behaviors and micturition characterizes dysfunctional voiding syndromes , such as enuresis that can occur both in children and older adults ( Torrens and Collins , 1975; Nevéus , 2017 ) . This temporal coordination is performed by circuits that convey reciprocal communication between the brain and bladder . The pontine micturition center ( PMC ) is central to a circuit that is positioned to coordinate neurobehavioral and visceral limbs of micturition ( Valentino et al . , 2011b ) . Through its projections to lumbosacral preganglionic neurons that regulate the parasympathetic innervation of the detrusor , the PMC is proposed to function as an all or none switch , whereby the neurons become activated at some threshold pressure and engage the descending limb that initiates bladder contraction ( de Groat , 2006 ) . In contrast to our knowledge of how the brain regulates bladder function , there is a paucity of information on the bottom-up processing of bladder information that is necessary for the neurobehavioral limb of the micturition reflex . Human brain imaging studies using SPECT , PET and fMRI have guided our current thinking about the structures involved in this complex process ( Griffiths , 2004; Griffiths et al . , 2009 ) . However , these approaches have limitations with respect to temporal and spatial resolution . Using anatomical and physiological approaches , we described one circuit that is positioned to convey bladder information to cortical regions involved in executive function . We identified a population of spinal-projecting PMC neurons that are retrogradely labeled from both the lumbosacral spinal cord and the pontine noradrenergic nucleus locus coeruleus ( LC ) ( Valentino et al . , 1996 ) . The LC has a widely distributed axonal network that extensively innervates the forebrain and particularly the cortex where norepinephrine serves as a neuromodulator of cortical activity ( Swanson and Hartman , 1975 ) . Selective pharmacological or optogenetic activation of LC neurons or activation using Gs-linked DREADDS is sufficient to produce cortical electroencephalographic ( EEG ) activation indicative of arousal ( Berridge and Foote , 1991; Carter et al . , 2010; Vazey and Aston-Jones , 2014 ) . Notably , LC neurons are activated by bladder and colonic distention and this is temporally linked to cortical EEG activation ( Elam et al . , 1986; Page et al . , 1992; Lechner et al . , 1997 ) . In addition to increasing arousal in response to salient stimuli , LC activation is thought to reset attentional networks to shift the focus of attention to salient stimuli ( Aston-Jones and Bloom , 1981a; Aston-Jones and Cohen , 2005; Bouret and Sara , 2005 ) . Integrating these findings , we proposed a model whereby increases in bladder pressure could activate PMC neurons that project to the LC and the impact on LC-cortical projections would increase arousal , reorient attention and promote the initiation of a voiding behavioral repertoire while suspending incompatible ongoing behaviors ( Valentino et al . , 1999; Valentino et al . , 2011a ) . Thus , through collateral projections to the LC and spinal cord based on dual retrograde labeling , the PMC could coordinate visceral and neurobehavioral limbs of the micturition reflex . To begin to test components of this model we recorded unit activity from PMC neurons and/or LC neurons along with local field potentials ( LFP ) from the medial prefrontal cortex ( mPFC ) simultaneously with urodynamic endpoints measured during in vivo cystometry in unanesthetized rats ( Figure 1A , B , Figure 1—figure supplement 1–1 ) . After the first micturition cycle , subsequent cycles in rats became regular , with bladder pressure gradually increasing to a peak , at which point micturition occurred and bladder pressure fell to 0 mm Hg ( Figure 1—figure supplement 1–2 ) . Single unit activity was recorded from 36 PMC neurons from three rats during in vivo cystometry . Figure 1C , D shows activity from three single units recorded on one channel in the PMC during two micturition cycles . Whether depicted as the raw waveform trace ( Figure 1C ) or rate ( Figure 1D ) , the pattern of PMC neuronal discharge was relatively homogeneous and this was true across all three rats ( Figures 1C , D and 2A ) . The PMC has been proposed to function as a switch such that neuronal activation results in micturition ( de Groat , 2006 ) . Consistent with this , in all rats , PMC neurons discharged during micturition until bladder pressure decreased to zero and the bladder was empty ( Figures 1C , D and 2A ) . Figure 2B shows the change in discharge rate associated with micturition for the individual cells shown in the example rasters . Most neurons increased their discharge rate and no recorded neurons were inhibited at this time . All PMC neurons also exhibited burst activity at different times with respect to micturition ( Table 1 ) . This was particularly apparent within the 20 s period following micturition during which bursts averaging 28 ± 1 Hz frequency and 606 ± 66 ms in duration occurred ( Table 1 , Figures 1C , 2A and C ) . These post-micturition bursts were sometimes associated with a very small , transient increase in bladder pressure ( Figure 2A asterisks ) . Notably , in all PMC cells similar bursts ( 28 ± 2 Hz , 432 ± 35 ms duration ) occurred regularly during inter-micturition intervals when bladder pressure was low ( Table 1 , Figure 1C ) . Burst activity was rare during the pre-micturition period ( 20 s prior to micturition ) . Notably , some cells were distinguished by tonic discharge between bursts ( Figure 1C , first and third cells ) . LC neuronal activity was recorded from 51 neurons of 4 rats . As described in previous studies in unanesthetized rats ( Aston-Jones and Bloom , 1981a ) , LC neurons were spontaneously active throughout the recordings although discharge rate fluctuated with the micturition cycle . Figure 3A , E shows how activity of LC units of one representative rat is temporally correlated to bladder pressure . In contrast to PMC neurons , LC neurons consistently increased their discharge rate 10–30 s prior to micturition and this activation ceased with micturition ( Figure 3A , B , C , E ) . The distribution of latencies between significant LC activation and micturition ranged between 0 and 65 s with most cells having a latency between 10–30 s ( Figure 3D ) . For one rat , the proximity of the LC and the PMC allowed for detection and recording of units from both regions during cystometry ( Figure 4 ) . Notably , the temporal relationship between LC activation , PMC neuronal activation and micturition in this particular case argued against the hypothesis that PMC neurons drive LC neurons prior to micturition because LC activation preceded PMC neuronal activation . The LC neuronal activation , occurring 10–30 s prior to micturition , was temporally correlated with mPFC desynchronization ( Figure 5A1 , A2 ) . Simultaneous local field potential recordings ( LFP ) in LC and mPFC demonstrate a shift from low frequency activity at baseline ( e . g . , Figure 5B1 , 0–47 s ) to a prominent theta oscillation in the LC during the pre-micturition period ( e . g . , Figure 5B1 , 47–77 s ) . This was accompanied by a desynchronization of the mPFC ( Figure 5A2 , B2 ) . Interestingly , although the mPFC LFP magnitude was small during the 30 s period prior to micturition relative to the preceding baseline interval , there was still evidence of a theta oscillation in the mPFC during this time as well ( white arrowheads in Figure 5B2 , Figure 6B , cf . , red and black lines ) . This effect was consistent as indicated by the mean power spectra of LC and mPFC LFP activity generated from each of 3 rats recorded across 8 micturition cycles ( 3 , 2 , and 3 for the individual rats ) ( Figure 6A , B ) . Thus , LC activity recorded from 60 to 30 s before micturition shifted from mixed low frequency activity to a prominent theta oscillation occurring 30–0 s before micturition . In concert , the mPFC power spectrum shifted from high amplitude , low frequency activity ( 0–5 Hz ) to much lower amplitude oscillations and a decrease of power in all frequencies , characteristic of desynchronization . An increase in coherence between the LC and mPFC in the theta frequency characterized this pre-micturition interval ( Figure 6C ) . This change in LC and mPFC LFP activity reverted with bladder emptying ( Figure 5 ) . The present findings should be interpreted with consideration of certain technical limitations . In vivo cystometry was used to monitor and regulate micturition cycles . However , it does not mimic natural micturition cycles , which would be longer in the absence of steady bladder infusion . Although the multiwire recording technique allows for sampling activity from multiple neurons within the same nucleus , the results are inevitably biased toward activity that the wires can record from and that can be reliably discriminated . We cannot rule out the possibility that certain neuronal populations in a nucleus were not sampled . This may be a particular problem for PMC which is neurochemically heterogeneous ( Sutin and Jacobowitz , 1988 ) . This is less of a caveat for the LC , which is relatively homogeneous in the rat ( Aston-Jones et al . , 1995 ) . Only females were used in the study because the bladder surgery is somewhat easier . The surgery was relatively complex because of the need to electrophysiologically localize brain electrodes as well as to implant the catheter , so that we chose to use females . Therefore , there is a caveat that our interpretations may not extend to both sexes . However , we previously demonstrated the same temporal correlation between cortical activation and peak bladder pressure in male rats ( Kiddoo et al . , 2006; Rickenbacher et al . , 2008 ) . Most previous studies of PMC neuronal activity with relation to bladder pressure were done in anesthetized rats or cats with the bladder distended under isotonic conditions such that reflex contractions would occur ( Willette et al . , 1988; Tanaka et al . , 2003; de Groat and Wickens , 2013 ) . Three general neuronal types were identified in these conditions , neurons that discharge before and during reflex bladder contractions , neurons that are inhibited during contractions and active between contractions and neurons that fire transiently at the beginning of the contractions . The majority of recorded neurons in those studies were of the second type . One study recorded activity throughout the pons and medulla during cystometry in the absence of anesthesia in the decerebrate cat ( Sugaya et al . , 2003 ) . This study also reported neurons that were either activated , inhibited or unaffected during the contraction with the majority being inhibited . Notably , in that study the mapping of these neuronal types demonstrated that nearly all neurons that were inhibited during bladder contraction were ventral to the region corresponding to the PMC , whereas most neurons within the PMC were activated during bladder contraction . The present study is unique in recording single unit activity from the PMC in the intact unanesthetized and freely moving rat during cystometry . PMC neurons recorded in this study were homogeneous in the temporal correlation of their discharge pattern with the micturition cycle and resembled the previously described excitatory neurons . Notably , the present results are consistent with recent studies using photometry to measure the calcium currents in genetically-identified CRF-PMC neurons in unanesthetized mice during cystometry ( Hou et al . , 2016 ) . These studies demonstrated a signal that was most prominent coincident with and up to 10 s after micturition . The present unit recordings have temporal resolution to detect changes in discharge rates and patterns such as bursts , that underlie the encoding of information . The post-micturition burst was consistent across all PMC neurons . If this results in increased glutamate release , it could be a mechanism to assure complete emptying . However , CRF is co-localized with glutamate in PMC neurons ( Hou et al . , 2016 ) and evidence suggests that CRF is inhibitory in this pathway and may counteract the excitatory effects of glutamate neurotransmission ( Pavcovich and Valentino , 1995 ) . Higher frequency bursts that favor neuropeptide release could be a mechanism for switching the function of the nucleus from one that facilitates active emptying to one that facilitates storage by countering detrusor contraction . Repetition of this high frequency burst during the intermicturition phases when bladder pressure is low would also be consistent with that . Future studies of simultaneous manipulation and recording of PMC neurons during cystometry would better address this question . Rat LC neurons are relatively homogeneous in that they are all noradrenergic and have similar electrophysiological properties ( Aston-Jones et al . , 1995 ) . A characteristic property of LC neurons of both anesthetized and unanesthetized rats is their activation by salient sensory stimuli of all modalities ( Aston-Jones and Bloom , 1981b ) . LC neurons are also activated by diverse physiological stimuli including hypotension and non-noxious distention of the bladder or colon ( Elam et al . , 1985; Elam et al . , 1986; Valentino and Wehby , 1988; Page et al . , 1992; Lechner et al . , 1997 ) . In the present study , LC neuronal recordings during in vivo cystometry in unanesthetized rats revealed an exquisite sensitivity of LC neurons in that they become tonically activated approximately 20 s prior to micturition and in most cases before the steady pressure rise to micturition threshold ( Figures 3 and 5 ) . Possibly , this is a response to bladder wall stretching rather than an increase in intravesicular pressure . The timing of activation of LC neurons and PMC neurons with respect to micturition argues against our hypothesis that PMC projections to the LC drive its activation by bladder stimuli . This was particularly apparent in the one case in which both LC and PMC neurons could be recorded in the same subject . However , because only a small subpopulation of PMC neurons project to the LC , it is possible that our neuronal sampling did not detect PMC-LC projection neurons ( Valentino et al . , 1996 ) . Previous lesion studies provided evidence that LC activation by colonic distention was mediated by PMC but bladder distention was not examined in that study ( Rouzade-Dominguez et al . , 2001 ) . Another route by which the LC could receive bladder information is through the ventrolateral periaqueductal gray ( vlPAG ) . Anterogradely labeled vlPAG fibers innervate both PMC and the rostromedial peri-LC and synaptically contact LC dendrites in this region ( Bajic et al . , 2000 ) . The PAG then could regulate both the LC and the PMC in response to bladder related stimuli . The apparent lack of temporal coordination between LC and PMC neuronal discharge could then be explained if they received distinct vlPAG afferents . The LC-norepinephrine system densely innervates the mPFC and modulates its activity and executive functions ( Arnsten and Li , 2005; Arnsten , 2011 ) . Regionally selective pharmacological activation of the LC or activation using Designer Receptor Exclusively Activated by Designer Drug ( DREADD ) technology produces a desynchronization of the cortical EEG ( an arousal response ) ( Berridge and Foote , 1991; Vazey and Aston-Jones , 2014 ) . Moreover , LC activation by certain stimuli is necessary for their ability to elicit cortical arousal ( Valentino et al . , 1991 ) . This includes bladder distention produced by injection of saline in anesthetized rats ( Page et al . , 1992 ) . In addition to increasing arousal , the LC-norepinephrine system has been implicated in shifting attention towards salient stimuli ( Aston-Jones and Bloom , 1981b; Aston-Jones and Cohen , 2005 ) . The present results identified an important temporal relationship between LC activity , mPFC activity and bladder pressure , with LC and cortical activation being synchronized to occur approximately 20 s prior to micturition . The examination of network activity in the LC and between the LC and mPFC revealed that during this pre-micturition interval when LC neuronal activity is elevated , network activity shifts from low frequency oscillations to a prominent theta oscillation . At the same time mPFC activity desynchronizes as characterized by a decreased power in all frequencies but LC-mPFC coherence , specifically in the theta frequency , is increased . This network shift that is temporally offset from micturition may be a neural code to increase arousal and reset behavior prior to micturition so that voiding occurs in appropriate environments . Taken together , the present results highlight the complex synchronization of pontine and cortical neurons with bladder afferent information that is required to maintain coordination between voiding behavior and urination . Just as normal bladder signals are relayed to the cortex , signals arising from bladder-related pathology could impact on cortical function through these same networks . Consistent with this , partial bladder outlet obstruction produces an abnormal urodynamic pattern that is associated with robust changes in cortical network activity ( Rickenbacher et al . , 2008 ) . This condition is characterized by frequent non-micturition contractions that are temporally correlated with persistent cortical theta oscillations . Selective lesion of LC-forebrain projections using the toxin , DSP-4 , prevented the cortical changes without affecting the abnormal urodynamics . This underscores the importance of the LC-cortical network in communicating bladder status to the cortex and suggests that dysregulation of this network could be a prominent signature of underlying bladder pathology . Female adult Sprague-Dawley rats ( Charles River , Wilmington , MA ) were maintained in a temperature and light controlled environment ( 20°C , 12 hr light-dark cycle , lights on at 0700 hr ) with food and water available ad libitum . Rats were housed 2/cage until the day of surgery . The care and use of animals were approved by the Children’s Hospital of Philadelphia Institutional Animal Care and Use Committee . Surgery was performed at least 1 week after arrival at the animal facility . Surgery for implantation of an 8-microwire bundle electrode ( NB Labs , Denison , TX ) into the locus coeruleus ( LC ) was identical to that previously described ( Curtis et al . , 2012 ) . Briefly , rats were anesthetized with an isoflurane-air mixture and positioned in a stereotaxic frame . Body temperature was maintained at 37 . 5o C by a feedback controlled heating unit . A hole ( 4 mm diameter ) was drilled in the skull centered at 3 . 7 mm caudal and 1 . 2 mm lateral to lambda for approaching the LC . Additional holes were drilled to insert skull screws for fixing the microwire bundle electrode to the skull with dental cement . Neuronal recordings with glass micropipettes ( 2–4 μm diameter tip , 4–7 MOhm ) filled with 0 . 5 M sodium acetate buffer were used to initially localize the LC . These were advanced toward the LC with a micromanipulator . Neuronal signals were amplified , filtered and monitored with an oscilloscope and a loudspeaker . LC neurons were tentatively identified during recording by their spontaneous discharge rates ( 0 . 5–5 Hz ) , entirely positive , notched waveforms ( 2–3 ms duration ) , and biphasic excitatory–inhibitory responses to contralateral hindpaw or tail pinch . Trajectories where LC units were encountered with the glass micropipette for at least 400 μm ( dorsal-ventral penetration ) were targeted for implantation with the microwire electrode bundle ( NB Labs , Denison , TX ) consisting of 8 Teflon insulated stainless steel wires ( 50 μm diameter ) that were gathered in a circular bundle ( 7–8 mm long ) and cut to produce bare wire tips for recording . A ground wire from the electrode encircled a skull screw and was in contact with brain tissue through another hole drilled next to the anchoring skull screw . The multiwire bundle was attached to a Microstar head stage and connected to a 16-channel data acquisition system ( AlphaLab; Alpha Omega; Nazareth Illit , Israel ) . Accurate placement was aided by recording neuronal activity through the multiwire bundle during the implantation procedure . After detecting LC activity , the multiwire bundle was affixed to the skull and screws with dental cement . The scalp wound was sutured closed . Surgery was similar for targeting of the electrode into the PMC , which is just rostral and medial to the ventral LC . As for the LC , the PMC was first localized with a glass micropipette as previously described . First the LC was localized and then the micropipette was repositioned rostrally until LC activity was no longer encountered . The multiwire bundle was then repositioned 200 μm medial to this point . Rats had an additional depth electrode ( tungsten microelectrode , 250 μm diameter ) implanted into the medial prefrontal cortex ( mPFC ) ( +3 . 2 AP , −0 . 6 ML , −3 . 0 DV ) for mPFC local field potential ( LFP ) recordings . Animals were allowed 3 days to recover before bladder catheter implantation . At least 2 days after the electrode implantation , rats were anesthetized with isoflurane and a small cut was made between the scapulae to provide an exit for the bladder catheter . A midline incision was made in the abdomen to access the bladder . A 5-French umbilical artery catheter was tunneled subcutaneously from the hole between the scapulae to the abdomen . The catheter end , which had been previously cauterized and flared , was brought intraperitoneally and inserted into the bladder dome and sutured above the flare . The exposed catheter end was connected to a port that allowed infusion of saline into the bladder . Three days after the bladder catheter implantation rats were placed into a cystometry chamber ( Med Associates , St . Albans , CT ) . Sterile saline was continuously infused ( 100 μl/min ) through the bladder catheter while urodynamic endpoints , including bladder pressure and the timing of micturition were recorded on-line for 1 hr using cystometry equipment ( Medical Associates , St . Albans , VT ) and software ( Cystometry Analysis Software , SOF-522 , Catamount R and D , St . Albans , VT ) . Neuronal activity was recorded from PMC neurons and/or LC neurons and LFPs were recorded simultaneously through micturition cycles . Figure 1A illustrates points along the pontine-cortical micturition circuit at which recordings were taken . Single unit LC waveforms were discriminated and sorted using the WaveMark template-matching algorithm in Spike2 ( Cambridge Electronic Design , CED , v7 . 09 , Cambridge , England ) , as described previously ( Curtis et al . , 2012 ) . Local field potentials ( LFPs ) from LC were obtained from one of the wires of the multiwire bundle ( sampled at 780 Hz , filtered from 1 to 150 Hz ) . Electrode recordings in the mPFC were amplified at a gain of 5000 and band pass filtered between 1–150 Hz . LC neuronal firing rate ( in Hz ) was estimated in 5 s time bins as the total number of spikes in each time bin divided by duration of time bin in seconds . For each neuron , the z-score was calculated by subtracting the mean average background firing rate from the firing rate and then divided by the standard deviation . The examples showing the spike rasters are accompanied by histograms showing the z-score firing rate in 1 s bins . The average across animals were estimated in 5 s bins . For burst analysis the following criteria were used: the maximum interspike interval defining burst onset was 80 ms , interspike intervals < 160 ms , number of spikes > 4 . To obtain time-frequency decomposition of the LFP signals , power spectral density was estimated using multitaper spectral estimators ( Chronux toolbox scripts with MATLAB; three tapers , moving windows of 2 s width and 0 . 5 s overlap , ±1 Hz bandwidth and displayed using interpolated smoothing between 0–20 Hz ) . At the end of the experiment , rats were anesthetized and current was passed through the electrode ( 10 mA , 15 s ) . Rats were transcardially perfused with 60 ml of 10% potassium ferrocyanide in 0 . 1 M phosphate buffered saline to form a Prussian blue reaction product for identification of the recording site . Frozen sections were cut on a cryostat and stained with neutral red for visualization of the Prussian blue labeled recording site ( Figure 1B ) . Data were used only from histologically identified cases . There was no removal of outliers .	How do we know when we need to find a bathroom ? As the bladder fills up , it sends signals to the brain to say that it needs emptying . But before the brain sends a message back to the bladder muscles telling them to contract to release urine , it first triggers a change in behavior . By increasing our alertness and arousing our senses , the brain ensures that we begin to look for a place where it is safe and appropriate to urinate . Only when we have found such a place will the brain tell the bladder to empty . Previous work has suggested that two brain regions play important roles in this process: the pontine micturition center ( PMC ) and its neighbor , the locus coeruleus . The PMC is thought to act as an on-off switch . When the bladder reaches a certain level of fullness the PMC activates , which tells the bladder muscles to contract . The locus coeruleus helps animals pay attention to important stimuli by making them more alert and energized whenever such stimuli are present . By recording the activity of neurons in the brains of rats while also measuring the pressure inside their bladders , Manohar et al . show that the PMC and the locus coeruleus work together to coordinate behavior and bladder emptying . Filling the bladder causes neurons in the locus coeruleus to activate in synchronized waves . This helps the locus coeruleus communicate with the brain’s outer layer , the cortex , leading to an increase in sensory alertness and arousal . This all happens before the bladder reaches the threshold fullness that activates the PMC , explaining why behavioral changes occur before urination . Manohar et al . show too that PMC neurons also activate when the rat is not urinating , suggesting that the PMC is more than an on-off switch . Healthy people experience the sensation of needing to empty their bladder well before the bladder is full , but people who do not receive these sensory signals may be unable to tell when they need to take action . This can lead to bedwetting in children and to incontinence in the elderly . Targeting the brain circuit that responds to bladder signals could lead to new treatments for these conditions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Brainstem network dynamics underlying the encoding of bladder information
Heterochromatin Protein 1 ( HP1 ) is a conserved chromosomal protein in eukaryotic cells that has a major role in directing heterochromatin formation , a process that requires co-transcriptional gene silencing mediated by small RNAs and their associated argonaute proteins . Heterochromatin formation requires erasing the active epigenetic mark , such as H3K4me2 , but the molecular link between HP1 and H3K4 demethylation remains unclear . In a fertility screen in female Drosophila , we identified ovaries absent ( ova ) , which functions in the stem cell niche , downstream of Piwi , to support germline stem cell differentiation . Moreover , ova acts as a suppressor of position effect variegation , and is required for silencing telomeric transposons in the germline . Biochemically , Ova acts to link the H3K4 demethylase dLsd1 to HP1a for local histone modifications . Therefore , our study provides a molecular connection between HP1a and local H3K4 demethylation during HP1a-mediated gene silencing that is required for ovary development , transposon silencing , and heterochromatin formation . Editorial note: This article has been through an editorial process in which the authors decide how to respond to the issues raised during peer review . The Reviewing Editor's assessment is that all the issues have been addressed ( see decision letter ) . In eukaryotic genomes , heterochromatin is mainly composed of repetitive sequences such as transposons that require active silencing ( Slotkin and Martienssen , 2007 ) . Heterochromatin is defined by the presence of repressive epigenetic methylation of histone H3 at lysine 9 ( H3K9me ) and by heterochromatin protein 1 ( HP1 ) , which binds to H3K9me sites ( Lachner et al . , 2001; Bannister et al . , 2001 ) . Heterochromatin formation is mediated by co-transcriptional gene silencing , a process that requires small RNAs and their associated argonaute proteins ( Martienssen and Moazed , 2015 ) . In Drosophila , the argonaute protein Piwi and Piwi-interacting RNAs ( piRNAs ) use base-pairing to target nascent transcripts to the corresponding transposon regions . The Piwi/piRNAs then recruit gene silencing machinery , including HP1a and the H3K9 methyltransferase Egg to form heterochromatin ( Yang and Xi , 2017; Czech and Hannon , 2016; Brower-Toland et al . , 2007 ) . The formation of heterochromatin also requires erasing of active epigenetic mark by the H3K4 demethylase dLsd1 ( Rudolph et al . , 2007 ) , but the molecular link between HP1a and local H3k4 demethylation remains elusive . Piwi , a founding member of the piRNA pathway in Drosophila , was initially identified as a fertility factor; its mutation results in germline degeneration and sterility ( Cox et al . , 1998; Lin and Spradling , 1997 ) . To identify new genes involved in Piwi/piRNA-mediated gene silencing , we here conducted a female fertility screen by EMS mutagenesis and identified a novel recessive mutation on the second chromosome . The homozygous mutant males are semi-lethal ( Supplementary file 1 , Table 1 ) , but females are viable but do not lay any eggs; other than sterility , these females do not have other notable defects . Dissection revealed that these females had rudimentary ovaries: rather than a normal ovary , each oviduct in these mutant females was connected to only a tiny mass of cells ( Figure 1a , b ) . Given this nearly ‘ovaryless’ phenotype , we named the gene associated with this mutation as ovaries absent ( ova ) and named this mutant allele ova1 . Complementation mapping with deficiency lines , followed by sequencing of candidate genes led us to identify a single nucleotide deletion in an exon of CG5694 , which results in a truncated protein of 387 amino acids ( aa ) rather than the predicted 623 aa full length protein ( Figure 1—figure supplement 1a , b ) . CG5694 encodes a protein with no obvious sequence similarity to any existing proteins in the NCBI database , but does have a conserved nuclear respiratory factor-1 ( NRF-1 ) - like domain at its N-terminus ( 15–105 aa ) ; this DNA-binding domain was initially identified in the mammalian transcription factor NRF-1 and are known to occur in at least one other Drosophila transcription factor , Erect Wing ( Ewg ) ( Figure 1—figure supplement 1c , d ) . We used CRISPR-Cas9 to generate a knock-out allele in which the entire coding region of CG5694 was deleted ( Figure 1—figure supplement 1e ) . Homozygous knock-out allele females are sterile and exhibit virtually identical ‘ovaryless’ phenotypes as the ova1 females ( Figure 1a ) . Additionally , transgenic expression of a genomic DNA fragment containing the ova gene region was able to effectively rescue the ovary defect and restore fertility of ova1 homozygous or CG5694 null females ( Figure 1a , b ) . Therefore , ova is allelic to CG5694 . Normally , oogenesis initiates in the germarium , an anterior part of the ovariole where germline stem cells ( GSCs ) reside . Each germarium normally harbors 2–3 GSCs that can be distinguished by spherically-shaped spectrosome and by their direct contact with the cap cell niche ( Figure 1c ) . The decedents of GSCs move posteriorly as they differentiate into germline cyst , and then bud off from the germarium to form egg chambers ( Xie , 2013; Spradling , 1993 ) . Immunostaining of ova1 homozygous and ova1/4 trans-heterozygous ovaries revealed that the mutant ovaries completely lacked vitellaria , and the germaria were either full of GSC-like cells [77% ( n = 117 ) of ova1 germaria] ( Figure 1c , d , g ) or entirely germless ( lacking Vasa expression ( Figure 1c , d , g ) , suggesting that ova is required for GSC differentiation and for germline survival . To determine whether ova functions cell-autonomously in the germline and/or non-cell-autonomously in somatic supporting cells to regulate GSCs , we conducted mosaic analysis by inducing mitotic clones using a FLP-FRT system ( Xu and Rubin , 1993 ) . Similar to wild-type control clones , ova1 mutant GSC clones behaved normally: the mutant GSCs were properly maintained in the niche , and their descendant cells were properly differentiated into germline cysts and egg chambers with properly specified oocytes ( Figure 1—figure supplement 2a–d ) , although germline mutant eggs failed to hatch ( Figure 1b ) . Similarly , germline-specific knocking down ova by UAS-Dcr2; nos-GAL4 ( thereafter referred as ova GLKD ) also did not cause any obvious defects in ovary morphology , and the number of GSCs and their immediate daughter cystoblasts ( collectively referred to as GSC-like cells ) per germarium remained largely normal ( Figure 1—figure supplement 2g , h ) . Collectively , these data demonstrate that ova is not cell-autonomously required for the early stages of GSC differentiation . We next used a temperature sensitive GAL4/UAS system ( Brand and Perrimon , 1993; McGuire et al . , 2004 ) to specifically deplete ova in somatic escort cells with c587-GAL4 ( c587 >ova RNAi ) ( Song et al . , 2004 ) . The somatic escort cells , which usually send out long protrusions that encapsulate the germline , are known to provide the niche environment required for germline cyst differentiation ( Kirilly et al . , 2011; Morris and Spradling , 2011 ) . After treatment at the restrictive temperature , c587 >ova RNAi germaria began to exhibit a significantly increased number of spectrosome-containing GSC-like cells in a time-dependent manner ( Figure 1e , g ) . The mutant escort cells were still able to send protrusions to the encapsulate the germline cells ( Figure 1e ) , indicating that the GSC differentiation defects is likely not caused by defects in escort cell morphology . The requirement for ova i in somatic escort cells for proper GSC differentiation was further supported by the observation that escort cell-specific expression of an ova transgene was sufficient to rescue the ovary defects of ova mutant females ( hereafter referred to as ova germline mutants ) ( Figure 1f ) . Therefore , ova functions in somatic escort cells and regulates germline differentiation in the germarium in a non-cell-autonomous manner . The ova mutant phenotype is reminiscent of the piwi mutant phenotypes: piwi mutant flies also have rudimentary ovaries that contain an abnormal number of differentiation-blocked GSC-like cells , and piwi also functions primarily in somatic escort cells to regulate GSC differentiation ( Jin et al . , 2013; Ma et al . , 2014 ) . Loss of piwi in escort cells causes de-repression of decapentaplegic ( dpp ) , a major self-renewal signal for GSCs , leading to GSC-like cell accumulation in the germarium ( Jin et al . , 2013; Ma et al . , 2014 ) . Interestingly , we found that the ova phenotype was also associated with increased dpp signaling . The extra GSC-like cells in c587 >ova RNAi germaria had dramatically increased expression of Dad-lacZ and phosphorylated Mad ( pMad ) ( Figure 1h , i ) , both of which are reporters of BMP pathway activity , and decreased expression of bam ( Figure 1h ) , a gene whose expression is normally suppressed by BMP signaling ( Song et al . , 2004; Chen and McKearin , 2003 ) . Collectively , these data suggest that loss of ova in escort cells leads to ectopic dpp signaling that blocks further GSC differentiation , leading to GSC-like cell accumulation in the germarium . The phenotypic and molecular similarities between the ova and piwi mutants led us to further test whether Ova and Piwi act via the same genetic pathway to regulate GSCs . As expected , we observed that homozygous piwi mutant females had rudimentary ovaries . As a positive control , escort cell-specific expression of a piwi transgene was sufficient to rescue the piwi mutant ovary phenotype ( Figure 2a , b ) . On the one hand , transgenic expression of ova in escort cells of piwi mutants also partially rescued the ovary morphology phenotype with the frequent appearance of developing germline cysts , including late stages of egg chambers , although oogenesis was still abnormal , GSC-like tumor still remained , and no mature eggs were produced ( Figure 2a , b , c ) . On the other hand , transgenic expression of piwi in escort cells of ova mutants could not rescue any ovary phenotypes ( Figure 2a , b ) . Consistent with previous reports of piwi phenotypes , escort cell-specific knocking down of other Piwi/piRNA pathway effectors , such as panx , also showed a similar GSC-l accumulation phenotype ( Figure 2d , e ) , further supporting the idea that Ova may participate in the same Piwi/piRNA pathway . Next , we tested whether overexpression of ova could rescue the germline TE upregulation phenotype caused by piwi mutation . As a control , ubiquitous expression of piwi , but not soma-only expression of piwi was able to effectively rescue the TE upregulation phenotype . However , neither ubiquitous nor soma-specific expression of ova could rescue the TE upregulation phenotype in piwi mutants ( Figure 2f ) . These observations suggest that , genetically , ova acts downstream of piwi , but there must be additional factors downstream of piwi that cooperatively function with ova to regulate GSC differentiation and transposon silencing . Given that Piwi is associated with a number of chromatin factors that are known to regulate heterochromatin formation and germline transposon silencing , and considering that dpp silencing in escort cells can be attributed to Piwi-dependent gene silencing , we asked whether Ova is also associated with these silencing machinery components and somehow participates in these processes . We performed a yeast two-hybrid ( Y2H ) screen for potential physical interactions among Ova and other known essential components of the heterochromatin silencing machinery ( Yu et al . , 2015; Sienski et al . , 2015; Sienski et al . , 2012 ) , including: Panoramix ( Panx ) , Arx , and Mael , which participates in Piwi/piRNA-mediated gene silencing ( Yu et al . , 2015; Sienski et al . , 2015; Sienski et al . , 2012; Muerdter et al . , 2013; Dönertas et al . , 2013; Ohtani et al . , 2013 ) ; HP1a , the H3K9me3 methyltransferase Eggless ( Egg ) , and the Egg cofactor Windei ( Wde ) ( Seum et al . , 2007; Tzeng et al . , 2007; Koch et al . , 2009 ) ; the H3K4me2 demethylase dLsd1 and its cofactor CoREST ( Rudolph et al . , 2007 ) ; and Piwi . The Y2H screen identified two positive interactions: Ova and HP1a , and Ova and dLsd1 ( Figure 3a and Figure 3—figure supplement 1 ) . Notably , the previously reported interaction between HP1a and Piwi was not observed in our screen here ( Figure 3—figure supplement 1 ) ( Brower-Toland et al . , 2007 ) , possibly due to different expression systems used in the studies . Co-immunoprecipitation experiments also showed positive interactions between Ova and HP1a and between Ova and dLsd1 in ovary lysates ( Figure 3b , c ) . Collectively , these results indicate that Ova is physically associated with the co-transcriptional silencing machinery and directly interacts with HP1a and with dLsd1 . Previous studies have reported that HP1a and dLsd1 function in the escort cell niche to restrict dpp signaling and to facilitate GSC differentiation ( Wang et al . , 2011; Eliazer et al . , 2011 ) . These reports , considered alongside the known role of Piwi-dependent gene silencing of the dpp gene locus in normal escort cells , further supporting the notion that these three Piwi-associated factors ( Ova , HP1a , dLsd1 ) function in a shared pathway in escort cells to establish a repressive chromatin state for the dpp gene locus . We next tested whether Ova , similar to HP1a and dLsd1 ( Wang and Elgin , 2011; Czech et al . , 2013 ) , is required for heterochromatin formation and germline transposon silencing . The white locus of In ( l ) wm4h chromosomal reversion flies is relocated to a position next to a heterochromatin region , and this relocation often causes heterochromatin-based silencing of this gene , resulting from a genomic phenomenon referred to as position effect variegation ( PEV ) , these flies typically display mosaic eyes with red and white facets as a result of this relocation based silencing ( Wallrath and Elgin , 1995; Schotta et al . , 2003 ) ( Figure 3—figure supplement 2a ) . Interestingly , removing one functional copy of ova from the In ( l ) wm4h background was sufficient cause fully-pigmented eyes ( Figure 3—figure supplement 2a ) . Analysis using several additional PEV reporter fly lines ( 118E-10 , 118E-15 , 39 C-72 , and 6 M-193 ) , each of which has its white gene locus relocated ( inserted ) into the heterochromatin rich fourth chromosome , showed that ova acts as a suppressor of PEV: the ova transheterozygous flies had fully-pigmented eyes with increased pigment level whereas the ova heterozygous flies from all three of the reporter lines had mosaic eyes ( P values by two-tailed Student t-test , Figure 3—figure supplement 2b , c ) . It thus appears that ova has a functional role in heterochromatic gene silencing . To test whether or not ova functions in germline transposon silencing , we performed germline-specific knock-down of ova using the UAS-Dcr2; nos-GAL4 driver ( ova GLKD ) , followed by RNA-seq analysis . Interestingly , nos > ova RNAi ovaries had dramatically up-regulated transcripts of a subset of transposons that included the LTR element 412 and the telomeric non-LTR repeats Het-A , TAHRE , Tart ( Figure 3d ) . By comparison , the expression of protein-coding genes and piRNAs was largely un-altered ( Figure 3—figure supplement 3a , b , e ) . Somatic cell-specific knock-down of ova ( tj-GAL4 >ova RNAi ) only caused mild , if any , TE upregulation ( Figure 3—figure supplement 3f ) . Consistent with a role in germline transposon silencing , a previously reported genetic screen for genes involved in germline transposon silencing identified ova ( CG5694 ) as one of the top hits ( Muerdter et al . , 2013; Czech et al . , 2013 ) . Notably , germline-specifc knock-down of either ova ( ova GLKD ) or dLsd1 ( dLsd1 GLKD ) exhibited de-repression of a similar subset of transposons ( R = 0 . 9463 by Pearson’s correlation coefficient , Figure 3e ) ; this subset is distinguished by enrichment for bivalent histone marks ( both H3K9me3 and H3K4me2 ) ( Czech et al . , 2013; Klenov et al . , 2014 ) . In mutants deficient in piRNA biogenesis , the inability to form Piwi/piRNA complexes typically results in retention of Piwi in the cytoplasm ( Wang and Elgin , 2011; Malone et al . , 2009; Olivieri et al . , 2010 ) . The fact that nuclear Piwi localization was largely unaffected in the ova mutant germline ( Figure 3—figure supplement 3c ) further supports our conclusion that ova is not required for piRNA biogenesis , but may function at the chromatin to mediate Piwi/piRNAs- induced transcriptional gene silencing , a phase that has been referred to as ‘effector step’ ( Czech et al . , 2013 ) . To explore the biochemical mechanisms underlying Ova function in greater detail , we used Y2H assays to identify the Ova domains required for its interactions with HP1a and/or dLsd1 . We constructed multiple truncated forms of Ova ( Figure 4a ) , and found that the Ova 250–486 fragment and the Ova 388–623 fragment were both able to interact with the chromo shadow domain ( CSD ) of HP1a ( Figure 4b ) ; neither of these Ova fragments could interact with the chromodomain ( CD ) of HP1a ( Figure 4b ) . We next constructed an Ova fragment composed of the overlapped 388–486 region and confirmed that this fragment was sufficient for interaction with the CSD domain of HP1a ( Figure 4b ) . Mapping the interaction domains of Ova with dLsd1 revealed that both Ova 1–388 and Ova 250–486 fragments , but not Ova 388–486 fragment , could interact with dLsd1 ( Figure 4c ) . Interestingly , transgene expression of the Ova 250–486 fragment , which is able to interact with both HP1a and dLsd1 , was sufficient to rescue both the ovary development defect and transposon silencing defect of ova mutant females , similar to the effect produced by transgene expression of a full length ova ( Figure 4d , f ) . In contrast , no rescue effect was observed with the transgenic expression of the Ova 388–623 fragment , which interacts with HP1a only , or with expression of the Ova 1–388 fragment , which interacts with dLsd1 only ( Figure 4d ) . Therefore , the domain that is sufficient to interact with both HP1a and dLsd1 is sufficient for Ova function in ovary development and transposon silencing . These biochemical and genetic experiments indicate that Ova may serve as a protein adaptor that links HP1a and dLsd1 . To functionally test this putative adaptor function in vivo , we generated a transgene expressing HP1a::dLsd1 fusion protein . If Ova merely functions as an adapter that bridges the two proteins , the HP1a::dLsd1 transgene should render Ova dispensable and therefore should be able to rescue the ova mutant phenotypes . Strikingly , transgenic expression of HP1a::dLsd1 in escort cells was sufficient to rescue the rudimentary ovary phenotype of ova mutants ( Figure 4e ) . Eighty percent of the HP1a::dLsd1 rescued germaria contained 2–5 GSC-l ( n = 41 ) and all the germaria had properly differentiating cysts . Moreover , ubiquitous expression of HP1a::dLsd1 also significantly rescued the transposon silencing defects of ova mutants and partially restored female fertility ( Figure 4f ) . Given that the genomic fragment transgene of ova ( ova-g ) , which includes the cis-elements of ova , could fully restore fertility ( Figure 1a ) , the incomplete rescue of fertility by the HP1a::dLsd1 fusion could be due to non-physiological levels of the transgene expression . Alternatively , ova could have additional roles beyond the adaptor role that are important for female fertility . In addition to increased expression of transposons , ova germline mutant ovaries also showed moderate upregulation of many protein-coding genes ( Figure 4—figure supplement 1 ) . Interestingly , this transgene expression also effectively brought the expression of many protein-coding genes back to wild-type levels ( Figure 4—figure supplement 1 ) . These observations indicate that HP1a and Ova may participate in transcriptional silencing of many regular protein-coding genes , in addition to transposons . We conclude that Ova acts as a protein adaptor to link HP1a and dLsd1 to promote HP1a-mediated gene silencing . Since dLsd1 catalyzes H3K4me2 demethylation , Ova may function to link dLsd1 and HP1a for local H3K4 demethylation during heterochromatic gene silencing . Indeed , ChIP-seq analysis revealed that the H3K4me2 density was specifically increased at Het-A and TAHRE loci but not other TE loci ( Figure 5—figure supplement 1 ) . Further analysis revealed that there was a significant increase in H3K4me2 levels and in RNA Pol II occupancy at the 3’UTR of the Het-A and TAHRE transposons in ova GLKD ovarian germline cells ( Figure 5a , b ) . Note that these telomeric transposons are arranged in a head-to-tail fashion; therefore , the 3’ UTR of one element likely directs the transcription of its downstream neighbor ( Danilevskaya et al . , 1997 ) . To further test this potential role of Ova in linking H3K4 demethylation during HP1a-mediated gene silencing in vivo , we used a clean lacI/lacO reporter system to tether lacI-HP1a to the promoter of a lacO-GFP reporter ( Sienski et al . , 2015 ) . We found that 26% of ovarioles examined ( n = 131 ) had reduced GFP signal in their germline upon lacI-HP1a induction ( Figure 5c , d ) , although there was no significant reduced in the overall level of GFP mRNA ( P value by two-tailed Student t-test , Figure 5e ) . Importantly , co-expression of ova in the germline caused a significant increase in the number of ovarioles with reduced or abolished GFP signal [86% ( n = 138 ) ] , and the overall GFP mRNA level was also significantly reduced in these samples ( P value by two-tailed Student t-test , Figure 5c–e ) . ChIP-seq analysis showed that this reduction in reporter expression was accompanied by significantly reduced H3K4m2 levels near the promoter region of the GFP gene reporter ( Figure 5f , h ) . To further confirm that the alteration of H3K4 deposition is a consequence of Ova recruitment , rather than a secondary effect following altered gene transcription , we performed a similar set of experiments , but with a modified lacO-terminator-GFP reporter that has a transcriptional terminator immediately following the promoter ( Figure 5g ) . This should result in blocked transcription no matter whether a transcriptional activator/repressor is present or not . As expected , this reporter showed a significant reduction of baseline transcription ( down to approximately 3 . 8% ) ( P value by two-tailed Student t-test , Figure 5i ) . We found that tethering lacI-HP1a to the promoter failed to alter the H3K4me2 level proximal to the tethering site . Co-expression of Ova , however , almost erased entirely the H3K4me2 marks in the proximal region ( Figure 5g , h ) . These observations further support the notion that Ova links HP1a and dLsd1 for local erasing of H3K4me2 marks . We also tested whether Ova itself can induce gene silencing by tethering Ova directly to DNA and to mRNA using in vivo reporter systems in the ovarian germline . We used a lacI-Ova and lacO-GFP binary system to tether Ova to genomic DNA ( Sienski et al . , 2015 ) and found that such tethering did not have any obvious effect on GFP expression ( Figure 5—figure supplement 2a–c ) . Similarly , tethering Ova to mRNA using a λN-Ova and GFP-boxB binary system did not cause any obvious effect on GFP expression ( Figure 5—figure supplement 2d–f ) . These results are consistent with the idea that Ova acts downstream of HP1a in heterochromatic gene silencing . Similar to other ‘effector step’ mutations , the loss of ova or dlsd1 only causes de-repression of a subset of transposons; this is in contrast with the widespread transposon de-repression that is common in mutations affecting piRNA biogenesis ( Czech et al . , 2013 ) . This disparity can possibly be explained by the existence of different silencing mechanisms for particular subsets of transposons . Illustrating this idea , our work suggests that transposons with bivalent histone marks may be preferential targets for Ova and dLsd1 . A bivalent pattern of histone methylation may help to regulate the expression of transposons that require a delicate On/Off balance , for example with the expression of telomeric repeats known to be required for normal telomere function ( e . g . , Het-A , TAHRE , and Tart ) . The results of our study establishes that Ova has an indispensable role in facilitating dLsd1’s H3K4 demethylation activity during HP1a-induced heterochromatic gene silencing and demonstrates that this Ova function is essential for germline development , heterochromatin formation , and Piwi/piRNA-mediated co-transcriptional gene silencing . Our study suggests that the Piwi/piRNA pathway may adapt a similar effector machinery to repress regular genes , such as the dpp gene in escort cells , in addition to TEs . A study in S . pombe reported a mechanism in which a RNAi protein complex links the activity of the H3K9 methyltransferase Clr4 with H3K4 demethylation by the H3K4 demethylase Lid2 ( Li et al . , 2008 ) , indicating an evolutionarily conserved interplay of epigenetic marks during transcriptional gene silencing . Given that the mechanisms underlying heterochromatic gene silencing are known to be strongly conserved from Drosophila to mammals , an equivalent functional module that links HP1a with H3K4 demethylation likely exists in mammals as well . Flies were cultured on standard media with yeast paste added to the food surface . The culture temperature was 25°C unless otherwise noted . Strains used in this study were as follows: ova1 is nucleotide loss allele ( A1045 ) generated in this study . ova4 is a knock-out allele generated in this study by CRISPR-Cas9 ( Ren et al . , 2013 ) . c587-GAL4 ( Song et al . , 2004 ) ; Dad-lacZ ( Tsuneizumi et al . , 1997 ) ; bam-GFP ( Chen and McKearin , 2003 ) ; piwi ( Lachner et al . , 2001 ) and piwi ( Bannister et al . , 2001 ) ( Lin and Spradling , 1997 ) ; GFP-piwi ( gift from Katalin Toth , California Institute of Technology ) ; 118E-10 , 118E-15 , 6 M-193 , and 39C . 72 ( gift from Lori Wallrath , University of Iowa ) ; dLsd1-GFP ( gift from Yang Yu , Institute of Biophysics IBP , Chinese Academy of Sciences ) ; from Bloomington Drosophila Stock Center ( BDSC ) :EGFP-RNAi ( #41553 ) RFP-HP1a ( #30562 ) ;; UAS-Dcr2; nos-GAL4 ( #25751 ) ; tub-GAL4 ( #5138 ) ; tub-GAL80ts ( #7016 , #7018 ) ; Df ( 2L ) BSC144 ( #9504 ) ; attP2 ( #25710 ) ; from Kyoto Stock Center: In ( 1 ) wm4h ( #101652 ) ; Df ( 2L ) ED737 ( #150520 ) ; from Vienna Drosophila Research Center: ova-RNAi ( #102156 ) ; piwi-RNAi ( #101658 ) ; panx-RNAi ( #102702 ) ; EGFP-5xBoxB ( #313408 ) ; lacO-GFP-Piwi ( #313394 ) ; lacI-HP1a; lacO-GFP-Piwi ( #313409 ) . To obtain ova knock-out allele , two gRNAs ( gRNA1: aagtctttacagccttgatc and gRNA2: cgttgggttgaggtacatac ) were designed that target ova 5’UTR and 3’UTR respectively and cloned into U6b vector . The plasmids were introduced into nos-Cas9 embryos ( Ren et al . , 2013 ) . Obtained flies were backcrossed with w1118 for at least three generations to eliminate potential off-target events . For ova-g transgenic fly , w1118 genomic region ( 2L: 10226867–10234857 ) was cloned intro pCasper4 vector . The attP-UASP vector was used to generate UASP-Flag-ova , UASP-Flag-ova1-388 , UASP-Flag-ova1-249 , UASP-Flag-ova250-486 , UASP-Flag-ova-388–623 , UASP-ova , UASP-piwi , and UASP-HP1a::dLsd1 . The GFP-ova construct was obtained using Gateway cloning technology ( Invitrogen ) and pUGW ( DGRC1283 ) vector . Ova cDNA was cloned into UASP-λN and UASP-lacI ( gifts from Julius Brennecke , Institute of Molecular Biotechnology ) to generate the UASP-λN-ova and UASP-lacI-ova transgenes respectively . For the lacO-terminator-GFP reporter , 555 bp VASA terminator was injected immediately following start codon of GFP in the lacO-GFP reporter . All the plasmids were purified using a Qiagen Plasmid Midi Kit ( #12145 ) and the DNA sequencing verified plasmids were introduced into embryos using either P-element or nos-phiC31 system to generate transgenic flies according to a standard procedure . Drosophila ovaries were dissected and immunostained as described previously ( Yang et al . , 2015 ) . Briefly , ovaries were fixed in 4% paraformaldehyde for 15 min , and blocked in 5% normal goat serum in PBT ( 10 mM NaH2PO4 , 175 mM NaCl , pH 7 . 4 , 0 . 1% Triton X-100 ) . The following primary antibodies were used: rabbit anti-pMad ( 1:1000 , gift from Ed Laufer , Columbia University Medical Center , New York ) , rabbit anti-β-galactosidase ( 1:3000; MP Biologicals , 0855976 ) , mouse anti-α-Spectrin ( 1:50; DSHB ) , rabbit anti-GFP ( 1:1000; Life , A11122 ) , mouse anti-Flag ( 1:300; Sigma , F1804 ) . Secondary antibodies , including goat anti-rabbit , anti-mouse IgGs , conjugated to Alexa ( 488 or 568 ) ( Molecular Probes ) were used at a dilution of 1:300 and tissues were also stained with 0 . 1 mg/ml DAPI ( 4’ , 6’-diamidino-2-phenylindole; Sigma ) for 5 min . Images were collected using either a Zeiss LSM510/LSM 800 or Nikon A1 confocal microscope system . All acquired images were processed in Adobe Photoshop and Illustrator . To test female fertility , for each vial , three newly enclosed females were collected and mated with three 5–7 days old w1118 males in cornmeal food with yeast paste for two days , then the flies were transferred to a cornmeal food vial without yeast paste . After another three days , the flies were dumped out . The number of offspring was accounted until 16 days after eclosion . Mean values are reported as SEM . To measure eye pigmentation , the heads of ten 5–7 days old flies of each genotype were manually dissected . The isolated heads were homogenized in 0 . 2 ml of methanol , acidified with 0 . 1% HCl and warmed at 50°C for 5 min; The homogenate was clarified by centrifugation , and the OD at 480 nm of 0 . 15 ml supernatant was recorded . Mean values are reported with SEM . Yeast Two-hybrid experiment was performed as described previously ( Yang et al . , 2015 ) . Briefly , cDNA encoding interesting genes were amplified from w1118 ovary cDNA and cloned into either pGBKT7 bait vector or pGAD prey vector ( Clontech ) . The pGBKT7 and pGAD plasmid carrying interesting genes were co-transformed into AH109 yeast cells according to a standard procedure . Colonies appearing on media lacking tryptophan and leucine ( SC-WL ) were picked onto selection plate lacking tryptophan , leucine and histidine ( SC-WLH ) or tryptophan , leucine , histidine and adenine ( SC-WLHA ) to determine proteins interaction . Co-IP was done as previously described ( Yang et al . , 2015 ) , with minor modifications . Female flies of appropriate genotypes were dissected in ice cold PBS . Ovaries were lysed in lysis buffer ( 10 mM Hepes pH 7 . 0 , 150 mM NaCl , 5 mM MgCl2 , 10% glycerol , 1% Triton X-100 , 1x complete protease inhibitor ( Roche ) , 1 mM DTT , 1 mM EDTA , 0 . 1 mM PMSF ) at 4°C for 30 min and spun for 10 min at max speed in a table top centrifuge at 4°C . The supernatant was incubated with tag-recognizing beads including anti-Flag resin ( Sigma ) , GFP-Trap agarose beads ( Chromoteck ) and RFP-Trap agarose beads ( Chromoteck ) . After incubation , the beads were washed three times with lysis buffer and eluted by boiling in SDS loading buffer , loaded onto SDS-PAGE gels , and analyzed by immunoblotting with indicated antibodies . The following primary antibodies were used: anti-Flag ( Sigma , 1:6000 ) , anti-GFP ( Life , 1:10000 ) , anti-mCherry ( BioVision , 1:2000 ) , anti-Tubulin ( DSHB , 1:2000 ) . Secondary antibodies , including: anti-mouse and anti-rabbit IgG-HRP ( ZSJQ-BIO , 1:10000 ) . The membrane was developed by Immobilon Western Chemiluminescent HRP Substrate Kit ( Millipore ) according to the manufacturer’s instructions . Total RNA from 10 to 20 ovaries was extracted using TRIzol reagent ( TaKaRa ) . After DNase treatment , complementary DNA ( cDNA ) was synthesized using HiScript II Q RT SuperMix ( Vazyme Biotech , R223-01 ) . RT-qPCR was performed in three duplicates using ChamQ SYBR qPCR master Mix ( Vazyme Biotech , Q331 ) on an ABI PRISM 7500 fast real-time PCR system ( Applied Biosystems ) . Fold changes for mRNA were calculated using the △△Ct method ( Livak and Schmittgen , 2001 ) . Primers used were shown in Supplementary file 1 , Table 2 . Total RNA from ovaries was isolated using TRIzol reagent ( TaKaRa ) . 10 μg of total RNA from each sample used for library preparation after poly ( A ) -containing mRNA molecule purification ( NEB , #S1419S ) , RNA amplification , double-strand cDNA synthesis , and adaptor ligation ( NEB , #E7645S ) . For the small RNA sequencing , 10 μg enriched small RNA were separated on a 15% denaturing polyacrylamide gel and 18- to 30-nt RNAs were purified according to RNA oligo markers . All the libraries were prepared by following the manufacturer’s instructions and subsequent sequencing on the Illumina GAII instrument ( Vazyme , NR801 ) . For CDS gene expression analysis , all the sequencing reads were mapped to the D . mel genome ( BDGP6 ) using STAR program ( options: --outFilterMultimapNmax 20 --alignIntronMin 20 --alignIntronMax 500000 ) . The mapped reads were used for expression analysis via Cufflinks package with reference gene annotation from Ensembl . And Cuffdiff was used to perform differential expression . For transposon expression analysis , sequencing reads were mapped to the transposon sequences which download from flybase website using STAR program with default parameters . Then alignment reads were used for calculating the expression level of transposons . Different transposons were combined together if they belong to the same one . The expression levels were normalized to reads per million ( RPM ) . For small RNA analysis , Cutadapt package was used to remove adapter from 3’ end . The reads were aligned to the genome sequence by Bowtie . The reads were discarded which mapped to rRNA , tRNA , snoRNA sequences . And retained reads were aligned to miRNA ( pre-miRNA sequences download from miRBase ) and whole genome sequences ( r5 . 42 ) with one mismatch and unique hit . Sequences in the 25–32 nt size range , not annotated as a previously known RNA were classified as candidate piRNAs . The expression levels of small RNA were normalized to RPM according to the total mapped reads number . ChIP was performed as previously described ( Sienski et al . , 2012 ) . Briefly , about 200 pairs of ovaries were dissected into cold PBS and washed once . Ovaries were cross-linked in 1 . 8% paraformaldehyde for 10 min at room temperature then quenched with glycine . Ovaries were homogenized by douncing . Pellet was resuspended in lysis buffer and incubated 10 min on ice . Chromatin was sonicated for immunoprecipitation and followed by reverse crosslink and DNA purification . Recovered DNA fragment was used to prepare libraries using VAHTS Universal DNA Library Prep Kit ( Vazyme , ND607 ) and TruePrep Index Kit ( Vazyme Biotech , TD202 ) sequencing was done on HiSeq2500 ( Illumina ) . Antibodies: polyclonal rabbit anti-H3K4me2 ( Abcam , ab7766 ) and monoclonal mouse anti-RNA polymerase II ( Abcam , ab817 ) . ChIP-seq reads were aligned using Bowtie ( version 1 . 1 . 2 ) to build version BDGP6 of the Drosophila melanogaster genome . MACS ( version 1 . 4 . 1 ) was used to identify regions of ChIP-seq enrichment . The density of reads in each region was normalized to 10 million reads library size . For lacO-GFP ChIP-seq , normalized reads were removed w1118 ChIP reads as the reporter unique mapped reads due to lacO-GFP reporter shared common sequences in fly genome . BigWig files were generated for visualization using Homer package . For transposons , all raw reads were mapped to the transposon database using Bowtie ( version 1 . 1 . 2 ) with –v 3 –-best parameters . The sum of the number reads that mapped to genome and transposon was used as a normalization factor for all samples , reporting all feature abundances as RPM mapped .	The complete set of genetic material within a cell is known as a genome . The genomes of human and other animal cells have regions of active genes interspersed with ‘dark’ regions known as heterochromatin , which contain genes and other types of genetic material that have been inactivated . Heterochromatin commonly contains sections of genetic material known as transposons . When a transposon is active it is able to move around the genome , therefore , inactivating ( or ‘silencing’ ) transposons helps to maintain the integrity of the genetic material in a cell . It is particularly important to silence transposons in the stem cells that produce sperm and egg cells – known as germline stem cells – to ensure genetic information is faithfully passed on to the next generation . A protein called HP1a plays a major role in directing where heterochromatin forms in the genome . This process requires an enzyme called dLsd1 to remove a small tag from the genetic material but it is not clear how HP1a regulates the activity of dLsd1 . To address this question , Yang et al . studied how egg cells form in fruit flies , which are often used as models of animal biology in experiments . The team screened a population of fruit flies that carried mutations in many different genes to identify genes that affect the fertility of female flies . This revealed a gene named as ovaries absent ( or ova for short ) is required for egg cells to form . In germline stem cells ova silences transposons and in the surrounding tissue it represses a specific signal that usually maintains stem cells to allow the stem cells to divide to make egg cells . Further experiments using biochemical techniques found that the protein encoded by ova acts as a bridge to bring HP1a and dLsd1 together to silence genes in heterochromatin . The next step would be to identify the functional counterpart of the ova gene in mammals , including humans , which may help to discover causes of infertility and develop new fertility treatment .	[ "Abstract", "Introduction", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine", "research", "communication" ]	2019	Ovaries absent links dLsd1 to HP1a for local H3K4 demethylation required for heterochromatic gene silencing
Events in early life contribute to subsequent risk of asthma; however , the causes and trajectories of childhood wheeze are heterogeneous and do not always result in asthma . Similarly , not all atopic individuals develop wheeze , and vice versa . The reasons for these differences are unclear . Using unsupervised model-based cluster analysis , we identified latent clusters within a prospective birth cohort with deep immunological and respiratory phenotyping . We characterised each cluster in terms of immunological profile and disease risk , and replicated our results in external cohorts from the UK and USA . We discovered three distinct trajectories , one of which is a high-risk ‘atopic’ cluster with increased propensity for allergic diseases throughout childhood . Atopy contributes varyingly to later wheeze depending on cluster membership . Our findings demonstrate the utility of unsupervised analysis in elucidating heterogeneity in asthma pathogenesis and provide a foundation for improving management and prevention of childhood asthma . Asthma is a global health problem , and there is a pressing need for better understanding of its pathogenesis ( Global Initiative for Asthma , 2015 ) . Asthma is strongly associated with allergy , and both genetic and environmental factors may be involved ( Ober and Yao , 2011; Dick et al . , 2014 ) . The ‘hygiene hypothesis’ proposes that modern changes to hygiene , sanitation and living environment have modified human exposures to microbes , with subsequent effects on early-life immune development ( Okada et al . , 2010 ) . However , the clinical presentation and prognosis of childhood wheeze is highly variable: some children remit; others remit but relapse; and yet others have wheeze persisting into adult asthma ( Morgan et al . , 2005 ) . These differences suggest that the underlying causes of disease also differ from person to person . For example , while asthma is commonly linked to allergy , not all individuals with wheeze are sensitised to allergen , and vice versa ( Spycher et al . , 2010 ) . As such , childhood asthma is a heterogeneous condition ( Hekking and Bel , 2014; Wenzel , 2012 ) , and this greatly complicates the study of its pathogenesis ( Anderson , 2008 ) . We postulate that there are subpopulations in early childhood , each sharing similar patterns of pathophysiology , disease susceptibility and phenotype that permit categorisation into clusters . If we can agnostically identify these clusters , then we may explore the biological mechanisms that underlie them , and find targets for early intervention that are specific for different asthma subtypes . Previous attempts at subtyping asthma susceptibility relied on supervised classification , using expert knowledge and cut-offs to define clusters . For example , criteria such as – specific immunoglobulin E ( IgE ) ≥0 . 35 kU/L; wheal diameter ≥3 mm in a skin prick test ( SPT ) ; or symptom score surpassing a threshold –may determine classification into a high-risk profile ( Castro-Rodríguez et al . , 2000; Frith et al . , 2011 ) . However , these cut-offs vary with age , gender or other parameters , and may not accurately reflect true attribution of risk ( Linden et al . , 2011 ) . Hence , they often continue to produce heterogeneous groups . Furthermore , previous studies tended to focus on a single ‘domain’ , for instance grouping only by immunological response ( Prescott et al . , 1999 ) , symptomatology or timing of disease ( Martinez et al . , 1995; Kurukulaaratchy et al . , 2003 ) . Recently , researchers have turned to unsupervised approaches , such as model-based cluster analysis and latent class analysis ( LCA ) ( Deliu et al . , 2016; Lazic et al . , 2013; Simpson et al . , 2010; Belgrave et al . , 2014; Belgrave et al . , 2013; Wu et al . , 2015 ) . These do not require experts to supply cut-offs , but can instead ‘learn’ boundaries from the data . They can potentially uncover patterns of similarity not immediately obvious to the human eye . Finally , these methods can cover a broader range of domains , incorporating measurements from multiple sources to determine clusters that are potentially informative of asthma risk . Here , we use a data-driven unsupervised framework together with a comprehensively phenotyped birth cohort , to define developmental trajectories during preschool years , a period known to be critical to asthma pathogenesis . Specifically , we ( 1 ) use non-parametric mixture models to discover latent clusters that define early childhood trajectories of immune function and susceptibility to respiratory infection; ( 2 ) investigate how these clusters relate to differential profiles of asthma susceptibility , and to existing definitions of atopy; ( 3 ) identify risk factors for asthma within each cluster; and ( 4 ) externally validate the clusters in independent cohorts . CAS1 was a low-risk cluster with infrequent and transient respiratory wheeze . Rates of wheeze declined from 33% at age 1% to 12% by age 10 ( Table 1; Figure 3 ) . In this cluster , Th2 cytokine responses of peripheral blood mononuclear cells ( PBMCs ) to allergen stimulation were minimal; and rates of allergen sensitisation ( as measured by IgE or skin prick test , SPT ) were the lowest among all groups ( Table 2; Figure 4; Supplementary file 1 – table supplement 3B-D ) . IgG and IgG4 were also low across all allergens . Frequency of respiratory infection in CAS1 was low ( Table 3 ) . However , high frequency of lower respiratory infections ( LRIs ) in childhood , especially wheezy LRIs ( wLRIs ) , was a risk factor for age-5 wheeze – even after adjusting for sex , body mass index ( BMI ) and parental history of asthma as demographic covariates ( Table 4 ) . Repeated-measures ANOVA identified that LRI and wLRI frequency in the first 3 years were predictors for age-5 wheeze ( Supplementary file 1 – table supplement 4 ) ; however , timepoint-specific analyses showed that differences were only noticeable from age 3 onwards ( Table 4; Figure 5A–B ) . A multiple regression model with stepwise elimination yielded three significant variables: age-three wLRI frequency ( odds ratio OR 5 . 6 per unit increase , p=0 . 0068 ) ; age-four LRI frequency ( OR 3 . 6 , p=0 . 018 ) ; and a protective effect from proportion of infection-associated microbiome profile groups ( MPGs; Streptococcus , Haemophilus , Moraxella ) in age-two-to-four healthy nasopharyngeal aspirate samples ( NPAs; OR 0 . 19 per quartile , p=0 . 014 ) . Similar to CAS1 , CAS2 was a low-risk cluster with infrequent allergic disease . Compared to CAS1 , Phadiatop and house dust mite ( HDM ) IgE were elevated at most timepoints ( Table 2; Figure 4A; Supplementary file 1 – table supplement 3B ) , with the exception of peanut IgE ( Wilcoxon , adjusted p=0 . 99 at all timepoints; Figure 4D ) . CAS2 IgG and IgG4 were intermediate between CAS1 and CAS3 levels; CAS2 IgG was closer to CAS1 , while CAS2 IgG4 was closer to CAS3 ( Table 2; Figure 4 ) . Despite these antibody differences , yearly rates of wheeze in CAS2 remained comparable to CAS1 ( 30% at age 1 , declining to 18% at age 10; Table 1; Figure 3 ) . Interestingly , compared to CAS1 , individuals in CAS2 had fewer older siblings living in the household at age 2 , as well as more frequent paternal history of asthma ( adjusted p=0 . 029 and 0 . 055 , respectively; Supplementary file 1 – table supplement 3A ) . Predictive factors for age-5 wheeze in CAS2 included: LRI , wLRI and febrile LRI ( fLRI ) frequency ( GLM; p=2 . 7 × 10−3 , 0 . 016 and 0 . 02 at age 3 , respectively ) ; HDM IgE ( p=0 . 016 and 0 . 011 at ages 2 and 4 , respectively ) ; and Phadiatop IgE ( p=0 . 01 at age 4 ) ( Table 4 ) . Repeated-measures ANOVA showed that HDM IgE and LRI-related variables ( LRI , wLRI , fLRI ) from the first 3 years were significant predictors of age-5 wheeze ( Supplementary file 1 – table supplement 4 ) . Timepoint-specific analyses showed that differences were observable in HDM IgE and fLRI from age 2 onwards , while in LRI and wLRI they were only noticeable from age 3 ( Table 4; Figure 5 ) . A multiple regression model with stepwise elimination identified three significant variables: age-2 fLRI ( OR eight per unit increase , p=0 . 0075 ) , age-4 wLRI ( OR 5 . 3 p=0 . 0016 ) , and age-4 Phadiatop IgE ( OR 3 . 3 , p=0 . 0088 ) . But although both IgE-related and infection-related risk factors contributed to age-5 wheeze , there was no significant evidence of interaction between them ( p=0 . 36 within CAS2 alone , p=0 . 92 across entire cohort , for age-4 wLRI frequency ×Phadiatop IgE ) . Overall , CAS2 represented a low-risk trajectory susceptible to , but not necessarily afflicted by , wheeze due to atopic and non-atopic risk factors . In this cluster , atopic determinants of age-5 wheeze were only active from age 2 onwards , suggesting delayed atopic wheeze in this cluster . This duality of atopic and non-atopic risk factors for wheeze in this cluster was further supported by decision tree analysis , which identified that wheezy LRI frequency and HDM IgE best separated wheezers from non-wheezers in CAS2 ( Figure 5—figure supplement 3 ) . CAS3 was a ‘high-risk’ cluster , where persistent respiratory wheeze and atopic disease was seen in more than half the group throughout the first 10 years of life ( Table 1; Figure 3 ) . This cluster was dominated by males ( 86% , Fisher exact test , unadjusted p=6 . 8 × 10−3 compared to CAS1 , Table 1 ) , and appeared to represent an early- and multi-sensitised atopic phenotype with persistent wheeze . CAS3 had elevated IgE , IgG , and IgG4 responses to common allergens , especially Phadiatop , HDM and peanut IgE from 6 months onwards ( Table 2; Figure 4; Supplementary file 1 – table supplement 3B ) . SPTs were also more frequently positive in CAS3 , especially to HDM and food allergens ( peanut , cow’s milk and egg white , Supplementary file 1 – table supplement 3D ) . No strong predictors for age-5 wheeze were identified within CAS3 ( Table 4 ) : only couch grass IgE at age 2 and acute respiratory infection ( ARI ) frequency at age 1 were weakly significant ( both p=0 . 046 ) . Neither of these reached statistical significance when incorporated in the same model . However , the prolific IgE response , and the frequency and severity of early-life LRIs in this cluster ( Table 3 ) , strongly suggest contribution from both atopic and non-atopic causes of wheeze . Hence , CAS3 primarily represented those with extreme levels of atopic sensitisation and infection . The relative paucity of identifiable predictors may be explained by the small size of CAS3 ( N = 22 ) , the intrinsically high rate of wheeze in the cluster ( 76% with age-5 wheeze ) , and saturation of risk from high levels of IgE and frequent infections . Unlike the antibody measurements , cytokine measurements were excluded as clustering features due to high missingness . Nonetheless , with post-hoc analyses , we found that in vitro stimulation of PBMCs with HDM antigen elicited stronger Th2 cytokine responses in CAS3 compared to other clusters ( Table 2 , Figure 6 ) . These cytokines ( IL-4 , IL-5 , IL-13 ) were elevated from a very young age ( Wilcoxon , adjusted p=4 . 6 × 10−5 for IL-4 mRNA at age 6 m , compared to CAS1 ) , coinciding with increase in HDM IgE and IgG4 responses . Weaker but similar differences were observed for peanut- and ovalbumin-stimulated PBMCs at 6 months ( unadjusted p<0 . 05 for all , Supplementary file 1 – table supplement 3C ) . There were no other significant differences for other non-Th2 cytokines ( IFN-γ , IL-10 ) , nor were there specific differences for CAS1 or CAS2 . Across all clusters , allergen-specific IgG4 and IgG were positively correlated with IgE for the same allergen ( especially HDM , Figure 4—figure supplement 1 ) . As noted previously , CAS2 and CAS3 were distinguished from CAS1 by high IgG4 , and CAS3 had greater IgG4 than either CAS1 or CAS2 ( Supplementary file 1 – table supplement 3B; Figure 4 ) . Decision tree analysis ( Figure 5—figure supplement 1 to 3 ) confirmed that IgG4-type variables strongly separated CAS2 and CAS3 from CAS1 , while IgE-type variables separated CAS3 from the others . Although previous literature suggests possible protection conferred by IgG4 ( Okamoto et al . , 2012 ) or IgG ( Holt et al . , 2016 ) , in this study there was no clear evidence of such protection against later wheeze ( Table 4 ) . Furthermore , the protected status of CAS2 relative to CAS3 was unlikely to be driven by IgG4 , given that CAS3 had greater quantities of both IgE and IgG4 . Although they were highly correlated , IgE , IgG , Th2 cytokine and SPT responses did not overlap perfectly . CAS3 was enriched for individuals with strong signals in all modalities , but there remained individuals within CAS3 and the rest of the cohort who were only responsive in some modalities and not others . Notably , the general direction of IgE , IgG4 , SPT and Th2 cytokine signals did not always coincide ( Figure 4—figure supplement 2 ) . The npEM-derived CAS clusters were partially consistent with traditional atopy thresholds ( i . e . any specific IgE ≥0 . 35 kU/L or SPT ≥ 2 mm at age 2 ) . When we compared CAS clusters with supervised groups created using traditional thresholds ( Supplementary file 1 – table supplement 5 ) , we found that CAS1 most closely matched a non-atopic phenotype ( 58 of 84 had no specific IgE greater than 0 . 35 kU/L by age 2 ) . Conversely , CAS2 and CAS3 partially matched traditional criteria for atopy , with CAS3 being an extreme phenotype ( all 22 children in CAS3 had some specific IgE ≥0 . 35 kU/L by age 2 ) . However , the CAS clusters outperformed IgE/SPT-defined atopy in terms of predicting for age-5 wheeze ( likelihood ratio test for clusters vs . IgE/SPT , Chi-squared = 23 , p=2 . 0 × 10−6 ) . In addition , at age 2 , 68% of CAS3 were ‘sensitised’ ( any specific IgE ≥0 . 35 kU/L ) to two or more allergens , compared to only 1% and 6% for CAS1 and CAS2 respectively . This emphasised CAS3 as an early- and multi-sensitised phenotype . Finally , fewer members of CAS1 and CAS2 who were IgE- or SPT-responsive prior to age 5 maintained atopic wheeze at age 5 ( 23% or 24 of 103 ) , compared to CAS3 ( 76% or 16 of 21 ) . Therefore , the association of IgE and SPT with disease risk varied across clusters . This suggests that fixed atopy thresholds are not sufficient to delineate risk profiles – instead , an unsupervised clustering approach may be more informative . We mapped the npEM-derived clusters to pre-defined wheezing phenotypes ( Figure 3C ) : no wheeze ( in the first 3 years of life , or at age 5 ) , transient wheeze ( only in first 3 years ) , late wheeze ( only at age 5 ) , and persistent wheeze ( both first 3 years and age 5 ) . We found that CAS3 was enriched for persistent wheeze , while individuals in CAS1 or CAS2 tended to have transient or no wheeze . There were rarely any members of CAS with late wheeze ( approximately 10% ) . In addition to persistent wheeze , CAS3 was also enriched for persistent food sensitisation ( peanut IgE ≥0 . 35 kU/L , or positive egg white or cow’s milk SPTs ) and persistent eczema: 44% of CAS3 experienced all three ( Figure 3—figure supplement 1 ) . Almost all individuals in CAS3 had both eczema and food sensitisation from age 6 m onwards , with rates of food sensitisation and wheeze increasing with time ( Figure 3D ) . In contrast , CAS1 and CAS2 had low rates of food sensitisation , and declining rates of both eczema and wheeze . These trends lend credence to recent suggestions that the ‘atopic march’ phenotype ( Bantz et al . , 2014; Han et al . , 2017 ) may only be present in a minority of the population ( e . g . CAS3 ) ( Belgrave et al . , 2014 ) . Previous studies suggest an association between asthma risk and early-life disruption of the respiratory microbiome , especially colonisation with Streptococcus spp . in the first 7 weeks of life ( Teo et al . , 2015 ) . In this study , using the same data and definitions , we found that CAS3 was overrepresented by individuals who had >20% relative abundance of Streptococcus in their first infection-naive healthy NPA , within the first 7 weeks of life ( 44% versus 11% and 15% in CAS1 and CAS2 , respectively; Fisher exact test , unadjusted p=0 . 042 and 0 . 065 , respectively; Table 3 ) . Furthermore , Teo et al and others ( Teo et al . , 2015; Bisgaard et al . , 2007 ) previously found that transient incursions with certain MPGs ( Streptococcus , Haemophilus , Moraxella spp . ) were associated with increased frequency and severity of subsequent LRIs and wheezing disease . Here , we found that proportion of these infection-associated MPGs in healthy samples from age 0 to 2 was greater in CAS3 ( 62% vs . 49% and 32% in CAS1 and CAS2 , respectively; Fisher exact test , unadjusted p=0 . 2 and 5 . 5 × 10−4 , respectively; Table 3 ) . This finding was independent of LRI and wLRI frequency ( GLM; p<0 . 05 for model predicting group membership , with age-2 LRI and wLRI as covariates ) . On the contrary , there were no associations between cluster membership and health-associated MPGs ( Corynebacterium , Alloiococcus , Staphylococcus spp . ; Supplementary file 1 – table supplement 3E ) . Recent work by Teo et al . , 2017 ) suggested that infection-associated MPGs in early life were predictive for age-5 wheeze in atopic children , while in non-atopic children they were predictive for transient wheeze . In this study , with the same cohort , we noted a similar trend for infection-associated MPGs from age 0 to 2 , in relation to transient wheeze in ‘non-atopic’ CAS1 ( GLM , OR 3 . 6 per percent , p=0 . 17 , with demographic covariates ) . Surprisingly , there was evidence that infection-associated MPGs in later samples ( from age 2 to 4 ) were protective against age-5 wheeze in CAS1 ( OR 0 . 086 per percent , 0 . 45 per quartile , p=0 . 034 and 0 . 035 , respectively; Table 4 ) . Infection- and health-associated MPGs were otherwise not associated with age-5 wheeze within the other clusters . The trajectories described by the CAS npEM clusters were replicated in two cohorts – the Manchester Asthma and Allergy Study ( MAAS ) ( N = 1085 ) ( NAC Manchester Asthma and Allergy Study Group et al . , 2002 ) from Manchester , UK , and the Childhood Origins of Asthma Study ( COAST ) ( N = 289 ) from Wisconsin , USA ( Lemanske , 2002 ) . After applying our npEM classifier to these external cohorts ( materials and methods ) , we found that individuals classified into ‘Cluster 3’ ( MAAS3/COAST3 ) had a persistent disease phenotype extending into late adolescence , with consistently high rates of parent-reported wheeze and physician-diagnosed asthma from birth to age 16 . The other two clusters ( Cluster 1 = MAAS1/COAST1; Cluster 2 = MAAS2/COAST2 ) appeared to be low-risk ( Figure 7A , B , D ) . MAAS3 and COAST3 exhibited stronger IgE expression ( total , HDM , cat , dog ) from ages 1 to 8 ( Figure 7C , E ) , compared to other clusters in each dataset . Like CAS3 , COAST3 demonstrated elevated PBMC expression of Th2 cytokine protein ( IL-5 and IL-13 ) in response to HDM stimulation at age 3 ( Figure 7F ) . This was not replicated in MAAS3 , but previous work in MAAS had identified that a strong PBMC Th2 response ( IL-5 , IL-13 ) to HDM stimulation at age 8 was associated with increased risk of HDM sensitisation and asthma ( Wu et al . , 2015 ) . Nonetheless , MAAS3 was overrepresented in ‘early-sensitised’ and ‘multiple sensitised’ phenotypes described by Lazic et al . ( 2013 ) from SPT and IgE data . Approximately 86% of individuals in MAAS3 belonged to either one of these two phenotypes , although only 13% of individuals in these two phenotypes were accounted for by MAAS3 . Furthermore , when we explored potential predictors of wheeze phenotypes and asthma diagnosis in later childhood , we found that the clusters in COAST were very similar to those in CAS . In COAST1 , LRI and wLRI frequency at age 2 were predictive of asthma diagnosis at age 6 ( GLMs with demographic covariates , p=0 . 02 and 0 . 02 , respectively ) , while in COAST2 , HDM IgE at age 3 , and LRI , wLRI and fLRI frequencies at age were all predictive ( GLMs , p<0 . 05 for all ) ( Figure 5—figure supplement 4 ) . Although the timing and magnitude of associations differed between cohorts , this reaffirmed wheeze in Cluster 1 as being primarily non-atopic in origin , while wheeze in Cluster 2 appeared to be driven by both non-atopic and atopic factors . We re-applied npEM classification to CAS using only those features present in MAAS or COAST . For MAAS and COAST features , the subsequent clusters bore 79% and 72% concordance with the original CAS clusters , respectively . In both cases , concordance was excellent for Cluster 3 – all 22 members of the original CAS3 were correctly assigned to Cluster three after re-applying npEM . Therefore , CAS3 , COAST3 and MAAS3 likely represent very similar phenotypes . We checked the stability and validity of the CAS clusters with leave-one-out ( LOO ) analysis , Jaccard indices and silhouette widths . The average Jaccard indices from leave-one-individual-out analysis were 0 . 77 , 0 . 76 , and 0 . 85 for CAS1 , 2 and 3 , respectively . For leave-one-feature-out analysis , the average indices were 0 . 65 , 0 . 60 , and 0 . 74 , respectively . This demonstrates that the clusters , especially CAS3 , were relatively resilient to minor changes in sampling or feature selection . In relation to internal validity of the CAS clusters , average silhouette widths were universally poor , at 0 . 05 , 0 . 06 and 0 . 002 for CAS1 , 2 , 3 , respectively , with an average for all three clusters of 0 . 05 ( Figure 1—figure supplement 2 ) . Silhouette widths were particularly suboptimal with CAS3 , with at least half of those classified having negative values . The overall poor internal validity of the clusters may be due to the large-scale and exploratory nature of our approach – the metric may have been obscured by intra-cluster heterogeneity in other variables that were not particularly important for determining cluster membership . However , it must be noted that all clusters on average yielded positive silhouette widths , and as observed in the rest of the results , they were all relatively homogeneous in terms of the outcomes of interest ( wheeze status , allergic disease phenotypes ) . Decision tree analysis on the CAS dataset , using all available predictors from all timepoints , created a ‘Simple Tree’ with two decision nodes and three end nodes ( Figure 5—figure supplement 1 ) . This tree had 89% accuracy in retrieving cluster memberships from the original npEM model , where accuracy is calculated as percentage overlap of tree clusters with original CAS clusters . We found that membership in the CAS3-equivalent tree cluster was a better predictor for age-5 wheeze ( likelihood ratio test , Chi-squared = 19 , p<1 × 10−5 ) than traditional thresholds for atopy based on IgE and SPT measurements at age 2 . IgG4-related variables best separated CAS1 from other clusters , while IgE-related variables best separated CAS2 and CAS3 . Explicitly forcing the exclusion of Phadiatop variables from tree analysis caused these thresholds to be replaced with allergen-specific assays ( HDM IgE for Phadiatop IgE , Figure 5—figure supplement 2 ) in a way that is consistent with correlation patterns amongst IgE and IgG4 variables ( Supplementary file 1 – table supplement 6 ) . We also constructed a ‘Comprehensive Tree’ that best split individuals into six groups , based on cluster membership crossed with age-5 wheeze status ( Figure 5—figure supplement 3 ) . We thus identified nodes that were consistent with predictors for wheeze found in the previous regression analyses ( Table 4 ) , combined with nodes from the Simple Tree ( Figure 5—figure supplement 1 ) . The Comprehensive Tree had 77% accuracy in recovering both cluster membership and wheeze status . In terms of identifying pure wheeze status at age 5 , the accuracy of the tree was 84% , with a positive predictive value ( PPV , or precision ) of 72% , negative predictive value ( NPV ) of 88% , sensitivity ( recall ) of 71% and specificity of 89% . The Comprehensive Tree was more successful in flagging age-5 wheeze ( likelihood ratio test , Chi-squared = 60 , p=6 . 1 × 10−13 ) , compared to the traditional atopy thresholds described previously . Cluster 3 represented a multi-sensitive or polysensitised phenotype ( Bousquet et al . , 2015 ) . In CAS3 , not only was total IgE elevated , but specific IgE were also raised for most allergens . Three in four CAS3 individuals were sensitised ( specific IgE ≥0 . 35 kU/L ) to two or more allergens . In our external replication with MAAS , we observed a large overlap between our predicted high-risk phenotype ( MAAS3 ) and the multiple atopy phenotype from Lazic et al . , 2013 ) . This was consistent with findings from other studies , where the severely atopic and polysensitised subpopulation was at greater risk of both wheezing disease and reduced lung function ( Hose et al . , 2017 ) . It is not currently known what is fundamentally producing the strong atopic predisposition in Cluster 3 . It is possible that inherited ( genetic/epigenetic ) or environmental factors ( including in utero or perinatal exposures ) may be involved , and these should be targets for future investigations . The overrepresentation of males in CAS3 is consistent with the consensus that young boys are at greater risk for asthma than young girls; this was traditionally believed to be due to intrinsic sex differences in airway diameter ( Almqvist et al . , 2008 ) . However , our cluster analysis did not employ any clustering features related to airway size . This suggests that other sex-related factors could be involved , such as differences in immunity and allergic susceptibility . Allergic sensitisation is more frequent amongst prepubescent boys than girls ( Gabet et al . , 2016; Kim et al . , 2014 ) , and this may be linked to differences in cytokine responsiveness . However , not all boys were clustered into Cluster 3; and sex was not found to be a determinant for either IgE levels or cytokine response in CAS . We did observe that CAS3 overlapped strongly with both persistent food sensitisation and eczema , and that persistent wheeze co-occurred with early sensitisation and eczema . This suggests that the ‘atopic march’ may play a role in CAS3 . Early disruption of the skin barrier and exposure to certain food allergens may act in concert to promote and entrench the atopic phenotype , through the activation of cytokine pathways involving TSLP , IL33 and IL25 ( Bantz et al . , 2014; Han et al . , 2017 ) . Although recent research has suggested that very few children actually follow the disease trajectory of the atopic march ( Belgrave et al . , 2014 ) , we hypothesise that it remains relevant to a small but important high-risk subpopulation , who may potentially benefit from early interventions targeted at halting the progression of disease . In all three cohorts ( CAS , MAAS , COAST ) , house dust mite ( HDM ) sensitivity was an important determinant of atopic disease risk . HDM was a strong predictor for both CAS3 membership and later childhood wheeze in CAS2 , as well as being a ‘dominant’ allergen in the Phadiatop Infant assays . CAS3 in particular exhibited early and extreme HDM hypersensitivity , with prematurely-elevated HDM IgE , as well as PBMC Th2 response ( IL-4 , 5 , 9 , 13 ) to HDM stimulation . Similar phenomena were seen with MAAS3 and COAST3 . The importance of HDM hypersensitivity in driving allergic disease in some populations is well-described in the literature ( Thomas et al . , 2010; Calderón et al . , 2015 ) . Previous findings from MAAS and a similar cohort RAINE ( Wu et al . , 2015 ) have shown a confluence of high HDM IgE , as well as PBMC Th2 cytokine levels such as IL-13 and IL-5 , in discrete subsets of the population . However , we did observe that in other clusters ( CAS1 and CAS2 ) , some individuals with purported HDM sensitisation ( IgE >0 . 35 kU/L ) did not produce detectable Th2 responses; the reverse was also true , where Th2 response did not necessarily result in high IgE . It may be the case that there is high intra-individual variation in IgE and cytokine responses , or stochastic variation in detectability of IgE or cytokine , which may obscure association analyses . Regardless , early and strong Th2 cytokine responses against HDM indicate a high-risk phenotype . Interestingly , early-life peanut IgE was a strong delineator between high-risk CAS3 and lower-risk CAS1 and 2 . There is evidence in the literature for transmission of peanut allergen in utero or via breastmilk ( Vadas et al . , 2001; DesRoches et al . , 2010 ) , as well as early sensitisation via home environmental exposure , especially in those with concurrent eczema or a predisposing filaggrin ( FLG ) mutation that may allow transcutaneous infiltration of allergen ( Brough et al . , 2013; Brough et al . , 2014 ) . The strong correlation between Phadiatop and peanut IgE in the first year of life suggests that either peanut reactivity is significant at this earlier timepoint , or that ‘peanut-specific IgE’ is cross-reactive and representative of some other allergen hypersensitivity . The fact that this correlation exists within each cluster ( Supplementary file 1 – table supplement 6 ) suggests that it is not caused solely by differences between low- and high-risk clusters ( CAS1/CAS2 vs . CAS3 ) . There is a possibility that peanut IgE is a marker for a broader phenotype of early and unremitting sensitisation to multiple food allergens ( peanut , cow’s milk , eggwhite ) , as we had observed in CAS3 . However , it is unlikely that premature exposure to food allergen is the lone driver for sensitisation and disease , given that well-timed oral exposures to common food allergens ( e . g . within 4 to 6 months of age ) may actually be protective ( Koplin et al . , 2010 ) . There is some evidence that quantity ( minute vs . abundant ) , route ( skin vs . oral ) and timing ( early vs . late ) of exposure are key modifiers of risk ( Han et al . , 2017 ) . Ultimately , an underlying atopic predisposition linked to early-life exposure to food allergen may be driving the high-risk phenotype in Cluster 3 . In our study , neither IgG nor IgG4 were strong predictors or protectors of wheeze . However , IgG4 was a strong delineator of cluster membership in CAS , with individuals from CAS2 and CAS3 having elevated IgG4 across all specificities compared to CAS1 . Vulnerability to early IgE-driven respiratory disease ( ‘atopic wheeze’ ) can be seen in these same individuals –in CAS2 where HDM IgE is predictive for later wheeze , and in CAS3 where both wheeze frequency and IgE are elevated . Hence , although there had previously been doubt about the efficacy of IgG4 as a marker for atopy ( EAACI Task Force et al . , 2008 ) , our study suggests that IgG4 is still relevant for determining atopic risk , especially when used in combination with IgE . The underlying biology behind the association of IgG4 with susceptibility to ‘atopic wheeze’ is unclear . Th2-related pathways drive production of both IgE and IgG4 , with IgG4 predominating when modified by concurrent IL-10 signalling ( Davies and Sutton , 2015 ) . In susceptible individuals , IgG4 production likely precedes isotype switching to frank IgE production ( Aalberse , 2011 ) . Multiple studies have reported that IgG4 is correlated with induced tolerance following desensitisation immunotherapy with high-dose allergen treatment ( Davies and Sutton , 2015 ) . However , based on this study alone , we cannot observe any protection from naturally elevated IgG4 levels . Our group had previously suggested , using data from another cohort ( Holt et al . , 2016 ) , that IgG and specifically IgG1 may provide endogenous protection against IgE-associated wheeze in children experiencing natural ( low-level ) exposure to aeroallergen . In this present study , IgG1 was not measured . The co-occurrence of elevated IgE and LRI frequency in CAS3 , as well as their predictive effect in CAS2 , are consistent with previous findings from CAS ( Holt et al . , 2010; Teo et al . , 2015; Kusel et al . , 2007 ) . They lend support to the theory that allergic and infective processes act additively to intensify airway inflammation during respiratory pathogen clearance , which in turn drives progression towards persistent wheeze ( Holt and Sly , 2012 ) . In addition , our cluster analysis suggests that the pathologic effect of this interaction may be stratified in discrete subpopulations , rather than acting in a strictly dose-dependent fashion across the entire cohort . There may be subsets of children ( CAS2 and CAS3 ) who are more susceptible to the effects of this viral-atopy interaction . On the other hand , pathogen clearance in infected non-atopic ( CAS1 ) subjects may be more efficient , due to lack of susceptibility to the pro-inflammatory effects of atopic co-stimuli . This produces lower levels of ‘bystander’ inflammatory damage to airway tissues , with opportunity for recovery , resulting in a less severe wheeze phenotype . Of particular note is that , while both CAS1 and CAS2 have LRI and wLRI frequencies as predictors for age-5 wheeze , CAS2 also has fLRI , particularly at age 2 . This , along with the general higher incidence of fLRI in CAS3 , is consistent with previous findings from CAS ( Holt et al . , 2010; Teo et al . , 2015 ) . It suggests that symptomatically severe infections , correlating with severe airway inflammation , may be more potent in causing persistence of wheeze , specifically among those who are ‘atopic’ ( CAS2 and CAS3 ) . In addition , even during periods of good health , the upper respiratory microbiome played a role in determining later childhood wheeze . Its effect interacted with cluster membership , as well as the age at which the microbiome changes occurred . CAS3 was enriched for early-life infection-associated MPGs ( Streptococcus , Moraxella , and Haemophilus ) . This was consistent with the previous finding by Teo et al . , 2017 ) that early-life infection-associated MPGs were predictive of age-5 wheeze only within atopic individuals ( as defined by IgE alone ) . Interestingly , in our current study , we found a protective effect of infection-associated MPGs from age 2 to 4 in CAS1 . We hypothesise that those without atopy-related immune dysfunction are able to maintain a healthy trajectory by responding appropriately to stimuli from potential pathogens that colonise the respiratory tract , thus achieving protection against future ( non-atopic ) wheeze . This is akin to the ‘hygiene hypothesis’: exposure to a greater repertoire of pathogen-derived antigens may facilitate maturation of immune functions against said pathogens . Meanwhile , individuals with a predisposing immune dysfunction ( i . e . ‘atopy’ manifesting in early-life allergic sensitisation ) may be responding in a maladaptive manner to these microbes ( Holt and Sly , 2012 ) . This may result in inability to clear potential pathogenic bacteria , or shaping of aberrant immune responses – with subsequent effects on airway inflammation and wheeze . In this study , we applied mixture modelling to generate clusters from biological data . Similar methods such as latent class analysis ( LCA ) have previously been used in asthma research – for instance , LCA was applied to SPT and IgE measurements from MAAS to determine different patterns of allergen sensitisation and subsequent disease ( Lazic et al . , 2013 ) . However , LCA is limited to categorical clustering features , so measures of sensitisation in that study were thresholded ( e . g . IgE levels were split into <0 . 35 kU/L , 0 . 35 to 100 kU/L , and >100 kU/L ) . The method also assumed that these thresholds have the same relevance across all timepoints; that thresholds applied equally to all allergens; and that all allergens contributed equally to disease susceptibility profiles . Mixture modelling is an extension of LCA in that it does not require categorical variables or predetermined thresholds . Furthermore , non-parametric mixture modelling ( npEM ) does not require input features to have Gaussian distributions . Previous studies have used mixture models to explore phenotypes in adult asthma based on clinical measurements ( Janssens et al . , 2012; Newby et al . , 2014; Burte et al . , 2015 ) , and one of our own studies previously looked at cytokine expression patterns of PBMCs from children in response to HDM stimulation ( Wu et al . , 2015 ) . Our study is the first to apply non-parametric mixture modelling to data representing immune and respiratory health in early childhood , and to investigate possible predictors of disease within each cluster . Currently , mixture models are limited by an unproven ‘track record’; a lack of consensus about best protocols for data processing and analysis; instability or inconsistency of clusters; difficulty in interpretation of results; and uncertainty regarding the validity of certain assumptions that accompany models ( Deliu et al . , 2016 ) . Other methods of cluster analysis have similar problems , and while they have been applied frequently to asthma research , they have also produced a confusing myriad of phenotypes . The nature of cluster phenotypes is highly dependent on the type of features entered into the clustering algorithm . Clustering features that represent final clinical endpoints , such as markers of severity , may produce more heterogeneous clusters , as different pathological trajectories can arrive at similar endpoints . Some cluster phenotypes may contradict with each other , or may not be easily interpreted . Recently , Schoos et al . ( 2017 ) identified that , unlike our study , asthma was not as strongly associated with prominent HDM or peanut hypersensitivity in a Danish birth cohort ( COPSAC ) as other patterns of sensitisation ( especially cat , dog and horse ) . However , we note that they used thresholded IgE >0 . 35 kU/L to build their clusters . Other differences may emerge due to heterogeneity across different populations; geographical differences in environmental exposures and allergen sensitisation; and differences in testing procedures and phenotype definitions at different sites . COPSAC , CAS and COAST were cohorts enriched for high-risk individuals – each child had at least one parent with a history of atopic disease – while MAAS had no such recruitment criterion . Because of variability in findings , there has been wariness and scepticism among clinicians regarding the utility of mixture models and machine learning ( Chen and Asch , 2017 ) . Ultimately , one may argue that discrepancies in our findings serve as a caution against the blind application of ‘algorithms’ without due consideration of subtleties in target population and environment . Nonetheless , what we have demonstrated here is the vast potential of cluster analysis . We have discovered clusters in an unsupervised and exploratory fashion , described them comprehensively , replicated our findings in multiple datasets , and compared our clusters with other existing phenotypes . In doing so , we have generated some new and interesting insights about the nature of atopy and asthma risk . Our results build on previous findings ( Frith et al . , 2011; Klink et al . , 1990 ) demonstrating that the concept of atopy , as an intrinsic or heritable predisposition to allergic disease , is more complicated than what could be described by dichotomies or thresholds . We have also demonstrated that addressing subgroup differences via cluster analysis allows for identification of intra-cluster disease predictors . In the future , clusters may be further characterised by other aspects of asthma pathophysiology , such as genomics , transcriptomics , and epigenomics . The results of our study strongly support the future use of predictive models with more precise and subgroup-driven representations of atopy or other relevant pathophysiology . We argue for ongoing collaboration between research groups in terms of refining methodology , answering questions unique to certain populations , and comparing cluster phenotypes arising from different algorithms and datasets . We believe that , as clustering methods become more frequently used , we will gradually develop better consensus on how such methods are best applied to biomedical phenomena . By continuing with these approaches , we can hopefully move away from fixed thresholds to more sophisticated formulations of risk , which will then improve future attempts at targeted screening , prevention and treatment of asthma . These approaches are already being applied to other heterogeneous diseases , and in the future computerised tools may be designed to embody the sum knowledge from these approaches . Such approaches can eventually help clinicians and scientists achieve a fuller understanding of pathophysiology , and hence better predict and manage human disease . Our discovery dataset was the Childhood Asthma Study ( CAS ) , a prospective birth cohort ( N = 263 ) operated by the Telethon Kids Institute from Perth , Western Australia ( Kusel et al . , 2005 ) . The goal of CAS was to describe the risk factors and pathogenesis of childhood allergy and asthma . Further details of CAS have been reported previously ( Kusel et al . , 2005; Hollams et al . , 2009; Holt et al . , 2010; Teo et al . , 2015; Hollams et al . , 2017 ) . In CAS , expectant parents were recruited from private paediatric clinics in Perth during the period spanning July 1996 to June 1998 . Each child who was born and subsequently recruited had at least one parent with physician-diagnosed asthma or atopic disease ( hayfever , eczema ) . The child was then followed from birth till age 10 at the latest , with routine medical examinations , clinical questionnaires , blood sampling at multiple time points ( 6–7 weeks , 6 months , 1 year , 2 , 3 , 4 , 5 , and 10 years ) and collection of nasopharyngeal samples . Parents also kept a daily symptom diary for symptoms of respiratory infection in their child . The data extracted from these samples and measurements covered multiple ‘domains’ of asthma pathogenesis , including respiratory infection , allergen sensitisation , and clinical or demographic background . For each child and visit , the investigators of CAS recorded metrics related to suspected or known modulators of asthma risk . These included markers of immune function: ( 1 ) IgG , IgG4 , and IgE Phadiatop ImmunoCAP antibodies ( ThermoFisher , Uppsala , Sweden ) , covering common allergens such as house-dust mite ( HDM , Dermatophagoides pteronyssinus ) , mould , couch grass , ryegrass , peanut , cat dander; ( 2 ) IgE and IgG4 Phadiatop Infant and Adult assays ( ThermoFisher , Uppsala , Sweden ) that target multiple allergens simultaneously ( Ballardini et al . , 2006 ) ; ( 3 ) skin prick or sensitisation tests ( SPT ) for HDM , mould , ryegrass , cat , peanut , cow’s milk and hen’s egg; and ( 4 ) cytokine responses ( IL-4 , 5 , 9 , 13 , 10 , IFN-γ ) following in vitro stimulation of extracted peripheral blood mononuclear cells ( PBMCs ) by multiple antigen and allergen stimuli , including phytohaemaglutinin ( PHA ) , HDM , cat , peanut and ovalbumin ( Hollams et al . , 2009; Holt et al . , 2010 ) . In addition , nasopharyngeal samples ( swabs or aspirates , NPAs ) were taken from each child during healthy routine visits ( healthy samples ) , and unscheduled visits where parents presented with their child if they have a suspected respiratory infection ( disease samples ) . Frequency and severity of respiratory infections were measured accordingly . NPAs were then screened for viral and bacterial pathogens using rtPCR and 16 s rRNA amplicon sequencing with Illumina MiSeq ( San Diego , US ) , respectively ( Teo et al . , 2015 ) . These NPAs had previously classified by Teo et al . ( 2015 ) ; Teo et al . ( 2017 ) , based on clustering of bacterial composition , into microbiome profile groups ( MPGs ) that were associated with healthy respiratory states ( health-associated MPGs , for example Alloiococcus- , Staphylococcus- or Corynebacterium-dominated ) or infectious respiratory states ( infection-associated MPGs , for example Moraxella- , Haemophilus- , or Streptococcus-dominated ) . Other collected data included: sex , height and weight; paternal and maternal history of atopic disease; blood levels of basophils , plasmacytoid and myeloid dendritic cells as measured by fluorescence-assisted cell sorting ( FACS ) ; and levels of vitamin D ( 25-hydroxycholecalciferol , 25 ( OH ) D ) ( Hollams et al . , 2017 ) . We adopted an exploratory approach to cluster analysis , whereby we attempted to interrogate as much of the existing dataset as possible , identifying latent clusters that arise from the underlying data structure of CAS . We then assessed how these latent clusters correlate with risk of asthma or other markers of pathophysiology , such as degree of allergic sensitisation . All data processing and analysis were done in R v3 . 3 . 1 ( RRID:SCR_001905 ) . A graphical overview of the analytic process is displayed in Figure 1—figure supplement 3 . To identify latent clusters , we applied non-parametric expectation-maximisation ( ‘npEM’ ) mixture modelling to our discovery cohort CAS , using functions from the R package ‘mixtools’ ( Benaglia et al . , 2009a ) . This method assumes that frequency distributions of each cluster can be represented by non-parametric density estimates learned from the data in an iterative process . npEM was used because: ( 1 ) it was plausible to consider a population as a mixture of subpopulations each with their own distributions; ( 2 ) it had advantages over other unsupervised approaches ( Tan et al . , 2005 ) – for example , with LCA , continuous variables cannot be handled appropriately; with hierarchical clustering , poor decisions made early in the classifying process are not easily amended; ( 3 ) many variables were categorical or non-Gaussian , so theoretically a non-parametric approach should be superior to a Gaussian mixture model or k-means approach; and ( 4 ) inherent within mixture models is an intuitive method for supervised classification of other datasets into similar clusters . We used a largely non-selective approach to choosing features for cluster analysis , in that we attempted to retain as many CAS individuals and variables as possible . However , we did enforce certain quality-control measures such as excluding variables ( ‘features’ ) that had missing data for >20% subjects ( 442 variables removed ) , and subjects with missing data for >30% of the remaining variables ( 39 subjects removed ) . Also excluded were features pertaining to our primary outcomes of interest: incidence of parent-reported wheeze , physician-diagnosed asthma and hayfever at all timepoints . We specifically excluded these from feature selection so we could determine how subsequent clusters differ in these outcomes , even when clustering was not explicitly driven by them . On the other hand , eczema was not excluded because of evidence that infantile eczema may itself influence the risk for subsequent sensitisation and asthma ( Gustafsson et al . , 2000 ) . Frequency of wheeze in the context of respiratory infection was also included , as it was a symptomatic marker of infection severity . Variable reduction resulted in M = 174 variables remaining out of an original 659 . The complete list of variables included as clustering features is provided in Supplementary file 1 – table supplement 1 , and importantly covers multiple domains including demographic ( family history of atopy , household size ) , clinical ( incidence of childhood eczema ) , immunological ( IgE , IgG , IgG4 , SPT ) and microbiological ( respiratory infections , viral pathogens associated with infection ) features . By virtue of study design and quality control measures , many of the clustering features were related to immunological function or respiratory infection in the first 3 years of life . Highly skewed features , such as antibody and cytokine levels , were subjected to logarithmic ( base 10 ) transformation . We also applied limited thresholding to some variables ( cytokine responses , antibody assays ) , based on best practice for the reported limit-of-detection ( LOD ) of the measuring devices . The LOD for IgE was 0 . 03 kU/L; for IgG4 , 0 . 0003 μg/L; for IgG , 0 . 4 mg/L . For these variables , we assigned any values below the LOD to half the LOD ( i . e . 0 . 015 kU/L , 0 . 00015 μg/L , and 0 . 2 mg/L , respectively ) . For stimulated cytokine expression above unstimulated control , any zero or negative values ( i . e . unstimulated control had equal , or greater , expression than stimulated ) , were converted to 0 . 000001 units or 0 . 01 pg/ml for mRNA and protein variables , respectively . Positional standardisation scaling was then applied across all variables , to equally weight the contributions of each feature to the mixture model . This involved replacing each value xij for individual i of feature j , by:xij−med ( xj ) max ( xj ) −min ( xj ) where functions med , max and min refer to the median , maximum , and minimum for the complete-case dataset for feature j , respectively . The processed and scaled CAS dataset was further split into those subjects with no missingness in the remaining variables ( ‘complete-case’ , 186 subjects , 174 variables ) ; versus those who had limited missingness of <30% variables ( ‘low-missingness’ , 36 subjects , 174 variables ) . Cluster analysis was performed initially in the complete-case CAS subset to generate an npEM model . The mathematical theory underpinning npEM has already been described extensively in other sources ( Benaglia et al . , 2009b ) . In brief , it involves three steps: ( 1 ) an expectation or E-step , which calculates the posterior probability of membership in cluster k , given the observed dataset , estimated mixing proportions λk , and probability distribution for k; ( 2 ) a maximisation or M-step , which calculates the mixing proportions λk from the cluster memberships determined above; ( 3 ) a non-parametric kernel density estimation step , which calculates the probability distribution based on a kernel density function for each cluster k and clustering feature j . These steps were then iterated until the model converged to a point where log-likelihood values were maximised . As with any EM algorithm , an initial state must first be set prior to commencing the iterative process . To do this , we used a constant seed state ( ‘set . seed ( 1 ) ' ) to allow reproducibility of results . Based on these pseudo-random centroids for a set number of clusters L , the initial state was then determined by k-means clustering as in Benaglia et al ( Benaglia et al . , 2009b ) . The other options in npEM were set to defaults . These included the use of non-stochastic ( deterministic ) as opposed to a stochastic method; the use of a standard normal density function as the kernel function; and the use of default constant bandwidths for estimating kernel densities ( Benaglia et al . , 2009b ) . The ideal number of clusters L was determined by two methods . Firstly , we performed hierarchical clustering on the complete-case dataset , and scrutinised the dendrogram as well as a scree plot for an optimal cut-off using the ‘knee method’ ( Tan et al . , 2005 ) . We observed that this occurred at around L = 3 or 4 . Secondly , we repeated npEM clustering for values of L = 1 , 2 , … , 20 , and calculated the Bayesian information criterion ( BIC ) for each of these , using the formula:BIC=−2log⁡ ( p^ ) +ν log ( N ) where P^ is the maximum likelihood , ν=L×M+ ( L-1 ) , and L , M , N are total number of clusters , clustering features , and individuals respectively . The optimal number of clusters was again determined to be around L = 3 or 4 , based on minimum BIC observed . For the sake of parsimony , we set the number of clusters to three . The density functions generated by the resultant npEM model were then used to classify as many subjects of the low-missingness subset as possible . This method relied on the assumption that distributions observed in the ‘training’ ( complete-case ) dataset were representative of distributions that existed in ‘test’ ( low-missingness or external ) datasets . Classification was performed as follows: consider individual i of N; clustering feature or coordinate j of M; and component or cluster k of L . For each individual i belonging to known cluster k=K , let the kernel density function for coordinate j be fjKxij . We now assume that the coordinates j were independent of each other . Although this was not truly the case – for instance , weak correlation exists between IgE and IgG4 of different allergen specificities in the CAS dataset [23] – we believed the assumption was justified given our theory-naive and exploratory approach . With this assumption , the joint distribution for individual i in cluster K should be the product of density functions for all j given K . and therefore the probability of individual with value xij belonging to cluster K was:P ( k=K\|xij ) =λK∏j=1MfjK ( xij ) ∑k=1Lλk∏j=1Mfjk ( xij ) In addition to this , we made two other assumptions: ( 1 ) if xij was missing , then the density was assumed to be one , or fjKxij=1; ( 2 ) else , if xij < min ( xj ) , the minimum value in feature j for which there was a non-zero density value , then the density was equal to that of the minimum value , that is fjKxij=fjKmin⁡xj . Likewise , if xij > max ( xj ) , then fjKxij=fjKmax⁡xj . Individuals with membership probability greater than 90% for cluster K were classified into K . Using this method , an additional 31 individuals from 36 were successfully classified into one of three clusters , for a total combined dataset of 217 classified individuals in CAS . Finally , we formally defined each CAS cluster using the composite of complete-case and low-missingness datasets , and described each cluster in terms of key characteristics and significant cluster-specific predictors for age-5 wheeze . Importantly , variables that were initially excluded from feature selection were treated as subsequent outcomes for post-hoc comparison of clusters . The study designs and measurements for the two replication cohorts – the Manchester Asthma and Allergy Study ( MAAS ) ( N = 1085 ) from Manchester , UK , and the Childhood Origins of Asthma Study ( COAST ) ( N = 289 ) from Wisconsin , USA – have been described elsewhere ( Belgrave et al . , 2014; Gern et al . , 2002; NAC Manchester Asthma and Allergy Study Group et al . , 2002; Lemanske , 2002 ) . COAST , like CAS , was comprised of high-risk individuals with a known family history of asthma or allergy; while MAAS included individuals without family history . In terms of matching variables for replication , all cohorts had measurements that covered the three major ‘domains’ of asthma pathogenesis: respiratory infection , allergen sensitisation , and clinical or demographic background . COAST had a comprehensive collection of respiratory infection and IgE-type measurements , but no IgG4 measurements . MAAS had multiple measurements of IgE and SPT-type variables . Following consultation with investigators from all three cohorts , clustering features were matched based on proximity of timepoint and phenotype . Respiratory infection phenotypes ( ARI , LRI , URI , fLRI , wLRI ) were generated in COAST and MAAS using recorded data , to approximate CAS infection phenotypes as closely as possible . Specifically , LRI was defined as respiratory infection with evidence of lower respiratory tract involvement in the form of chest sounds ( wheeze , rattle , whistle ) , or increased respiratory effort ( retractions , tachypnea , cyanosis ) ; URI was defined as a cold-like infection limited to the upper respiratory tract , without signs of LRI . IgE and IgG4 assays for MAAS and COAST were performed using ImmunoCAP and UniCAP , respectively . Both replication cohorts recorded basic demographic data , and exposures to pets , childcare , and tobacco smoke . The complete list of clustering features and the matching scheme across cohorts is provided in Supplementary file 1 – table supplement 1 . The npEM clusters were described and validated in MAAS and COAST . This replication was performed by applying the density-function-derived classifier used previously for the low-missingness CAS subjects . Because these external cohorts did not necessarily share the same clustering features or variables as CAS ( Supplementary file 1 – table supplement 1 ) , we assumed that the respective densities for these variables were fjKxij=1 for the jth feature and Kth cluster . In doing so , this was effectively the same as using a model where the missing features were excluded , and only those features common to both CAS and MAAS ( or COAST ) were used; or equivalently , where we assumed that each member of MAAS or COAST was missing values in those particular features . Because these 'CAS-derived' npEM models were non-identical to the original npEM models in CAS , we tested whether 'MAAS-like' and 'COAST-like' algorithms ( CAS-derived model as applied to MAAS or COAST , respectively ) generated similar clusters to the original CAS clusters , when applied back onto CAS ( Results ) . Internal validation of the clusters in the complete-case CAS dataset was performed by use of silhouette widths . Briefly , we calculated the silhouette widths for each cluster as per Rousseeuw ( 1987 ) . For an individual , the closer the silhouette width is to one , the more appropriate the cluster membership; while the closer it is to negative one , the more likely it has been misclassified . Cluster stability was assessed by performing leave-one-out ( LOO ) analysis – that is , we applied the npEM algorithm to a subset of the complete-case dataset – an N-1 by M dataset ( N = 186 , M = 174 ) for a total of N times , leaving out an individual each time . A similar process was repeated M times on an N by M-1 dataset , leaving out one clustering feature at a time . The Jaccard indices for each iteration were then calculated in comparison to known clusters from the original complete-case N by M dataset , and averaged across each assigned cluster . Cluster labels for each iteration were assigned based on whichever complete-case cluster yielded the smallest Jaccard index . This whole process was then repeated with 10 random seeds ( ‘set . seed ( 1 ) ’ through to ‘set . seed ( 10 ) ' ) for determining the initial state for npEM . The final averaged Jaccard indices for each cluster thus represented the mean stability of each cluster . Decision tree analysis was performed using a number of different partitioning schemes . Classification trees with recursive partitioning were built from CAS clusters using the R package ‘rpart’ ( Therneau and Atkinson , 2015 ) , an open-source implementation of CART . The motivation for decision trees was to identify the variables that most strongly separated the clusters and wheezing status , and not necessarily variables that were most predictive . For tree outcomes ( end-nodes ) , we investigated both cluster membership and presence of age-5 wheeze given cluster membership . That is , decision trees were generated to identify the biological features that most strongly distinguished each npEM cluster ( ‘Simple Tree’ ) , as well as npEM cluster ×age-5 wheeze status ( ‘Comprehensive Tree’ ) . We used two different schemes for selecting predictors on which to base the partitions: 1 ) include all predictors that were used as clustering features in the original npEM model; 2 ) include only predictors from one timepoint ( variables from age 6 m , 1 , 2 or 3 ) . The motivation for the latter was that we wanted to see whether measurements taken at a specific timepoint in early infancy could strongly distinguish between clusters . For the former scheme , we excluded all age-5 features related to wheeze ( e . g . LRIs , wheezy LRIs at age 5 ) as decision nodes , because of definitional overlap with our primary outcome of interest ( age-5 wheeze ) . Decision trees were then pruned based on the complexity parameter that minimised cross-validated error . Final classification into tree clusters was manually performed based on the pruned tree , and not by automatic classification using the ‘predict’ function for the ‘rpart’ tree object – this was because , for the latter , individuals who are missing key variables were re-classified based on the next best , non-missing , surrogate variable ( Therneau and Atkinson , 2015 ) . Thus , it resulted in children being erroneously classified into a tree cluster even when they were missing key classifier variables . The decision tree analyses generated thresholds which were then compared with existing thresholds for atopy ( any specific IgE at age 2 ≥ 0 . 35 kU/L , and/or any specific SPT at age 2 ≥ 2 mm ) ( Frith et al . , 2011 ) in terms of predicting disease outcomes of interest . We performed statistical analyses comparing clusters in terms of multiple variables , especially those not used as clustering features . Of interest to us were the primary outcomes of asthma diagnosis and parent-reported wheeze at each timepoint . Where appropriate , we used t-tests , Mann-Whitney-Wilcoxon tests , ANOVAs , Kruskal-Wallis tests , chi-squared and Fisher exact tests; and logistic and linear regression . For summary statistics , multiple testing adjustment was performed using the Benjamini-Yekutieli ( BY ) method , for all across-cluster tests ( Cluster × trait ) ; and for all comparisons between clusters ( CAS1 vs . 2 , 1 vs . 3 , and 2 vs . 3 ) . The BY method was chosen as it accounted for positive dependency across the highly correlated variables in the CAS dataset ( Benjamini and Yekutieli , 2001 ) . For variables that underwent logarithmic transformation for statistical analysis , we used geometric mean to describe central tendency . We then determined the predictors for age-5 wheeze within each cluster . Repeated-measures ANOVAs were performed for selected predictors of age-5 wheeze . For each potential predictor , generalised linear regression models ( GLMs ) were generated with and without a base set of covariates ( sex , family history of asthma , BMI where available ) . The pool of variables found to be statistically significant ( at least p<0 . 05 ) in the above analyses were further restricted , such that strongly collinear predictors were avoided , and at most one timepoint was considered for each predictor type . Targeted multiple regression models were then built by selecting predictors from this constrained pool . Stepwise backward elimination was applied , in which the predictor with the largest p-value was eliminated at each step , until all remaining predictors have significant p<0 . 05 . Using the ‘lrtest’ function from the R package ‘Epidisplay’ ( Chongsuvivatwong , 2015 ) , likelihood ratios were examined to check how much cluster membership or classification improved upon prediction of age-5 wheeze compared to traditional makers of atopy .	Asthma causes wheezy and troubled breathing , and can be life-threatening . Scientists and doctors understand that asthma begins in early childhood . Chest infections , exposure to bacteria , viruses , and allergies may cause or trigger asthma . One person with asthma may not have the same origins as another . But it is not yet clear how various triggers may interact to trigger or exacerbate asthma . To disentangle how these factors contribute to asthma , experts have tried to group people with asthma into subgroups . Unfortunately , the groups often vary from expert to expert . Now , some scientists are using computers to sort patients with asthma . The scientists let the computers decide the best criteria for sorting patients . This way the machines may identify patterns that are not obvious to humans . Using this computer-based approach , Tang et al . sorted Australian children with asthma into 3 groups based on their early life allergies and respiratory health . One group has high-risk asthma with frequent chest infections and strong allergic responses . The other two groups are low-risk , but they respond differently to allergy and infection . Common tests used by doctors to diagnose patients with allergy or asthma may not work the same with all three groups . The bacteria found in the nose influence the risk of asthma , even in patients who are well , and the way this occurs varies by group . Similar groups were also found among children with asthma in the United States and the United Kingdom . Learning more about subgroups of patients with asthma may help other scientists and doctors design better ways to diagnose , treat , or prevent asthma . Working together with scientists around the world to determine how to best describe subgroups of people according to asthma type and risk is a critical step in the process . Tang et al . hope other scientist will test whether these three groups are also found in people from other parts of the world .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology" ]	2018	Trajectories of childhood immune development and respiratory health relevant to asthma and allergy
While the aging process is central to the pathogenesis of age-dependent diseases , it is poorly understood at the molecular level . We identified a mouse mutant with accelerated aging in the retina as well as pathologies observed in age-dependent retinal diseases , suggesting that the responsible gene regulates retinal aging , and its impairment results in age-dependent disease . We determined that a mutation in the transmembrane 135 ( Tmem135 ) is responsible for these phenotypes . We observed localization of TMEM135 on mitochondria , and imbalance of mitochondrial fission and fusion in mutant Tmem135 as well as Tmem135 overexpressing cells , indicating that TMEM135 is involved in the regulation of mitochondrial dynamics . Additionally , mutant retina showed higher sensitivity to oxidative stress . These results suggest that the regulation of mitochondrial dynamics through TMEM135 is critical for protection from environmental stress and controlling the progression of retinal aging . Our study identified TMEM135 as a critical link between aging and age-dependent diseases . One explanation for why age-dependent diseases manifest themselves in an age-dependent manner is that disease-causing mechanisms interact with age-dependent cellular changes that normally occur in aging . The common phenomena observed in both aging and age-dependent diseases may provide clues to this interaction . For example , one of the major age-dependent changes that generally occur in the tissue is accumulation of damages caused by oxidative stress ( Harman , 1956 , 1972a , 1972b ) . As by-products of normal cellular respiration , reactive oxygen species ( ROS ) are constantly generated in cells mainly in the mitochondria . When cellular production of ROS overwhelms its antioxidant capacity ( state referred to as 'oxidative stress' ) , ROS damages cellular macromolecules such as lipids , protein , and DNA/RNA . In the course of aging , such damages caused by ROS are thought to accumulate and contribute to the development of age-dependent tissue dysfunctions . Increase in oxidative damage has been observed in a number of age-dependent diseases as well , and its involvement in the pathogenesis of these diseases has been widely suggested ( Davies , 1995 ) . Related to the oxidative damage , another phenomenon that is observed in both aging and age-dependent diseases is the decline in mitochondrial function ( Lenaz , 1998 ) . Mitochondria are the organelle that consumes over 90% of cellular oxygen and generates ROS ( Harman , 1981; Murphy , 2009 ) . Due to its proximity to the site of ROS generation , mitochondrial components are particularly susceptible to ROS-mediated oxidative damage ( Cadenas and Davies , 2000 ) . Prolonged exposure to ROS during aging is thought to result in mitochondrial dysfunctions and significantly contribute to the development of pathologies associated with aging . There is also strong evidence that mitochondrial dysfunction occurs early and acts causally in the pathogenesis of age-dependent neurodegenerative diseases ( Lin and Beal , 2006 ) . While these phenomena indicate some of the common aspects between aging and age-dependent diseases , the mechanisms linking these two processes have not been elucidated at the molecular level . Given the complexity in both the aging process and age-dependent diseases , as well as countless variables ( including genetic an environmental variables ) that exist among human population , it is extremely challenging to study the mechanisms underlying the aging process and how they relate to the disease-causing mechanism in humans . An animal model that shows accelerated aging as well as age-dependent disease symptoms could provide a useful experimental system for this purpose . Furthermore , a forward genetics approach starting with an animal model with these symptoms offers a potential of identifying a responsible gene that is not previously known to be associated with the aging process nor age-dependent diseases . We isolated an N-ethy-N-nitrosourea ( ENU ) -induced mutant mouse line , FUN025 , that exhibits age-dependent retinal abnormalities with a trajectory similar to that found with retinal aging observed in wild-type ( WT ) mice ( Higuchi et al . , 2015 ) but with an early onset and faster progression . In addition , we found that the FUN025 mutation leads to pathologies observed in age-dependent retinal diseases such as age-related macular degeneration ( AMD ) . These phenotypes in FUN025 mice suggest that the responsible gene is involved in regulating the rate of aging in the retina , and that its impairment leads to development of age-dependent disease . In this study , we identify a gene mutation that is responsible for retinal abnormalities in FUN025 mice and characterize the novel molecular functions of this gene/protein associated with regulation of mitochondria as well as sensitivity to oxidative stress . Our findings reveal a molecular link between the aging process and age-dependent diseases , and a molecular mechanisms leading to age-dependent disease pathologies . FUN025 mice were isolated through fundus examination in an ENU mouse mutagenesis project ( Pinto et al . , 2004; Vitaterna et al . , 2006 ) , and were found to exhibit retinal abnormalities similar to those observed in aged WT mice ( Higuchi et al . , 2015 ) . It has been shown that , in the WT retina , age-dependent abnormalities including retinal degeneration , increased number of ectopic synapses and increased retinal stress all start from the peripheral retina by 8 months of age on an aging-susceptible A/J background but later on a less susceptible B6 background , which progress to the central retina with age ( Higuchi et al . , 2015 ) . Histological analysis of homozygous FUN025 mutant and C57BL/6J ( B6 ) WT retina at two and seven months of age revealed that a decrease in the ONLT Index [the thickness of outer nuclear layer ( ONL ) normalized by the thickness of inner nuclear layer ( INL ) ] in FUN025 mice compared to WT mice is observed by two months of age and becomes more pronounced by seven months of age , indicating progressive loss of photoreceptor cells in the FUN025 retina ( Figure 1A , B ) . Immunohistochemical analysis demonstrated that , while most of the presynaptic photoreceptor terminals ( PSD95 , green ) line up in the outer plexiform layer ( OPL ) in close opposition to the bipolar cell postsynaptic structures ( PKC , red ) in the WT retina , ectopic localization of presynaptic terminals and abnormal extension of bipolar cell dendrites into the ONL were observed in the FUN025 retina ( Figure 1C , D ) . Quantification of ectopic synapses indicated that a significantly increased number of ectopic synapses are observed in the peripheral retina of FUN025 mice by two months of age , and in the central retina by seven months of age , indicating that this phenotype progresses from the peripheral to the central retina ( Figure 1D ) . We also observed that the sign of retinal stress , up-regulation of glial fibrillary acidic protein ( GFAP ) ( Higuchi et al . , 2015; Lewis and Fisher , 2003 ) , was significantly increased in the peripheral retina of FUN025 mice compared to the WT mice by two months of age , which later progresses toward the central retina ( Figure 1E ) . Thus , FUN025 retina exhibits age-dependent retinal abnormalities with a trajectory similar to that found with retinal aging observed in WT mice ( Higuchi et al . , 2015 ) but with an early onset and faster progression . 10 . 7554/eLife . 19264 . 003Figure 1 . Age-dependent retinal abnormalities in FUN025 mice . ( A–B ) A significant decrease of the ONLT index occurred by two months of age in FUN025 retina . Mo = months . Data from n = 10 WT ( 2 Mo ) , n = 4 FUN025 ( 2 Mo ) , n = 20 WT ( 7 Mo ) , n = 8 FUN025 ( 7 Mo ) mice . Scale bar = 20 μm . ( C–D ) Ectopic synapses were observed as bipolar cell neurites ( PKC , red ) and photoreceptor synaptic terminals ( PSD95 , green ) extending into the ONL indicated by asterisks ( C ) . Scale bar = 10 μm . Significant increase of ectopic synapses were found earlier in the peripheral retina , and later in the central retina of FUN025 compared to WT mice . Data for central retina from n = 3 WT ( 2 Mo ) , n = 3 FUN025 ( 2 Mo ) , n = 3 WT ( 7 Mo ) , n = 3 FUN025 ( 7 Mo ) mice; data for peripheral retina from n = 5 WT ( 2 Mo ) , n = 6 FUN025 ( 2 Mo ) , n = 6 WT ( 7 Mo ) , n = 6 FUN025 ( 7 Mo ) mice . ( E ) GFAP ( green ) upregulation was progressively observed in the FUN025 retina . ONL: outer nuclear layer . INL: inner nuclear layer . Outer nuclear layer thickness ( ONLT ) index = ONL thickness/INL thickness . p<0 . 05 , Student’s t-test . All data are mean ± s . e . m . Scale bar = 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 19264 . 003 We further investigated whether the early-onset and accelerated aging process in the FUN025 retina is accompanied by pathologies observed in age-dependent diseases . Punctate light deposits were found in the fundus photography of the eyes from FUN025 mice ( Figure 2A ) , which may be due to the accumulation of autofluorescent cells and aggregates observed between photoreceptors and retinal pigment epithelium ( RPE ) ( Figure 2B ) . These cells/aggregates and RPE exhibit autofluorescence detected using a DPS 561 laser ( Figure 2B , C ) , which resembles that of lipofuscin , protein and lipid rich aggregates known to accumulate in aging tissues and suggested to be involved in age-dependent diseases such as AMD and Alzheimer's disease ( Giaccone et al . , 2011; Nowotny et al . , 2014 ) . The hyper-autofluorescent aggregates ( Figure 2B , arrowhead ) resemble subretinal drusenoid deposits ( SDD ) ( Rudolf et al . , 2008 ) , a type of extracellular lesion between the photoreceptors and RPE often observed in AMD patients ( Curcio et al . , 2013; Zweifel et al . , 2010 ) . To determine the cell type of these autofluorescent cells , immunostaining was performed in 7-month old retinal sections . The subretinal autofluorescent cells were positive for a microglia marker , Iba1 , and a macrophage marker , F4/80 , in the FUN025 retina ( Figure 2C ) . The Iba1+ cells near the apical surface of the RPE have very few processes and show immunoreactivity to F4/80 , suggesting that they have transformed from microglia to macrophages ( Figure 2C ) . These data also suggest that the RPE layer is the origin of stress , toward which the microglia migrate ( Jonas et al . , 2012; Mcglade-Mcculloh et al . , 1989 ) . Long-term , low-grade inflammation ( innate immunity ) has been widely postulated as a part of the aging process , and is also enhanced in many age-dependent diseases ( Salminen et al . , 2012 ) . To test if innate immunity is increased in FUN025 mice , immunostaining was performed with an inflammasome marker , NLR Family , Pyrin Domain Containing 3 ( NLRP3 ) and a functional caspase-1 subunit marker , caspase-1 p10 on retinal sections from seven-month old FUN025 and WT mice . Immunoreactivity to both makers was increased in the neural retina as well as RPE of FUN025 mice compared to WT mice ( Figure 2D ) , indicating elevated innate immunity ( Franchi et al . , 2009 ) . In addition , by seven months , the RPE ( highlighted by anti-CRALBP antibody ) is uniformly thickened in FUN025 mice compared to that of controls while each RPE cell is still intact ( the nuclei of RPE highlighted by anti-Otx1+Otx2 antibody ) ( Figure 2E ) . Similar increases in innate immunity markers ( Tarallo et al . , 2012 ) and thickness of RPE ( Acton et al . , 2012; Karampelas et al . , 2013; Zhao and Vollrath , 2011 ) have been observed in AMD patients . To determine if early-onset of retinal aging as well as retinal pathologies observed in FUN025 mice affect the ability of the retina to respond to light , we performed transcorneal electroretinograms ( ERG ) recordings from seven-month-old FUN025 mice and their WT littermates . ERG data revealed significantly reduced scotopic ( dark-adapted ) a-wave and b-wave from the rod pathway , and a modest reduction in photopic ( light-adapted ) b-wave from the cone pathway in FUN025 mice compared to WT controls ( Figure 2F ) indicating impaired visual function in FUN025 mice . In conclusion , FUN025 mice provide a mouse model with accelerated aging process and age-dependent disease pathologies in the retina . 10 . 7554/eLife . 19264 . 004Figure 2 . FUN025 mice show AMD-like pathologies . ( A ) Punctate light deposits were found in the fundus photography of the eyes from WT and FUN025 mice . ( B ) Autofluorescent cells/aggregates ( indicated by arrows ) and lipofuscin-like autofluorescence were observed in proximity to the apical surface of the RPE in FUN025 mice . Scale bar = 60 μm . ( C ) Iba1 ( microglia/ macrophage marker ) and F4/80 ( macrophage marker ) positive cells were found in the FUN025 retina at seven months , whereas very few Iba1 positive cells were found in the WT retina at seven months . Scale bar = 20 μm . ( D ) Signals for inflammasome markers , NLRP3 and caspase1 , increased in the RPE in both peripheral and central retina from FUN025 mice compared to WT control . ( E ) At seven months of age , the RPE ( highlighted by CRALBP staining ) thickness is significantly increased in both central and peripheral retina of FUN025 mice compared to control mice . The RPE nuclei were highlighted with Otx1+Otx2 . Data from n = 4 mice per genotype . ( F ) Transcorneal electroretinograms ( ERG ) recordings from seven-month-old FUN025 mice and their WT littermates . Both scotopic ( dark-adapted ) ERG a- and b-waves from the rod pathway were markedly reduced in FUN025 mice . A reduction was also observed in photopic ( light-adapted ) ERG b-wave from the cone pathway with higher flash intensity , while no difference between FUN025 and WT was observed in the photopic a-wave , majority of which is postreceptoral in origin . Data from n = 5 mice per genotype . p<0 . 05 , two-way analysis of variance ( ANOVA ) . All data are mean ± standard error of the mean ( s . e . m . ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19264 . 004 To identify the causative gene/mutation for retinal phenotypes in FUN025 mice , we performed a genome-wide linkage analysis . Genetic mapping of the FUN025 mutation was performed using F2 intercrosses ( C57BL6-FUN025 x C57BR/cdJ and C57BL6-FUN025 x 129S1/SvImJ ) and genetic markers to distinguish between the alleles of FUN025 ( C57BL/6J ) and C57BR/cdJ or 129S1/SvImJ ( Figure 3A ) . The F2 mice were phenotyped by fundus photography and histological analysis at 3 months . Genotype and phenotype data from 70 F2 mice generated a minimal genetic region flanked by SNPOlf294T>C ( 93764448 ) and Rs31011252 , containing 25 genes . Using a SureSelectXT custom library ( Agilent Technologies ) , we performed a sequence capture array on FUN025 genomic DNA followed by paired end sequencing . Standard bioinformatic analyses of our sequencing data revealed a point mutation ( T>C ) in the splice-donor site adjacent to exon 12 of Tmem135 in FUN025 mice ( Figure 3B , C ) . The consensus sequence of mouse splice donor sites depicts the necessity of the GT sequence at positions 1 and 2 downstream of the exon boundary for the functionality of the site ( Carmel et al . , 2004 ) . The mutation disrupts the splice donor site ( Figure 3C ) , resulting in skipping of exon 12 and a frame shift creating an early stop codon in FUN025 mice ( Figure 3D , F ) . The probability of forming transmembrane helices predicted by the program TMHMM ( v . 1 . 0 ) ( http://www . cbs . dtu . dk/ ) suggested that WT TMEM135 contains five transmembrane helices ( Figure 3G and Figure 3—figure supplement 1 ) , while the 4th and 5th transmembrane helices are abolished and the orientation of the remaining 3 transmembrane helices in the membrane is reversed in mutant TMEM135 compared to WT TMEM135 ( Figure 3G and Figure 3—figure supplement 1 ) . The c-terminal region of the mutant TMEM135 is also shorter due to the early stop codon ( Figure 3E , G ) . Thus , the FUN025 mutation in Tmem135 which results in the shorter c-terminal region with amino acid sequence changes , predicted loss of two transmembrane helices and reversed orientation within the membrane likely impairs the normal functions of the TMEM135 protein . The TMEM135 antibody recognizing the N-terminus of TMEM135 protein detected the mutant TMEM135 protein in the FUN025 brains as well as the WT TMEM135 protein in WT brains by western blot analysis ( Figure 3H ) . 10 . 7554/eLife . 19264 . 005Figure 3 . Identification of a Tmem135 mutation in FUN025 mice . ( A ) Minimal genetic region of FUN025 on chromosome 7 determined by genetic mapping . ( B ) A point mutation ( T > C ) in the splice-donor site adjacent to exon 12 of Tmem135 in FUN025 mice . ( C ) The consensus sequence of mouse splice donor sites , depicting the necessity of the GT sequence at positions 1 and 2 downstream of the exon boundary for the functionality of the site . The C57BL/6J and FUN025 sequences are shown below , demonstrating the disrupted site in FUN025 mice . ( D ) RT-PCR spanning exon 12 and sequencing the product revealed the absence of this exon in the FUN025 retina . ( E ) Amino acid sequences of the C-terminus of TMEM135 in C57BL/6J and FUN025 mice . The WT protein is 458 amino acids long , whereas the truncated mutant protein is 406 . The change in amino acid sequence is highlighted in red . ( F ) Consequences of the mutation in the genomic sequence of Tmem135 . The mutation adjacent to exon 12 of Tmem135 ( red star ) results in a non-functional splice donor site , causing skipping of exon 12 . This results in a frameshift and an early stop codon ( chr7: 96 , 290 , 429 , NCBI build 37 ) . Locations of the stop codons are highlighted in red . ( G ) A predicted structure of TMEM135 having five transmembrane domains . The FUN025 mutation is predicted to result in a protein with only three transmembrane domains , whose orientation in the membrane is reversed . The rest of the c-terminal region is absent due to the early stop codon ( asterisk ) . ( H ) Western blot for TMEM135 in WT and FUN025 brains . GAPDH was used as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 19264 . 00510 . 7554/eLife . 19264 . 006Figure 3—figure supplement 1 . Identification of the causative gene , Tmem135 , for the FUN025 mutation . Wild-type TMEM135 protein is predicted to form five transmembrane helices by the program TMHMM ( v . 1 . 0 ) ( http://www . cbs . dtu . dk/ ) . The FUN025 mutation decreases the probability of forming the 4th and 5th transmembrane helix . DOI: http://dx . doi . org/10 . 7554/eLife . 19264 . 00610 . 7554/eLife . 19264 . 007Figure 3—figure supplement 2 . A T > C mutation in Tmem135 fails to complement FUN025 . Using the CRISPR/Cas9/sgRNA system , we generated C57BL/6J mice that carry a point mutation ( T > C ) in intron 12 of Tmem135 that is the same as observed in FUN025 mice ( see Materials and methods ) . We refer to these mice as T > C /+ mice . The T > C heterozygous mice were crossed with the FUN025 homozygous mice to produce F1 ( T > C/FUN025 compound heterozygous ) mice , which were analyzed for retinal phenotypes . At 11 weeks , T > C/FUN025 mice showed increased ectopic synapses ( upper panel ) , increased Iba1+ cells ( middle panel ) , and increased GFAP immunoreactivity ( lower panel ) in peripheral retina compared to the age-matched FUN025 heterozygous control mice , which are similar to those observed in FUN025 homozygous mice . These results indicate non-complementation between FUN025 and T > C mutations , demonstrating that the point mutation ( T > C ) in Tmem135 is the FUN025 causative mutation . Scale bar = 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 19264 . 007 To confirm that this mutation , rather than any other unknown mutations that occurred in the ENU-induced FUN025 mutant line , is responsible for the retinal phenotypes in FUN025 mice , a complementation test was performed . We introduced the same point mutation ( T > C ) as observed in FUN025 mice in the intron 12 of Tmem135 ( Chr7:96 , 296 , 478 ) in C57BL/6J mice using the CRISPR/Cas9 system ( T > C mice ) . The T > C heterozygous mice were crossed with the FUN025 homozygous mice to produce F1 ( T > C/FUN025 compound heterozygous ) mice , which were analyzed for retinal phenotypes . The F1 mice exhibited retinal phenotypes similar to FUN025 homozygous mice , indicating non-complementation ( Figure 3—figure supplement 2 ) . This result demonstrates that the point mutation ( T > C ) in Tmem135 is indeed the FUN025 causative mutation . Therefore , we now designate homozygous FUN025 mice as Tmem135FUN025/FUN025 . TMEM135 protein was suggested to be involved in fat storage and the regulation of longevity in C . elegans ( Exil et al . , 2010 ) , but the function of TMEM135 is not yet clearly characterized . In order to elucidate the mechanistic role of TMEM135 , we characterized the localization of the TMEM135 protein in cultured cells and mouse retina in vivo . Primary WT mouse fibroblast cells ( MFs ) were co-transfected with the green fluorescent protein ( GFP ) -tagged vector containing Tmem135 and DsRed2-tagged mitochondria vector ( pDsRed2-Mito Vector , Clontech , Mountain View , CA ) . Immunofluorescence data showed that the GFP-TMEM135 protein displays an intracellular vesicular expression pattern in the cytoplasm , and a proportion is found in punctate structures on mitochondria ( Figure 4A ) . Colocalization of TMEM135 to mitochondria was further confirmed in WT MFs using an anti-TMEM135 antibody and an AcGFP1-tagged mitochondria vector ( pAcGFP1-Mito Vector , Clontech , Mountain View , CA ) ( Figure 4A ) as well as in WT MFs with anti-TMEM135 antibody and a red-fluorescent dye that stains mitochondria ( MitoTracker Red CMXRos ) ( Figure 4B , upper panel ) . Notably , the proportion of TMEM135 signals tend to distribute to small foci along the surface of mitochondria , at mitochondrial constriction sites , and at the tips of individual mitochondria ( Figure 4A , B ) , and there appears to be less colocolization of TMEM135 on mitochondria in FUN025 MFs ( Figure 4B , lower panel ) . Immunoelectron microscopy revealed that transfected GFP-TMEM135 localizes on the surface of mitochondria ( Figure 4C ) . Furthermore , colocalization of TMEM135 to mitochondria was observed in monkey kidney fibroblast-like cells ( Cos-7 ) ( Figure 4D ) and mouse primary hippocampal neurons ( Figure 4E ) . The mitochondria fraction isolated from mouse brains showed TMEM135 signals by immunoblotting , also indicating that TMEM135 is associated with mitochondria ( Figure 4F ) . In the retina from WT and FUN025 mice , stronger TMEM135 signals were detected in the ganglion cell layer ( GCL ) , inner plexiform layer ( IPL ) , outer plexiform layer ( OPL ) , inner segments of photoreceptor cells , and RPE , which colocalized with mitochondria labeled with anti-TOMM20 antibody ( Figure 4G ) . The colocalization of TMEM135 with mitochondria was also observed in the primary mouse RPE cell culture ( Figure 4H ) . Similar to the observation in fibroblasts ( Figure 4B ) , less colocalization of TMEM135 with mitochondria was observed in FUN025 RPE cells compared to WT RPE cells ( Figure 4H ) . In conclusion , a proportion of TMEM135 is strongly associated with mitochondria in vivo , although other intracellular organelles and small vesicles may be also associated with TMEM135 . 10 . 7554/eLife . 19264 . 008Figure 4 . Localization of TMEM135 to the mitochondria . ( A ) Mitochondrial localization of TMEM135 in MFs co-transfected with GFP tagged TMEM135 vector ( green ) and DsRed2 tagged mitochondria vector ( red ) . GFP-TMEM135 signals were detected as puncta to the mitochondria as well as in the cytoplasm . Colocalization of TMEM135 and mitochondria in MFs transfected with AcGFP1 tagged mitochondria vector ( green ) and immunostained with anti-TMEM135 antibody ( red ) . Scale bar = 10 μm . ( B ) Colocalization of TMEM135 ( anti-TMEM135 antibody , green ) and mitochondira ( MitoTracker , red ) in wild-type and FUN025 mouse fibroblasts . Scale bar = 10 μm . ( C ) Immuno-EM revealed localization of GFP-tagged TMEM135 to the mitochondria . ( D–E ) Colocalization of TMEM135 ( anti-TMEM135 antibody , green ) and mitochondria ( MitoTracker and TOMM20 , red ) in Cos-7 cells and primary mouse hippocampal neuron . Scale bar = 10 μm . ( F ) The mitochondrial fraction isolated from the WT mouse brain show TMEM135 signals by immunoblotting . Following proteins were used as organelle markers: MFN2–mitochondria; LAMP2–lysosome; Lamin B1– nucleus; PDI– endoplasmic reticulum ( ER ) . ( G ) Strong TMEM135 signals ( green ) in GCL , IPL , OPL , inner segments of photoreceptor cells , and RPE from wild-type and FUN025 mouse retina . Throughout the retina , TMEM135 is colocalized with mitochondria ( anti-TOMM20 antibody , red ) . Scale bar = 10 μm . ( H ) Colocalization of TMEM135 ( anti-TMEM135 antibody , green ) and mitochondria ( MitoTracker , red ) in wild-type and FUN025 primary mouse RPE cell culture . Scale bar = 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 19264 . 008 We next investigated what roles TMEM135 might play in mitochondria . TMEM135 punctate structures were often observed at mitochondrial constriction sites and at the tips of individual mitochondria . This unique localization pattern suggests a role of TMEM135 in the regulation of mitochondrial morphology . We isolated MFs from WT , Tmem135FUN025/FUN025 and transgenic mice overexpressing WT Tmem135 ( Tg-Tmem135 ) ( Figure 5—figure supplement 1A–B ) , and stained the mitochondria with MitoTracker Red . Compared to WT cells with mitochondria of all different sizes and shapes , Tmem135FUN025/FUN025 cells showed over-fused mitochondrial networks whereas Tg-Tmem135 cells exhibited over-fragmented mitochondrial networks ( Figure 5A ) . We used a morphology scoring assay ( Loson et al . , 2013 ) in which each cell was categorized as having fragmented , tubular , elongated or aggregated mitochondria . Among Tmem135FUN025/FUN025 MFs , more cells were found to have elongated mitochondria relative to cells isolated from WT mice , indicating that more fusion occurs compared to fission in Tmem135FUN025/FUN025 cells ( Figure 5B ) . In contrast , among Tg-Tmem135 MFs , more cells were found to have fragmented mitochondria relative to WT cells , suggesting that more fission than fusion takes place in Tg-Tmem135 cells ( Figure 5B ) . Next , we compared the number and size of mitochondria between Tmem135FUN025/FUN02 , Tg-Tmem135 and WT MFs ( Figure 5C–E ) . The size of mitochondria increases and its number decreases in Tmem135FUN025/FUN025 cells indicating that more mitochondrial fusion occurs than fission leading to elongated mitochondria . In contrast , the size and mass of mitochondria decrease in Tg-Tmem135 cells indicating that more fission than fusion takes place producing over-fragmented mitochondria . Observation that mitochondrial size increases in Tmem135FUN025/FUN025 cells was further confirmed by knocking down Tmem135 using RNA interference ( RNAi ) . Knocking down 68% of Tmem135 RNA in WT MFs using a short interfering RNA ( siRNA ) against Tmem135 resulted in a significant increase in mitochondrial size compared to WT MFs treated with scrambled siRNA ( Figure 5F ) . We next analyzed the retinal mitochondrial morphology in the RPE as well as inner segments of photoreceptor cells from 12-month-old WT and FUN025 mice using electron microscopy ( Figure 5G ) . We observed enlarged mitochondria in both the RPE and inner segments of photoreceptor cells from FUN025 mice compared to those from WT mice ( Figure 5G , H ) . Taken together , these results indicate that TMEM135 is involved in the regulation of the balance between mitochondrial fission and fusion ( mitochondrial dynamics ) . 10 . 7554/eLife . 19264 . 009Figure 5 . TMEM135 is involved in the balance of mitochondrial fission and fusion . ( A ) Morphology of mitochondria ( MitoTracker; red ) in WT , Tmem135FUN025/FUN025 ( FUN025 ) and Tg-Tmem135 MFs . Scale bar = 10 μm . ( B ) Scoring of mitochondrial network morphologies in WT , FUN025 and Tg-Tmem135 MFs . Data from n = 290 WT , n = 304 FUN025 , and n = 372 Tg-Tmem135 cells; 3 mice per genotype . ( C–E ) Quantification of size , number and coverage of mitochondria in WT , FUN025 and Tg-TMEM135 MFs . Data from n = 49 WT , n = 104 FUN025 , and n = 29 Tg-Tmem135 cells; 3 mice per genotype . ( F ) Knocking down Tmem135 by siRNA against Tmem135 results in increased mitochondrial size in WT MFs . Data from n = 72 cells with scrambled siRNA , n = 81 cells with Tmem135 siRNA . Two tailed , unpaired , Student’s t-test . ( G ) EM revealed the morphology of mitochondria in RPE and inner segments of photoreceptor cells from wild-type and FUN025 mice . m = mitochondria; m . g . = melanin granules; n = necleus; o . s . = outer segments; i . s . = inner segments . Scale bar = 1 μm . ( H ) Quantification of mitochondria size in RPE and inner segments of photoreceptor cells from wild-type and FUN025 mice . RPE data from 300 mitochondria from three wild-type mice and 400 mitochondria from four FUN025 mice . Inner segments data from 200 mitochondria from four wild-type mice and 200 mitochondria from four FUN025 mice . ( I ) Immunofluorescence of DRP1 ( anti DRP1 antibody , green ) and mitochondria ( MitoTracker , red ) in WT and FUN025 MFs . Scale bar = 10 μm . ( J–K ) Quantification of DRP1 puncta density and fluorescence intensity on mitochondria . ( L ) Time-lapse fluorescence imaging ( modified as described in Materials and methods ) of a living HT22 cell expressing TMEM135-GFP and labeled with MitoTracker at the indicated time points . TMEM135 puncta that are located relatively close to the mitochondrial surface are shown in magenta , whereas those relatively far from the mitochondrial surface are shown in turquoise . Mitochondria ( gray ) was made partial transparent in order to see TMEM135 puncta located on the back of mitochondria . p<0 . 05 , p<0 . 01 , p<0 . 001 , *p<0 . 0001 , two-way ANOVA . All data are mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 19264 . 00910 . 7554/eLife . 19264 . 010Figure 5—figure supplement 1 . Generation of Tg-TMEM135 mice . ( A ) Transgenic construct for generating the transgenic mice overexpressing Tmem135 ( Tg-TMEM135 ) . ( B ) Western blot analysis of TMEM135 expression in WT and Tg-TMEM135 ( TG ) mouse brain . GAPDH was used as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 19264 . 01010 . 7554/eLife . 19264 . 011Figure 5—figure supplement 2 . TMEM135 does not inhibit mitochondrial fusion . ( A ) WT and Tg-Tmem135 MFs were transfected with expression vectors encoding MFNs-YFP ( green ) at low and high levels and immunostained with TMEM135 antibody ( red ) . Mitochondria were highlighted by MitoTracker ( blue ) . Nuclei were labeled by DAPI ( cyan ) . Scale bar = 10 μm . ( B ) Western blot analysis of OPA1 , MFN1 , and MFN2 in WT , FUN025 and Tg-TMEM135 ( TG ) fibroblasts . GAPDH was used as a loading control . ( C–E ) Quantification of western blotting results from ( B ) . Protein levels were normalized using GAPDH expression levels . DOI: http://dx . doi . org/10 . 7554/eLife . 19264 . 01110 . 7554/eLife . 19264 . 012Figure 5—figure supplement 3 . A portion of TMEM135 is colocalized with DRP1 on the mitochondria . Confocal fluorescence microscopy of TMEM135 localization ( red ) and DRP1 localization ( green ) on mitochondria ( blue ) . Some of mitochondrial-bound DRP1 puncta colocalize with mitochondrial-bound TMEM135 puncta indicated by arrows . Scale bar = 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 19264 . 012 Two possible mechanisms underlie the mitochondrial morphological changes in Tmem135FUN025/FUN025 and Tg-Tmem135 MFs . One possible mechanism is that TMEM135 may be involved in inhibition of mitochondrial fusion . We tested this hypothesis by promoting mitochondrial fusion through overexpression of mitochondrial fusion factor , mitofusin 2 ( MFN2 ) in WT and Tg-Tmem135 MFs . If TMEM135 inhibits mitochondrial fusion , we would see less elongated mitochondria in Tg-Tmem135 cells compared to WT cells upon MFN2 overexpression . However , we observed similar number of elongated ( with lower MFN2 expression ) and aggregated ( with higher MFN2 expression ) mitochondria in WT and Tg-Tmem135 MFs transfected with pMFN2-YFP ( Figure 5—figure supplement 2A ) . The results indicate that overexpression of Tmem135 in Tg-Tmem135 MFs does not inhibit the mitochondrial morphological changes caused by overexpression of Mfn2 , suggesting that TMEM135 does not inhibit fusion . While it is also possible that downregulation of mitochondrial fusion proteins is responsible for over-fragmented mitochondria in Tg-TMEM135 MFs , which can be overcome by MFN2 overexpression , we found that this is not the case . The protein levels of mitochondrial fusion proteins , optic atrophy 1 ( OPA1 ) , MFN1 and MFN2 were not changed in Tg-TMEM135 MFs compared with WT MFs ( Figure 5—figure supplement 2B–E ) . Additionally , the levels of these proteins are either unchanged ( OPA1 and MFN2 ) or decreased ( MFN1 ) in FUN025 MFs compared with WT MFs ( Figure 5—figure supplement 2B–E ) , indicating that the overly fused mitochondrial network observed in FUN025 MFs was not caused by up-regulation of these mitochondrial fusion proteins . Another possible mechanism is that TMEM135 may promote mitochondrial fission . Dynamin related protein 1 ( DRP1; also known as dynamin 1-like or DNM1L ) is a mitochondrial dynamin-like GTPase that is essential for mitochondrial fission ( Mears et al . , 2011 ) . During mitochondrial fission , DRP1 is recruited from the cytosol onto the mitochondrial outer membrane , where it is assembled into puncta ( Mears et al . , 2011 ) . These puncta consist of oligomeric DRP1 complexes that wrap around and constrict the mitochondrial tubule to mediate fission ( Mears et al . , 2011 ) . We found that some of the punctate DRP1 signals colocalize with TMEM135 puncta on mitochondria ( Figure 5—figure supplement 3 ) , suggesting that TMEM135 may be involved in DRP1-dependent mitochondrial fission . We first tested if DRP1 binding to mitochondria is affected in the Tmem135FUN025/FUN025 MFs using immunofluorescence to visualize DRP1 puncta on mitochondria . In MFs , much of the DRP1 staining was diffused in the cytosol , but a proportion could be found in punctate structures on mitochondria ( Figure 5I ) . We used image analysis to quantify the density ( number of puncta/mitochondrial area ) of DRP1 puncta on mitochondria ( Figure 5J ) . The density of DRP1 puncta detected on mitochondria in Tmem135FUN025/FUN025 MFs was not different from what was detected in WT MFs ( Figure 5J ) , indicating that DRP1 translocation to the mitochondria was not affected in Tmem135FUN025/FUN025 MFs compared to WT MFs . The average total fluorescence per puncta ( Figure 5K ) was even increased in Tmem135FUN025/FUN025 MFs while the mitochondria stayed fused . These results indicate that DRP1 is translocated to mitochondria normally and assembled into higher order structures but appears to stay inactive in Tmem135FUN025/FUN025 MFs . DRP1 has been shown to be activated or inactivated through various post-translational modifications at different amino acids ( Knott et al . , 2008 ) . These results support the notion that TMEM135 may be required for DRP1 activation . To further characterize the role of TMEM135 in morphological changes of mitochondria , we monitored their dynamics in relation to mitochondria in living WT MFs and HT22 cells expressing GFP-TMEM135 . We observed that a proportion of TMEM135 localized at mitochondrial constriction sites where mitochondria later divided ( Figure 5L; Video 1; Video 2 ) . Based on these data , we conclude that TMEM135 is more likely involved in mitochondrial fission rather than fusion . 10 . 7554/eLife . 19264 . 013Video 1 . Live imaging of HT22 cells expressing TMEM135-GFP . Live imaging of a HT22 cell expressing TMEM135-GFP and labeled with MitoTracker Red . TMEM135 puncta that are located relatively close to the mitochondrial surface are shown in magenta , whereas those relatively far from the mitochondrial surface are shown in turquoise . Mitochondria ( gray ) was made partial transparent in order to see TMEM135 puncta located on the back of mitochondria . DOI: http://dx . doi . org/10 . 7554/eLife . 19264 . 01310 . 7554/eLife . 19264 . 014Video 2 . Live imaging of WT fibroblasts expressing TMEM135-eGFP . Live imaging of fibroblasts derived from B6 ( WT ) mice transfected with TMEM135-eGFP plasmid and labeled with MitoTracker Red . TMEM135-eGFP are shown in green and mitochondria are shown in red . DOI: http://dx . doi . org/10 . 7554/eLife . 19264 . 014 We then examined if the structural changes of mitochondria in Tmem135FUN025/FUN025 and Tg-Tmem135 mice affect their mitochondrial functions . First , we analyzed the mitochondrial membrane potential ( ΔΨM ) , and found that Tg-Tmem135 MFs showed a significantly increased ΔΨM and Tmem135FUN025/FUN025 MFs retained a similar ΔΨM as WT MFs ( Figure 6A; Figure 6—figure supplement 1 ) . These results indicate that Tmem135FUN025/FUN025 and Tg-Tmem135 MFs do not lose the ΔΨM , suggesting that the individual unit of the mitochondrial membrane is functional . Next we determined whether impaired mitochondrial dynamics lead to total mitochondrial respiratory impairment within the cells ( Figure 6B ) . Compared to WT MFs , Tmem135FUN025/FUN025 and Tg-Tmem135 MFs showed a lower basal oxygen consumption rate ( OCR ) ( Figure 6C ) . A significantly lower ATP production was detected in Tmem135FUN025/FUN025 and Tg-Tmem135 MFs ( Figure 6D ) . Tmem135FUN025/FUN025 MFs showed a decrease in the maximal respiration and spare respiratory capacity ( SRC ) , whereas Tg-Tmem135 showed no difference in these parameters ( Figure 6E , F ) . The proton leak was not changed in Tmem135FUN025/FUN025 and Tg-Tmem135 MFs ( Figure 6G ) . Our results suggest that the differences in mitochondrial basal and maximal respiration could be due to the change of mitochondrial dynamics . SRC is a measure of the extra capacity available in cells to produce energy in response to increased stress or work ( van der Windt et al . , 2012 ) . Our observation that Tmem135FUN025/FUN025 MFs have significantly reduced SRC suggests that the mutant TMEM135 not only affects oxidative phosphorylation at the normal condition but also decreases the potential ability for the mitochondria to produce more energy under stressed conditions . 10 . 7554/eLife . 19264 . 015Figure 6 . Tmem135 plays a role in mitochondrial metabolism . ( A ) Mitochondrial membrane potential ( ΔΨM ) in Tg-Tmem135 MFs was higher than that in WT MFs , whereas ΔΨM in Tmem135FUN025/FUN025 ( FUN025 ) MFs was comparable to that in WT MFs . Data from n = 3 mice per genotype . n = 1 for WT+CCCP . ( B–G ) Oxygen consumption rates ( OCR ) of MFs were determined with a Seahorse XFe24 Extracellular Flux Analyzer in basal and stimulated conditions ( n = 3 mice per genotypes ) . The areas under the curve from different sections of the experiment ( B ) are shown as individual histograms for basal respiration , ATP production , maximal respiration , spare respiratory capacity , and proton leak ( C–G ) . Data from n = 3 mice per genotype . ( H–I ) Total ROS and superoxide in FUN025 and Tg-Tmem135 MFs were higher compared to WT cells . Data from n = 3 mice per genotype . ( J ) Western blot analysis for SOD1 , SOD2 , SOD3 , GPx1and CAT in WT and FUN025 MFs with and without the hydrogen peroxide treatment . Protein levels were normalized using GAPDH expression levels . p<0 . 05 , p<0 . 01 , p<0 . 001 , **p<0 . 0001 , two-way ANOVA . All data are mean ± s . e . m . . DOI: http://dx . doi . org/10 . 7554/eLife . 19264 . 01510 . 7554/eLife . 19264 . 016Figure 6—figure supplement 1 . Mitochondrial membrane potential in FUN025 and Tg-Tmem135 MFs . ( A–D ) Representative flow cytometry data showing ΔΨM differences between MFs derived from WT ( A ) , FUN025 ( B ) , and Tg-Tmem135 ( C ) mice . CCCP treatment caused decrease of fibroblast ΔΨM and therefore serves as the negative control for fibroblast ΔΨM ( D ) . Unstained fibroblasts were used to gate the quadrants 1–4 ( Q1-Q4 ) which Q4 represents background fluorescence intensity . ( D ) Population with detectable ΔΨM was gated according to CCCP treated WT MFs . DOI: http://dx . doi . org/10 . 7554/eLife . 19264 . 01610 . 7554/eLife . 19264 . 017Figure 6—figure supplement 2 . Increased ROS and superoxide in FUN025 and Tg-Tmem135 MFs . Representative flow cytometry data of ROS and superoxide levels in MFs derived from WT ( A , D , G ) , FUN025 ( B , C , H ) , and Tg-Tmem135 ( C , F , I ) mice . Panels A–C show unstained controls for each condition . WT MFs treated with pyocyanin were used as a positive control for ROS and superoxide detection ( D–F ) . Both FUN025 ( H ) and Tg-Tmem135 ( I ) MFs showed increased levels of ROS and superoxide when compared with WT MFs ( G ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19264 . 017 Dysfunctional mitochondria can generate more ROS , if not being eliminated instantly , leading to accelerated oxidative damage . We measured both total ROS and superoxide , which is the principal ROS formed by the mitochondrial electron transport chain ( ETC ) , in the MFs . Both Tmem135FUN025/FUN025 and Tg-Tmem135 MFs showed higher signals for total ROS and superoxide compared to WT MFs ( Figure 6H , I; Figure 6—figure supplement 2 ) . We then compared the protein level of major antioxidant enzymes , including superoxide dismutases ( SODs ) , glutathione peroxidases 1 ( GPx1 ) and catalase ( CAT ) , in WT and Tmem135FUN025/FUN025 MFs under normal and stress conditions ( by adding hydrogen peroxide at 200 μM for 2 hr ) . The stress condition was created by exogenous exposure to hydrogen peroxide , which has been reported to activate the cells to produce superoxide via NAD ( P ) H oxidase ( van Klaveren et al . , 1997 ) . The western blot results indicate that under the normal condition , the protein levels of SOD1 , SOD2 , SOD3 , GPx1 and CAT are all increased in Tmem135FUN025/FUN025 MFs compared to WT MFs ( Figure 6J ) reflecting an increase of ROS in Tmem135FUN025/FUN025 MFs . As a positive control , WT MFs treated with hydrogen peroxide showed an increase in SODs and GPx1 but not CAT ( Figure 6J ) . In contrast , none of the tested antioxidant enzymes except GPx1 showed an increase in hydrogen peroxide-treated Tmem135FUN025/FUN025 MFs compared to non-treated Tmem135FUN025/FUN025 MFs ( Figure 6J ) . These results suggest that the antioxidant system is upregulated in response to the increased ROS in Tmem135FUN025/FUN025 MFs . However , they may have reached the maximal capacity to upregulate the antioxidant system , and unable to further upregulate it upon exposure to exogenous oxidative stress . Our cellular data showed that Tmem135FUN025/FUN025 MFs had increased ROS , decreased SRC and the antioxidant system incapable of responding to additional environmental ROS . For that reason , we hypothesized that Tmem135FUN025/FUN025 mice may be more sensitive to oxidative stress in vivo due to the higher basal ROS level and reduced capacity to produce more energy under a stressed condition . Hyperoxia has been used as a condition to induce oxidative stress in different organs ( Jamieson et al . , 1986 ) . We exposed two-month-old Tmem135FUN025/FUN025 and control mice to the hyperoxic condition ( 75% O2 , 14 days ) to test the effect of oxidative stress to the mutant phenotypes . TMEM135FUN025/FUN025 retina were more susceptible to oxidative stress-induced cell death than WT mice as indicated by the ONLT index and the number of terminal dUTP nick end labeling ( TUNEL ) positive cells ( Figure 7A , B ) . While retinal stress marker , GFAP increased in both control and Tmem135FUN025/FUN025 retina under the hyperoxic condition ( Figure 7C ) , Tmem135FUN025/FUN025 mice under the hyperoxic condition had the most severe phenotype ( Figure 7C ) . In both WT and Tmem135FUN025/FUN025 retina , hyperoxia resulted in upregulation of 4-hydroxy-2-nonenal ( 4-HNE ) ( Figure 7D ) , a commonly used marker of lipid peroxidation ( Cingolani et al . , 2006 ) . In addition , the TMEM135 protein level increased in the Tmem135FUN025/FUN025 retina compared to WT control in the normal air with normal inspired PO2 ( Figure 7D ) . In Tmem135FUN025/FUN025 but not WT retina , hyperoxia resulted in an upregulation of the TMEM135 protein level ( Figure 7D ) . Together , these data suggest that TMEM135 plays a role in protecting cells from increased oxidative stress . 10 . 7554/eLife . 19264 . 018Figure 7 . Tmem135FUN025/FUN025 mice are more sensitive to hyperoxic condition . ( A–B ) Tmem135FUN025/FUN025 ( FUN025 ) mice raised in 75% O2 for two weeks show significant decrease of ONLT and increase of TUNEL positive cells , indicating accelerated photoreceptor cell death by apoptosis . Data from n = 4 WT mice in control air , n = 3 WT mice in 75% O2 , n = 3 FUN025 mice in control air , and n = 3 FUN025 mice in 75% O2 . Scale bar = 20 μm . ( C ) Upregulation of GFAP ( green ) indicating retinal stress is observed in FUN025 mice raised in the normal air , as well as WT mice and FUN025 mice raised in 75% O2 for two weeks . FUN025 mice raised in 75% O2 have the highest increase of GFAP signals in the retina . Scale bar = 20 μm . ( D ) Western blotting showing that hyperoxia results in upregulation of 4-HNE in both WT and FUN025 retina , and an increase of TMEM135 in FUN025 retina but not in WT retina . Data from n = 3 WT mice in control air , n = 3 WT mice in 75% O2 , n = 3 FUN025 mice in control air , and n = 3 FUN025 mice in 75% O2 . p<0 . 05 , *p<0 . 01 , two-way ANOVA . All data are mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 19264 . 018 In this study , we attempt to address a central question regarding age-dependent diseases: 'Why do age-dependent diseases manifest themselves in an age-dependent manner ? ' While it can be postulated that disease-causing mechanisms interact with molecular/cellular changes that occur with aging , molecules and pathways linking the aging process and age-dependent diseases have not been identified . By using a forward genetics approach , we identified a novel mouse mutation that leads to accelerated aging as well as pathologies observed in age-dependent retinal disease ( Figure 1; Figure 2 ) . Since this 'phenotype-driven' approach is unbiased and requires no prior knowledge of the gene functions , it allows the discovery of unsuspected mechanisms underlying biological phenomena and diseases . The mutation was discovered in transmembrane protein 135 ( TMEM135 ) , which has been associated with fat storage and longevity in C . elegans but has no defined molecular functions . Our results suggest that TMEM135 is critical for the regulation of the onset and progression of the aging process , a defect in which leads to age-dependent diseases . Thus , TMEM135 presents a novel molecular link between the aging process and age-dependent diseases . As stated above , the unique characteristics of Tmem135FUN025/FUN025 mice is that it displays phenotypic features associated with normal retinal aging at much younger ages as well as age-dependent disease pathologies . The normal aging phenotypes in the mouse retina includes photoreceptor cell degeneration , ectopic synapse formation , and increased retinal stress ( Higuchi et al . , 2015 ) . While these age-dependent abnormalities are observed at 8 months on an aging-susceptible A/J background and later in a less susceptible B6 background ( Higuchi et al . , 2015 ) , Tmem135FUN025/FUN025 mice on the B6 background have an early onset ( as early as 2 months of age ) and more rapid advancement of the retinal aging phenotypes ( Figure 1 ) . We previously found that these retinal aging phenotypes all start from the peripheral retina and progress toward the central retina ( Higuchi et al . , 2015 ) . This spatial pattern of progression is unique to the retinal aging phenotypes in mice , and is maintained in the Tmem135FUN025/FUN025 retina albeit with earlier onset and faster progression , indicating that the retinal aging process is accelerated in Tmem135FUN025/FUN025 mice . On the other hand , these mice also display pathologies observed in an age-dependent retinal disease , AMD , including autofluorescent aggregates , increased inflammation and thickened RPE ( Figure 2 ) . The fact that a single gene mutation leads to both accelerated aging and age-dependent disease pathologies in this mouse model provides a confirmation that these processes are closely associated with each other at the molecular level , and supports the idea that accelerating the aging process may lead to age-dependent diseases . Consistent with this idea , pathologies similar to those in the aging human retina including drusen formation , ectopic synapse formation and photoreceptor cell degeneration are observed in AMD patients with earlier onset and/or increased severity ( Eliasieh et al . , 2007; Hageman et al . , 2001; Liets et al . , 2006; Sullivan et al . , 2007; Terzibasi et al . , 2009 ) . Based on our finding , TMEM135 may be a key molecule that could tip the normal aging process toward age-dependent diseases . Our observations associated with the RPE in Tmem135FUN025/FUN025 mice indicate that this cell layer may be the primary site affected by the Tmem135FUN025mutation . We observed subretinal autofluorescent cells in the Tmem135FUN025/FUN025 retina that are positive for microglia and macrophage markers . Microglia have been reported to sense and react to different stimuli , and move toward the origin of stress or the injury site ( Mcglade-Mcculloh et al . , 1989 ) while going through multiple morphological stages of microglial activation ( Jonas et al . , 2012 ) . Upon evaluating the morphology and orientation of microglia in the Tmem135FUN025/FUN025 retina , we found that the microglia located closer to the RPE ( and thus farther from the IPL ) showed increased body size and decreased cell processes compared to those located closer to the IPL ( Figure 2C ) . Similar morphological changes of microglia were documented in the injured rat olfactory bulb , and were classified as stages of microglial activation response to stimuli ( Jonas et al . , 2012 ) . The Iba1+ cells near the apical surface of the RPE also show immunoreactivity to F4/80 , indicating that they have transformed from microglia to macrophages ( Figure 2C ) . These data suggest that the RPE layer is the origin of stress , toward which the microglia migrate . In addition , in the Tmem135 mutant retina , RPE abnormalities were observed earlier than the onset of photoreceptor cell degeneration . Since the RPE is located between the choriocapillaris/Bruch's membrane and the photoreceptors , and maintains the health of photoreceptor cells by delivering oxygen and metabolites and phagocytizing the shed photoreceptor outer segment discs ( Sparrow et al . , 2010 ) , primary RPE defects could result in secondary loss of photoreceptor . Interestingly , this is consistent with the current hypothesis for pathogenesis of AMD ( Bonilha , 2008; Roth et al . , 2004 ) . Together , our data suggest that the RPE and the subretinal space are likely affected first and become the origin of stress followed by photoreceptor cell loss in Tmem135FUN025/FUN025 mice . In the cell culture study , we revealed that the Tmem135FUN025/FUN025 mutant cells exhibit over-fused mitochondrial networks whereas Tg-Tmem135 cells ( transgenic cells overexpressing WT Tmem135 ) show over-fragmented mitochondrial networks . Since mitochondrial networks are the net product of fusion and fission events that occur within cells , there are two possible mechanisms TMEM135 can be involved in: inhibition of mitochondrial fusion or promotion of mitochondrial fission . Following observations in our study strongly indicated that TMEM135 in involved in promotion of mitochondrial fission rather than inhibition of mitochondrial fusion . First , experimentally promoting mitochondrial fusion through overexpression of mitochondrial fusion factor MFN2 in WT and Tg-Tmem135 cells resulted in similarly over-fused mitochondrial networks in WT and Tg-Tmem135 cells ( Figure 5I ) indicating that TMEM135 does not inhibit mitochondrial fusion . Second , we observed that some of the TMEM135 signals are colocalized with punctate signals of the mitochondrial fission factor , DRP1 , on mitochondria ( Figure 5—figure supplement 3 ) . Additionally , we found that DRP1 is normally translocated onto mitochondria which stay fused in Tmem135FUN025/FUN025 mutant cells , indicating that DRP1 is not properly activated to promote fission . It has been shown that DRP1 activity is regulated delicately by a number of post-translational modifications including phosphorylation , S-nitrosylation , SUMOylation , ubiquitination , and O-GlcNAcylation ( Cereghetti et al . , 2008; Chang and Blackstone , 2010; Knott et al . , 2008; Zunino et al . , 2007 ) . Hence , our results suggest an important role for TMEM135 to promote DRP1-dependent fission possibly by activating DRP1 through post-transcriptional modification . Another possible cause for over-fragmented mitochondrial networks observed in Tg-Tmem135 cells is increased oxidative stress , which has been shown to induce mitochondrial fragmentation in cultured cells ( Iqbal and Hood , 2014; Wu et al . , 2011 ) . Although we cannot completely rule out this possibility , it is unlikely that oxidative stress is the major cause of mitochondrial fragmentation in Tg-TMEM135 cells . While ROS is increased in both Tmem135FUN025/FUN025 mutant cells and Tg-TMEM135 cells ( Figure 6H , I ) , mitochondrial networks are over-fused in mutant cells and over-fragmented in Tg-TMEM135 cells ( Figure 5A–E ) , suggesting the TMEM135 function rather than oxidative stress as the primary factor affecting the mitochondrial networks in these cells . In addition , our live imaging of GFP-TMEM135 in WT MFs and HT22 cells captured the direct interaction of TMEM135 with mitochondria at fission sites , providing additional evidence that the TMEM135 function in mitochondrial fission rather than oxidative stress is the cause of altered mitochondrial networks . We report here that defects in a novel regulator of mitochondrial dynamics , TMEM135 , lead to impaired mitochondrial respiration . We found that adult mouse fibroblasts with either over-fused ( Tmem135FUN025/FUN025 mutant ) or over-fragmented ( Tg-TMEM135 ) mitochondria show impairment of respiration , indicating that the balance between mitochondrial fission and fusion is important for proper respiratory functions . In addition to lower basal respiration and ATP production , Tmem135FUN025/FUN025 mutant fibroblasts also showed decreased SRC . SRC or ‘spare respiratory capacity’ refers to the difference between basal and maximal respiration , which reflects the extra mitochondrial capacity available in a cell to produce energy under conditions of increased work or stress ( Nicholls , 2009 ) . Exhaustion of the SRC has been associated with heart diseases , neurodegenerative disorders and smooth muscle death ( Hill et al . , 2010; Nicholls , 2008; Sansbury et al . , 2011; Yadava and Nicholls , 2007 ) . It has been also hypothesized that mitochondria contributes to aging and age-dependent pathologies through a life-long continued decrease of the SRC ( Desler et al . , 2012 ) . The Tmem135FUN025/FUN025 mutation that leads to decreased SRC ( Figure 6F ) and accelerated retinal aging phenotypes provides new evidence to support this hypothesis . In addition , both Tmem135FUN025/FUN025 and Tg-Tmem135 MFs have increased total ROS and superoxide compared to WT MFs ( Figure 6H , I; Figure 6—figure supplement 2 ) , indicating increased oxidative stress and damage . Based on the decreased SRC and increased ROS in Tmem135FUN025/FUN025 MFs , we hypothesized that the Tmem135FUN025/FUN025 retina may be more vulnerable to the increase of environmental stress . Our animal study showed that Tmem135FUN025/FUN025 mice under the hyperoxic condition had more severe retinal degeneration , apoptosis in the retina and retinal stress compared to WT mice under the hyperoxic condition ( Figure 7A–C ) . These results suggest that while WT mice can tolerate certain amount of oxidative stress , Tmem135FUN025/FUN025 mutant mice with a decline in mitochondrial functions have lower tolerance to such stress . Thus , Tmem135FUN025/FUN025 mutant mice are more susceptible to damages caused by environmental stressors , accumulation of which leads to accelerated aging and age-dependent diseases with early onset . In conclusion , both our cell and animal studies suggest a role for TMEM135 in protecting cells from increased environmental stress . In summary , through genetic analysis of an ENU-induced mutant mouse strain , we have identified that TMEM135 plays a critical role in maintaining the health of the retina by keeping the balance of mitochondrial dynamics , defects in which result in accelerated retinal aging and development of age-dependent disease pathologies . In particular , our data suggest that TMEM135 may promote DRP1-dependent fission through activation of DRP1 . Our study suggests that TMEM135 or consequences of its defect such as unbalanced mitochondria dynamics and increased oxidative stress could be potential therapeutic targets for retinal aging and age-dependent diseases . All experiments were performed in accordance with the National Institute of Health Guide for the Care and Use of Laboratory Animals and were approved by the Animal Care and Use Committee at the University of Wisconsin-Madison and Northwestern University . FUN025 mice were generated by ENU mutagenesis in the Northwestern University Center for Functional Genomics as described previously ( Pinto et al . , 2004; Vitaterna et al . , 2006 ) and isolated from a screen designed to detect recessive mutants with vision phenotypes . These mice on the C57BL/6J genetic background were imported to University of Wisconsin-Madison . Affected FUN025 mice were crossed to C57BL/6J wild-type mice and the offspring were crossed to affected mice . Subsequently , the mutant line was maintained by crossing affected ( based on phenotyping by fundus photography ) mice to non-affected siblings , which were presumably heterozygous for the recessive mutation . C57BL/6J ( RRID:IMSR_JAX:000664 ) mice obtained from The Jackson Laboratory were used as control mice in the experiments . Tg-Tmem135 mice were generated at University of Wisconsin-Madison . We replaced the EGFP sequence in the pCX-EGFP vector ( kindly provided by Dr . Junichi Miyazaki ) ( Niwa et al . , 1991 ) with the full length Tmem135 cDNA and named it pCX-TMEM135 . We used pCX-TMEM135 for the transgene construct after linearization with HindIII and SalI ( New England Biolabs , Ipswich , MA ) . The construct was micro-injected into pronuclei of FVB/NJ embryos at the Transgenic Facility of the University of Wisconsin-Madison Biotechnology Center . Transgene-positive founders were crossed to C57BL/6J mice for two generations and subsequently maintained by intercrossing . The Ped6brd1 mutation in the FVB/NJ background was removed during this process . Mice with a point mutation ( T > C ) in Tmem135 that is the same as observed in FUN025 mice were generated using the CRISPR/Cas9 system ( T > C mice ) . The T > C mutation was introduced into intron 12 of Tmem135 ( Chr7:96 , 296 , 478 ) in C57BL/6J mice at Translational Genomics Facility of University of Wisconsin-Madison Biotechnology Center . ‘Optimized CRISPR Design’ ( http://crispr . mit . edu/ ) was used to choose the sequence ( 5’-GCCAAGCACACAGGGTTTGC-3’ ) that is located at 28 to 47 nucleotides downstream of the mutation site . This DNA sequence was incorporated into the px330 plasmid ( Cong et al . , 2013 ) ( kindly provided by Feng Zhang ) for generation of the single guide RNA ( sgRNA ) . In vitro transcription template was generated by PCR using the plasmid and an oligo with a T7 adapter end , and the sgRNA was transcribed using the MEGAshortscript T7 Transcription Kit ( Thermo Fisher Scientific , Waltham , MA ) and purified with the MEGAclear Kit ( Thermo Fisher Scientific , Waltham , MA ) . In addition , we used a 200 nucleotide oligo DNA containing the point mutation ( T > C ) in the middle of the sequence as a donor oligo ( 5’-AGTTTTTCCTTGTCATTTCAGGGTTTTTGGCAGGTGTGTCGATGATGTTTTATAAAAGCACAACAATTTCCATGTACCTAGCTTCCAAGCTGGTGGAGGcAAGCACAGCTCTTATGCCTGAGAAGTTGCCAAGCACACAGGGTTTGCAGGTGGTGTGGAGTTGCTTCAGTTGCAAGAATGACTGCTACCAAAGCAGCTCT -3’ ) . The Cas9 protein ( 40 ng/μl ) , sgRNA ( 50 ng/μl ) and donor oligo ( 50 ng/μl ) were injected into C57BL/6J embryos at Transgenic Facility of University of Wisconsin-Madison Biotechnology Center . Following microinjection , the embryos were transferred into the oviducts of pseudo-pregnant recipients . We obtained 4 mutants with the point mutation . The thickness of retinal layers was measured in H&E stained sections by using the Measure function of ImageJ software ( available at http://rsb . info . nih . gov/ij; developed by Wayne Rasband , National Institutes of Health ) . ONLT index was calculated as the ONL thickness normalized to the INL thickness . Following asphyxiation of mice by CO2 administration , eyes were immediately removed and immersion fixed in Bouin’s fixative overnight at 4°C . Eyes were then rinsed , dehydrated , and embedded in paraffin . Paraffin blocks were sectioned 6 μm thick on an RM 2135 microtome ( Leica Microsystems , Wetzlar , Germany ) and mounted on glass slides . The slides were then stained with hematoxylin and eosin ( H&E ) to visualize the retinal structure . H&E-stained sections were imaged on an Eclipse E600 microscope ( Nikon , Tokyo , Japan , using a SPOT camera ( Spot Diagnostics , Sterling Heights , MI ) . For cryostat sections , eyes were fixed in 4% paraformaldehyde ( PFA ) for 2 hr at 4°C , then cryoprotected at 4°C in a graded series of sucrose . Eyes were embedded in optimal cutting temperature compound ( OCT ) ( Sakura Finetek USA , Torrance , CA ) and sectioned at 12 μm thickness . For immunohistochemistry on cryostat sections , sections were blocked in PBS with 0 . 5% Triton X-100 and 2% normal donkey serum for 1 hr at room temperature . Next , sections were incubated overnight with the primary antibody against TMEM135 ( Sigma-Aldrich Cat# SAB2102454 , RRID:AB_10611002 , St . Louis , MO . This antibody has been discontinued recently from Sigma but the same antibody is available from Aviva Systems Biology Cat# ARP49773_P050 , RRID:AB_2048451 , San Diego , CA ) , PKCα ( Sigma-Aldrich Cat# P4334 , RRID:AB_477345 , St Louis , MO ) , PSD95 ( UC Davis/NIH NeuroMab Facility Cat# 75–028 , RRID:AB_2292909 , Davis , CA ) , Iba1 ( Wako Cat# 019–19741 , RRID:AB_839504 , Richmood , VA ) , F4/80 ( Abcam Cat# ab6640 , RRID:AB_1140040 , Cambridge , MA ) , GFAP ( Lab Vision Cat# RB-087-A0 , RRID:AB_60417 ) , CRALBP ( Abcam Cat# ab15051 , RRID:AB_2269474 , Cambridge , MA ) , TOMM20 ( Sigma-Aldrich Cat# WH0009804M1 , RRID:AB_1843992 , St . Louis , MO ) , NLRP3 ( Abcam Cat# ab4207 , RRID:AB_955792 Cambridge , MA ) , and caspase1 ( Santa Cruz Biotechnology Cat# sc-514 , RRID:AB_2068895 , Dallas , TX ) . Sections were rinsed in PBS , and incubated with a 1:200 diluted Alexa 488 conjugated secondary antibody ( Thermo Scientific , Rockford , IL ) and/or Cy3 conjugated secondary antibody ( Jackson ImmunoResearch Laboratories , West Grove , PA ) for 45 min at room temperature . All sections were imaged on the Nikon A1R+ confocal microscope ( Nikon Instruments , Melville , NY ) equipped with high sensitivity GaAsP detectors; high-speed resonant scanner; six lasers at 405 , 440 , 488 , 514 , 561 , and 640 nm . NIS-Elements AR software ( Nikon Instruments , Melville , NY ) was used for image acquisition and image analysis . Frequencies of ectopically localized bipolar cell dendrites extending into the ONL were quantified in sections immunostained with the PKCα antibody , using the Measure and Label function of ImageJ software . We counted the number of PKCα fiber that extended beyond the OPL and the length was measured along the outer plexiform layer ( OPL ) , using the Measure function of ImageJ software . Frequency was calculated as the number of ectopic bipolar cell dendrites per millimeter of retina length . Fundus photography was performed as previously described ( Pinto et al . , 2004 ) . The iris was dilated with 1% Mydriacil and the corneas kept moist with saline solution during photography . The mouse was positioned on a heating pad during the procedure to maintain the body temperature . Photographs were made with a Kowa small animal fundus camera ( 2 . 5 megapixels ) equipped with a 66 diopter supplemental lens ( Kowa American Corporation , Torrance , CA ) . ERG was performed on seven-month-old mice housed in standard diurnal cycling . Mice were dark adapted overnight and ERG was performed as previously detailed ( Pattnaik et al . , 2015 ) . To determine cone response to light , we used a background light exposure of 30 cd . s/m2 during 10 min and stimulus flash intensities ranged from 0 . 01 to 25 cd . s/m2 . To prevent cataracts , we used tear supplements during handling of mice . Animals were maintained at 37°C during the entire procedure . All ERG data was stored and exported in digital format for post-hoc analysis using Microsoft Excel . To map the FUN025 gene , we performed a whole genome scan using F2 animals from mating ( C57BL6-FUN025 x C57BR/cdJ ) F1 mice . We initially used 30 microsatellite markers , which distinguish C57BR/cdJ alleles from C57BL/6J alleles across the whole genome . All F2 animals were phenotyped by the presence or absence of drusen-like spots in the image produced by fundus photography , as previously described ( Pinto et al . , 2007 ) . Once the chromosomal locus on chromosome 7 was identified for the FUN025 mutation , we narrowed the genetic region by genotyping and phenotyping F2 recombinant mice from mating ( C57BL7-FUN025 x 129S1/SvImJ ) F1 mice . We used microsatellite markers and SNPs to differentiate C57BL/6J and 129S1/SvImJ alleles . All marker positions reported are based on the NCBI mouse genome build 37 . 1 reference assembly . All genotyping was carried out by polymerase chain reaction ( PCR ) . For FUN025 genotyping , PCR primers , mTmem135 F1 ( GGTTTTTGGCAGGTGTGTC ) and mTmem135 R1 ( TGTGTGCTTGGCAACTTCTC ) , were used for amplification of the wild-type ( WT ) allele and FUN025 allele ( 118 bp ) . Cac8I ( NewEngland Biolabs ) was used to digest the FUN025 allele specifically to generate two bands ( 80 bp and 38 bp ) . Using a SureSelect custom DNA bait library representing the FUN025 region ( Chr 7: 86 , 816 , 656 – 98 , 732 , 541 ) created by Agilent Technologies , sequence capture was performed on genomic DNA isolated from the spleen of FUN025 mice ( Qiagen DNEasy Blood and Tissue Kit , QIAGEN , Valencia , CA ) followed by paired end sequencing on the Illumina HiSeq platform at DNA Sequencing Facility of University of Wisconsin-Madison Biotechnology Center . Alignment of the sequence reads to the C57BL/6J reference genome ( NCBI build 37 ) was performed by the Bioinformatics Research Center at University of Wisconsin-Madison Biotechnology Center using Bowtie Software ( http://bowtie-bio . sourceforge . net/index . shtml ) . The sequence reads cover 93% of bases within the candidate region , with an average coverage of greater than 100x . Highly repetitive regions were excluded from the capture array and are not represented here . Standard bioinformatic analysis of our sequencing data revealed 20 single nucleotide polymorphisms ( SNPs ) within five genes in the candidate region , all of which were intronic . All the primary cells and cell lines used in this study were subjected to morphology check by microscope and used at low passages in our laboratory . Authentication and quality control tests on distributed lots of cell lines were performed by the distributors . No cell lines used in this study are cross-contaminated or misidentified by International Cell Line Authentication Committee ( ICLAC ) . Primary fibroblasts were harvested from three-month-old mouse ears . Briefly , two pieces of ear punched tissues were collected into a 1 . 5 ml microcentrifuge tube containing 70% ethanol , and the tissue pieces were rinsed with PBS containing penicillin and streptomycin . The tissues were diced into small pieces using a razor blade in a 6 cm Petri dish , and gathered into a microcentrifuge tube with 0 . 5 ml Trypsin-EDTA ( 0 . 25% Trypsin , 0 . 1% EDTA ) ( Thermo Fisher Scientific , Waltham , MA ) and 0 . 5 ml Dispase ( 5 U/ml ) ( STEMCELL Technologies , Vancouver , Canada ) . The tissues were incubated at 37° C for 30 min , followed by centrifugation for 5 min at 3000 rpm . The supernatant was discarded and the tissues were washed with 2 ml HBSS . After centrifugation for 5 min at 3000 rpm , the supernatant was discarded . Then , 0 . 5 ml trypsin-EDTA was added to the precipitated cells . After mixing thoroughly , cells were incubated at 37° C for 20 min . Following incubation , the solution was centrifuged and the supernatant was decanted . The pellet was resuspended in 0 . 5 ml fibroblast culture media: Dulbecco’s Modified Eagle’s Medium ( DMEM , ATCC , Manassas , VA ) with 10% Fetal Bovine Serum ( FBS , ATCC , Manassas , VA ) , 1% Penicillin Streptomycin ( Thermo Fisher Scientific , Waltham , MA ) . Cell aggregates were triturated and cell suspension was plated into a 3 cm tissue culture dish with 2 ml fibroblasts culture medium . Cells were incubated at 37° C with 5% CO2 . Fibroblasts were sub-cultured every 2–4 days at 1:4 to 1:6 ratio . Cells were split similarly for expansion into T75 flasks for final harvest . Fibroblasts were tested for mycoplasma contamination and the results were negative . Primary hippocampal cell culture was performed using SPOT culture kit ( University of Illinois at Chicago ) . COS-7 cells ( ATCCCRL-1651 , passage number 2–4; RRID:CVCL_0224 ) were obtained from a biological resource center , ATCC and used at low-passage . ATCC perform authentication and quality-control tests on all distribution lots of cell lines . COS-7 cells were maintained and sub-cultured according to the suggested culture methods from ATCC ( www . atcc . org/ ) . COS-7 cells were cultured in DMEM with 10% FBS . All of the medium and supplements mentioned above unless otherwise indicated were purchased from ATCC . Hippocampal cells were cultured in Neurobasal medium ( Invitrogen , Thermo Fisher Scientific , Waltham , MA ) with B27 supplement ( Invitrogen , Thermo Fisher Scientific , Waltham , MA ) and Glutamine ( Invitrogen , Thermo Fisher Scientific , Waltham , MA ) as the product manual recommendation . Immortalized mouse hippocampal cell line ( HT22; RRID:CVCL_0321 ) was a gift from Dr . Kiren Rockenstein ( Salk Institute , San Diego , CA ) . A stable cell line expressing TMEM135-GFP ( pDEST53B6TMEM135 ) was established using HT22 cells . HT22-TMEM135-GFP stable cell line was maintained and sub-cultured in DMEM with 10% FBS . In the H2O2 study , fibroblasts were plated in 100 mm dishes at a concentration of ~1 × 106 cells per dish in the complete fibroblasts culture medium described above . Two days later , the cultured cells were exposed to 200 μM H2O2 ( Sigma-Aldrich , St . Louis , MO ) at 37°C for 2 hr and were immediately washed with cold PBS , lysed with cold RIPA buffer , protease inhibitor and phosphatase inhibitor ( Thermo Scientific , Rockford , IL ) , centrifuged , and stored at −80°C until use . Full length Tmem135 constructs were cloned into vector by pENTRTM Directional TOPO Cloning Kits ( invitogen , Thermo Fisher Scientific , Waltham , MA ) and were purified using a QIAprep Spin Miniprep Kit ( QIAGEN , Valencia , CA ) after culturing on the LB agar plate containing 10 mg/mL of kanamycin and in the LB liquid medium . pcDNADEST53 ( GFP-attR1-CmR-ccdB-attR2; invitrogen , Thermo Fisher Scientific , Waltham , MA ) was used as the destination vector for TMEM135 . The LR recombination reaction between the entry clone and a destination vector was carried out using LR Clonase Enzyme ( invitrogen , Thermo Fisher Scientific , Waltham , MA ) according to the protocols recommended in the product manual . The expression constructs were then purified using a QIAfilter Plasmid Midi Kit ( QIAGEN , Valencia , CA ) . The TMEM135 expression constructs were transfected into mouse fibroblasts using SuperFect Transfection Reagent ( QIAGEN , Valencia , CA ) following the manufacture’s protocol and cultured for 48 hr . Mitochondria were stained with MitoTracker Red CMXRos ( Thermo Fisher Scientific , Waltham , MA ) according to the manufacturer’s protocol . Briefly , cells were incubated with 200 nM MitoTracker Red CMXRos in cultured medium for 20 min at 37°C . After a rinse with pre-warmed PBS , Cells were ready to be imaged live , or to be fixed with 4% PFA . Cells were cultured on coverslips and were fixed with 4% PFA for 10 min at 4°C . Cells were permeabilized using 0 . 5% Triton-X in PBS for 30 min followed by blocking in 2% normal donkey serum for 30 min . Then the cells were incubated with primary antibody against TMEM135 ( Sigma-Aldrich Cat# SAB2102454 , RRID:AB_10611002 , St . Louis , MO . This antibody has been discontinued recently from Sigma but the same antibody is available from Aviva Systems Biology Cat# ARP49773_P050 , RRID:AB_2048451 , San Diego , CA ) , GFP ( Synaptic Systems Cat# 132 002 , RRID:AB_887725 , Germany ) and TOMM20 ( Sigma-Aldrich Cat# WH0009804M1 , RRID:AB_1843992 , St . Louis , MO ) in PBS with 0 . 5% Triton X-100 and 2% normal donkey serum overnight at 4°C . Cells were washed in PBS and then incubated with Alexa 488 conjugated secondary antibody ( Thermo Scientific , Rockford , IL ) and Cy3 conjugated secondary antibody ( Jackson ImmunoResearch Laboratories , West Grove , PA ) for 1 hr at room temperature . All immunocytochemistry slides were imaged on a Nikon A1R+ confocal microscope ( Nikon Instruments , Melville , NY ) . Cells , mouse brains and retina were homogenized in RIPA buffer ( 1x PBS with 1% NP-40 and 0 . 1% SDS , Thermo Scientific , Rockford , IL ) ) containing a protease and phosphatase inhibitor cocktail ( Thermo Scientific , Rockford , IL ) . Protein concentrations were determined using the Pierce BCA Protein Assay ( Thermo Scientific , Rockford , IL ) according to the manufacturer's instructions . High- purity mitochondrial fraction was isolated from the whole brain lysate using Qproteome Mitochondria Isolation Kit ( QIAGEN , Valencia , CA ) according to the manufacturer’s protocol . The cell lysate , retinal lysate or the whole brain lysate and isolated mitochondrial fraction containing equal amounts of protein were subjected to SDS–PAGE using 10% Bis-Tris gels and antibodies against HNE ( Millipore Cat# 393206-100UL , RRID:AB_211975 , Billerica , MA , ) , β-actin ( Cell Signaling Technology Cat# 4970 , RRID:AB_2223172 , Danvers , MA ) , TOMM20 ( Sigma-Aldrich Cat# WH0009804M1 , RRID:AB_1843992 , St . Louis , MO ) , TMEM135 ( Sigma-Aldrich Cat# SAB2102454 , RRID:AB_10611002 , St . Louis , MO . This antibody has been discontinued recently from Sigma but the same antibody is available from Aviva Systems Biology Cat# ARP49773_P050 , RRID:AB_2048451 , San Diego , CA ) , SOD1 ( Abcam Cat# ab13498 , RRID:AB_300402 , Cambridge , MA ) , SOD2 ( Abcam Cat# ab13533 , RRID:AB_300434 , Cambridge , MA ) , SOD3 ( Santa Cruz Biotechnology Cat# sc-32222 , RRID:AB_2191977 , Dallas , TX ) , GPx1 ( Abcam Cat# ab22604 , RRID:AB_2112120 , Cambridge , MA ) , CAT ( Novus Cat# NB100-79910 , RRID:AB_2071872 , Littleton , CO ) , MFN1 ( Abnova Corporation Cat# H00055669-A01 , RRID:AB_529865 , Taipei City , Taiwan ) , MFN2 ( Abnova Corporation Cat# H00009927-A01 , RRID:AB_1204675 , Taipei City , Taiwan ) , LAMP2 ( NBP2-22217 , Novus Biologicals , Littleton , CO ) , Lamin B1 ( Abcam Cat# ab16048 , RRID:AB_10107828 , Cambridge , MA ) , PDI ( Cell Signaling Technology Cat# 3501 , RRID:AB_2156433 , Danvers , MA ) and GAPDH ( EnCor Biotechnology Cat# MCA-1D4 , RRID:AB_2107599 , Gainesville , FL ) . Horseradish peroxidase conjugated secondary antibodies were used ( Jackson Immunoresearch , West Grove , PA ) prior to detection with a chemiluminescent reagent ( Amersham ECL Plus Western blotting detection system , General Electric , Buckinghamshire , UK ) and exposure to X-ray film ( Thermo Scientific , Rockford , IL ) . IRDye 800CW or IRDye 680RD secondary antibodies were used ( LI-COR Biotechnology , Lincoln , NE ) prior to detection with Odyssey CLx imaging system ( LI-COR Biotechnology , Lincoln , NE ) . Deeply anesthetized mice [12 –month old C57BL/6J ( n = 4 ) and FUN025 ( n = 4 ) ] were used for imaging RPE and inner segment of photoreceptors by electron microscopy . The sample preparation and imaging were performed as previously described ( Johnson et al . , 2006 ) . WT fibroblasts were seeded in coated glass coverslips in a 24-well plates overnight . The cells were than transfected with a GFP-TMEM135 vector ( pDEST53B6TMEM135 ) for 48 hr . The cells were fixed with a 4% PFA , 0 . 1% Glutaraldehyde fixative solution in 0 . 1 M Phosphate Buffer ( PB ) for 1 hr at room temperature . Next , the cells were washed with 0 . 1 M PB ( 3 × 5 min ) , incubated with a solution of 0 . 1% NaBH4 in 0 . 1 M PB for 10 min , rinsed in 0 . 1 M PB ( 4 × 5 min ) , and permeabilized with a solution of 0 . 5% Triton-X100 in PBS for 30 min followed by PBS washes ( 3 × 10 min ) . The specimens were then transferred to AURION Blocking Solution ( Product code 905 . 002: contains Normal Goat serum , AURION Immuno Gold Reagents & Accessories , Wageningen , The Netherlands ) for 30 min , and washed with incubation buffer ( 0 . 1% AURION BSA-c , 10 mM NaN3 in PBS at pH 7 . 4 ) for 3 × 10 min . Next , the specimens were incubated with anti-GFP antibody ( Abcam Cat# ab290 , RRID:AB_303395 , Cambridge , MA ) at 5 μg/ml in incubation buffer at 4°C for overnight . After 6 times washes ( 10 min each ) , the specimens were incubated with the secondary antibody Ultra Small gold conjugate reagent [F ( ab’ ) 2 Fragment of Goat-anti-Rabbit IgG ( H&L ) , Cat . # 25361 , EMS , Hatfield , PA] at 1/100 in incubation buffer overnight . The specimens were then washed with incubation buffer ( 6 × 10 min ) and with PBS ( 6 × 10 min ) followed by post-fixation in 2% glutaraldehyde in 0 . 1 M PB for 30 min and washes on 0 . 1 M PB for 2 × 5 min . Next , the specimens were washed with 1X AURION Enhancement Conditioning Solution ( ECS ) ( dilute from 10x concentrated solution , product code 500 . 055 , EMS , Hatfield , PA ) for 6 × 5 min and incubated with enhancement mixture ( 1:20 of DEVELOPER and ENHANCER , EMS , Hatfield , PA ) for 2 . 5 hr in RT in dark . The silver enhancement reaction was stopped by washing the specimens with 1X AURION ECS for 2 × 5 min . The cells were post-fixed with 4% OsO4 for 30 min followed with three times quick washes with 0 . 1 M PB . The coverglasses were each transferred into a plastic jar and the cells were dehydrated in a graded series of ethanol . Next , the cells were embedded with 1:1 mix of resin and 100% ethanol , overnight in RT . The samples were then incubated in a 60°C oven for 20 min , and moved to a new plastic jar with fresh 100% plastic resin and incubated for 30 min at 60°C . After that , the samples were moved to another new plastic jar with fresh 100% plastic resin and incubated for another 30 min at 60°C . Next , the samples were polymerized with 100% plastic resin at 60°C for two days . All sectioning was performed on a Reichert Ultracut E ( Reichert/Leica Microsystems; Wetzlar , Germany ) . Ultra-thin sections were cut with a Microstar type SU diamond knife ( Microstar Technologies , Huntsville , TX ) . Ultra-thin sections were stained en drop with Uranyl Acetate and Lead Citrate at 60°C . Sections were imaged on a Philips CM120 Scanning Transmission Eletron Microscope ( Philips Electron Optics , Eindoven , Netherlands ) using Analysis Software ( Soft Imaging System Corp . , Lakewood , CA ) . Mitochondria were isolated from the WT mouse brain following the manufacturer’s protocol for Mitochondrial isolation kit ( MITOISO1 , Sigma , Saint Louis , MO ) . Mitochondria were prepared from four fresh brain tissues following homogenization by an overhead electric motor ( ~200 rpm ) . The BSA was added to the extraction buffer to remove lipids . The homogenate was centrifuged at 1000 X g for 5 min . The supernatant liquid was moved to a new tube and centrifuged at 3500 x g for 10 min . Next , the pellet was resuspended in the extraction buffer . The low and high speed centrifugation steps were repeated and the pellet was suspended in the storage buffer ( ~40 μl per 100 mg tissue ) . For mitochondrial morphometric analysis using ImageJ ( RRID:SCR_003070 ) software , Z-stacks were collected from cells labeled with MitoTracker , and summed projections were generated . Images were thresholded to select mitochondria . From the thresholded fluorescence , binary images were generated , and the size of mitochondria , the number of mitochondria and the total mitochondrial area were measured . Mitochondria number and coverage were generated by dividing the number of mitochondria and total mitochondrial area by the size of cells . Mitochondrial DRP1 puncta were analyzed using NIS-Elements ( RRID:SCR_014329 , Nikon Instruments , Melville , NY ) . A binary mask of the mitochondrial channel ( MitoTracker ) was generated and used it to substrate all extra-mitochondrial DRP1 fluorescence . To select mitochondrial DRP1 puncta for analysis , mitochondrial DRP1 fluorescence was thresholded , and the thresholded image was converted to a binary image . To measure mitochondrial DRP1 puncta intensity , binary area of mitochondrial DRP1 was divided by binary area of mitochondria . To measure DRP1 puncta fluorescence , puncta identified by thresholding were analyzed for fluorescence intensities . The live imaging of TMEM135 and mitochondria were done in both WT MFs transfected with eGFP-TMEM135 and a stable cell line we generated in HT22 cells expressing GFP-TMEM135 . The cells were plated at a density sufficient to reach confluence in two days on glass bottom culture dishes ( MatTek Corporation , Ashland , MA ) in DMEM with 10% FBS . 30 min before imaging , the cells were live-stained with MitoTracker Red CMXRos ( Thermo Fisher Scientific , Waltham , MA ) , and the culture medium was replaced with DMEM without phenol red ( Thermo Fisher Scientific , Waltham , MA ) with 10% FBS . Live imaging of TMEM135 and mitochondria was performed on the Revolution XD spinning-disk microscopy system ( Andor Technology , Belfast , UK ) equipped with the Yokogawa CSU-X1 confocal spinning-disk head; Nikon Eclipse Ti inverted microscope surrounded by an Okolab cage incubator . Andor IQ2 ( RRID:SCR_014461 ) software was used for image acquisition and Imaris ×64 ( RRID:SCR_007370 , Bitplane AG , Zurich , Switzerland ) for image analysis . For live imaging , rapid Z-stacks were acquired using the 100×/1 . 49 NA Apo TIRF objective ( Nikon Instruments , Melville , NY ) at 37°C . After background subtraction and smoothing , videos were recorded for WT MFs . For images captured in HT22 cells , Spots algorithm was used to identify TMEM135 spots and follow them through time and cell space , and Surface algorithm was used to identify mitochondria . TMEM135 spots were identified as two groups ( mitochondria-bound TMEM135 spots and TMEM135 spots that are not bound to mitochondria ) according to their distance to the mitochondria using Spots Close To Surface XTension . One set of TMEM135 Spots ( mitochondria-bound TMEM135 ) , located inside the threshold defined region are show in magenta , and another group of TMEM135 spots ( TMEM135 not bound to mitochondria ) , whose shortest distance to the mitochondria surface exceeds the specified threshold are shown in turquoise . ΔΨm was measured using the Mito ID membrane potential detection kit ( Enzo Life Sciences , Lorrach , Germany ) . Cells were seeded overnight and the positive control cells were pretreated with CCCP at a final concentration of 2 uM for 30 min in 37°C , and then harvested by trypsinization , washed , and incubated at room temperature for 15 min with reagent followed by flow cytometry FACSAria II ( BD Biosciences , San Jose , CA ) . Loss of MMP was observed as a decrease in orange and increase in green fluorescence . Results are analysized by FlowJo 7 . 1 ( RRID:SCR_008520 , Treestar , Ashland , OR ) and presented as a change in ratio of two fluorescence means that correlates to changes in ΔΨm . Mitochondrial respiration was determined using the Seahorse XFe24 Extracellular Flux Analyzer and the XF Cell Mito Stress Test Kit according to the manufacturer’s instruction ( Agilent Technologies , Santa Clara , CA ) . Briefly , WT , FUN025 and Tg-Tmem135 fibroblasts were seeded at 5 × 104 cells per well in a XF24 cell culture microplate . Cells were washed with XF Assay Media , pre-incubated in a non-CO2 incubator at 37°C for 1 hr in XF Assay Media . For the assay , cells were treated sequentially with 1 μM Oligomycin , 5 μM FCCP ( carbonyl cyanide-p-trifluoromethoxyphenylhydrazone ) , and 1 μM Rotenone plus 1 μM Antimycin A . The Seahorse Wave ( RRID:SCR_014526 ) software is used to design , run and collect the results for all XF assays ( all reagents are purchased from Agilent Technologies ) . ROS levels were analyzed using Total ROS/Superoxide Detection kit from Enzo Life Sciences , Inc . ( Plymouth Meeting , PA ) . The non-fluorescent , cell-permeable total ROS detection dye was added to cells followed by incubation for 45 min . The dye reacted directly with a wide range of reactive species , such as hydrogen peroxide , peroxynitrite and hydroxyl radicals , yielding a green fluorescent product indicative of cellular production of different ROS/RNS types . The cells were washed twice with PBS in a volume sufficient to cover the cell monolayer and analyzed using a flow cytometry equipped with standard green filter ( 490/525 nm or 488 nm laser ) . Appropriate positive control samples induced with Pyocyanin exhibit bright green fluorescence in the cytoplasm . Cells pre-treated with the ROS inhibitor do not demonstrate any green fluorescence signal upon induction . 1 μM of ROS and superoxide-sensitive fluorescent dyes , and subsequently assayed by flow cytometry FACSAria II ( BD Biosciences , San Jose , CA ) . Double staining with fluorescein isothiocyanate and phycoerythrin was carried out , and data were assessed by FlowJo software version 7 . 6 ( RRID:SCR_008520 , Tomy Digital Biology Co . , Tokyo , Japan ) . Hyperoxic-exposure of mice was conducted as previously described ( Bozyk et al . , 2012 ) . Briefly , mice were exposed to 75% oxygen for 14 days using a chamber coupled to an oxygen controller and sensor ( BioSpherix , Lacona , NY ) , while the control mice were maintained at normoxia ( room air ) conditions for the duration of the experiment . Terminal deoxynucleotidyl transferase dUTP nicked-end labeling ( TUNEL ) staining was performed with an Apoptag kit using fluorescein detection ( Millipore , Billerica , MA ) , according to the manufacturer's instructions . Nuclei were counterstained with DAPI , specimens mounted in ProLong Gold antifade reagent ( Thermo Fisher Scientific , Waltham , MA ) . The number of apoptotic cells per retina was counted in three mice retina from three consecutive serial sections and averaged to obtain a mean single TUNEL positive cells value/retina section/mouse . Sample size was chosen empirically following previous experience in the assessment of experimental variability . No statistical methods were used to predetermine sample size . No animals were excluded . Samples were not randomized and the investigators were not blinded . Statistical analyses were performed in GraphPad Prism 6 ( RRID:SCR_002798 , GraphPad Software , La Jolla , CA ) . Significance of the difference between groups was calculated by unpaired Student’s two-tailed t test , for experiments comparing two groups , and one-way or two-way analyses of variance ( ANOVA ) with the Bonferroni-Dunn multiple comparison posttest for experiments comparing three or more groups using p<0 . 05 , p<0 . 01 , p<0 . 001 . ***p<0 . 0001 . All data are presented as the mean ± the standard error of the mean ( s . e . m . ) of three or more independent experiments , with three or more replicates per condition per experiment . P < 0 . 05 was considered to be statistically significant .	Older people have an increased risk of developing many diseases , such as diabetes and age-related macular degeneration ( which is often shortened to AMD ) . This suggests that changes that occur during normal aging may some how be linked to how such diseases develop . However , the molecular mechanisms responsible for these links are not clear . AMD causes damage to the retina of the eye , which can lead to visual loss in older people . To investigate the link between aging and age-dependent diseases , Lee et al . used mutant mice whose retina of the eye ages more quickly than normal mice and are prone to developing an eye condition that is similar to AMD . The experiments show that these mice have a mutation in a gene called Tmem135 that is responsible for these visual problems . Tmem135 regulates the size of cell compartments called mitochondria , which produce energy for the cell . This affects the ability of the mitochondria to work properly and makes the cells more sensitive to environmental stress , which in turn makes the retina age more quickly . The findings of Lee et al . show that Tmem135 is a critical link between aging and an AMD-like condition in mice . Furthermore , the experiments suggest that defects in mitochondria may accelerate the normal pace of aging and lead to AMD and other age-dependent diseases . Further studies are needed to find out exactly what role Tmem135 plays in mitochondria and whether it also contributes to the aging of other parts of the body .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "neuroscience" ]	2016	Mouse Tmem135 mutation reveals a mechanism involving mitochondrial dynamics that leads to age-dependent retinal pathologies
Abscisic acid ( ABA ) is a plant hormone that regulates plant growth and development and mediates abiotic stress responses . Direct cellular monitoring of dynamic ABA concentration changes in response to environmental cues is essential for understanding ABA action . We have developed ABAleons: ABA-specific optogenetic reporters that instantaneously convert the phytohormone-triggered interaction of ABA receptors with PP2C-type phosphatases to send a fluorescence resonance energy transfer ( FRET ) signal in response to ABA . We report the design , engineering and use of ABAleons with ABA affinities in the range of 100–600 nM to map ABA concentration changes in plant tissues with spatial and temporal resolution . High ABAleon expression can partially repress Arabidopsis ABA responses . ABAleons report ABA concentration differences in distinct cell types , ABA concentration increases in response to low humidity and NaCl in guard cells and to NaCl and osmotic stress in roots and ABA transport from the hypocotyl to the shoot and root . Plant hormones control plant growth and development . Knowledge of the locations of hormone synthesis and transport , and the resulting hormone gradients and distributions in plants is important for understanding how plants respond to their environment via hormone signaling and cross talk of hormone and other small molecule signaling pathways ( Hetherington and Woodward , 2003; Israelsson et al . , 2006; Nemhauser et al . , 2006; Muday et al . , 2012 ) . Among plant hormones , auxin is best characterized in terms of its distribution and transport ( Vanneste and Friml , 2009 ) , analyzed using reporter constructs for auxin-induced gene expression or protein degradation ( Ulmasov et al . , 1997; Brunoud et al . , 2012; Wend et al . , 2013 ) . Reporter genes developed for ABA-induced gene expression ( pRD29A/B , pRAB18 and pAtHB6 ) are also used ( Lång and Palva , 1992; Yamaguchi-Shinozaki and Shinozaki , 1993; Ishitani et al . , 1997; Christmann et al . , 2005; Kim et al . , 2011; Duan et al . , 2013 ) . Despite their potential , such promoter-linked reporters respond indirectly and slowly to their respective plant hormone . To unequivocally investigate dynamic models of hormone distribution and dissect the complex functions and interconnection of these signaling molecules , direct plant hormone reporters that act instantaneously and reversibly are essential . Optogenetic reporters provide a potential solution . These genetically engineered chromogenic proteins ( often fluorescent proteins ) respond to a specific environmental change via conformationally linked changes in their spectral properties measurable with optical instruments ( Giepmans et al . , 2006; Alford et al . , 2013 ) . Such reporters have been developed for a whole palette of molecules and physiochemical processes ( Okumoto et al . , 2012 ) . However , no reporters for direct visualization of any plant hormone have yet been developed . During land colonization , plants adopted ABA as a hormone to signal stress due to limited water supply ( Cutler et al . , 2010; Raghavendra et al . , 2010; Hauser et al . , 2011 ) . ABA is integrated into a complex signaling network that transcriptionally and post-translationally regulates seed germination , root development and stomatal aperture ( Hetherington and Woodward , 2003; Cutler et al . , 2010; Kim et al . , 2010; Raghavendra et al . , 2010; Tanaka et al . , 2013 ) . ABA biosynthesis is a multi-step reaction involving Zeaxanthin epoxidation , isomerization and cleavage to Xanthoxin in the plastid , followed by conversion to abscisic aldehyde and oxidization to ABA in the cytoplasm ( Nambara and Marion-Poll , 2005 ) . In Arabidopsis , ABA is synthesized primarily in vascular tissues of roots and leaves , in guard cells and in seeds ( Sauter et al . , 2001; Endo et al . , 2008; Seo and Koshiba , 2011; Bauer et al . , 2013; Boursiac et al . , 2013 ) . ABA catabolism includes hydroxylation pathways and glucose conjugation leading to less- or inactive compounds ( Nambara and Marion-Poll , 2005; Kepka et al . , 2011 ) . ABA-glucose ester is stored in the vacuole and was reported to be rapidly hydrolyzed by β-glucosidases ( Lee et al . , 2006 ) . However , direct measurements of rapid ABA release are missing . ABA moves throughout the plant and crosses cell borders as a function of pH . This ‘ionic trap model’ explains the movement , but excludes cellular efflux of ABA due to low apoplastic pH ( Sauter et al . , 2001; Seo and Koshiba , 2011; Boursiac et al . , 2013 ) . Recently identified ABA transporters contribute to ABA export from the vasculature and import into guard cells ( Kang et al . , 2010; Kuromori et al . , 2010 , 2011; Kanno et al . , 2012; Boursiac et al . , 2013 ) . Two non-mutually exclusive current models describe how water limitations in the root induce stomatal closure in the leaf: ( 1 ) ABA acts as a chemical signal synthesized in the root and transported to the shoot ( Sauter et al . , 2001; Wilkinson and Davies , 2002 ) , and ( 2 ) a hydraulic signal from the root induces ABA synthesis in the shoot ( Christmann et al . , 2005 , 2007; Ikegami et al . , 2009 ) . In response to water limitations ABA concentrations increase ( Harris et al . , 1988; Harris and Outlaw , 1991; Christmann et al . , 2007; Forcat et al . , 2008; Ikegami et al . , 2009; Geng et al . , 2013 ) and decrease upon stress relief ( Harris and Outlaw , 1991; Endo et al . , 2008 ) . Despite recent progress on ABA synthesis and transport , direct evidence for conditionally triggered changes in local ABA concentrations and time-resolved data for ABA re-distribution in vivo are lacking . ABA is perceived by members of a protein family designated as PYRABACTIN RESISTANCE 1 ( PYR1 ) /PYR1-LIKE ( PYL ) /REGULATORY COMPONENT OF ABA RECEPTOR ( RCAR ) , which in the presence of ABA negatively regulate Clade A TYPE 2C PROTEIN PHOSPHATASES ( PP2Cs ) ( Ma et al . , 2009; Park et al . , 2009 ) . Inhibition of PP2Cs enables the activation of SNF1-RELATED KINASES 2 ( SnRK2s ) ( Fujii et al . , 2009; Umezawa et al . , 2009; Vlad et al . , 2009 ) , that target transcription factors , ion channels and NADPH oxidases ( Kobayashi et al . , 2005; Furihata et al . , 2006; Geiger et al . , 2009; Lee et al . , 2009; Sato et al . , 2009; Sirichandra et al . , 2009 , 2010; Brandt et al . , 2012 ) . PYR/PYL/RCARs contain an internal ligand-binding pocket flanked by two conserved loops , the ‘Pro-cap/gate’ and the ‘Leu-lock/latch’ ( Melcher et al . , 2009; Nishimura et al . , 2009; Santiago et al . , 2009a ) . ABA binding triggers these loops to rearrange , closing the lid over ABA . The conformational change and rearranged protein surface favors PP2C binding over receptor dimerization ( Melcher et al . , 2009; Nishimura et al . , 2009; Yin et al . , 2009 ) . In the resultant ABA receptor–phosphatase complex , a conserved Trp residue from the PP2C inserts between the ‘Pro-cap/gate’ and the ‘Leu-lock/latch’ to further enclose the ABA ( Melcher et al . , 2009; Miyazono et al . , 2009; Yin et al . , 2009 ) . Here , we report the design , development and application of optogenetic FRET-based reporters for ABA ( ABAleons ) , in which covalently attached PYR1 and the PP2C ABI1 are modulated upon ABA binding , triggering changes in fluorescence emission from attached fluorescent proteins . ABAleons can affect ABA responses at high concentrations and enable the analysis of time-dependent changes in ABA concentration , distribution and transport in live plants with appropriate resolution to monitor endogenous ABA concentration changes . Based on structural analyses of PYR1 ( Nishimura et al . , 2009 ) and the PYL1-ABA-ABI1 complex ( Miyazono et al . , 2009 ) and a FRET cassette consisting of the fluorescent proteins mTurquoise ( Goedhart et al . , 2010 ) and Venus circularly permutated at amino acid 173 ( cpVenus173; Nagai et al . , 2004; Piljić et al . , 2011 ) , FRET-based reporters for ABA , named ABAleons , were designed ( Figure 1A ) . Full length PYR1 and ABI1 truncated at amino acid S125 ( ΔNABI1 ) were fused via a flexible ASGGSGGTS ( GGGGS ) 4-linker ( Arai et al . , 2004; Nagai et al . , 2004 ) and inserted into the mTurquoise-cpVenus173 FRET cassette using short two amino acid GP- and PG-linkers ( Piljić et al . , 2011 ) resulting in the FRET reporter ABAleon1 . 1 ( Figure 1A ) . Due to the long flexible linker between PYR1 and ΔNABI1 ( > 120 Å; Arai et al . , 2004; Figure 1A ) , ΔNABI1 might be free for substrate access in the ABA unbound conformation . Therefore the wild type ABAleon1 . 1 was mutated to abolish phosphatase activity of ABI1 by introducing a D413L mutation in the catalytic metal-binding site ( ABAleon2 . 1; http://www . uniprot . org/uniprot/P49597 ) ( Figure 1A ) . 10 . 7554/eLife . 01739 . 003Figure 1 . In vitro characterization of ABAleons . ( A ) mTurquoise ( cyan ) is fused through a GP-linker to PYR1 ( gold ) , which is separated by a flexible ASGGSGGTS ( GGGGS ) 4 linker from ΔNABI1 ( green ) fused to cpVenus173 ( yellow ) through a PG linker . Structural features of the PYR1-ΔNABI1 complex including ABA ( blue and red balls ) , ABI1 D413 ( purple ball ) and loops controlling access to the ABA binding site are highlighted . Dashed lines indicate linkers and unresolved structures . ( B ) Without ABA , ABAleon flexibility enables FRET from mTurquoise ( mT ) to cpVenus173 ( cpV173 ) . ABA triggered PYR1-ΔNABI1 binding increases the distance or orientation between the fluorescent probes , thereby reducing FRET efficiency . ( C and D ) Normalized ( nu ) emission spectra of ( C ) ABAleon1 . 1 and ( D ) ABAleon2 . 1 in absence ( unbound ) and in presence ( bound ) of ABA with indicated dynamic range ( DR ) . ( E ) Emission ratios and ( F ) ΔR/ΔRmax plotted against increasing [ABA] , with indicated ABA affinity K′d of each ABAleon calculated from the respective plot . ( G ) Time-dependent normalized emission ratios of ABAleon2 . 1 in response to 0 and 1 µM ABA . ( H ) Phosphatase activity assays of equimolar PYR1 and ΔNABI1 combinations and indicated ABAleons in presence of 0 and 5 µM ABA ( mean ± SD , n = 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01739 . 00310 . 7554/eLife . 01739 . 004Figure 1—figure supplement 1 . ABA does not affect ABAleon absorbance and ABAleon emission is stable at physiological pH conditions . ( A ) Normalized absorbance spectrum of ABAleon2 . 1 in absence ( cyan ) and in presence of 100 µM ABA ( yellow ) . ( B ) pH titration of ABAleon1 . 1 emission ratios ( left scale ) in absence ( cyan ) and in presence of 10 µM ABA ( yellow ) . Curves were fitted by a four parameter Hill equation and ratio change ( red , right scale ) was calculated by subtraction of the 10 µM ABA from the 0 µM ABA curve . Note the dramatic ABA-independent ratio change of ABAleon1 . 1 below pH 6 . 2 . However in the physiological range ( pH 7 . 2–7 . 8 ) ABAleon1 . 1 emission is stable . DOI: http://dx . doi . org/10 . 7554/eLife . 01739 . 004 In vitro application of ABA had no impact on ABAleon2 . 1 absorbance ( Figure 1—figure supplement 1A ) . However , analyses of fluorescence emission spectra after application of ABA revealed an increase in mTurquoise emission ( peak at 476 nm ) and a decrease of cpVenus173 emission ( peak at 527 nm ) , indicating that the distance between both fluorescent proteins is increased , or their orientation to each other is changed by the ABA-dependent interaction of PYR1 and ΔNABI1 ( Figure 1B–D ) . Apparent ABA affinities were calculated by curve fitting of ABA-dependent emission ratio plots ( Figure 1E ) or from fits of the ABA-dependent ratio change ( ΔR ) relative to the maximum ratio change ( ΔRmax ) ( Figure 1F ) . For ABAleon1 . 1 a dynamic range of ∼ 15 % was recorded with an apparent ABA affinity K′d of ∼ 300 nM ( Figure 1C , E , F ) . Moreover , ABA-induced emission ratio changes of ABAleon1 . 1 were stable in the range of physiological pH conditions ( e . g . , pH 6 . 6–8 . 2; Figure 1—figure supplement 1B ) . ABAleon2 . 1 exhibited a dynamic range of ∼ 9 % and a K′d of ∼ 100 nM ( Figure 1D–F ) . In plate reader-based analyses , application of ABA rapidly and clearly induced emission ratio changes of ∼ 8 % when analyzing ABAleon2 . 1 at 1 µM ABA ( Figure 1G ) . ABAleon1 . 1 exhibited phosphatase activity comparable to PYR1 and ΔNABI1 when combined in a 1:1 molar ratio ( Figure 1H ) . In the presence of 5 µM ABA phosphatase activity was inhibited to 50 % of initial activity ( Figure 1H; Ma et al . , 2009; Park et al . , 2009; Santiago et al . , 2009b ) . However , the predicted phosphatase inactive ABI1D413L mutation in ABAleon2 . 1 enabled the design of an ABA-reporter without detectable phosphatase activity ( Figure 1H ) , which was considered to be preferable for use in plants . ABAleon2 . 1 was transformed into the Arabidopsis Columbia 0 accession ( Col-0 ) to determine whether it can detect ABA level changes in planta . In a macroscopic view ABAleon2 . 1 emission ratio maps were recorded from 5 day-old seedlings before ( Figure 2A ) or 2 h after ABA application ( Figure 2B ) . Emission ratios were also recorded from guard cells of 33-day-old soil-grown plants ( Figure 2C ) . As indicated in the color code of the calibration bar ( Figure 2 ) , low ABAleon2 . 1 ratios ( blue ) indicate high ABA concentrations and high ratios ( red ) indicate low ABA concentrations . Comparison of the emission ratio maps before and after ABA application revealed visible ABA uptake into the entire seedling ( Figure 2A , B ) . However the most prominent ratio changes were observed in the root elongation- and early maturation zone , where the yellow-coded regions showed an ABA concentration increase , which is indicated by the yellow-to-blue color shift ( Figure 2A , B ) . Visible ratio changes were also observed in the lower hypocotyl . Here an upward-directed ABA accumulation was detected ( Figure 2A , B ) . 10 . 7554/eLife . 01739 . 005Figure 2 . ABA-induced ABAleon2 . 1 responses in whole seedlings . ( A and B ) Manually assembled ABAleon2 . 1 ratio images ( A ) before and ( B ) 2 h after application of 50 µM ABA . ( C ) Ratio image of untreated guard cells from lower epidermis of 33-day-old soil grown plants . Images were calibrated to the indicated calibration bar . Blue colors indicate low ABAleon2 . 1 emission ratios , corresponding to high ABA concentrations , and red colors indicate high ABAleon2 . 1 emission ratios corresponding to low ABA concentrations . Shown is a representative of four experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 01739 . 00510 . 7554/eLife . 01739 . 006Figure 2—figure supplement 1 . pRAB18-GFP expression in guard cells . ( A–C ) Confocal images of pRAB18-GFP expression in the intact lower epidermis of 27-day-old plants show highest expression ( GFP emission ) in guard cells . ( A ) Representative image of plants grown at 70 % relative humidity ( RH ) conditions , ( B ) 4 h after leaf floating in 50 µM ABA and ( C ) 2 days after plant transfer to 25 % RH conditions . Note that pRAB18-GFP expression appears in the epidermal cells only after ABA application ( B ) . ( D ) Images were calibrated to background fluorescence ( lowest value ) and to the maximum value recorded in guard cells ( highest value ) . ( E ) Quantified pRAB18-GFP emission in guard cells from the same analyses as in ( A–D ) ( means ± SEM , n = 3 with ≥ 34 guard cells/n ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01739 . 006 ABAleon2 . 1 emission ratios in guard cells were low ( Figure 2C , blue ) , indicating elevated ABA concentrations under the imposed conditions . High ABA concentrations in guard cells were consistent with the constitutive guard cell expression of the ABA-induced reporter pRAB18-GFP when plants were grown at 70 % relative humidity ( Figure 2—figure supplement 1A ) . Expression of pRAB18-GFP was further induced by ABA ( Figure 2—figure supplement 1B ) , however with stronger induction 2 days after transfer to 25 % relative humidity ( Figure 2—figure supplement 1C , E ) . ABA-induced ABAleon2 . 1 responses were then analyzed with higher time- and spatial resolution in guard cells of 45-day-old plants ( Figure 3A–C ) and in three differentially color-coded regions of the hypocotyl ( Figure 3D–F ) , root differentiation– ( Figure 3G–I ) , root maturation– ( Figure 3J–L ) and root elongation-zone ( Figure 3M–O ) of 5-day-old seedlings . Application of ABA induced increases in mTurquoise ( Figure 3A , D , G , J , M , mT , solid lines ) and decreases in cpVenus173 emission ( Figure 3D , G , J , M , cpV , dashed lines ) , resulting in an up to 12 % decrease in the emission ratios ( Figure 3B , E , H , K , N ) . The ABA-induced ABAleon2 . 1 emission ratio changes indicate increases in the ABA concentration in all investigated Arabidopsis tissues . These in planta analyses were consistent with in vitro analyses ( Figure 1E , G ) . The ABA-induced ABAleon2 . 1 ratio changes in guard cells ( 3–6 %; Figure 3B ) and in the root differentiation zone ( 6 %; Figure 3H ) were low compared to changes in the hypocotyl and lower root tissues ( 9–12 %; Figure 3E , K , N ) , consistent with data indicating elevated ABA concentrations prior to ABA application ( Figure 2A , C ) . While ABA uptake into guard cells ( Figure 3B ) and into the root occurred simultaneously ( Video 1 ) in all three analyzed regions ( Figure 3H , K , N ) , a delay in the ABAleon2 . 1 response was observed in the hypocotyl ( Figure 3E; Video 1 ) . These data suggest a directional ‘wave-like’ ABA transport in the hypocotyl , which was also indicated in the ratio images ( Figure 3F , Video 1 ) . To describe the ABA transport in the hypocotyl more quantitatively , ABAleon2 . 1 response curves of all three analyzed regions were fitted by a four parameter logistic curve R=Rmin+Rmax−Rmin1+ ( tt1/2 ) n . From these fits the t1/2 values measure the time point when ABAleon2 . 1 is half-saturated . These values were used to calculate the delay in ABAleon2 . 1 responses between the analyzed regions which is a measure for the speed of ABA transport . From three independent experiments the rate of ABA transport in the hypocotyl was calculated as 16 . 4 ± 0 . 8 µm/min . 10 . 7554/eLife . 01739 . 007Figure 3 . ABA-induced ABAleon2 . 1 responses in Arabidopsis tissues . Time-resolved ABAleon2 . 1 responses to 10 µM ABA in ( A–C ) guard cells of 45-day-old plants and ( D–F ) the hypocotyl , ( G–I ) the root differentiation- , ( J–L ) maturation- and ( M–O ) elongation-zone of 5-day-old seedlings . ( A , D , G , J , M ) Time course of mTurquoise ( mT , solid lines ) and cpVenus173 emission ( cpV , dashed lines ) and ( B , E , H , K , N ) the corresponding normalized emission ratios colored according to the analyzed regions boxed in the in initial t = 0 min images ( C , F , I , L , O ) . Each analysis is a representative of 3–4 experiments . Note , that there is a slight sample drift , which causes cpVenus173 emission increases in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01739 . 00710 . 7554/eLife . 01739 . 008Figure 3—figure supplement 1 . ABAleon2 . 1 but not the empty FRET cassette responds specifically to ABA . Time-resolved responses of ABAleon2 . 1 in ( A–C ) the hypocotyl and ( D–F ) the root maturation zone , and of ( G–I ) the empty FRET cassette in the hypocotyl . ( A , D , G ) Time course of mTurquoise ( solid lines ) and cpVenus173 emission ( dashed lines ) and ( B , E , H ) the respective normalized emission ratios colored according to the analyzed regions ( blue , yellow and red boxes ) given in the t = 0 min ratio images ( C , F , I ) . ( C , F , I ) Emission ratio images from indicated time points after solvent control ( 0 . 05 % EtOH ) , buffer or ABA application calibrated to the scale given in the final ratio images . Shown are representative analyses of one to three experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 01739 . 00810 . 7554/eLife . 01739 . 009Video 1 . 10 µM ABA-induced ABAleon2 . 1 responses in Arabidopsis . Video of 10 µM ABA-induced ABAleon2 . 1 responses in ( A ) guard cells , ( B ) the hypocotyl , ( C ) the root maturation- and ( D ) elongation-zone . ABA was applied at timepoint 00:00:00 of the indicated timescale . Single videos represent data of analyses in Figure 3 . Emission ratio changes to blue color indicate ABA concentration increase . ( B ) Emission ratio changes in the hypocotyl propagate gradually from the hypocotyl base towards the shoot . DOI: http://dx . doi . org/10 . 7554/eLife . 01739 . 009 As ABA was dissolved in EtOH , responses to EtOH as solvent control were analyzed in the hypocotyl ( Figure 3—figure supplement 1A–C ) and the root maturation zone ( Figure 3—figure supplement 1D–F ) . These treatments did not induce measurable emission ratio changes of ABAleon2 . 1 ( Figure 3—figure supplement 1B , E ) . In contrast , subsequent application of ABA induced ABAleon2 . 1 emission ratio changes similar to previous data ( Figure 3E , K , Figure 3—figure supplement 1B , E ) . In control seedlings expressing only the empty FRET-cassette no response to ABA was detected ( Figure 3—figure supplement 1G–I ) . These data support , that PYR1-ΔNABI1D413L incorporated in ABAleon2 . 1 ( Figure 1A ) are responsible for the ABA-induced emission ratio changes in Arabidopsis . Taken together , these data clearly demonstrate the direct and instantaneous ABA detection by ABAleon2 . 1 in various tissues and the visualization of ABA transport in planta . ABAleon2 . 1 was transformed into a PYR/PYL/RCAR ABA receptor quadruple mutant pyr1-1/pyl1-1/pyl2-1/pyl4-1 ( pyl4ple; Park et al . , 2009; Nishimura et al . , 2010 ) . ABA response curves of Col-0 wild type ( Figure 4A ) and pyl4ple ( Figure 4B ) were fitted by a four parameter logistic curve using data of four single measurements ( Figure 4A , B ) or the combined datasets ( Figure 4C ) . Data show that in the pyl4ple mutant ABAleon2 . 1 exhibited a faster response to 10 µM ABA in the root maturation zone when compared to Col-0 wild type ( Figure 4A–C ) . This finding is also reflected in the t1/2 values ( Figure 4D ) . Half saturation of ABAleon2 . 1 was reached 16 min after ABA application in Col-0 , while this appeared within 7 min in the pyl4ple mutant ( Figure 4C , D ) . 10 . 7554/eLife . 01739 . 010Figure 4 . Accelerated ABAleon2 . 1 responses in roots of the pyl4ple mutant . Normalized 10 µM ABA-induced ABAleon2 . 1 emission ratio changes in the root maturation zone of Col-0 ( A , C cyan line ) and pyr1-1/pyl1-1/pyl2-1/pyl4-1 ( pyl4ple ) ( B , C yellow line ) . ( A and B ) Data points from single measurements fitted by the respective four parameter logistic curve . ( C ) Combined data from four experiments in ( A and B ) fitted by the respective four parameter logistic curve . ( D ) t1/2 values ( means ± SEM , n = 4 ) calculated from the fitted curves in ( A and B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01739 . 010 To investigate whether ABAleon2 . 1 might affect ABA responses in general , two ABAleon2 . 1 lines ( line 3 and line 10 ) were compared to Col-0 wild type , YFP-PYR1 and abi1-3/YFP-ABI1 ( Nishimura et al . , 2010 ) over-expression lines . Analyses of the cpVenus173/YFP fluorescence emissions of the investigated lines indicated , that ABAleon2 . 1 ( line 3 ) exhibited a ∼ fivefold higher fluorescence emission ( expression ) when compared to ABAleon2 . 1 ( line 10 ) ( Figure 5A ) while emission of YFP-ABI1 was ∼ 7 % compared to YFP-PYR1 ( Figure 5A ) . 10 . 7554/eLife . 01739 . 011Figure 5 . ABAleon2 . 1-expressing plants show an ABA hyposensitivity . From left to right , Col-0 wild type , ABAleon2 . 1 ( line 3 ) , ABAleon2 . 1 ( line 10 ) , YFP-PYR1 and abi1-3/YFP-ABI1 . ( A ) Analyses of cpVenus173/YFP fluorescence emission in the leaf epidermis . Numerical fluorescence intensity values in the images represent means ± SEM of n = 4 images . ( B and C ) 7-day-old seedlings germinated and grown on 0 . 5 MS media supplemented with ( B ) 0 and ( C ) 0 . 8 µM ABA . ( D and E ) 9-day-old seedlings 5 days after transfer to 0 . 5 MS media supplemented with ( D ) 0 and ( E ) 10 µM ABA . ( F–H ) 7 day time course of ( F and G ) seed germination and ( H ) cotyledon expansion in presence of ( F ) 0 and ( G and H ) 0 . 8 µM ABA normalized to the seed count of each replicate ( means ± SEM , n = 4 technical replicates with 49 seeds/n ) . ( I ) Fresh weight of seedlings from ( D and E ) normalized to the 0 µM ABA control conditions ( means ± SEM , n = 5 technical replicates with seven seedlings/n ) . ( J ) Stomatal aperture of 20-23-day old seedlings exposed to 10 µM ABA normalized to the 0 µM ABA control conditions ( means ± SEM , n = 3 with ≥ 24 stomata/n ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01739 . 011 In seed germination ( Figure 5F , G ) and cotyledon expansion assays ( Figure 5B , C , H ) ABAleon2 . 1 lines were hyposensitive to 0 . 8 µM ABA when compared to Col-0 wild type and YFP-PYR1 . However , the ABAleon2 . 1 lines exhibited a less pronounced ABA hypersensitivity when compared to abi1-3/YFP-ABI1 ( Figure 5G , H ) . Interestingly , the degree of ABA hyposensitivity of both ABAleon2 . 1 lines correlated with ABAleon2 . 1 expression levels ( fluorescence emission ) , as the stronger expressing line 3 exhibited a reduced ABA sensitivity when compared to the lower expressing line 10 ( Figure 5 ) . In seedling assays 10 µM ABA inhibited growth of all investigated lines ( Figure 5D , E , I ) . Fresh weight of ABAleon2 . 1 and abi1-3/YFP-ABI1 plants , when grown on media supplemented with 10 µM ABA , was less reduced compared to Col-0 wild type ( Figure 5I ) . Again , the degree of ABA sensitivity of the ABAleon2 . 1 lines correlated with ABAleon2 . 1 expression levels ( Figure 5A , I ) . Interestingly , growth and root length of YFP-PYR1 plants was drastically reduced , when grown on 10 µM ABA media ( Figure 5D , E , I ) , suggesting a strong ABA hypersensitivity when over-expressing this receptor . In ABA-induced stomatal closure assays ABA sensitivity of ABAleon2 . 1 lines was comparable to Col-0 wild type responses ( Figure 5J ) , suggesting potential tissue or cell specific effects of ABAleon2 . 1 on ABA responses , which may be linked to a higher basal ABA concentration in these cells ( Figure 2C , Figure 2—figure supplement 1 ) . Taken together , ABAleon2 . 1 plants exhibit a reduced ABA sensitivity in seed germination and seedling growth which correlated with ABAleon2 . 1 expression levels . To study long-distance ABA transport , modeling clay was placed into the middle of each imaging chamber to generate two isolated chambers ( Figure 6A ) . Seedlings were placed such that the hypocotyl base and root differentiation zone laid over the isolating modeling clay , which isolated the shoot ( top chamber ) from the root ( bottom chamber ) . Ratio images of the shoot/upper hypocotyl ( Figure 6E ) and the root maturation zone ( Figure 6F ) were recorded before and after application of 50 µM ABA . Three regions in the upper hypocotyl and the root maturation zone , indicated by boxes in the t = 0 min ratio images ( Figure 6E , F ) , were used to measure time dependent ratio changes ( Figure 6B ) . Upon ABA application to the upper chamber , ABA concentrations increased in the shoot/upper hypocotyl , as seen by an immediate decrease of the ABAleon2 . 1 emission ratios ( Figure 6B , E ) . In the root maturation zone a decrease in the emission ratio was observed starting ∼ 90 min after the treatment ( Figure 6B , F ) . 10 . 7554/eLife . 01739 . 012Figure 6 . Visualization of long-distance ABA transport . ( A ) ABAleon2 . 1 seedlings were transferred to microscope dishes , which were divided into two isolated experimental chambers by a horizontal block of modeling clay . ( B , E , F ) Shoot-to-root , ( C , G , H ) hypocotyl-to-root and ( D , I , J ) root-to-hypocotyl ABA transport after application of 50 µM ABA . ( B–D ) Time-dependent normalized emission ratios ( means ± SEM , n = 3 ) in the hypocotyl ( cyan ) and root ( yellow ) were quantified in three regions indicated by boxes in the initial images ( E–J ) . The calibration bar in the final t = 180 min image indicates the scale of the emission ratios . Decreasing ratios indicate ABA accumulation . Shown are representative analyses of 3–4 experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 01739 . 01210 . 7554/eLife . 01739 . 013Figure 6—figure supplement 1 . Solvent control for long-distance ABA transport . ( A ) Time-dependent normalized ABAleon2 . 1 emission ratios ( mean ± SEM , n = 3 ) in the hypocotyl ( cyan ) and root ( yellow ) in response to 0 . 05 % EtOH as solvent control for ABA were quantified in three regions indicated by boxes in the initial images of ( B and C ) . The calibration bar in the t = 180 min image indicates the scale of the emission ratios . Shown are representative analyses of four experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 01739 . 013 In additional experiments seedlings were placed such that the hypocotyl and root differentiation zone were located in the top chamber and the root maturation zone in the bottom chamber ( Figure 6A , left seedling ) . In control experiments where 0 . 05 % EtOH as solvent control was added to the top chamber , no emission ratio changes were detected in the hypocotyl and the root maturation zone , indicating that in these experimental conditions endogenous ABA concentrations were stable ( Figure 6—figure supplement 1 ) . However , ABA application to the top chamber induced an ABAleon2 . 1 emission ratio change in the hypocotyl after 30 min which did not drastically change for the remaining time period ( Figure 6C , G ) . ABA application to the hypocotyl ( top chamber ) led to a gradual decrease of the ABAleon2 . 1 emission ratios in the root maturation zone ( bottom chamber; Figure 6C , H ) , providing evidence that ABA is actively transported to the root maturation zone . When ABA was applied to the root maturation zone ( bottom chamber ) , ABAleon2 . 1 emission ratios rapidly dropped , indicating ABA uptake into the root ( Figure 6D , J ) . However , ABA was not transported upwards to the hypocotyl ( top chamber ) within 180 min under the imposed conditions ( Figure 6D , I ) . The above data indicate a shoot to root ABA transport ( Figure 6B , C ) . A root to shoot ABA transport could not be detected within 180 min after ABA application ( Figure 6D ) possibly due to low transpiration when whole seedlings were perfused with buffer solution ( ‘Materials and methods and Discussion’ ) . Based on structural models ( Figure 7A , Figure 7—figure supplement 1A ) , mutations in PYR1 and ΔNABI1D413L of ABAleon2 . 1 were selected that potentially reduce but not abolish PYR1-ΔNABI1D413L interaction ( Melcher et al . , 2009; Miyazono et al . , 2009; Mosquna et al . , 2011; Zhang et al . , 2012 ) resulting in the new constructs ABAleon2 . 11–ABAleon2 . 17 ( Figure 7—figure supplement 1A; Table 1 ) . In addition shorter linker versions between PYR1 and ΔNABI1D413L were generated ( ABAleon2 . 2 and ABAleon2 . 3; Table 1 ) . Recombinant empty FRET control ( F3 ) and all ABAleon versions were purified ( Figure 7—figure supplement 2 ) and biochemical characteristics of these proteins are provided in Table 1 . 10 . 7554/eLife . 01739 . 014Figure 7 . In vitro analyses of ABAleon2 . 1 mutants . ( A ) Structural model of the PYR1 ( gold ) -ABA ( purple ) -ABI1 ( green ) complex , indicating mutations in ABAleon2 . 1 that were analyzed in ( B–G ) : H60P monomer-inducing , V83H in Pro-Cap and H115A in Leu-Lock of PYR1 ( grey balls ) and D413L phosphatase-inactivating in ABI1 ( purple ball ) . Emission spectra of ( B ) ABAleon2 . 11 , ( D ) ABAleon2 . 13 and ( F ) ABAleon2 . 15 in the absence ( unbound ) and presence of ( + ) -ABA ( bound ) with indicated dynamic range ( DR ) . ( B ) ABAleon2 . 11 exhibited no clear response to ABA . ( C ) ( + ) -ABA titrations of ABAleon2 . 11 compared to ABAleon2 . 1 suggest saturation of ABAleon2 . 11 in the absence of ABA . ( E and G ) ΔR/ΔRmax plots of ABAleon titrations with ( E ) naturally occurring ( + ) -ABA and ( G ) its enantiomer ( − ) -ABA , which binds more weakly . The respective apparent ABA affinities ( K′d ) are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 01739 . 01410 . 7554/eLife . 01739 . 015Figure 7—figure supplement 1 . ( + ) - and ( − ) -ABA titrations of selected ABAleons . ( A ) Structural model of the PYR1 ( gold ) -ABA ( purple ) -ABI1 ( green ) complex with indicated mutations in PYR1 ( grey balls ) and ABI1 ( purple balls ) of ABAleon2 . 1 ( ABI1D413L ) , ABAleon2 . 11 ( PYR1H60P ) , ABAleon2 . 12 ( PYR1F61L ) , ABAleon2 . 13 ( PYR1V83H ) , ABAleon2 . 14 ( PYR1L87F ) , ABAleon2 . 15 ( PYR1H115A ) , ABAleon2 . 16 ( PYR1E141Q ) and ABAleon2 . 17 ( ABI1E142Q ) . ( B and E ) Emission ratios , ( C and F ) ratio changes and ( D and G ) ΔR/ΔRmax of ABAleon1 . 1 , ABAleon2 . 1 , ABAleon2 . 13 and ABAleon2 . 15 plotted against ( B–D ) increasing ( + ) -ABA or ( E–G ) ( − ) -ABA concentrations . The color code and the respective mutations in the analyzed ABAleons are given above the graphs . Values in the graphs give ( B and E ) the dynamic range and ( C and F ) the apparent affinities ( K′d ) calculated form a four parameter logistic fit or ( D and G ) a three parameter sigmoidal Hill fit . DOI: http://dx . doi . org/10 . 7554/eLife . 01739 . 01510 . 7554/eLife . 01739 . 016Figure 7—figure supplement 2 . Purification of ABAleons after expression in E . coli . ( A–D ) Purification of recombinant ABAleon2 . 1 . ( A ) anti-GFP immuno-detection , ( B ) PageBlue stain of SDS-gel after blotting , ( C ) Instant Blue stain of indicated fractions after gel filtration ( GF ) run and ( D ) normalized absorbance at 280 nm ( protein; cyan ) and 516 nm ( cpVenus173; yellow ) measured during GF-run . Ex , extract; FT , flow through; W , wash; E , elution; W2 , Amicon filter wash; E2 , Amicon filter elution; GF , gel filtration with numbered fractions . ( E ) anti-GFP immuno-detection and ( F ) PageBlue stain of 1 µg empty FRET cassette ( F3 ) and ABAleon proteins after purification . DOI: http://dx . doi . org/10 . 7554/eLife . 01739 . 01610 . 7554/eLife . 01739 . 017Table 1 . Biochemical properties of ABAleonsDOI: http://dx . doi . org/10 . 7554/eLife . 01739 . 017ABAleonMutations/DeletionsRminRmaxDR [%]K’d ( 3 parameter Hill ) [nM]K’d ( 4 parameter logistic ) [nM]empty FRETΔ ( PYR1-ΔNABI1 ) 2 . 502 . 51−0 . 69––ABAleon1 . 1–0 . 870 . 98−14 . 83266 ± 55332 ± 45ABAleon2 . 1ΔNABI1D413L0 . 910 . 97−8 . 9879 ± 29114 ± 32ABAleon2 . 2Δ[ ( GGGGS ) 3] linker0 . 981 . 04−8 . 14121 ± 38156 ± 47ABAleon2 . 3Δ ( GGSGGTS ) linker0 . 940 . 99−7 . 5372 ± 1884 ± 22ABAleon2 . 11PYR1H60P , ΔNABI1D413L0 . 910 . 91−2 . 39––ABAleon2 . 12PYR1F61L , ΔNABI1D413L1 . 051 . 12−8 . 2987 ± 20107 ± 22ABAleon2 . 13PYR1V83H , ΔNABI1D413L0 . 910 . 97−7 . 098600 ± 71002900 ± 1500ABAleon2 . 14PYR1L87F , ΔNABI1D413L0 . 890 . 91−2 . 801200 ± 1200–ABAleon2 . 15PYR1H115A , ΔNABI1D413L0 . 921 . 01−10 . 09488 ± 45510 ± 41ABAleon2 . 16PYR1E141Q , ΔNABI1D413L0 . 971 . 02−7 . 30194 ± 46229 ± 48ABAleon2 . 17ΔNABI1D413L , E142Q0 . 780 . 82−4 . 9035 ± 1048 ± 8Biochemical properties of the empty FRET cassette and ABAleons with indicated mutations or deletions compared to the wild type ABAleon1 . 1 . Shown are minimum ( Rmin ) and maximum emission ratios ( Rmax ) , the dynamic range ( DR ) calculated as Rmin−RmaxRmin⋅100 and the apparent ABA affinity ( K′d ) calculated from a three parameter Hill fit or a four parameter logistic fit . Analyses showed that linker deletions slightly changed ABA affinity but also reduced the dynamic range ( Table 1 ) . Compared to other investigated ABAleon2 . 1 mutants , PYR1H60P in ABAleon2 . 11 strongly impaired ABA-induced emission changes ( Figure 7B ) , which were comparable to ABA-bound ABAleon2 . 1 ( Figure 7C; Table 1 ) . Of the seven analyzed ABAleon2 . 1 mutants , PYR1V83H in ABAleon2 . 13 and PYR1H115A in ABAleon2 . 15 were of particular interest , as these mutants exhibited a reduced ABA affinity ( K′d ∼ 3–4 µM of ABAleon2 . 13 and ∼ 500–600 nM of ABAleon2 . 15; Figure 7E; Table 1 ) . Also the dynamic range of these ABAleons was not drastically affected ( Figure 7D , F; Table 1 ) . Unless otherwise stated , all of the above analyses have been conducted using the natural ( + ) -enantiomer of ABA . In additional experiments , apparent affinities for ( + ) -ABA ( Figure 7E , Figure 7—figure supplement 1B–D ) were compared to binding of its unnatural enantiomer ( − ) −ABA ( Figure 7G , Figure 7—figure supplement 1E–G ) . In these analyses the dynamic ranges and ( + ) −ABA affinities of ABAleon1 . 1 ( K′d ∼ 300 nM ) and ABAleon2 . 1 ( K′d ∼ 100 nM ) ( Figure 7—figure supplement 1B–D ) were comparable to previous analyses ( Figure 1E , F; Table 1 ) . However , the binding affinity for ( − ) −ABA was reduced by about twofold in ABAleon1 . 1 ( K′d ∼ 600 nM ) and ABAleon2 . 1 ( K′d ∼ 180 nM ) ( Figure 7—figure supplement 1E–G ) . In case of ABAleon2 . 13 , affinities for ( + ) - and ( − ) -ABA were comparably low ( K′d ∼ 3–4 µM ) and ABAleon2 . 13 did not reach saturating conditions at 200 µM ( + ) - or ( − ) -ABA ( Figure 7E , G , Figure 7—figure supplement 1B–G ) . Remarkably , ABAleon2 . 15 harboring the PYR1H115A mutation exhibited an apparent affinity for ( − ) -ABA of ∼ 30 µM ( Figure 7G , Figure 7—figure supplement 1F , G ) , which was 50-fold reduced compared to the ( + ) -ABA affinity ( K′d ∼ 0 . 6 µ M ) ( Figure 7E , Figure 7—figure supplement 1C , D ) suggesting , that PYR1H115 has an important function in ABA stereospecificity . From these analyses , ABAleon2 . 13 and ABAleon2 . 15 could be good candidates for analyzing ABA concentration changes in cell types that have higher basal ABA concentrations . To investigate the utility of low affinity ABAleons , ABAleon2 . 13 , ABAleon2 . 14 and ABAleon2 . 15 were transformed into Arabidopsis Col-0 wild type plants . ABAleon2 . 14 was included in these analyses , as it exhibited an apparent K′d of ∼ 1 . 2 µM for ABA , however with a reduced dynamic range ( Table 1 ) . Initial analyses were performed in the root maturation zone of T2 lines and compared with ABAleon2 . 1 ( line 10 ) at comparable expression levels ( Figure 8 , Figure 8—figure supplement 1 ) . Examples of single representative measurements are presented in Figure 8—figure supplement 1 . All investigated ABAleons responded to externally applied 10 µM ABA with a negative emission ratio change ( Figure 8 , Figure 8—figure supplement 1 ) . While ABAleon2 . 13 and ABAleon2 . 14 exhibited a relatively low dynamic range in planta ( Figure 8B , C , E , Figure 8—figure supplement 1E , H ) , ABAleon2 . 15 responded comparable to ABAleon2 . 1 ( line 10 ) ( Figure 8A , D , E , F , Figure 8—figure supplement 1B , K ) . Thus , ABAleon2 . 15 is the best candidate for a low affinity ( K′d ∼ 600 nM ) ABA-reporter . 10 . 7554/eLife . 01739 . 018Figure 8 . ABA-induced ABAleon2 . 1 ( line 10 ) , ABAleon2 . 13 , ABAleon2 . 14 and ABAleon2 . 15 responses in the root maturation zone . 10 µM ABA-induced normalized emission ratio changes in the root maturation zone of ( A , E dark blue line ) ABAleon2 . 1 ( line 10 ) , ( B , E cyan line ) ABAleon2 . 13 , ( C , E yellow line ) ABAleon2 . 14 , and ( D , E orange line ) ABAleon2 . 15 . ( A–D ) Data points from single measurements fitted by the respective four parameter logistic curve . ( E ) Combined data from three to four experiments in ( A–D ) fitted by the respective four parameter logistic curve . ( F ) t1/2 values ( means ± SEM , n = 3–4 ) calculated from the fitted curves in ( A–D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01739 . 01810 . 7554/eLife . 01739 . 019Figure 8—figure supplement 1 . ABA-induced ABAleon2 . 1 ( line 10 ) , ABAleon2 . 13 , ABAleon2 . 14 and ABAleon2 . 15 responses in the root maturation zone ( examples ) . Responses to 10 µM ABA in the root maturation zone of ( A–C ) ABAleon2 . 1 ( line 10 ) , ( D–F ) ABAleon2 . 13 , ( G–I ) ABAleon2 . 14 and ( J–L ) ABAleon2 . 15 . ( A , D , G , J ) Time course of mTurquoise ( solid lines ) and cpVenus173 emission ( dashed lines ) and ( B , E , H , K ) the corresponding normalized emission ratios colored according to the analyzed regions boxed in the in initial t = 0 images ( C , F , I , L ) . Each analysis is a representative of 3–4 experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 01739 . 019 It is well established that plants synthesize ABA in response to water stress ( Seo and Koshiba , 2011 ) . Recent studies reported ABA increases in shoots and roots after 3 h of water stress ( Ikegami et al . , 2009; Geng et al . , 2013 ) . In addition , older studies reported ABA concentration increases within 15 min in guard cells of Vicia faba ( Harris and Outlaw , 1991 ) . 15 min after a drop in humidity ABAleon2 . 1 emission ratios decreased ∼ 3 % in guard cells and did not change when analyzed at the 30 min time point ( Figure 9A , B ) , indicating fast ABA concentration adjustments in response to humidity changes . In 4 h stress treatments of detached leaves 100 mM NaCl and 10 µM ABA induced a 5–6 % ABAleon2 . 1 emission ratio change in guard cells ( Figure 9C , D ) . In contrast , 300 mM sorbitol did not induce any detectable changes ( Figure 9C , D ) . 10 . 7554/eLife . 01739 . 020Figure 9 . ABAleon2 . 1 reports ABA concentration changes in response to low humidity , salt and osmotic stress . ABAleon2 . 1 emission ratios in response to ( A and B ) low humidity and ( C–H ) 4–6 h treatments with 0 . 01 % EtOH ( control ) , 100 mM NaCl , 300 mM sorbitol and 10 µM ABA in ( C and D ) guard cells , ( E and F ) the root maturation- and ( G and H ) elongation zone . ( A , C , E , G ) Representative emission ratio images with indicated calibration bars . ( B and D ) Normalized emission ratios in guard cells ( means ± SEM , n = 3 with ≥ 24 guard cell pairs/n ) . ( F and H ) Normalized emission ratios analyzed from two boxed regions ( cyan and yellow ) color-coded in the left images of ( E and G ) ( means ± SEM , n = 8–10 seedlings ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01739 . 020 In root tissues two separate regions labeled by cyan and yellow boxes ( Figure 9E , G , left ) were measured after 6 h stress treatments . In these analyses 100 mM NaCl or 300 mM sorbitol induced ABAleon2 . 1 emission ratio changes of 3–6 % in the root maturation zone compared to 7 % changes in response to 10 µM ABA ( Figure 9F ) . In the root elongation zone , responses to 100 mM NaCl and 300 mM sorbitol were 7–8 % while 10 µM ABA induced ABAleon2 . 1 emission ratio changes of 5–6 % ( Figure 9H ) . These analyses demonstrate the utility of ABAleon2 . 1 to report endogenous ABA concentration changes in response to low humidity , salt- and osmotic stress ( Figure 9 ) . Genetically encoded fluorescent protein-based reporters are powerful tools in cell biology ( Giepmans et al . , 2006; Okumoto et al . , 2012; Alford et al . , 2013 ) . Here we report the design , engineering and application of ABAleons , FRET-based reporters that enable the direct analysis of instantaneous ABA concentration changes in vitro and in planta . ABAleons were built on the strictly ABA-dependent interaction of PP2Cs with PYR1 and its phylogenetically close relatives ( Park et al . , 2009; Santiago et al . , 2009b; Nishimura et al . , 2010 ) , rather than the other members of the PYR/PYL/RCAR family , which have a higher probability of residing in a ‘monomeric’ state , and which can interact with PP2Cs even in the absence of ABA ( Ma et al . , 2009; Park et al . , 2009; Dupeux et al . , 2011; Hao et al . , 2011 ) . PYR1 and close homologs exhibit lower ABA affinities ( Kd ∼ 50–100 µM ) than those of the ‘monomeric’ homologs ( Kd ∼ 1 µM ) ( Ma et al . , 2009; Miyazono et al . , 2009; Santiago et al . , 2009b; Dupeux et al . , 2011 ) . In comparison , the ABA affinity of PYR/PYL/RCARs when bound to PP2Cs ranges between Kd ∼ 20–125 nM ( Ma et al . , 2009; Joshi-Saha et al . , 2011 ) . ABAleons exhibit similar ABA affinities to endogenous PYR/PYL/RCARs or PYR/PYL/RCARs in complex with PP2Cs ( Figure 1E , F , Figure 7E , Figure 7—figure supplement 1; Table 1 ) , consistent with our findings , that ABAleon2 . 1 is able to detect changes in endogenous ABA concentrations ( Figure 9 ) . ABAleons exhibit higher affinity to the naturally occurring ( + ) -ABA than to ( − ) -ABA ( Figure 7E , G , Figure 7—figure supplement 1B–G ) , and could potentially also bind to the synthetic ABA mimics pyrabactin and quinabactin ( Park et al . , 2009; Okamoto et al . , 2013 ) . One of the ABAleon derivatives , ABAleon2 . 15 , carrying the PYR1H115A mutation , exhibited strongly reduced affinity for ( − ) -ABA ( Figure 7G , Figure 7—figure supplement 1E–G ) , suggesting an important role of this amino acid in ABA stereospecificity . These results are consistent with recent findings that the homologous H139 in PYL3 was important for ABA stereospecificity ( Zhang et al . , 2013 ) . ABAleon2 . 11 , carrying the PYR1H60P mutation ( Dupeux et al . , 2011 ) , exhibited spectral characteristics comparable to ABA-bound ABAleon2 . 1 ( Figure 7C ) . This is consistent with the notion , that the PYR1H60P protein may form an alternative interaction interface with PP2Cs , even in the absence of ABA ( Dupeux et al . , 2011 ) . In Arabidopsis , over-expression of the ABA receptors PYR1 , PYL2 and PYL5 induces ABA hypersensitivity ( Figure 5; Santiago et al . , 2009b; Mosquna et al . , 2011 ) , while ectopic expression of the PP2Cs ABI1 and HAB1 decreases ABA sensitivity ( Figure 5; Santiago et al . , 2009b; Nishimura et al . , 2010 ) . Transgenic Arabidopsis plants expressing ABAleon2 . 1 exhibited a reduced sensitivity to exogenously applied ABA in seed germination , cotyledon expansion and growth assays , but not in stomatal closure assays ( Figure 5 ) . The degree of reduced ABA sensitivity correlated with increasing ABAleon2 . 1 expression levels in two independent transgenic lines ( Figure 5 ) . Because ABAleon2 . 1 did not exhibit any phosphatase activity in vitro ( Figure 1H ) , it can be speculated , that ABAleon2 . 1 might sequester a certain amount of physiologically relevant ABA in certain tissues and cell types due to its high affinity ( Figure 1E , F , Figure 7—figure supplement 1C , D; Table 1 ) , thus causing ABA hyposensitivity ( Figure 5 ) . Future generations of ABAleons and improved imaging sensitivity at lower ABAleon concentrations ( Figure 8 ) could enable the reduction of ABAleon side effects . By definition , a hormone transmits a signal from the site of hormone synthesis to its place of action . Although long-distance ABA transport has been studied for many years ( Sauter et al . , 2001; Wilkinson and Davies , 2002; Seo and Koshiba , 2011; Boursiac et al . , 2013 and references therein ) , no method for the direct detection of ABA concentration changes and ABA transport rates in planta has been available . The detailed characterizations of ABAleons demonstrate the utility of ABAleon2 . 1 ( K′d ∼ 100 nM ) and ABAleon2 . 15 ( K′d ∼ 600 nM ) , which exhibit a sufficient ABA-specificity and dynamic range ( 9–10 % ) upon ABA binding to monitor instantaneous ABA-induced or environmentally-triggered ABA concentration changes . ABAleon2 . 1 and ABAleon2 . 15 enabled measurements of rapid ABA-induced changes in ABA concentrations in various tissues ( Figure 3 , Figure 8 ) , ABA uptake into whole seedlings ( Figure 2A , B ) , directional ABA transport from the hypocotyl base towards the shoot ( Figure 3E , F; Video 1 ) and from the hypocotyl or shoot to the root ( Figure 6B , C , E-H ) . Under the imposed conditions the speed of ABA transport within the hypocotyl was ∼ 16 µm/min . Furthermore , ABA transport from the root maturation zone to the shoot could not be detected within three hours ( Figure 6D , I , J ) . Because the experimental setup ( Figure 6 ) , in which both shoot and root were perfused with buffer , would compromise the transpirational stream , the present data do not exclude concomitant ABA transport from roots to shoots , as has been found in other plant species ( Wilkinson and Davies , 2002 ) . In response to water stress , Arabidopsis plants synthesize ABA in the shoot , which has been reported to be transported to the root ( Ikegami et al . , 2009 ) . ABA accumulation in roots and leaves was detected 2 . 5–3 h after stress initiation ( Ikegami et al . , 2009; Geng et al . , 2013 ) . Older studies , using manually dissected guard cells , measured ABA concentration increases in guard cells 15 min after passive dehydration of leaves ( Harris and Outlaw , 1991 ) . ABAleon2 . 1 enabled the rapid detection of ABA concentration changes in guard cells in response to a humidity drop ( Figure 9A , B ) and the visualization of long term ABA accumulation in response to salt in guard cells ( Figure 9C , D ) and in response to salt- and osmotic stress in roots ( Figure 9E–H ) . Surprisingly , ABA-induced ABAleon2 . 1 ratio changes in the root maturation zone were accelerated in the pyr1-1/pyl1-1/pyl2-1/pyl4-1 mutant ( Figure 4 ) . Mutants defective in ABA signaling may compensate by up-regulating ABA levels , as reported previously ( Nakashima et al . , 2009 ) and could also up-regulate ABA transport activity ( Figure 4C , D ) . Alternatively , knock out of ABA receptors may also affect ABA buffering capacity ( Figure 4C , D ) . ABA reporter analyses of ABA uptake ( Figure 4 ) or long-distance ABA transport ( Figure 3D–F , Figure 6 ) could be utilized for the characterization or identification of ABA transporters and their regulation mechanisms in planta or in heterologous systems ( Jones et al . , 2014 ) . ABA induces cytoplasmic alkalinization of guard cells ( Blatt and Armstrong , 1993; Islam et al . , 2010 ) . In guard cells of Vicia faba , cytoplasmic pH was found to be 7 . 67 and increased 0 . 27 units upon ABA treatment ( Blatt and Armstrong , 1993 ) . Cytoplasmic pH in roots was 7 . 3 and could increase to 7 . 7 ( Bibikova et al . , 1998 ) . In these pH ranges ABA-induced ABAleon1 . 1 emission ratio changes were stable in vitro ( Figure 1—figure supplement 1B ) . In vivo ABA concentrations in cellular compartments of specific plant cells and tissues are currently unknown . Overall ABA levels range from 30–50 ng/g dry-weight in non-stressed plants ( Forcat et al . , 2008; Geng et al . , 2013 ) , which can increase up to 30-fold in response to limited water conditions ( Harris et al . , 1988; Harris and Outlaw 1991; Christmann et al . , 2007; Ikegami et al . , 2009; Geng et al . , 2013 ) . ABAleon2 . 1 and ABAleon2 . 15 exhibit a sufficient dynamic range for in vitro calibrations ( Figure 1E , F , Figure 7E , Figure 7—figure supplement 1B–D ) that permit approximations of cellular ABA concentrations , for example in the range of ≤ 25 nM in the root elongation zone . In Vicia faba guard cells ABA concentrations were ∼ 0 . 7 fg/cell pair in unstressed and ∼ 17 . 7 fg/cell pair in stressed guard cells ( Harris et al . , 1988 ) . ABA concentrations in stressed guard cells were estimated to be in the range of ∼ 15 µM ( Harris et al . , 1988; Harris and Outlaw 1991 ) . Extrapolating from these values , unstressed guard cell ABA concentration would be ∼ 500 nM . Such approximations would be consistent with the partial saturation and reduced response of ABAleon2 . 1 in guard cells ( Figure 3A–C ) and with strong expression of the ABA-induced reporter pRAB18-GFP ( Figure 2—figure supplement 1 ) . Our results demonstrate that ABAleons affect ABA signaling to certain extent , but can analyze changes in ABA concentrations in diverse tissues and cell types and measure ABA transport in vivo . ABAleons will thus allow hitherto challenging investigations of ABA synthesis and transport in planta , in response to changes in environmental conditions or treatment with synthetic compounds designed to improve plant survival and crop yields under adverse climate conditions . Further , in combination with other genetically encoded reporters , ABAleons can be used to decipher the cross talk between ABA and other signaling molecules . During the course of our ABAleon research we found , that Jones et al . developed ABACUS-type ABA reporters , however with biochemical properties complementary to ABAleons ( Jones et al . , 2014 ) . Thus , ABAleons and ABACUS could be utilized to study novel aspects of ABA signaling in intact plants . Fluorescent protein coding sequences were re-amplified from the pF40 plasmid ( Piljić et al . , 2011 ) and ligated into Xba I/Apa I ( mTurquoise ) or Xma I/Sac I sites ( cpVenus173 ) of a modified pUC19 plasmid ( Walter et al . , 2004; Waadt et al . , 2008 ) resulting in the pUC-F3 and pUC-F3_II empty FRET-cassettes with the latter containing a Nde I-site downstream of the Xba I-site and a StrepII-tag fusion of cpVenus173 at its free end . PYR1-GGSGG and ( GGGGS ) 4-ΔNABI1 and mutants were cloned and inserted between Apa I/Spe I and Spe I/Xma I sites of pUC-F3 and pUC-F3_II to obtain pUC-ABAleon and pUC-ABAleon_II , respectively . Escherichia coli expression vectors were obtained by sub-cloning ABAleon_II versions Nde I/Sac I into pET-24b ( + ) ( Novagen , Darmstadt , Germany ) . For expression in plants , the pUBQ10 promoter ( AT4G05310; Norris et al . , 1993; Krebs et al . , 2012 ) , inserted between Hind III/Spe I sites of a modified pUC19 plasmid ( kindly provided by Jörg Kudla , University of Münster ) , was mutated to remove a Sac I site within the pUBQ10 . In addition , the HSP18 . 2 terminator ( T ) ( AT5G59720; Nagaya et al . , 2010 ) was inserted between the Sac I/Eco RI sites of pBluescript II ( Stratagene , La Jolla , CA ) and the Hind III site within the HSP18 . 2T was deleted resulting in pKS-HSP18 . 2TΔHind III . Both pUBQ10ΔSac I and HSP18 . 2TΔHind III were sub-cloned into plant compatible vectors pGPTVII . Bar , which confers glufosinate ( BASTA ) resistance , and pGPTVII . Hyg , which confers hygromycin resistance ( Walter et al . , 2004 ) , resulting in the barII-UT and hygII-UT plasmids . Finally , ABAleon2 . 1 , ABAleon2 . 1 mutants and the empty FRET-cassette were sub-cloned from pUC-ABAleon2 . 1 , pUC-ABAleon2 . 1x_II and pUC-F3 plasmids into the barII-UT and hygII-UT plasmids to obtain the barII-UT-ABAleon2 . 1 , barII-UTF3 and hygII-UT-ABAleon2 . 1 and mutant plasmids for expression in plants . More detailed information about oligo-nucleotides and plasmids used and generated in this work is provided in Supplementary file 1A and Supplementary file 1B . pET-F3_II ( empty FRET ) and pET-ABAleon_II versions in BL21-CodonPlus ( DE3 ) -RIL cells ( Stratagene ) were grown at 37 °C in 2 L Luria Broth ( LB ) medium containing 25 µg/mL Kanamycin . At an optical density at 600 nm ( OD600 ) of 0 . 5 , cells were induced with 0 . 5 mM Isopropyl β-D-1-thiogalactopyranoside ( IPTG ) and shaken for additional 4-6 h at 24 °C . Cells were collected by centrifugation ( 15 min 5 . 000×g and 4 °C ) and stored at −80 °C . Proteins were extracted by sonification after 60 min incubation in 20 mL extraction buffer ( 30 mM Tris–HCl [pH 7 . 4] , 250 mM NaCl , 1 mM Ethylenediaminetetraacetic acid [EDTA] , 1 mM Phenylmethylsulfonyl fluoride [PMSF] , 1x protease inhibitor [Roche , USA] and 1 mg/mL Lysozym ) . Cell debris was removed by centrifugation ( 2 × 30 min , 20 . 000×g and 4 °C ) and by filtration through 0 . 45 µm syringe filters . Protein extracts were mixed with 2 . 5 ml 50 % Strep-Tactin Macroprep resin ( IBA , Göttingen , Germany ) pre-equilibrated in wash buffer I ( 30 mM Tris–HCl [pH 7 . 4] , 250 mM NaCl , 1 mM EDTA ) and protein/resin mix was incubated for 1 h at 4 °C while shaking in a 50 ml tube . The suspension was run twice through a 20 mL gravity column ( BioRad , Hercules , CA ) followed by two washes of the remaining protein/resin mix with 10 column volumes ( CV ) of wash buffer I and one wash with 10 CV of wash buffer II ( 30 mM Tris–HCl [pH 7 . 4] , 250 mM NaCl , 10 mM MgCl2 and 1 mM MnCl2 ) . Proteins were eluted 3x in 1 CV wash buffer II supplemented with 2 . 5 mM Desthiobiotin ( Sigma , USA ) and concentrated to ∼ 1 mL volume by centrifugation at 3 . 000×g and 4 °C using Amicon Ultra-4 30K or 10K centrifugal filters ( Millipore , Billerica , MA ) . Purified proteins were run through a Superdex 200 HiLoad 16/60 column ( GE Healthcare ) in wash buffer II using an ÄKTA purifier fast protein liquid chromatography ( FPLC ) system ( GE Healthcare ) with 0 . 8 MPa column pressure limit , 1 mL/min flow rate and 2 mL fraction size volume . Fractions exhibiting 280 nm , 445 nm and 516 nm absorbance were analyzed by SDS-PAGE and Instant Blue ( Cole–Parmer , USA ) protein staining and selected for further concentration using Amicon Ultra-4 centrifugal filters . Protein aliquots were flash frozen in liquid N2 and stored at −80 °C . Protein purity was analyzed by SDS-PAGE , immuno-blotting using anti-GFP antibody ( Life Technologies , Darmstadt , Germany ) and PageBlue staining ( Thermo Scientific , Rockford , IL; Waadt et al . , 2014 ) . Results of F3 empty FRET and ABAleon purifications are provided in Figure 7—figure supplement 2 . The coding sequence of ΔNABI1 corresponding to amino acid residues 125–429 was inserted via Nde I/Bam HI into pET28a ( Novagen; Supplementary file 1A , Supplementary file 1B ) and transformed into E . coli BL21 ( DE3 ) . E . coli were grown at 37 °C , and protein expression was induced by 1 mM IPTG at OD600 of 0 . 6–0 . 8 . After overnight incubation at 25 °C , cells were harvested by centrifugation ( 15 min 5 . 000×g and 4 °C ) and re-suspended in 50 mM Tris–HCl ( pH 8 . 0 ) and 500 mM NaCl . Cells were sonicated on ice and lysates were obtained after centrifugation at 12 , 000×g for 1 h 6xHis-ΔNABI1 extracts were applied to a Ni-NTA column ( Qiagen , Hilden , Germany ) and washed with five bed volumes of 50 mM Tris–HCl ( pH 8 . 0 ) , 500 mM NaCl and 10 mM imidazole . Bound proteins were eluted in 50 mM Tris–HCl ( pH 8 . 0 ) , 500 mM NaCl and 300 mM imidazole . 6xHis-ΔNABI1 was further purified using Sephacryl S-200 ( GE Healthcare ) in 50 mM Tris–HCl ( pH 8 . 0 ) and 150 mM NaCl . 6xHis-PYR1 ( Nishimura et al . , 2009 ) and 6xHis-ΔNABI1 proteins were re-buffered and concentrated into wash buffer II using Amicon Ultra-4 centrifugal filters . Protein purity and concentrations were analyzed by SDS-PAGE and PageBlue staining and quantified according to a 0–2000 ng BSA ( NEB , Ipswich , MA ) standard calibration . ABA titration experiments were conducted in a TECAN Infinite M1000 PRO ( TECAN , Männedorf , Switzerland ) using 1 µM ABAleon protein in wash buffer II with 0 . 1 % EtOH or DMSO as solvent for ( + ) -ABA ( TCI , Portland , OR ) , generally used in assays unless otherwise stated , or ( − ) -ABA ( Sigma ) . Protein samples were excited with 440 ± 5 nm and emission 450–700 nm was measured in 1 nm steps with 5 nm bandwidth and 10 flashes of 20 µs and 400 Hz . Gain settings to obtain optimal emission spectra were calculated by the TECAN software from unbound ABAleon emission . Emission bands of mTurquoise ( 470–490 nm ) and cpVenus173 ( 518–538 nm ) were used to calculate cpVenus173/mTurquoise emission ratios . Apparent ABA affinities ( K′d ) were calculated from emission ratio plots by fitting a four parameter logistic curve R=Rmin+Rmax−Rmin1+ ( [ABA]K'd ) n or from ΔR/ Δ Rmax plots by fitting a three parameter sigmoidal Hill equation ΔRΔRmax=Rmax·[ABA]nKd′n·[ABA]n ( Palmer et al . , 2006 ) using the SigmaPlot 10 . 0 version ( Systat , San Jose , CA ) . Dynamic ranges were calculated from experimentally determined values as Rmin−RmaxRmin·100 . Absorbance spectra ( 275–700 nm , slit 0 . 2 nm ) of ABAleons were analyzed with a UV-VIS-Spectrophotometer ( UV-2700 ) ( Shimadzu , Columbia , MD ) . Absorbance at 434 nm ( mTurquoise ) was used to calculate concentrations of the empty FRET and ABAleon proteins ( Goedhart et al . , 2010 ) and the ratio of Abs515 ( cpVenus173 ) and Abs434 was used to estimate protein purity . pH titrations were performed by addition of 1 µL concentrated ABAleon1 . 1 protein ( final concentration 200 nM ) to 100 µL wash buffer II adjusted to a pH range between pH 5 . 0–8 . 2 with 1 M HCl and MES powder or with 2 M NaOH . After recordings of ABA-free ABAleon1 . 1 emission spectra , using the TECAN Infinite M1000 PRO as mentioned above , 1 µL of 1 mM ABA ( final concentration 10 µM ABA ) was added and emission spectra were recorded using identical settings . Experiments were performed in duplicate and fitted by a four parameter Hill equation R=R0+ ( a·xncn+xn ) in SigmaPlot 10 . 0 ( Systat , San Jose , CA ) . Ratio change was calculated by subtraction of the ABA-free from the ABA-bound equation values . ABA-induced ABAleon2 . 1 kinetics were analyzed using 2 . 77 µM ABAleon2 . 1 in a Berthold Mithras LB 940 ( Berthold Technologies , Bad Wildbad , Germany ) with the following settings: Lamp energy 5000 , counting time 0 . 05 s , excitation 440 ± 10 nm , emission 470 ± 10 nm ( mTurquoise ) and 530 ± 20 nm ( cpVenus173 ) measured in cycles of 6 . 12 s . At cycle 25 , 50 µL 3 µM ABA in wash buffer II and 0 . 3 % EtOH was applied with low injector speed to result in the final 1 µM ABA in 0 . 1 % EtOH . Phosphatase assays were performed using the serine/threonine phosphatase assay system ( Promega , Madison , WI ) . In brief , 50 µL reactions containing wash buffer II , 10-50 pmol protein , 5000 pmol Ser/Thr phosphopeptide ± 5 µM ABA with 0 . 005 % EtOH as solvent were incubated for 10–30 min at room temperature . Reactions were stopped by addition of 50 µL of the supplied molybdate dye/additive mixture and phosphate release was measured according to a standard curve with a Berthold Mithras LB 940 plate reader ( Absorbance 600 ± 10 nm , lamp energy 50 , 000 and counting time 2 s ) . Three-dimensional coordinates of major components of ABAleon were built with known crystal structures of mTurquoise ( pdb: 2YE0 , Goedhart et al . , 2012 ) , PYL1-ABA-ABI1 ( pdb: 3JRQ , Miyazono et al . , 2009 ) and Venus ( pdb: 1MYW , Rekas et al . , 2002 ) . Each component of ABAleon was manually assembled using COOT ( Emsley et al . , 2010 ) . In the assembly , PYL1 was replaced by PYR1 ( pdb: 3K3K , Nishimura et al . , 2009 ) by tracing the Cα backbone . The unstructured C-terminus of mTurquoise was placed in distance corresponding to the PG linker and the PYR1 N-terminus . Venus was placed between ABI1 and mTurquoise with the N-terminus of Venus facing towards the C-terminus of ABI1 . All structural figures were drawn with PyMOL ( The PyMOL Molecular Graphics System , Version 1 . 5 . 0 . 4 Schrödinger , LLC . ) . Seeds were sterilized in 70 % EtOH and 0 . 04 % sodium dodecyl sulfate ( SDS ) , washed three times in 100 % EtOH and sown on 0 . 5 Murashige and Skoog ( MS ) media ( Sigma ) adjusted to pH 5 . 8 with 1 M KOH and supplemented with 0 . 8 % phytoagar . After at least 4 days of stratification in the dark at 4 °C plants were cultivated in a growth room ( 16 h day/8 h night cycle , 25 °C , 50–100 µEm−2s−1 and 26 % relative humidity ) or in a CMP4030 plant growth chamber ( 16 h day/8 h night cycle , 22 °C , 50 µEm−2s−1 and 25 % relative humidity; Conviron , Winnipeg , Manitoba ) . 6-day-old seedlings were transferred to soil and grown either in the growth room or in a CMP3244 plant growth chamber ( 16 h day , 22 °C/8 h night , 18 °C cycle , 50–100 µEm−2s−1 and 30–50 % relative humidity; Conviron ) . barII-UTF3 empty FRET , barII-UT-ABAleon2 . 1 , hygII-UT-ABAleon2 . 13 , hygII-UT-ABAleon2 . 14 and hygII-UT-ABAleon2 . 15 were transformed into Arabidopsis Columbia 0 accession and hygII-UT-ABAleon2 . 1 was transformed into pyl4ple [pyr1-1 ( Q169 stop ) /pyl1-1 ( SALK_054640 ) /pyl2-1 ( GT2864 ) /pyl4-1 ( SAIL_517_C08 ) ] ( Park et al . , 2009; Nishimura et al . , 2010 ) by the floral dip method ( Clough and Bent , 1998 ) . Transformants were selected on 0 . 5 MS media supplemented with either 10 µg/mL glufosinate or 25 µg/mL hygromycin and further cultivated in soil in a CMP3244 plant growth chamber . Positive transformants were further selected by fluorescence intensity at a confocal microscope ( see below ) . A list of transgenic Arabidopsis lines generated and used in this work is provided in Supplementary file 1C . For ABA seed germination assays seeds were sown on 0 . 5 MS agar media supplemented with 0 . 08 % EtOH as solvent control or 0 . 8 µM ( + ) -ABA ( TCI , Portland , OR ) . After stratification , plants were grown in the growth room . Germinated seeds and seedlings with expanded green cotyledons were counted for a time period of 7 days with blinded genotypes . Analyses represent mean values ± SEM of four technical replicates normalized to the seed count ( 48–50 seeds ) of each experiment . For growth assays , 4-day-old seedlings grown on 0 . 5 MS agar media were transferred to 0 . 5 MS agar media supplemented with 0 . 1 % EtOH as solvent control or 10 µM ( + ) -ABA and grown vertically in the growth room . Images were acquired 5 days after seedling transfer . Fresh weight was measured from pools of seven seedlings/experiment ( means ± SEM , n = 5 ) normalized to the 0 µM ABA control conditions . ABA-induced stomatal closure analyses were performed with 20-23-day-old plants grown vertically on 0 . 5 MS agar media in the CMP4030 plant growth chamber . Six detached leaves were floated in assay buffer ( 5 mM KCl , 50 µM CaCl2 and 10 mM MES-Tris pH 5 . 6 ) at 22 °C and 100 µE m−2s−1 for 2 h . Subsequently , ( + ) -ABA or EtOH as solvent control was added to a final concentration of 10 µM ( + ) -ABA in 0 . 1 % EtOH followed by additional 2 h incubation . Leaves were blended 4x ∼10 s in ∼50 ml deionized water and leaf epidermal tissue was collected through a 100 µm nylon mesh ( Millipore , Billerica , MA ) and mounted on a microscope slide for imaging . Images were acquired using an inverted light microscope ( Nikon Eclipse TS100 ) equipped with a 40x/0 . 65 ∞/0 . 17 WD . 0 . 57 objective and connected to the Scion camera and Scion VisiCapture Application Version1 . 3 ( Scion Corporation , Frederick , MD ) . Experiments were performed with blinded genotypes and treatments and ≥ 24 stomatal apertures were measured per experiment using Fiji ( Schindelin et al . , 2012 ) . Data represent mean stomatal apertures ± SEM of three experiments normalized to the solvent control . For guard cell imaging , 4-week-old detached leaves without mid vein were glued with the abaxial side on a cover glass using medical adhesive ( Hollister , Libertyville , IL ) and upper cell layers were dissected away with an industrial razor blade . Epidermal strips were incubated in assay buffer ( 5 mM KCl , 50 µM CaCl2 , 10 mM MES-Tris , pH 5 . 6 ) and 0 . 01 % EtOH , as solvent control for ABA , for 1 h . Glass cover slips were mounted on a microscope slide with a central hole ( Ø = 13 mm ) using vacuum grease silicone ( Beckman , Pasadena , CA ) and analyzed in 200 µL of buffer mentioned above . For ABA application , epidermal strips were perfused by washing ( pipetting ) four to five times with assay buffer supplemented with 10 µM ABA in 0 . 01 % EtOH . Low humidity drop experiments were conducted on 19-27-day-old seedlings grown vertically on 0 . 5 MS agar media in the CMP4030 plant growth chamber . Low humidity was induced by opening the lid of the 0 . 5 MS agar plates . At time points 0 , 15 and 30 min after plate opening two seedlings were blended 4x ∼10 s in ∼50 mL deionized water and leaf epidermal tissue was collected through a 100 µm nylon mesh ( Millipore , Billerica , MA ) and mounted on a microscope slide for imaging . Experiments were performed in triplicates and ≥ 27 guard cell pairs were analyzed per experiment . Long term stress treatments were conducted on detached leaves of 20-24-day-old plants grown vertically on 0 . 5 MS agar media in the CMP4030 plant growth chamber . Four leaves were pre-incubated for 1–4 h in assay buffer at 22 °C and 100 µEm−2s−1 and treatments were performed by addition of assay buffer supplemented with 10-fold concentrated EtOH ( as solvent control for ABA ) , ABA , NaCl or sorbitol to obtain final concentrations of 0 . 01 % EtOH , 10 µM ABA , 100 mM NaCl or 300 mM sorbitol . 4 h after the treatments leaves were blended ( see above ) and leaf epidermal tissue was collected for imaging . Experiments were performed in triplicates with blinded treatments and epidermal fractions used for ABAleon2 . 1 emission ratio imaging were selected using the bright field channel . 24–40 guard cell pairs were analyzed per experiment . For seedling imaging , 4-day-old seedlings were transferred to glass bottom dishes ( MatTek , Ashland , MA ) supplemented with 200 µL 0 . 25 MS , 10 mM MES-Tris ( pH 5 . 6 ) and 0 . 7 % low melting point agarose ( Promega ) and grown vertically for an additional day in the CMP4030 plant growth chamber . Before microscopic analyses , 90 µL assay buffer was added . Treatments were conducted by pipetting 10 µL ABA solution or the respective amount of EtOH as solvent control diluted in assay buffer to reach a final concentration of 10–50 µM ABA and 0 . 01–0 . 05 % EtOH . To apply ABA to defined tissues , transparent modeling clay was used to divide each glass bottom dish into two isolated chambers before application of the growth media . Seedlings were placed on top of the growth media and modeling clay , such that either the hypocotyl base and root differentiation zone ( shoot to root transport ) or the root maturation zone ( hypocotyl to root and root to hypocotyl transport ) laid on the dry modeling clay . Long term treatments were performed on 5-day-old seedlings , which were transferred to glass bottom dishes ( MatTek , Ashland , MA ) supplemented with 200 µL 0 . 25 MS , 10 mM MES-Tris ( pH 5 . 6 ) and 0 . 7 % low melting point agarose with addition of 0 . 01 % EtOH ( as solvent control for ABA ) , 10 µM ABA , 100 mM NaCl or 300 mM sorbitol . 6 h after transfer 100 µL assay buffer supplemented with 0 . 01 % EtOH , 10 µM ABA , 100 mM NaCl or 300 mM sorbitol was added before the root maturation- and elongation zone were imaged . Regions of the root maturation zone with similar distance from the root tip were selected in the bright field channel before ratio images were acquired . Treatments were performed blinded and 8–10 seedlings were analyzed per treatment . pRAB18-GFP plants ( Kim et al . , 2011 ) were grown in soil in the CMP3244 plant growth chamber . To ensure high relative humidity ( RH ) conditions ( 70% ) plants were kept under a plastic cover and sprayed with water twice a day . 2 days before the analyses , plants were removed from the growth chamber , placed in 25 % RH conditions and withheld from water supply . For the ABA treatment , detached leaves were floated for 4 h in 50 µM ABA prior to microscopic analyses . Expression analyses based on cpVenus173 , YFP or GFP emission were performed using an Eclipse TE2000-U microscope equipped with Plan 20x/0 . 40 ∞/0 . 17 WD 1 . 3 and Plan Apo 60x/1 . 20 WI ∞/0 . 15–0 . 18 WD 0 . 22 objectives ( Nikon ) , a CascadeII 512 camera ( Photometrics ) , a MFC2000 z-motor ( Applied Scientific Instruments , Eugene , OR ) , a QLC-100 spinning disc ( VisiTech international , Sunderland , UK ) , a CL-2000 Diode pumped crystal laser ( LaserPhysics Inc . , West Jordan , UT ) , a LS 300 Kr/Ar laser ( Dynamic Laser , Boston , MA ) and guided by Metamorph software version 7 . 7 . 7 . 0 ( Molecular Devices ) . Images were analyzed , processed and calibrated in Fiji ( Schindelin et al . , 2012 ) . ABAleon ratio-imaging was conducted according to Allen et al . ( 1999 ) using an Eclipse TE300 microscope equipped with a Plan Fluor 10x/0 . 30 DIC L ∞/0 . 17 WD 16 . 0 for seedlings or a Plan Fluor 40x/1 . 30 Oil objective DIC H ∞/0 . 17 WD 0 . 2 for guard cells ( Nikon , Tokyo , Japan ) , a Cool SNAP HQ camera ( Photometrics , Tucson , AZ ) , a Mac 2002 System automatic controler , a CAMELEON filter set 71007A ( D440/20 , D485/40 , D535/30; Chroma , Bellows Falls , VT ) and guided by the MetaFluor software version 7 . 0r3 ( Molecular Devices , Sunnyvale , CA ) . Images were acquired in intervals of 6 s using 200-250 ms exposure , Binning 4 , Gain 2x ( 4x ) and 20 MHz transfer speed . Images were analyzed and processed using Fiji ( Schindelin et al . , 2012 ) . Analyses of seedlings were standardized as treatments were performed 4 min after the experiments started and regions used for emission ratio analyses had identical areas and distances from each other . ABA response curves in the root maturation zone and the hypocotyl were analyzed by fitting R=Rmin+Rmax−Rmin1+ ( tt1/2 ) n , using SigmaPlot 10 . 0 ( Systat , San Jose , CA ) , to either data of single measurements or combined datasets .	Plants are able to respond to detrimental changes in their environment—when , for example , water becomes scarce or the soil becomes too salty—in ways that minimize stress and damage caused by these changes . Hormones are chemicals that trigger the plant’s response under these circumstances . Abscisic acid is the hormone that regulates how plants respond to drought and salt stress and that controls the plant growth in these conditions . In the past , it was possible to measure the average level of this hormone in a given tissue , but not the level in individual cells in a living plant . Moreover , it was difficult to follow directly how abscisic acid moved between the plant cells , tissues or organs . Now , Waadt et al . ( and independently Jones et al . ) have developed tools that can measure the levels of abscisic acid within individual cells in living plants and in real time . The plants were genetically engineered to produce sensor proteins with two properties: they can bind to abscisic acid in a reversible manner , and they contain two ‘tags’ that fluoresce at different wavelengths . Shining light onto the plant at a specific wavelength that is only absorbed by one of the tags actually causes both of the tags on the sensor proteins to fluoresce . However , the sensors fluoresce more at one wavelength when they are bound to abscisic acid , and more at the other wavelength when they are not bound to abscisic acid . Hence , measuring the ratio of these two wavelengths in the light that is given off by the sensor proteins can be used as a measure of the concentration of abscisic acid in a plant cell . Waadt et al . developed sensor proteins called ‘ABAleons’ , and used one of these to analyze the uptake , distribution and movement of abscisic acid in different tissues in the model plant Arabidopsis thaliana . Changes in the level of abscisic acid could be detected at the level of an individual plant cell , and over time scales of fractions of seconds to hours . ABAleons also revealed that the concentration of abscisic acid in guard cells—specialized cells that help stop the loss of water vapor from a leaf—increases when humidity levels are low , or when salt levels are high . Low water levels , or high salt levels , also slowly increased the concentration of abscisic acid in the roots of the plant . Furthermore , Waadt et al . saw that abscisic acid moved long distances from the base of the stem up into the shoot , and down to the root . Waadt et al . also report that the ABAleons made plants less responsive to abscisic acid , possibly because binding to the ABAleons reduced the amount of abscisic acid that was available to perform its role as a hormone . The next challenge is to engineer ABAleons that minimize this unwanted side effect , and then go on to use ABAleons to study environmental conditions and proteins involved in plant hormone responses .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology", "cell", "biology" ]	2014	FRET-based reporters for the direct visualization of abscisic acid concentration changes and distribution in Arabidopsis
The cytosolic antiviral innate immune sensor RIG-I distinguishes 5′ tri- or diphosphate containing viral double-stranded ( ds ) RNA from self-RNA by an incompletely understood mechanism that involves ATP hydrolysis by RIG-I's RNA translocase domain . Recently discovered mutations in ATPase motifs can lead to the multi-system disorder Singleton-Merten Syndrome ( SMS ) and increased interferon levels , suggesting misregulated signaling by RIG-I . Here we report that SMS mutations phenocopy a mutation that allows ATP binding but prevents hydrolysis . ATPase deficient RIG-I constitutively signals through endogenous RNA and co-purifies with self-RNA even from virus infected cells . Biochemical studies and cryo-electron microscopy identify a 60S ribosomal expansion segment as a dominant self-RNA that is stably bound by ATPase deficient RIG-I . ATP hydrolysis displaces wild-type RIG-I from this self-RNA but not from 5' triphosphate dsRNA . Our results indicate that ATP-hydrolysis prevents recognition of self-RNA and suggest that SMS mutations lead to unintentional signaling through prolonged RNA binding . The innate immune system provides a rapid initial reaction to invading pathogens and also stimulates the adaptive immune system ( Iwasaki and Medzhitov , 2015 ) . Pattern recognition receptors ( PRRs ) of the innate immune system sense pathogen- or danger-associated molecular patterns ( PAMPs or DAMPs ) and trigger molecular cascades that together initiate and orchestrate the cellular response through activation of e . g . interferon regulatory factors and nuclear factor κB ( Brubaker et al . , 2015; Pandey et al . , 2015; Wu and Chen , 2014 ) . Retinoic-acid inducible gene I ( RIG-I ) , melanoma differentiation-associated gene 5 ( MDA5 ) and laboratory of physiology and genetics 2 ( LGP2 ) are three structurally related PRRs – denoted RIG-I like receptors ( RLRs ) – that recognize cytosolic foreign RNA . RIG-I senses RNA from a broad range of viruses including measles virus and Sendai virus ( both paramyxoviridae ) , Influenza A virus , Japanese encephalitis virus and Hepatitis C virus , whereas MDA5 is activated for example by picornavirus RNA . LGP2 has augmenting and regulatory roles in MDA5 and RIG-I dependent signaling ( Bruns et al . , 2014; Satoh et al . , 2010; Sparrer and Gack , 2015 ) . RIG-I preferentially detects base-paired double-stranded RNA ( dsRNA ) ends containing either 5′ triphosphate ( ppp ) or 5′ diphosphate ( pp ) moieties ( Goubau et al . , 2014; Hornung et al . , 2006; Pichlmair et al . , 2006; Schlee et al . , 2009; Schmidt et al . , 2009 ) and not 2’ OH methylated at the first 5’ terminal nucleotide ( Schuberth-Wagner et al . , 2015 ) . ppp-dsRNA arises , for example , at panhandle structures of influenza virus nucleocapsids , or during measles or Sendai virus transcription ( Liu et al . , 2015; Weber et al . , 2013 ) . 5′ diphosphates are found on genomic RNA of reoviruses ( Banerjee and Shatkin , 1971 ) . RIG-I can also detect poly-U/UC-rich dsRNA ( Schnell et al . , 2012 ) . Ligands of MDA5 are less well characterized but include dsRNA longer than 1000 bp ( Kato et al . , 2008 ) , higher-order dsRNA structures ( Pichlmair et al . , 2009 ) , or AU-rich RNA ( Runge et al . , 2014 ) . RLRs are members of the superfamily II ( SF2 ) of ATPases , helicases or nucleic acid translocases . RIG-I and MDA5 consist of two N-terminal tandem caspase activation and recruitment domains ( 2CARD ) , a central ATPase/translocase domain and a C-terminal regulatory domain ( RD ) . LGP2 lacks the 2CARD module but otherwise has a similar domain architecture . Binding of RNA induces a conformational change in RIG-I . If activated , the RD binds the ppp- or pp-dsRNA end , while the SF2 domain interacts with the adjacent RNA duplex and forms an active ATPase site ( Civril et al . , 2011 ) . In this conformation , the 2CARD module is sterically displaced from its auto-inhibited state ( Jiang et al . , 2011; Kowalinski et al . , 2011; Luo et al . , 2011 ) and can be K63-linked poly-ubiquitinated ( Gack et al . , 2007 ) . Multiple Ub-2CARD complexes assemble to form a nucleation site for the polymerization of mitochondrial antiviral-signaling adaptor protein ( MAVS ) into long helical filaments ( Hou et al . , 2011; Wu et al . , 2014; Xu et al . , 2014 ) . Instead of recognizing terminal structures like RIG-I , MDA5 cooperatively polymerizes along dsRNA ( Berke and Modis , 2012 ) , which is suggested to trigger MAVS polymerization . The SF2 ATPase domain plays a critical part in RIG-I activation , although the role of the ATPase activity is still debated . Mutation of the seven SF2 “helicase” motifs resulted in RLRs that are either inactive or signal constitutively ( Bamming and Horvath , 2009; Louber et al . , 2015 ) . On the other hand , overexpression of the 2CARD module alone is sufficient for signaling ( Yoneyama et al . , 2004 ) . Further studies revealed that the SF2 domain is an ATP-dependent dsRNA translocase ( Myong et al . , 2009 ) that can help enhance signaling by loading multiple RIG-I on dsRNA ( Patel et al . , 2013 ) and may execute anti-viral “effector” functions through displacement of viral proteins ( Yao et al . , 2015 ) . Finally , RIG-I ATPase activity promotes recycling of RIG-I:dsRNA complexes in vitro , suggesting a kinetic discrimination between self and non-self RNA ( Anchisi et al . , 2015; Louber et al . , 2015 ) . Several autoimmune diseases , including the Aicardi-Goutières and Singleton-Merten syndromes ( SMS ) , were linked to single amino acid mutations in the SF2 domains of MDA5 and RIG-I ( Funabiki et al . , 2014; Jang et al . , 2015; Rice et al . , 2014; Rutsch et al . , 2015 ) . Two point mutations within the Walker A ( motif I ) or Walker B ( motif II ) of RIG-I are linked to atypical SMS and functional studies indicated constitutive RIG-I activation ( Jang et al . , 2015 ) . Thus , these mutations have been described as a gain of function , which is puzzling considering previous mutations in motif I led to loss of RIG-I function , while mutations in motif II led to either gain or loss of function , depending on the type of mutation ( Bamming and Horvath , 2009; Louber et al . , 2015 ) . In order to clarify the role of RIG-I’s ATPase in antiviral signaling and RLR associated human diseases , we engineered structure-derived and patient-identified mutations into RIG-I and tested the resulting proteins in different types of cell-based and in vitro analyses . Collectively , we find that SMS mutations phenocopy the structure-derived E373Q mutation in motif II , which is designed to trap RIG-I in an ATP-bound state . Freezing this state results in a dramatic autoimmune response because the enzyme binds self-RNA and signals . An unexpected , strongly enriched self-RNA is the ribosomal large subunit , which contains large , dsRNA expansion segments . Collectively , our results suggest that a biomedical and functional critical role of RIG-I’s ATPase is to prevent spontaneous and unintended activation by self-RNA . Thus , the SF2 translocase likely increases the sensitivity of the system by reducing background signaling . Furthermore , our studies suggest that in SMS , RIG-I is trapped in an ATP-bound state and signals through self-ligands . To address the roles of ATP binding and hydrolysis by the SF2 domain of RIG-I , we studied RIG-I variants containing structure-based mutations designed to i ) prevent ATP binding and formation of a functional ATP-bound complex , ii ) allow ATP binding and ATP-induced conformational changes but prevent ATP hydrolysis , or iii ) disable interaction of the RNA with either the 1A or 2A domain of SF2 ( Figure 1A , B ) . The structure of RIG-I in complex with RNA and ADP·BeFx served as guide for these mutations ( [Jiang et al . , 2011] , PDB code 3TMI , Figure 1B ) . 10 . 7554/eLife . 10859 . 003Figure 1 . Cellular studies of RIG-I ATPase mutants in infected or non-infected cells . ( A ) Location of amino acid substitutions of RIG-I SF2 domain variants used in this study ( orange lines ) within different RLR helicase motifs ( orange squares ) . ( B ) Single amino acid substitutions ( orange ) within the RIG-I 3D structure ( PDB: 3TMI ) . ( C ) Fold change of interferon-β ( IFNβ ) promoter driven luciferase activity in uninfected HEK 293T RIG-I KO cells or in cells challenged with Sendai virus defective interfering particles ( SeV DIs ) . Cells were co-transfected with RIG-I expression vectors and p125-luc/ pCMV-RL reporter plasmids , and infected with SeV DIs 6 hr post transfection . Firefly ( FF ) luciferase activities were determined in respect to Renilla ( Ren ) luciferase activities 24 hpi . All ratios were normalized to the empty vector control . n=3–12 , error bars represent mean values ± standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 10859 . 00310 . 7554/eLife . 10859 . 004Figure 1—figure supplement 1 . Assay for defining of the impact of RLR variant expression on RLR signaling in infected or non-infected cells . ( A ) HEK 293T RIG-I KO cells were co-transfected with different expression and control vectors as indicated . RLR signaling induces an interferon-β ( IFNβ ) promoter driven expression of firefly luciferase ( FF ) . Renilla luciferase ( Ren ) is constitutively expressed via a CMV promoter and serves as transfection control . ( B ) Western Blot analysis of virus-induced RIG-I expression in HEK 293T and HEK 293T RIG-I KO cells . ( C ) Western Blot control of overexpressed FLAG/HA tagged RIG-I variants in HEK 293T RIG-I KO cells from Figure 1 , panel C . DOI: http://dx . doi . org/10 . 7554/eLife . 10859 . 00410 . 7554/eLife . 10859 . 005Figure 1—figure supplement 2 . RIG-I E373Q mutation does not confer constitutive activity due to an exposed 2CARD module . ( A ) Small-angle X-ray scattering ( SAXS ) intensity curves of RIG-I and RIG-I E373Q in presence and absence of ATP . ( B ) Distance distribution functions derived from SAXS data in panel A . Calculated radii of gyration ( Rg ) are indicated within the legend . ( C ) Thermal shift assays in presence and absence of ATP . Melting temperatures I are indicated within the legend . ( D ) Fold change of interferon-β promoter driven luciferase activity of HEK 293T RIG-I KO cells co-transfected with a RIG-I E373Q expression vector , varying concentrations of a RIG-I Δ2CARD , E373Q expression vector and p125-luc/ pCMV-RL reporter plasmids . Firefly luciferase ( FF ) activities were determined in respect to Renilla luciferase ( Ren ) activities 24 hr after transfection . All ratios were normalized to an empty vector control . n=3 , error bars represent mean values ± standard deviation . ( E ) Control Western Blot analysis of FLAG/HA-tagged constructs from panel D . DOI: http://dx . doi . org/10 . 7554/eLife . 10859 . 005 In order to dissect the influence of these mutations on the ability of RIG-I to elicit downstream signaling , we used an interferon-β ( IFNβ ) promoter activity assay carried out in HEK 293T RIG-I KO cells ( Figure 1—figure supplement 1A , B ) . Overexpressed wild-type RIG-I ( wtRIG-I ) is able to induce a slight activation of the IFNβ promoter , which can be further amplified by stimulation with Sendai virus defective interfering particles ( SeV DIs ) ( Figure 1C ) . The 2CARD module ( RIG-I 1-229 ) induced a strong activation in both non-infected and SeV DI-stimulated cells and is crucial since constructs lacking these domains ( Δ2CARD , RIG-I 230-925 ) cannot conduct any downstream signaling . RIG-I K270I , carrying a mutation in the motif I lysine that reduces ATP binding ( Rozen et al . , 1989 ) , signaled in neither uninfected nor SeV DIs stimulated cells , consistent with previous studies . Remarkably , the E373Q substitution in motif II had a strikingly different effect . RIG E373Q , which has a stabilized ATP-bound state by slowed-down ATP hydrolysis , strongly signaled in both non-infected and SeV DIs stimulated cells . Western blots validated correct expression of all mutants ( Figure 1—figure supplement 1C ) . To rule out a “constitutive” active conformation of RIG-I E373Q due to an exposed 2CARD module ( e . g . from an unfolded SF2 ) we performed small angle X-ray scattering with purified wtRIG-I and RIG-I E373Q demonstrating that both proteins have the same solution structure ( Figure 1—figure supplement 2A , B ) . In addition , thermal unfolding assays show that the E373Q mutation does not destabilize RIG-I ( Figure 1—figure supplement 2C ) . Finally , RIG-I Δ2CARD , E373Q has a dominant negative effect on signaling by RIG-I E373Q ( Figure 1 , Figure 1—figure supplement 2D , E ) . Taken together , these data show that RIG-I E373Q is neither destabilized nor constitutively active , suggesting it needs productive RNA interactions . To test whether E373Q signals in non-infected ( and perhaps also infected cells ) because of interaction with self-RNA , we additionally introduced mutations in various RNA binding sites , in particular a ΔRD variant ( RIG-I 1-798 ) and mutations in two RNA-interacting residues in domains 1A ( T347A ) and 2A ( V699A ) of SF2 . The single mutation RIG-I T347A did not signal in either infected or non-infected cells , showing that the interaction of RNA with this specific amino acid is critical for signaling ( Figure 1C ) . Interestingly , we find that the single mutation V699A slightly increases the signaling activity of RIG-I in non-infected cells ( Figure 1C ) , which could be explained by a putative reduction of translocation activity instead of a prevention of RNA binding to SF2 ( see discussion ) . Finally , deletion of the regulatory domain ( ΔRD ) inactivates signaling in both infected and non-infected cells as previously observed ( Cui et al . , 2008 ) . As expected , both combination mutants RIG-I E373Q , T347A and RIG-I E373Q , ΔRD failed to signal in both SeV DIs infected and non-infected cells . These data show that the increased immunostimulatory effect of E373Q requires a productive RNA interaction of SF2 and RD . Since RD is also required for the displacement of the 2CARD module from SF2 , we additionally analyzed a point mutation in RD . K888 mediates triphosphate binding in RD and mutations in this residue inactivate recognition of viral RNA ( Cui et al . , 2008; Wang et al . , 2010 ) . Of note RIG-I E373Q , K888T is still constitutively active in non-infected cells . This effect indicates that the increased signaling capacity on endogenous RNA is independent from the ppp-dsRNA or pp-dsRNA epitopes that RIG-I recognizes on viral RNA via the RD . Finally , we addressed the effect of the Singleton-Merten mutations C268F and E373A . E373A is at the same position as our structure-derived E373Q mutant . Consistent with this , we observed that this substitution leads to a constitutive activation of the IFNβ promoter ( Jang et al . , 2015 ) ( Figure 1C ) . Interestingly , although C268 is located in motif I , it also leads to constitutive signaling , whereas motif I mutation of K270 ( which coordinates the β-phosphate of ATP ) blocks ATP binding and renders RIG-I inactive . Thus , mutation of the non-ATP binding C268 in motif I appears to phenocopy a mutation that prevents ATP hydrolysis . In summary , our studies show that signaling of RIG-I requires both ATP and RNA binding . ATP hydrolysis , on the other hand , appears to be critical to dissolve the signaling state and to prevent activation of RIG-I by self-RNA . We hypothesized that E373Q traps RIG-I in an ATP bound high affinity conformation that is activated already by self-RNA . To test this idea , we immunoprecipitated RIG-I and its mutants from non-infected HEK 293T RIG-I KO cells or cells infected with measles or Sendai virus and analyzed the co-purified RNA molecules . Regardless of whether co-purified from infected or non-infected cells , the amount of RNA recovered from RIG-I E373Q was about 3 times higher than that from RIG-I ( Figure 2A ) . Similarly increased amounts of RNA co-purified with the SMS mutants C268F and E373A from uninfected cells , reflecting the same altered RNA binding properties as in RIG-I E373Q ( Figure 2—figure supplement 1A ) . 10 . 7554/eLife . 10859 . 006Figure 2 . RIG-I ATP hydrolysis defective mutant E373Q recognizes the 60S ribosomal subunit in vivo . ( A ) Relative RNA amount co-purified with overexpressed RIG-I or RIG-I E373Q from virus infected or non-infected HEK 293T RIG-I KO cells . n=4 ( infected ) or n=10 ( non-infected ) , error bars represent mean values ± standard deviation . ( B ) Bioanalyzer evaluation and agarose gel separation of RNA co-purified with overexpressed RIG-I or RIG-I E373Q from non-infected HEK 293T RIG-I KO cells . Curves are normalized in respect to 18S rRNA peaks . ( C ) Bioanalyzer evaluation and agarose gel separation of total RNA content of non-infected HEK 293T RIG-I KO cells overexpressing RIG-I or RIG-I E373Q . Curves were normalized as in panel B . ( D ) Immunostimulatory potential of co-purified RNA from RIG-I , RIG-I E373Q or GFP overexpressed in measles virus ( MeV ) , MeV-Cko-ATU-Cs or Sendai virus Cantell ( SeV ) infected HEK 293T RIG-I KO cells . RNA was back-transfected into HEK 293T ISRE-FF/RFP cells together with pTK-RL transfection control . Firefly luciferase ( FF ) activities were determined 24 hr after transfection in respect to Renilla luciferase ( Ren ) activity and were normalized to the immunostimulatory potential of RIG-I associated RNA . n=4 , error bars represent mean values ± standard deviation . ( E ) Immunostimulatory potential of endogenous RNA in cells overexpressing RIG-I E373Q . RNA was co-transfected into HEK 293T RIG-I KO cells together with a RIG-I E373Q expression vector and p125-luc/ pCMV-RL reporter plasmids . FF luciferase activities were determined in respect to Ren luciferase activities 24 hr after transfection . All ratios are normalized to the RIG-I E373Q control without RNA stimulation . Purified RNA was in addition analyzed on agarose gels . n=3 , error bars represent mean values ± standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 10859 . 00610 . 7554/eLife . 10859 . 007Figure 2—figure supplement 1 . Analysis of RNA co-purified with RIG-I SMS or MDA5 variants . ( A ) Relative RNA amount co-purified with overexpressed RIG-I , RIG-I E373Q or RIG-I SMS variants from non-infected HEK 293T RIG-I KO cells . ( B ) Relative RNA amount co-purified with overexpressed MDA5 from non-infected HEK 293T cells . n=3 , error bars represent mean values ± standard deviation . ( C ) Bioanalyzer evaluation of RNA co-purified with overexpressed MDA5 from non-infected HEK 293T cells . Curves are normalized in respect to 18S rRNA peaks . DOI: http://dx . doi . org/10 . 7554/eLife . 10859 . 00710 . 7554/eLife . 10859 . 008Figure 2—figure supplement 2 . Assay for defining the immunostimulatory potential of different RNAs . ( A ) Endogenous RLRs in HEK 293T ISRE-FF/RFP cells ( stably express firefly luciferase ( FF ) and RFP under control of an interferon stimulated response element ( ISRE ) promoter ) induce a downstream signaling cascade upon binding to transfected RNA . Subsequent interferon ( IFN ) expression results in activation of the STAT signaling pathway which in return induces ISRE promoter driven expression of FF luciferase . DOI: http://dx . doi . org/10 . 7554/eLife . 10859 . 00810 . 7554/eLife . 10859 . 009Figure 2—figure supplement 3 . Immunostimulatory potential of co-purified RNA from Sendai virus Cantell ( SeV ) infected cells . ( A ) HEK 293T RIG-I KO cells were transfected with the indicated RIG-I mutant or GFP expression vector . RNA co-purified with the respective overexpressed protein was back-transfected into HEK 293T ISRE-FF/RFP cells ( compare with Figure 2—figure supplement 2 ) . Firefly luciferase activities were determined 24 h after transfection and normalized to the RIG-I sample . DOI: http://dx . doi . org/10 . 7554/eLife . 10859 . 009 When analyzed on a Bioanalyzer RNA chip or on agarose gels , we found that the increased amount of RNA is to a large extent due to the presence of 28S rRNA , while 18S rRNA remains unaltered ( Figure 2B ) . Control analysis of the total RNA content ruled out an alteration of ribosome subunit ratio in RIG-I E373Q transfected cells ( Figure 2C ) . Both increased amount of RNA and specific enrichment of 28S rRNA were also observed for the equivalent MDA5 E444Q Walker B mutant ( Figure 2—figure supplement 1B , C ) . In order to determine the immunostimulatory potential of the RNA co-purified from virus-infected cells , we back-transfected the RNA into HEK 293T ISRE-FF/RFP reporter cells ( which contain endogenous RIG-I , see Figure 2—figure supplement 2A ) . RNA co-purified with wtRIG-I and RIG-I lacking the 2CARD module induced an immune response in these cells ( Figure 2—figure supplement 3A ) . RNA co-purified with RIG-I K270I ( ATP binding deficient ) and V699A ( putative translocation deficient ) was also able to stimulate the ISRE reporter in an amount comparable to wtRIG-I , indicating no altered RNA binding properties in these mutants under virus infected conditions . In contrast , RNA that co-purified with the RNA-binding deficient RIG-I T347A ( mutation in SF2 domain ) , RIG-I K858E ( mutation in RD domain that reduces triphosphate recognition ) or RIG-I ΔRD poorly stimulated the ISRE promoter and probably represents background RNA ( Figure 2—figure supplement 3A ) . These data suggest that RIG-I recognizes immunostimulatory RNA via the SF2 and RD domains , but does not require ATP binding for this process . ATP binding is necessary , however , because RIG-I K270I expression alone does not stimulate the IFNβ promoter ( compare with Figure 1C ) . Interestingly , RNA co-purified with RIG-I E373Q failed to induce reporter gene expression ( Figure 2D , Figure 2—figure supplement 3A ) . Thus , despite the observation that RIG-I E373Q co-purifies with approximately threefold more RNA than wtRIG-I from infected cells , the co-purified RNA is not immunostimulatory in a wtRIG-I background . However , cells that transiently express RIG-I E373Q can be further stimulated by transfection of total RNA extracts and purified ribosomal RNA ( Figure 2E ) , suggesting that ribosomal RNA can activate RIG-I E373Q . Cells lacking wtRIG-I or RIG-I E373Q on the other hand do not respond to those RNAs . We conclude that host-RNA , which does not activate wtRIG-I , can apparently compete with viral RNA for RIG-I E373Q . In order to verify a higher affinity of the RIG-I ATP hydrolysis defective mutant towards ribosomal RNA , we purified full-length human RIG-I and RIG-I E373Q , as well as human 80S ribosomes , and tested for a direct interaction . We confirmed that while both RIG-I E373Q and the wild-type protein are able to bind ATP , only wtRIG-I can hydrolyze ATP ( Figure 3A , B ) . We subsequently conducted sedimentation assays via ultra-centrifugation of sucrose cushions loaded with 80S ribosomes that have been pre-incubated with wtRIG-I or RIG-I E373Q in presence or absence of ATP or the non-hydrolysable ATP analogue ADP·BeF3 . In presence of ATP a minor binding of wtRIG-I to the ribosome could be observed , whereas RIG-I E373Q bound in a near stoichiometric manner . In absence of ATP or in presence of ADP·BeF3 binding of wtRIG-I was greatly enhanced and showed similar levels compared to RIG-I E373Q ( Figure 3C ) . 10 . 7554/eLife . 10859 . 010Figure 3 . RIG-I ATP hydrolysis defective mutant E373Q recognizes the 60S ribosomal subunit in vitro . ( A ) DRaCALA ATP binding assay of RIG-I or RIG-I E373Q in presence or absence of RNA . ( B ) ATP hydrolysis assay of RIG-I or RIG-I E373Q in presence and absence of RNA . ( C ) Binding studies of human 80S ribosomes with RIG-I or RIG-I E373Q in presence or absence of ATP or ADP·BeF3 . Pre-formed complexes were separated on sucrose cushions via ultracentrifugation and pellet ( P ) as well as supernatant ( SN ) fractions were analyzed by SDS-PAGE . ( D ) Side views of a cryo-EM reconstruction of RIG-I E373Q ( blue ) bound to the human 80S ribosome ( yellow: 40S subunit , gray: 60S subunit ) . Data was low pass-filtered at 15 Å . ( E ) Side views of a cryo-EM reconstruction of the human 80S ribosome without prior RIG-I E373Q incubation . Data filtering and color coding as in panel D . ( F ) Statistical difference map ( left , σ = 2 ) of cryo-EM reconstructions in panels D and E reveals a significant additional density at expansion segment 7L A ( ES7L-A , pink ) into which RIG-I ( PDB 3TMI ) can be fitted ( right , σ = 1 . 51 ) . ( G ) Secondary structure map of the 28S rRNA ES7L ( derived from ( Anger et al . , 2013 ) and zoom into RIG-I E373Q binding area . ES7L-A is indicated in pink ( as in panel F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10859 . 010 We next analyzed RIG-I E373Q:80S ribosome complexes by cryo-electron microscopy and single particle 3D reconstruction ( Figure 3D ) . The average resolution was estimated to be 17 . 7 Å based on the Fourier shell correlation cut-off criterion at 0 . 5 . When compared with the reconstruction of the human 80S ribosome alone ( Figure 3E ) , the ribosome:RIG-I E373Q complex revealed an additional density located at rRNA expansion segment ( ES ) 7L , which is located at the back of the large ribosomal subunit . Calculation of a statistical difference map between the two reconstructions confirmed that this distinct region contained significant additional density ( Figure 3F ) . Human ribosomes contain several long , G:C rich , base-paired RNA expansion segments forming large tentacle-like hairpin structures of substantial double-stranded nature ( Anger et al . , 2013 ) . A large part of the double-stranded RNA in these segments is not covered by ribosomal proteins and accessible for cytosolic proteins . The crystal structure of ADP·BeFx-bound RIG-I △2CARD:RNA complex ( ( Jiang et al . , 2011 ) , PDB code 3TMI ) fits well into the density observed at ES7L and is located at the root of the solvent exposed portion of helix A of ES7L that contains a contiguous stretch of seven G:C/C:G base pairs ( Figure 3G ) . In summary , we conclude that stabilizing the ATP-bound state of RIG-I induces a conformation where RIG-I binds to ribosomes , presumably at exposed dsRNA expansion segments . To further evaluate the role of ATP binding and hydrolysis of RIG-I we performed electrophoretic mobility shift assays ( EMSAs ) , fluorescence anisotropy experiments and ATP hydrolysis assays in presence and absence of ATP or ADP·BeF3 with different RNAs . These RNAs mimic different types of endogenous or viral RNAs and help dissecting contributions of RD’s binding to the RNA end and SF2’s binding to the stem . In addition to a 24mer or 12mer blunt-ended dsRNA or ppp-dsRNA ( Goldeck et al . , 2014 ) , we also used a 60 nucleotide hairpin RNA ( denoted as ES hairpin ) derived from the ribosomal expansion segment ES7L , which contains several bulges and a non-pairing end ( Figure 4—figure supplement 1A ) . The hairpin at one end and the added Y-structure at the other end are used to minimize RNA end binding by RIG-I’s RD because RD has a high affinity for blunt RNA ends . RIG-I and RIG-I E373Q bound to the 24mer blunt ended dsRNA with a slightly higher affinity in presence of ATP or ADP·BeF3 than in its absence ( Figure 4A ) , suggesting that ATP binding to the SF2 domain positively contributes to the overall affinity in addition to RD . A similar result was obtained when we used a 12mer dsRNA in fluorescence anisotropy experiments in order to further dissect the influence of different RNA ends ( Figure 4B ) . Interestingly , the positive effect of ATP was not observed when we used the corresponding ppp-dsRNA 12mer ( Figure 4C ) , most likely because the RD dominates RNA binding under these conditions . Thus , it is plausible that RIG-I dissociates from unphosphorylated RNA termini with an increased rate after ATP hydrolysis than from triphosphorylated termini . 10 . 7554/eLife . 10859 . 011Figure 4 . RIG-I’s ATP hydrolysis enhances RNA end recognition and removes RIG-I from RNA stems . ( A ) Quantification of electrophoretic mobility shift assays of RIG-I or RIG-I E373Q incubated with 24mer dsRNA in presence or absence of ATP , ADP or ADP·BeF3 ( compare with Figure 4—Figure supplement 1B ) . ( B ) Fluorescence anisotropy changes measured by titrating RIG-I or RIG-I E373Q in presence or absence of ATP into solutions containing fluorescently labeled 12mer dsRNA . ( C ) Fluorescence anisotropy changes measured by titrating RIG-I or RIG-I E373Q in presence or absence of ATP into solutions containing fluorescently labeled 12mer ppp-dsRNA . ( D ) Quantification of electrophoretic mobility shift assays of RIG-I , RIG-I E373Q or RIG-I T347A , E373Q incubated with an RNA hairpin derived from helix A of the human ribosome expansion segment 7L ( ES hairpin ) in presence or absence of ATP , ADP or ADP·BeF3 ( compare with Figure 4—Figure supplement 1C ) . All binding curves were fitted using the LL . 2 function of the R drc package ( Cedergreen et al . , 2005 ) . n=3-6 , error bars represent mean values ± standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 10859 . 01110 . 7554/eLife . 10859 . 012Figure 4—figure supplement 1 . Design of the ribosomal expansion segment derived hairpin RNA , EMSA raw figures and control experiments with RIG-I C268F SMS mutant . ( A ) RIG-I E373Q binding site at ES7L-A was used to design a 60b hairpin RNA ( ES hairpin ) . RNA secondary structure was determined with the RNAfold webserver ( Gruber et al . , 2008 ) . ( B ) Electrophoretic mobility shift assays of RIG-I or RIG-I E373Q incubated with 24mer dsRNA . Complexes were pre-formed at 37 °C for 20 min , separated on agarose gels and stained with GelRed . Free RNA bands were quantified using ImageJ . Protein concentrations ( from left to right ) : 0 , 0 . 1 µM , 0 . 3 µM , 0 . 5 µM , 0 . 7 µM , 1 µM , 1 . 5 µM and 2 µM . : unbound RNA , : protein:RNA complexes . ( C ) Electrophoretic mobility shift assays of RIG-I , RIG-I E373Q or RIG-I C268F incubated with ES hairpin RNA . Complexes were pre-formed , separated and stained as in panel B . Protein concentrations ( from left to right ) : 0 , 0 . 5 µM , 1 µM , 2 µM , 3 µM , 4 µM , 5 µM and 10 µM . : unbound RNA , **: protein:RNA complexes . DOI: http://dx . doi . org/10 . 7554/eLife . 10859 . 012 We next tested the role of ATP on binding of wtRIG-I , RIG-I E373Q , RIG-I T347A , E373Q and the SMS variant RIG-I C268F to the ES hairpin RNA mimicking the base of the ribosomal ES7L . In presence of ATP we observed moderately increased binding of RIG-I E373Q and of RIG-I C268F to this hairpin , however wtRIG-I displayed a strikingly opposing effect ( Figure 4D , Figure 4—figure supplement 1C ) . For this RNA , ATP reduced rather than increased the affinity of wtRIG-I . The addition of ADP·BeF3 to RIG-I could reconstitute the high affinity state of RIG-I E373Q . The RIG-I T347A , E373Q double mutant , on the other hand , showed binding affinities similar to RIG-I in presence of ATP , probably caused by residual binding of RD ( Figure 4D ) . Consistent with this , the ES hairpin RNA could induce signaling in RIG-I E373Q transfected HEK 293T RIG-I KO cells ( Figure 2F ) and could also stimulate the ATPase activity of RIG-I △2CARD , and to a lesser extent wtRIG-I ( which is auto-inhibited by the 2CARD module ) ( Figure 5A , Figure 5—figure supplement 1A ) . A comparable stimulatory effect on the ATPase activity of RIG-I could also be detected with whole human ribosomes ( Figure 5A ) . Control assays with the ATP hydrolysis defective mutants RIG-I E373Q and RIG-I T347A , E373Q confirmed the lacking ability of those proteins to hydrolyze ATP even in the presence of triphosphorylated RNA ( Figure 5A , Figure 5—figure supplement 1B ) . 10 . 7554/eLife . 10859 . 013Figure 5 . RIG-I’s ATPase activity correlates with its RNA binding affinity . ( A ) Quantification of hydrolyzed [γ-32P]ATP by RIG-I or RIG-I E373Q in presence of different RNA substrates . Reactions were allowed to proceed for 20 min at 37 °C and free phosphate was separated from ATP via thin layer chromatography . Spots corresponding to labeled ATP and labeled Pi were quantified using ImageJ . All curves were fitted using the LL . 2 function of the R drc package . n=3 , error bars represent mean values ± standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 10859 . 01310 . 7554/eLife . 10859 . 014Figure 5—figure supplement 1 . RIG-I’s 2CARD module reduces the ATP hydrolysis activity . ( A ) Measurement of ES hairpin or ppp-dsRNA stimulated [γ-32P]ATP hydrolysis of RIG-I or RIG-I Δ2CARD . Reactions were monitored over 3 hr at room temperature and free phosphate was separated from ATP via thin layer chromatography . ( B ) Quantification of hydrolyzed [γ-32P]ATP by RIG-I T347A , E373Q in presence of 24mer ppp-dsRNA . Reactions were allowed to proceed for 20 min at 37 °C and free phosphate was separated from ATP via thin layer chromatography . Spots corresponding to labeled ATP and labeled Pi were quantified using ImageJ . Curves were fitted using the LL . 2 function of the R drc package . n=3 , error bars represent mean values ± standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 10859 . 014 In summary , our results show that ATP hydrolysis leads to a moderately increased binding of RNA containing base-paired ends , but decreased binding of RNA lacking base-paired ends . These in vitro data are also consistent with our co-immunopurification studies of RNA from cells , where we observed that the ATP hydrolysis deficient RIG-I E373Q mutant co-purified with increased amounts of endogenous RNA . Here we show that mutations that slow down or inhibit RIG-I’s ATPase lead to an increased interaction of RIG-I with endogenous RNA , including double-stranded RNA expansion segments of the human large ribosomal subunit . Our results suggest that RIG-I’s ATPase confers specificity to viral RNA by preventing signaling through the abundant background of self-RNA and provide a molecular framework for understanding the pathology of atypical Singleton-Merton syndrome . Recently , several autoimmune diseases , including the Aicardi-Goutières and Singleton-Merten syndromes , have been linked to RLRs through whole exome sequencing , which discovered single amino acid mutations that are mostly found within the ATPase domain of RLRs ( Jang et al . , 2015; Rice et al . , 2014; Rutsch et al . , 2015 ) . Increased interferon levels suggest that an increased activation of MDA5 or RIG-I underlies the molecular pathology of these diseases . Indeed we find that not only E373Q , consistent with recent results , leads to an increased activation of RIG-I in non-infected cells , but also the SMS mutations E373A and C268F ( Jang et al . , 2015 ) ( Figure 1C ) . While this could have been expected for E373A , because of its similarity to E373Q , the increased immunostimulatory effect of C268F in motif I comes as a surprise . Prior mutations in motif I studied by others and us led to an inactivation of RIG-I , rather than constitutive activation . The precise structural reason for the increased signaling of C268F needs to be addressed in future studies , but our co-immunoprecipitation and in vitro binding assay results suggest that this mutation may also lock RIG-I in an RNA-bound , active conformation ( Figure 2—figure supplement 1A , Figure 4—figure supplement 1C ) . Mutational and biochemical analyses previously suggested a kinetic model for RIG-I’s specificity towards viral RNA , where the ATP-dependent recycling helps to discriminate ppp-dsRNA from endogenous RNA ( Anchisi et al . , 2015; Louber et al . , 2015; Runge et al . , 2014 ) ( Figure 6A ) . Our studies show that , in case of base-paired triphosphate containing RNA ends , the RIG-I RD dominates binding . Although RIG-I’s ATPase is very active , we do not see a strong effect of ATP on the affinity for the RNA ( Figure 4C , Figure 5A ) . ATP hydrolysis may under the assayed conditions not efficiently displace RIG-I from ppp-dsRNA because RD might prevent full dissociation even after ATP-hydrolysis displaced SF2 . Importantly , ATP reduces the affinity towards self-RNA containing a duplex region but not a “proper” ppp-dsRNA end ( Figure 4D ) . Thus , if RD is unable to tether RIG-I to ppp-dsRNA ends the ATPase could rapidly remove RIG-I from RNA duplex regions via its translocase and therefore prevents an autoimmune response towards self-RNA ( Figure 6B ) . Our cellular studies are consistent with this biochemical observation , because a point mutation in K888 , a residue that is critical for recognizing ppp-dsRNA ends , did not reduce the constitutive activation of ATP hydrolysis-deficient RIG-I ( Figure 1C ) . However , RD and ATP binding are clearly important for signaling , as shown by △RD and K270I mutations by us and others ( Louber et al . , 2015 ) ( Figure 1C ) , suggesting that a ring-like , ATP-bound structure is also involved in signaling caused by self-RNA ( Figure 6C ) . In this conformation , the RD likely helps to displace the 2CARD module from the SF2 domain but may not have a high affinity for the RNA itself . Of note , the mutation in V699 of motif V also leads to increased constitutive signaling ( Figure 1C ) . A plausible explanation could be that this mutation in RecA2 decouples RNA-binding induced ATP hydrolysis from translocation or displacement of RNA . In summary , our results suggest a model where RIG-I’s translocase removes SF2 from dsRNA , perhaps at nearby bulges , unless high-affinity binding by the RD on RNA ends containing di- or triphosphates tethers RIG-I despite ATP-hydrolysis and leads to repeated or prolonged exposure of the 2CARD module . 10 . 7554/eLife . 10859 . 015Figure 6 . Proposed model for impact of ATP on RIG-I signaling on different RNAs . ( A ) RIG-I recognizes tri- or diphosphorylated double-stranded RNA and preferentially binds to the RNA end through its regulatory domain ( RD , green ) . Binding of ATP-SF2 ( purple ) to the dsRNA releases the 2CARD module ( yellow ) and activates the downstream signaling process . ATP hydrolysis displaces the SF2 domain from dsRNA leading to either rebinding at the RNA end ( tethered by RD ) or to translocation along the RNA . ( B ) In healthy cells , sustained binding of RIG-I to self-RNA containing dsRNA stretches is prevented by ATP hydrolysis . The SF2 domain can be sufficiently displaced because the RD does not provide a high affinity tether . ( C ) Mutations that allow ATP promoted binding of dsRNA and displacement of the 2CARD module , but prevent ATP hydrolysis dependent dissociation of SF2 from dsRNA , such as those underlying atypical Singleton-Merten Syndrome , will result in an unintended signaling through self-RNA . DOI: http://dx . doi . org/10 . 7554/eLife . 10859 . 015 An unexpected finding was that trapping the ATP state of RIG-I leads to a particularly increased interaction with the large ribosomal subunit via the expansion segment ES7L ( Figure 3D , F ) . This expansion segment is present in metazoan ribosomes , however its length is substantially increased in human compared to drosophila ribosomes . The function of these expansion segments is not understood , but since helix E ( ES7L-E ) was recently found to interact with the selenoprotein synthesis factor SBP2 , it is likely that the RNA in these elements is accessible to cytosolic proteins ( Kossinova et al . , 2014 ) . The specific enrichment of the large ribosomal subunit under conditions where ribosomal subunits disengage argues for rather specific interactions of RIG-I E373Q with RNA present on the large but not the small subunit . The dominant binding of ribosomes by RIG-I E373Q can be explained by the high abundance of ribosomal RNA compared to other potential RIG-I ligands in the cytosol . We could directly visualize RIG-I E373Q on the ribosome at the solvent exposed root of ES7L-A ( Figure 3F , G ) . This site contains a stretch of seven G:C/C:G base pairs , which approximately matches the footprint of dsRNA across the two SF2 RecA domains in the crystal structure of ADP·BeFx-bound RIG-I ( Jiang et al . , 2011; Kohlway et al . , 2013; Kowalinski et al . , 2011; Luo et al . , 2011 ) and also meets the requirements for activation of RIG-I’s ATPase ( Anchisi et al . , 2015 ) . Since 40% of the particles had this additional density , it is conceivable that additional binding sites could contribute to the interaction with RIG-I E737Q as well . However , the peripheral parts of the expansion segments are flexible and not visible in the 3D reconstructions , preventing us from observing RIG-I at other regions . The RNA corresponding to the observed binding region of ES7L-A is also bound by RIG-I in vitro and can moderately stimulate RIG-I’s ATPase ( Figure 4D , Figure 5A ) . The much more efficient stimulation of RIG-I’s ATPase by ppp-dsRNA is likely due to the high affinity towards RD , which could repeatedly “present” the RNA to SF2 ( i . e . increasing the “local” concentration of RNA at SF2 ) . Of note , while the addition of ATP to RIG-I reduces the interaction with the ES hairpin RNA , consistent with a role of the ATPase in preventing interaction with self-RNA , RIG-I E373Q binds with a moderately increased affinity to the ES-hairpin RNA in presence of ATP . Because of the large number of ribosomes in the cytosol it is therefore conceivable that RIG-I binds to double-stranded ribosomal RNA , including ES7L-A , under conditions where the ATPase is not able to efficiently displace the protein , such as those arising in patients with atypical SMS . In addition , the high local concentration of ribosomes in polysomes as well as a potential binding of RIG-I to other expansion segments could bring multiple RIG-I E373Q in contact , such that their exposed 2CARD module could interact for downstream signaling ( Peisley et al . , 2014; Wu et al . , 2014 ) . We do not , however , want to rule out contributions by other self-ligands as well . For instance , RIG-I can bind to endogenous mRNA ( Zhang et al . , 2013 ) or RNase-L cleavage products ( Malathi et al . , 2007 ) , while MDA5 was shown to be activated by mRNA stem loop structures under conditions where reduction of A:T base-paired RNA is not prevented by ADAR1 ( Liddicoat et al . , 2015 ) . In any case , there are two levels of control to limit RLR mediated signaling to viral RNA . On one hand , RNA editing ( Liddicoat et al . , 2015 ) and methylation ( Schuberth-Wagner et al . , 2015 ) modifies particular types of self-RNA that would otherwise form reasonable ligands for RIG-I or MDA5 . On the other hand , the intrinsic ATPase and translocase activity removes RLRs from short , but abundant endogenous dsRNA stretches , thereby reducing background signaling and increasing the sensitivity of the system . Luciferase assays and RIG-I:RNA co-immunopurifications were carried out in HEK 293T cells ( purchased from ATCC , CRL-11268 ) or HEK 293T RIG-I KO cells ( Zhu et al . , 2014 ) . HEK 293T ISRE-FF/RFP reporter cells ( stable expression of firefly luciferase and RFP under the control of an ISRE promoter , kindly provided by Luis Martinez-Sorbid , University of Rochester , Rochester , NY ) were used for interferon stimulated luciferase reporter gene assays of recovered RNA . HEK cells were maintained in high glucose Dulbecco's Modified Eagle Medium supplemented with GlutaMAX , pyruvate and 10% FBS ( all purchased from Gibco , UK ) . Human ribosomes were purified from HeLa S3 cells cultured in SMEM ( Sigma , Germany ) supplemented with 10% FBS , Penicillin ( 100 U/mL ) / Streptamycin ( 100 µg/mL ) and 1x GlutaMAX ( all purchased from Gibco , UK ) using a spinner flask at 40 rpm . All cell lines were routinely checked for Mycoplasms by PCR and were , except for the HEK 293T ISRE-FF/RFP cell line , tested to be negative . Mycoplasm contaminations were suppressed using Plasmocin ( InvivoGen , France ) according to the manufacturer's protocol . Viruses used for infections were Sendai virus Cantell , Sendai virus defective interfering particles H4 ( kindly provided by Dominique Garcin , Geneva , Switzerland ) , recombinant measles virus ( MeV ) with a sequence identical to the vaccine strain Schwarz ( AF266291 . 1 . ) ( del Valle et al . , 2007; Devaux et al . , 2007 ) and recombinant MeV-Cko-ATU-Cs . MeV-Cko-ATU-Cs expresses the C Schwarz protein from an additional transcription unit ( ATU ) located between the M and the P gene , while expression of C from the P gene is abrogated . Specifically , three stop codons were introduced into the P gene for the C ORF while leaving P and V protein expression intact . Cloning was done as described previously ( Pfaller and Conzelmann , 2008; Sparrer et al . , 2012 ) . Additionally , an ATU was introduced between the P and M gene by duplicating the gene borders of the P gene . The ORF of the C ( Schwarz ) protein was cloned into that ATU and the virus rescued from cDNA using helper plasmids in 293-3-46 cells ( Radecke et al . , 1995 ) and propagated on Vero cells as described previously ( Parks et al . , 1999; Pfaller et al . , 2014 ) . Primary antibodies to human MDA5 ( AT113 ) and RIG-I ( Alme-1 ) were purchased from Enzo Life Science ( Loerrach , Germany ) . Antibodies to FLAG ( M2 ) , HA ( HA-7 ) and β-tubulin ( TUB 2 . 1 ) were obtained from Sigma-Aldrich ( Saint Luis , MO , USA ) . Secondary antibodies were supplied by GE Healthcare ( Buckinghamshire , UK ) . Sequences encoding full-length human RIG-I or MDA5 with N- or C-terminal FLAG/HA-tag were cloned into pcDNA5 FRT/TO ( Invitrogen , Carlsbad , CA , USA ) . Mutants were generated by site-directed mutagenesis with PfuUltra polymerase ( Agilent , Santa Clara , CA , USA ) . 6x106 HEK 293T or HEK 293T RIG-I KO cells were transfected with 10 µg pcDNA5 vector coding for different FLAG/HA tagged RLR proteins . Non-infected cells were harvested 24 h after transfection . Infections were carried out 6h after transfection with an MOI of 0 . 05 for measles virus or high MOI for Sendai virus and were allowed to proceed for 40 or 24 hr , respectively . Cells were harvested and incubated in Nonidet P-40 lysis buffer ( 50 mM HEPES , 150 mM KCl , 1 mM NaF , 0 . 5% NP-40 , 0 . 5 mM DTT , protease inhibitor ( Sigma , Saint Luis , MO , USA ) , pH 7 . 5 ) for 10 min on ice . Lysates were cleared by centrifugation and proteins were immunoprecipitated for 2 . 5 - –4 hr with anti-DDK magnetic beads ( OriGene , Rockville , MD , USA ) or anti-FLAG ( M2 ) bound to magnetic protein G Dynabeads ( Novex , Life Technologies , Carlsbad , CA , USA ) . Beads were washed five times with washing buffer ( 50 mM HEPES , 300 mM KCl , 0 . 05% NP-40 , 0 . 5 mM DTT , protease inhibitor , pH 7 . 5 ) and incubated with proteinase K ( Thermo Scientific , Vilnius , Lithuania ) for 30 min at 50 °C . RNA was isolated by phenol/ chloroform/ isoamyl alcohol extraction using Phase Lock Gel Heavy tubes ( 5 PRIME , Germany ) . The quality of the isolated RNA was validated on an Agilent RNA 6000 Nano chip . Immunoactivity experiments were carried out in 24-well plates seeded with 2 . 5×105 HEK 293T RIG-I KO or 2 . 5×105 HEK 293T ISRE-FF/RFP reporter cells per well using Lipofectamine 2000 ( Invitrogen , Carlsbad , CA , USA ) as transfection reagent according to the manufacturer's protocol . For downstream signaling assays HEK 293T RIG-I KO cells were co-transfected with 500 ng protein expression vector , 100 ng p125-luc , 10 ng pCMV-RL and 50 ng empty expression vector . For RIG-I E373Q/RIG-I Δ2CARD , E373Q competition assays HEK 293T RIG-I KO cells were co-transfected with 100 ng RIG-I E373Q expression vector , varying concentrations of the RIG-I Δ2CARD , E373Q expression vector , 100 ng p125-luc and 10 ng pCMV-RL . DNA concentrations were held constant by adding empty expression vector if necessary . For determination of the immunostimulatory potential of recovered RNA from co-immunoprecipitations , HEK 293T ISRE-FF/RFP cells were transfected with 250 ng RNA in Opti-MEM ( Gibco , UK ) . For RNA stimulation of cells overexpressing RIG-I E373Q 2 . 5×105 HEK 293T RIG-I KO cells were transfected with 100 ng RIG-I E373Q expression vector , 100 ng p125-luc , 10 ng pCMV-RL and 1000 ng total RNA/ rRNA or ES hairpin RNA in Opti-MEM . All cells were harvested 24 h after transfection using 200 µL PLB ( Promega , Madison , WI , USA ) and subjected to immunoactivity experiments using the Dual-Glo luciferase assay system ( Promega , Madison , WI , USA ) as previously described ( Runge et al . , 2014 ) . The luciferase activity was determined with a Berthold Luminometer in 96-well plates using 20 µL cell lysate . RIG-I and RIG-I E373Q were expressed and purified from insect cells as described previously ( Cui et al . , 2008 ) . Briefly , sequences encoding RIG-I were cloned into pFBDM vectors and transformed into E . coli DH10MultiBac cells . Bacmids were extracted for transfection into SF9 insect cells and propagated virus was used for protein expression in High Five insect cells . Seventy-two hours after infection cells were harvested and flash frozen in liquid nitrogen . RIG-I Δ2CARD was expressed in E . coli BL21 Rosetta ( DE3 ) , using pET expression vectors as described earlier ( Cui et al . , 2008 ) . All recombinant proteins were purified using metal affinity ( QIAGEN , Germany ) , heparin affinity and gel filtration chromatography ( both GE Healthcare , Buckinghamshire , UK ) . Fractions containing RIG-I were concentrated to 6 mg/mL and flash-frozen in liquid nitrogen . Thermal stability of RIG-I or RIG-I E373Q in presence or absence of ATP was analyzed by fluorescence thermal shift assays . Proteins ( 20 µM ) were incubated in 25 mM HEPES pH 7 , 150 mM NaCl , 10 mM MgCl2 , 5 mM TCEP , 5% glycerol and 5 mM ATP . After addition of SYPRO orange ( Invitrogen , Carlsbad , CA , USA , final concentration: 2 . 5x ) the fluorescence signal was detected using a gradient from 5 °C to 100 °C with 0 . 5 K/30 s and one scan each 0 . 5 K in a real-time thermal cycler ( Biorad , Germany , CFX96 touch ) using the FRET mode . SAXS experiments were conducted at the PETRA3 P12 beamline of the European Molecular Biology Laboratory/ Deutsches Elektronen-Synchrotron , Hamburg , Germany . Samples were measured in absence or presence of 5 mM ATP in size exclusion buffer ( 25 mM HEPES pH 7 , 150 mM NaCl , 5 mM MgCl2 , 5 mM β-Mercaptoethanol , 5% glycerol ) . RIG-I samples were measured at protein concentrations of 1 . 28 , 2 . 65 and 8 . 35 mg/mL and RIG-I E373Q samples with concentrations of 0 . 87 , 2 . 13 and 6 . 84 mg/mL . The respective scattering of the corresponding buffer was used for buffer subtraction . The samples did not show signs of radiation damage , which was assessed by automatic and manual comparison of consecutive exposure frames . The data was processed using PRIMUS from the ATSAS package ( Konarev et al . , 2006 ) and the radius of gyration was determined by Guinier plot [ln I ( s ) versus s2] analysis obeying the Guinier approximation for globular proteins ( s x Rg < 1 . 3 ) . HeLa S3 cells were harvested ( 2 min , 650 x g ) , washed with PBS ( Invitrogen , Carlsbad , CA , USA ) and incubated with 1 . 5x vol Buffer 1 ( 10 mM HEPES/KOH , pH 7 . 2/4 °C , 10 mM KOAc , 1 mM Mg ( OAc ) 2 and 1 mM DTT ) for 15 min on ice , followed by disruption with nitrogen pressure ( 300 psi , 30 min , 4 °C ) in a cell disruption vessel ( Parr Instrument , Moline , IL , USA ) . The cell lysate was cleared ( 10 min , 14 , 000 rpm , Eppendorf 5417R , 4 °C ) and the resulting supernatant was loaded onto a sucrose cushion ( Buffer 1 supplemented with 35% sucrose ) . Subsequent spinning ( 98 min , 75 . 000 rpm , TLA 120 . 2 , 4 °C ) was performed . After resuspension of the ribosomal pellet , a high-salt purification by centrifugation through a 500 mM sucrose cushion ( 50 mM Tris/HCl , pH 7 . 0/4 °C , 500 mM KOAc , 25 mM Mg ( OAc ) 2 , 5 mM β-mercaptoethanol , 1 M sucrose , 1 µg/mL cycloheximide and 0 . 1% Nikkol ) was conducted ( 45 min , 100 , 000 rpm , TLA120 . 2 , 4 °C ) . The ribosomal pellet was resuspended in Ribosome Buffer ( 50 mM Tris/HCl , pH 7 . 0/4 °C , 100 mM KOAc , 6 mM Mg ( OAc ) 2 , 1 mM DTT , 1/200 EDTA-free Complete protease inhibitor ( Roche , Germany ) , 0 . 2 U/mL RNasin ( Promega , Madison , WI , USA ) ) , quickly centrifuged , frozen in liquid nitrogen and stored at -80 °C . For total RNA isolation 2 . 5 x 105 HEK 293T were seeded per well of 24 well plates . After 24 h cells were harvested in PBS , collected by centrifugation and lysed in Nonidet P-40 lysis buffer for 10 min on ice . Supernatant was cleared by centrifugation and DNA was digested with TURBO DNase ( Ambion , Life Technologies , Carlsbad , CA , USA ) for 3 min at 37 °C . Proteins were digested and RNA was extracted as described above . For ribosomal RNA isolation purified human ribosomes were proteinase K digested and RNA was extracted accordingly . Human 80S ribosomes were incubated with or without 2 . 5x molar excess of RIG-I or RIG-I E373Q in binding buffer ( 50 mM HEPES/KOH , pH 7 . 5/ 4 °C , 100 mM KCl , 2 . 5 mM Mg ( OAc ) 2 , 2 mM DTT , 1 mM ATP , 0 . 1% DDM , 10% Glycerol ) for 15 min at room temperature and then for 15 min at 4 °C . The mixture was loaded onto a sucrose cushion ( binding buffer with 750 mM sucrose ) and spun ( 3 h , 40 , 000 rpm , SW55Ti , 4 °C ) . Supernatant and pellet fractions were separated and TCA precipitated . The resulting samples were analyzed by SDS-PAGE and visualized using SYPRO Orange Staining ( Molecular Probes , Eugene , OR , USA ) . 5 OD/mL human 80S ribosomes were incubated with or without 2 . 5x molar excess of RIG-I E373Q . Each sample ( 50 mM HEPES / KOH , pH 7 . 5 / 4 °C , 100 mM KCl , 2 . 5 mM Mg ( OAc ) 2 , 2 mM DTT , 1 mM ATP , 0 . 1% DDM , 5% glycerol ) was applied to 2 nm pre-coated Quantifoil R3/3 holey carbon supported grids and vitrified using a Vitrobot Mark IV ( FEI Company , Germany ) . Data were collected on a 120 keV TECNAI SPIRIT cryo-electron microscope with a pixel size of 2 . 85 Å/pixel at a defocus range between 1 . 4 µm and 4 . 6 µm ( with RIG-I E373Q ligand ) or between 1 . 8 µm and 5 . 3 µm ( without ligand ) under low dose conditions . Particles were detected with SIGNATURE ( Chen and Grigorieff , 2007 ) . Initial alignment resulted in 61 , 067 particles ( with ligand ) and 29 , 959 particles ( without ligand ) . Subsequent data processing and single particle analysis was performed using the SPIDER software package ( Frank et al . , 1996 ) . Non-ribosomal particles ( 19 , 080 particles , 31% ( with ligand ) and 10 , 663 particles , 35% ( without ligand ) ) were removed from each data set by unsupervised 3D sorting ( Loerke et al . , 2010 ) . The remaining particles were further sorted , resulting in a volume with additional density ( with ligand: 23 , 715 particles , 39% ) . The identical sorting scheme was applied to the control 80S ribosome without ligand , resulting in final 11 , 727 particles ( 39% ) . The final 80S structures with and without ligand were refined to an overall resolution ( FCS0 . 5 ) of 17 . 7 Å and 21 . 9 Å , respectively . For comparison of the two final volumes , a statistical difference map between the two reconstructions was calculated . We used the crystal structure of the human RIG-I protein ( PDB code 3TMI ) ( Jiang et al . , 2011 ) and the human ribosome ( PDB 4V6X ) ( Anger , et al . , 2013 ) for rigid-body fitting into the additional density . Figures depicting atomic models with and without density were prepared using UCSF Chimera ( Pettersen et al . , 2004 ) . ATP binding was determined by DRaCALA using [α-32P]ATP ( Hartmann Analytik , Germany ) . 12 µM RIG-I or RIG-I E373Q were incubated in 50 mM HEPES , pH 7 . 5 , 150 mM KCl , 5 mM MgCl2 , 2 . 5 mM TCEP , 0 . 1 mg/mL BSA supplemented with 2 . 5 nM [α-32P]ATP for 10 min at room temperature in presence or absence of 100 nM RNA . 2 . 5 µL of reaction mixture was spotted on nitrocellulose membranes ( 0 . 22 µM pores , GE Healthcare , Buckinghamshire , UK ) , air-dried and [α-32P]ATP was detected using a phosphor-imaging system ( GE Healthcare , Germany ) . Proteins at different concentrations were pre-incubated with ATP , ADP or ADP·BeF3 ( all 3 mM end concentration , ADP·BeF3 was generated using ADP , NaF and BeCl2 in a 1:1:5 molar ratio ) and added to 0 . 5 µM ES hairpin RNA or 0 . 2 µM 24mer RNA in EMSA buffer ( 50 mM Tris pH 7 . 5 , 50 mM KCl , 5 mM MgCl2 , 5 mM TCEP , 7 . 5 µM ZnCl2 , 3 mM ATP , 5% glycerol ) . Reactions were incubated for 20 min at 37 °C . Samples were separated on TB agarose gels ( 89 mM Tris , 89 mM boric acid , 0 . 8% agarose ) and stained with Gel-Red ( Biotium , Hayward , CA , USA ) . Unbound RNA bands were quantified with ImageJ . Different RIG-I or RIG-I E373Q protein concentrations were titrated into EMSA buffer without ATP and glycerol . Reactions were started by addition of 5 mM ATP and 20 nM Cy3- or Cy5-labeled RNA and fluorescence anisotropy was measured with a TECAN M1000 plate reader after incubation at room temperature for 20 min . ATPase hydrolysis activity was determined using [γ-32P]ATP ( Hartmann Analytik , Germany ) . Proteins at different concentrations were pre-incubated with 100 nM RNA or purified ribosomes for 10 min at room temperature in EMSA buffer without ATP . The reaction was initiated by addition of 1 . 5 mM unlabeled and 10 nM [γ-32P]ATP and incubated for 20 min at 37 °C . Free phosphate was separated from ATP by thin layer chromatography in TLC running buffer ( 1 M formic acid , 0 . 5 M LiCl ) on polyethyleneimine cellulose TLC plates ( Sigma-Aldrich , Germany ) . [γ-32P]Pi and [γ-32P]ATP were detected using a phosphor-imaging system ( GE Healthcare , Germany ) and quantified using ImageJ .	Living cells produce long , strand-like molecules of RNA that carry the instructions needed to make proteins . Viruses also make use of RNA molecules to hijack an infected cell’s protein-production machinery and create new copies of the virus . RNA molecules from viruses have a number of features that distinguish them from a cell’s own RNAs , and human cells contain receptors called RLRs that can start an immune response whenever they detect viral RNAs . All of these receptors break down molecules of ATP , a process that releases useable energy . However , so far it is not understood how this activity helps the receptors to distinguish viral RNA from the cell’s own RNA molecules ( called self-RNA ) . Recently , some autoimmune diseases ( including Singleton-Merten Syndrome ) were linked to mutations in the parts of RLRs that allow the receptors to break down ATP . Now , Lässig et al . have studied the effects of specific mutations in an RLR called RIG-I in human cells . The experiments showed that mutations that disrupt RIG-I’s ability to bind to ATP also prevented the receptor from becoming activated . However , mutations linked to Singleton-Merten Syndrome don’t stop ATP from binding but instead slow its breakdown; this effectively locks the receptor in an ATP-bound state . Lässig et al . found that similar mutations in RIG-I caused human cells to trigger a constant immune response against the self-RNAs . Further experiments then suggested that the breakdown of ATP helps to remove RIG-I that has bound to double-stranded sections of self-RNAs . This activity frees the receptor , making it more able to detect double-stranded viral RNAs and preventing unintentional signaling . Lässig et al . also identified a specific double-stranded section of a human RNA that may be recognized by the mutated version of RIG-I in people with Singleton-Merten Syndrome . The next steps following on from this work are to extend the analysis to also include other RLRs and further explore the underlying mutations within the three-dimensional structures of the receptors and RNA molecules involved .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "immunology", "and", "inflammation" ]	2015	ATP hydrolysis by the viral RNA sensor RIG-I prevents unintentional recognition of self-RNA
In abstinent drug addicts , cues formerly associated with drug-taking experiences gain relapse-inducing potency ( ‘incubate’ ) over time . Animal models of incubation may help develop treatments to prevent relapse , but these models have ubiquitously focused on the role of conditioned stimuli ( CSs ) signaling drug delivery . Discriminative stimuli ( DSs ) are unique in that they exert stimulus-control over both drug taking and drug seeking behavior and are difficult to extinguish . For this reason , incubation of the excitatory effects of DSs that signal drug availability , not yet examined in preclinical studies , could be relevant to relapse prevention . We trained rats to self-administer cocaine ( or palatable food ) under DS control , then investigated DS-controlled incubation of craving , in the absence of drug-paired CSs . DS-controlled cocaine ( but not palatable food ) seeking incubated over 60 days of abstinence and persisted up to 300 days . Understanding the neural mechanisms of this DS-controlled incubation holds promise for drug relapse treatments . The risk of relapse is a major obstacle for effective treatment of drug addiction ( O'Brien , 2005; Wikler , 1973 ) . In abstinent drug users , several factors contribute to drug relapse , including exposure to cues and contexts previously associated with drug use ( O'Brien et al . , 1992 ) , stressors ( Sinha , 2001 ) , or acute exposure to the drug itself ( Jaffe et al . , 1989 ) . Preclinical studies have recapitulated these effects in relapse models using mice , rats , and nonhuman primates ( Venniro et al . , 2016; Weiss , 2010 ) . A major finding across these studies is that cue-induced drug-seeking ( in the absence of the drug ) increases progressively during abstinence , a phenomenon termed incubation of drug craving ( Grimm et al . , 2001; Neisewander et al . , 2000 ) . Time-dependent increases in drug-seeking have been demonstrated in cocaine ( Lu et al . , 2004a ) , heroin ( Shalev et al . , 2001 ) , methamphetamine ( Shepard et al . , 2004 ) , alcohol ( Bienkowski et al . , 2004 ) , and nicotine ( Abdolahi et al . , 2010 ) , as well as non-drug rewards such as sucrose ( Grimm et al . , 2002 ) . These findings in rodents mirror incubation of cue-induced drug craving and physiological responses in human addicts ( Bedi et al . , 2011; Li et al . , 2015a; Wang et al . , 2013; Parvaz et al . , 2016 ) , and have been important for studying neural mechanisms contributing to drug relapse ( Dong et al . , 2017; Marchant et al . , 2013; Pickens et al . , 2011; Wolf , 2016 ) . Preclinical incubation models have shown how cues presented after performance of a drug-taking response and paired with subsequent drug delivery during training potentiate drug-seeking when presented response-contingently during abstinence . These ‘confirmatory’ conditioned stimuli ( CSs ) inform the laboratory animal that the drug-taking response has been completed during training . Early preclinical studies of incubation showed that it could also occur in the absence of such discrete drug-paired CSs ( Lu et al . , 2004a; Grimm et al . , 2002 ) . This suggests that incubation could also be induced by other stimuli associated with drug-taking , such as the contextual cues ( e . g . the chamber used for operant training ) or discriminative stimuli ( DSs ) that signal drug availability ( e . g . the house-light that is illuminated during the training session , the retractable lever that serves as the operant manipulandum ) . Surprisingly , little is known about the factors underlying incubation in the absence of previously drug-paired CSs . A recent study suggested that it is not mediated by contextual cues ( Adhikary et al . , 2017 ) , leaving DSs as a likely culprit . DSs are different from cues typically investigated in these studies in that they are neither response-contingent like CSs , nor ever-present like contextual cues . Rather , DSs signal drug availability—or unavailability—thereby preceding and guiding the performance of drug-taking behavior . Previous studies have shown that a DS signaling drug availability ( DS+ ) can promote persistent drug-seeking behavior while a DS signaling drug unavailability ( DS- ) can inhibit drug-taking behavior and drug-priming-induced reinstatement of drug seeking ( Weiss , 2010; Ettenberg , 1990; Gutman et al . , 2017; Katner et al . , 1999; McFarland and Ettenberg , 1997; Mihindou et al . , 2013; Yun and Fields , 2003; Pitchers et al . , 2017 ) . Further , DS control of drug seeking persists for many months and is highly resistant to extinction ( Ciccocioppo et al . , 2004; Ghitza et al . , 2003; Martin-Fardon and Weiss , 2017 ) . Despite the importance of DSs in stimulus control of drug taking and relapse , it is unknown whether DS-controlled drug-seeking incubates during abstinence . In this study , we sought to directly assess the contribution of DSs to incubation , in the absence of drug-paired CSs . To this end , we first designed a trial-based procedure to train male and female rats to discriminatively self-administer cocaine ( 0 . 75 mg/kg/infusion ) during trials in which a DS+ signaled cocaine availability , and to suppress responding on the same lever during trials in which a DS- signaled cocaine unavailability during the same session . Drug infusions were not paired with CSs . We then tested for the ability of DSs to control cocaine seeking at multiple time points extending up to 400 days of abstinence . Further , after complete cessation of cocaine-seeking behavior , we assessed whether a priming dose of cocaine would reinstate DS-controlled cocaine seeking in the same rats . Finally , to determine whether DS-controlled incubation under our experimental conditions was specific to cocaine , we trained a separate group of rats on an analogous procedure using palatable food ( 45 mg high-carbohydrate pellets ) as the operant reward and assessed the time course of DS-controlled food seeking . The goal of Experiment 1 was to determine the persistence of non-reinforced discriminated cocaine seeking ( relapse to DS-controlled cocaine seeking ) and to test for the potentiation of this seeking response during abstinence ( incubation of DS-controlled cocaine seeking ) . We trained male and female rats to press a central retractable lever only during trials in which lever entry was preceded by the illumination of a light stimulus that signaled cocaine availability ( DS+ trials ) and to suppress responding during trials when availability of the same lever was preceded by a second light stimulus signaling absence of cocaine reward ( DS- trials ) . There were no additional reward-paired discrete cues . Once trained , we used a within-subjects design to test for discriminated cocaine seeking ( extinction conditions ) after varying durations of abstinence extending up to 400 days . After complete cessation of cocaine-seeking behavior on abstinence day 400 , we used a within-subjects design and an ascending cocaine dose-response procedure to assess the ability of priming injections of cocaine to reinstate DS-controlled cocaine-seeking . All behavioral data pertaining to Experiment 1 are shown in Figure 1 ( collapsed across sex ) , and Figure 1—figure supplement 1 ( disaggregated by sex ) . Statistical outputs for all analyses pertaining to the experiment are provided in tabular format as Figure 1—source data 1 . The goal of Experiment 2 was to determine whether the persistence and potentiation of DS-controlled seeking observed in Experiment one would generalize to a palatable food reward . We trained male and female rats to lever press for palatable food reward using a training procedure similar to that in Experiment 1 . Following training , we used a within-subjects design to test the rats for discriminated palatable food-seeking after varying durations of abstinence extending up to 200 days . All behavioral data pertaining to Experiment 2 are shown in Figure 2 ( collapsed across sex ) , and Figure 2—figure supplement 1 ( disaggregated by sex ) . Statistical outputs for all analyses pertaining to the experiment are provided in tabular format as Figure 2—source data 1 . We used a trial-based procedure to investigate incubation of cocaine or palatable-food seeking controlled by discriminative stimuli that signal availability ( DS+ ) or unavailability ( DS- ) of the rewards in the absence of reward-paired CSs . Rats readily learned to respond to the DS+ for either cocaine ( Experiment 1 ) or food ( Experiment 2 ) and to inhibit responding to the DS- within the same session . DS-controlled cocaine seeking was maximal after 60 days of abstinence ( reflecting incubation of DS-controlled cocaine seeking ) and persisted for up to 300 days . Additionally , when DS-controlled cocaine seeking was fully extinguished after 400 days of abstinence , priming injections of cocaine reinstated cocaine seeking . In contrast , DS-controlled food seeking was maximal at 1 day of abstinence , progressively decreased over time , and was no longer observed after 60 abstinence days . Thus , incubation of DS-controlled reward seeking under our experimental conditions was specific for cocaine . In previous studies , DSs paired with cocaine self-administration have been shown to promote drug seeking that is highly resistant to extinction across multiple non-reinforced test sessions ( Ciccocioppo et al . , 2004; Martin-Fardon and Weiss , 2017; Weiss et al . , 2000 ) . Reward deliveries in these studies were paired with additional discrete CSs , and the contrasting DSs were paired with different levers and presented in separate sessions , making it difficult to disentangle the potential contribution of DSs from CSs and contextual stimuli . In our experiments , reward deliveries were not paired with additional CSs during training , and the two contrasting DSs were paired with a common retractable lever and presented in a pseudorandomized order within the same session . Following training , the rats were tested under non-reinforced conditions for DS-controlled drug seeking , using the same DS presentation schedule as during training . Because the same operant manipulandum and response was required to seek reinforcement in response to each DS within the same test session , we know that discriminated drug seeking in our model was exclusively controlled by the DSs and not by contextual stimuli , classically conditioned spatial cues , presentation of the operant manipulandum , or even performance of the drug-seeking response . Under these conditions , we observed persistent non-reinforced drug seeking during DS+ presentations but not DS- presentations , up to 300 days after the last DS-drug pairing . These data extend previous studies of DS-controlled drug-seeking and suggest that in addition to setting the occasion for drug-seeking behavior , the DS+ is sufficient to motivate drug-seeking in the absence of explicit drug-paired CSs . It has long been appreciated in basic behavioral research that learning about operant DSs involves both classical and operant conditioning ( Mowrer , 1960; Rescorla and Solomon , 1967; Weiss , 1978; Weiss , 2014 ) . As explained by Rescorla and Solomon ( 1967 ) , all conditions necessary for classical conditioning are present during discriminated operant responding . Thus , in addition to setting the occasion for reward-taking it should be expected that a DS+ will come to elicit classically conditioned responses ( CRs ) . Such CRs can include the induction of motivational states ( i . e . drug craving ) . In the present study , rats learned to lever press during the DS+ for cocaine ( i . e . they learned a stimulus-response-outcome relation and their behavior came under stimulus control ) . Because they received cocaine only during the DS+ , they should have also learned a Pavlovian stimulus-outcome relation that would imbue the DS+ with incentive motivational properties ( Weiss , 2014 ) . These excitatory motivational properties , acquired through Pavlovian processes , likely contributed to the incubation effect observed in this study . We demonstrate that DS-controlled cocaine seeking is potentiated during abstinence ( that is , we show incubation of DS-controlled cocaine craving ) even in the absence of explicit drug-paired CSs . Incubation studies have typically employed between-subjects testing procedures in which rats previously trained to self-administer an addictive drug are returned to the same chambers after varying periods of abstinence and tested for drug-seeking with or without the previously drug-paired CSs ( Pickens et al . , 2011; Li et al . , 2015b; Lu et al . , 2004b ) . Following cocaine self-administration , the response to cocaine-paired CSs progressively increased ( ‘incubated’ ) over the first 60–90 days of withdrawal ( Grimm et al . , 2001; Grimm et al . , 2003 ) . However , incubation has also been observed in the absence of drug-paired CSs; extinction responding in the absence of the drug-paired CSs also progressively increased for up to 90 days ( Grimm et al . , 2003 ) and persisted up to 180 days ( Lu et al . , 2004a ) . A recent study demonstrated that this potentiation is not mediated by contextual cues ( Adhikary et al . , 2017 ) . However , the factors controlling incubation in the absence of previously drug-paired CSs were not elucidated . In the present study , we found a time-dependent increase in drug seeking ( incubation ) during DS+ presentations in the absence of any explicit drug-paired CSs during abstinence . DS-controlled seeking continued to increase up to 60 days into abstinence and persist up to 300 days ( about half the lifespan of a rat ) . This time course of DS-controlled cocaine-seeking is especially remarkable when considering that the same group of rats were exposed to repeated relapse tests under extinction conditions . It is possible that in a between-subjects design , DS-controlled incubation would show a longer rise phase than the one observed here and persist beyond 300 days , in the absence of extinction learning over repeated relapse tests . In contrast , cocaine seeking in DS- trials did not incubate – rats continued to suppress responding in DS- trials during all relapse tests and maintained discrimination up to 300 days of abstinence . DSs signaling cocaine unavailability have been shown to inhibit ongoing cocaine self-administration and to suppress cocaine priming-induced reinstatement ( Mihindou et al . , 2013 ) . From the perspective of translation and treatment development , the inhibition of cocaine seeking may be just as important as its potentiation . The behavior guided by each DS in our study was able to survive multiple extinction sessions and subsequent tests of cocaine priming-induced reinstatement – after DS+ responding was extinguished to DS- levels – priming injections of cocaine reinstated cocaine seeking specifically during DS+ , but not DS- , trials . Future studies with this procedure will dissociate the neurobiological mechanisms that allow these two functionally orthogonal DSs to mediate incubation of DS-controlled cocaine seeking . Using a similar format of DS and lever presentation , we also trained rats to lever press for palatable food reward during DS+ , but not DS- , trials . We found that food-DS rats made more total responses than cocaine-DS rats during training , maintained their discrimination responding under non-reinforced conditions , and also showed higher seeking responses than cocaine-DS rats during the initial relapse test on day 1 . However , under the same repeated-testing schedule used for cocaine relapse , they quickly extinguished their DS-controlled responding in the absence of food and progressively decreased food seeking during abstinence . It is possible that DS-controlled food seeking would have incubated in the absence of repeated relapse testing in extinction . Indeed , incubation has been observed using the classical procedure with oral sucrose reward; sucrose reward-seeking during presentation of the previously reward-paired CSs ( cue-induced reinstatement ) progressively increased during abstinence , peaked at 30 days and abated at 90 days of abstinence ( Lu et al . , 2004b; Grimm et al . , 2003 ) . It is unlikely that the differences in non-drug seeking in response to CSs and DSs are the result of the choice of non-drug reinforcer as a recent study demonstrated incubation of CS-induced reward seeking using the same palatable food pellets ( Krasnova et al . , 2014 ) . The greater persistence of seeking in response to drug- over food-DSs observed here is more likely a result of an inherent difference in the strength of stimulus-control exerted by drug- over food-DSs . Our findings are in agreement with earlier studies directly comparing drug and food paired-DSs ( Ciccocioppo et al . , 2004; Martin-Fardon and Weiss , 2017 ) but also more broadly , with studies comparing drug- and food-paired CSs ( Tunstall and Kearns , 2016; N . Kearns et al . , 2011 ) . Future studies are required to determine whether this divergence of DS effects on drug versus food seeking is due to differences in the strength of the initial DS-reward associations during training or due to drug-specific neuroadaptations that emerge during abstinence ( Wolf , 2016; Grimm et al . , 2003; Shaham and Hope , 2005 ) . Taken together , the results of the present experiments show that DS-controlled operant drug seeking incubates during prolonged abstinence and persists up to 300 days of abstinence despite repeated relapse testing . However , using a similar repeated testing procedure , we observed an abatement of DS-controlled palatable food seeking . As we noted above , DS-controlled behaviors offer an especially promising path to treatment development because DSs are always present before and during human drug taking; they do not merely accompany or follow it . They can play a critical role in relapse; for example , a study measuring flight attendants’ cigarette craving showed that craving peaked toward the end of flights as the opportunity to smoke a cigarette neared , regardless of flight duration or time since the last cigarette ( Dar et al . , 2010 ) . Animal models of other aspects of addictive behavior have been questioned , by us and other authors , because the timing or sequencing of events does not reflect the typical experiences of human drug users ( Epstein and Kowalczyk , 2018; Vanderschuren et al . , 2017 ) . The procedure we describe here addresses those concerns in the realm of cue reactivity and its incubation , and is well suited to disentangle the complex array of behavioral and neural mechanisms underlying the contributions of DSs to relapse ( Bradfield and Balleine , 2013; Colwill and Rescorla , 1990; Rescorla , 1990; de Wit and Dickinson , 2009 ) . The goal of this study was to test for the ability of discriminative stimuli signaling cocaine availability to potentiate cocaine-seeking after withdrawal and then determine if this effect would generalize to non-drug rewards . A detailed description of experimental subjects , apparatus and procedures is included in the following subsection . We first provide an overview of the specific behavioral experiments . We used male ( n = 16 ) and female ( n = 16 ) Sprague-Dawley rats ( Charles River , USA; RRID: RGD_70508 ) , weighing 250–350 g prior to surgery and training . In experiment 1 with cocaine self-administration training , we pair-housed rats of the same sex for 1 week ( n = 8 each male and female ) prior to surgery and individually housed them after intravenous surgery , during training and abstinence phases . In Experiment 2 with food self-administration training , we pair-housed rats of the same sex for 1 week ( n = 8 each male and female ) prior to the start of behavioral training and individually housed them during training and abstinence . For both experiments , we maintained the rats in the animal facility under a reverse 12:12 hr light/dark cycle with free access to standard laboratory chow and water in their home cages throughout the experiment . All procedures followed the guidelines outlined in the Guide for the Care and Use of Laboratory Animals ( 8th edition; http://grants . nih . gov/grants/olaw/Guide-for-the-Care-and-Use-of-Laboratory-Animals . pdf ) . In Experiment 1 , 14 rats successfully completed discrimination training . We excluded one female rat due to catheter patency failure and one male rat due to failure to acquire drug self-administration . Two male rats and one female rat died during the abstinence period . In Experiment 2 , all 16 rats successfully completed discrimination training . One male rat died during the abstinence period . For both experiments , we used maximum-likelihood-based multilevel models ( SAS Proc Mixed ) to account for missing data . We received 100 mg/ml cocaine-HCl ( cocaine ) diluted in sterile saline from the NIDA pharmacy . We chose a unit dose of 0 . 75 mg/kg per infusion for self-administration training based on previous studies ( Koya et al . , 2009 ) and maintained the same unit dose during discrimination training . For Experiment 1 , we implanted the rats with silastic catheters in their right jugular vein using previously described methods ( Adhikary et al . , 2017 ) . We anesthetized the rats with isofluorane gas ( 5% induction , 1–3% maintenance ) and inserted silastic catheters into the jugular vein . We passed the catheters subcutaneously to the mid-scapular region and attached them to modified 22-gauge cannulae ( PlasticsOne , USA ) cemented in polypropylene mesh ( Small Parts Inc , USA ) placed under the skin . We administered ketoprofen ( 2 . 5 mg/kg , subcutaneous injection; Henry Schein Inc , USA ) after surgery to relieve pain and allowed rats to recover for 5–7 days prior to drug self-administration training . We flushed the catheters daily with sterile saline containing gentamicin ( 4 . 25 mg/ml; Fresenius Kabi , USA ) during the recovery and training phases . We weighed rats prior to each daily behavioral session , over the course of each experiment . Subject body weights for each experiment ( disaggregated by sex ) are shown as figure supplements linked to the main figures . We trained and tested all rats in standard Med Associates ( Med Associates Inc , USA ) self-administration chambers ( Med Associates ENV-007 ) enclosed in a ventilated , sound-attenuating cabinet with blacked out windows . Each chamber was equipped with a stainless steel grid floor and two side-walls , each with three modular operant panels . For Experiment 1 , we equipped the right-side wall of the chamber with a single retractable lever in the center panel , 7 . 5 cm above the grid floor . We positioned a discriminative stimulus ( light , Med Associates ENV-221M ) that signaled cocaine availability on the left panel and another discriminative stimulus ( light , Med Associates ENV-221M ) that signaled unavailability of cocaine on the right panel of the same side wall , equidistant from the central retractable lever and 14 . 0 cm above the grid floor . In addition to location , we used red or white lens caps to differentiate between the two discriminative cues and counterbalanced them across the 14 boxes used for Experiment 1 . We connected the rat’s catheter to a liquid swivel ( Instech Laboratories Inc , USA ) via polyethylene-50 tubing that was protected by a metal spring and used a 20-ml syringe driven by a single speed syringe pump ( Med Associates PHM-100 , 3 . 33 RPM ) placed outside the sound-attenuating cabinet to deliver intravenous cocaine infusions . In Experiment 2 , we used eight different self-administration chambers . We equipped the left side wall of these chamber with a single retractable lever in the center panel , 7 . 5 cm above the grid floor . We positioned a discriminative stimulus ( light , Med Associates ENV-221M ) that signaled availability of palatable food reward on the right panel and another discriminative stimulus ( light , Med Associates ENV-221M ) that signaled unavailability of food reward on the left panel of the same side-wall , equidistant from the central retractable lever and 14 . 0 cm above the grid floor . We again used red or white lens caps to differentiate between the two discriminative cues and counterbalanced them across the boxes used for this experiment . We equipped the central panel of the opposite ( right ) wall with a pellet receptacle ( Med Associates ENV-200R2M-6 . 0 ) connected to a 45-mg pellet dispenser ( Med Associates ENV-203–45 ) to deliver palatable food-reward . Experimental timelines for each experiment are shown in Figures 1A and 2A . The self-administration , trial and discrimination training phases for cocaine and food experiments are described separately below . The subsequent abstinence and relapse test phases are the same for both experiments and described together . As described earlier , not all rats completed all phases of the experiments . In Experiment 1 , two of the 16 rats failed to complete discrimination training and were excluded from the study . Three of the 14 remaining test rats died during abstinence and did not complete all relapse tests . In Experiment 2 , all 16 rats completed discrimination training and were tested repeatedly for relapse during abstinence . One test rat died during abstinence and did not complete all relapse tests . Therefore , we used maximum-likelihood-based multilevel models ( SAS Proc Mixed ) rather than ordinary-least-squares repeated-measures analyses of variance to account for missing data . Both approaches achieve the same objectives , but maximum-likelihood models obviate imputation of missing data and permit more accurate modeling of nonhomogeneity of variance across unevenly spaced time points . We conducted all statistical analysis on two behavioral measures - ( O'Brien , 2005 ) the total number of trials of each DS type with at least one lever press ( denoted as ‘successful’ trials ) and ( Wikler , 1973 ) the total number of responses made during each DS trial type over the entire session ( denoted as lever presses ) . We followed up on statistically significant main effects or interactions with post-hoc tests as described below . Because some of our models yielded multiple main effects and interactions , we report only those that are critical for data interpretation . In preliminary analyses controlling for sex , we saw sex differences in the acquisition of discrimination , but not in the effects of interest ( e . g . the intensity of potentiated seeking or time course of incubation of DS-controlled responding ) . Therefore , we collapsed our analyses across sex for both experiments . Sex-disaggregated data for all phases of each experiment are provided as figure supplements linked to the main figures ( Figure 1—figure supplement 1 and Figure 2—figure supplement 1 ) and statistical output for all analyses , including those controlling for sex are provided as associated source data files ( Figure 1—source data 1 and Figure 2—source data 1 ) . In Experiment 1 , for the analysis of discrimination ( Figure 1C , n = 14 ) , we used a two-way factorial model with within-subject factors of discrimination training session ( sessions 5–14 ) and DS type ( DS+ , DS- ) , accompanied by Tukey's Honest Significant Difference ( HSD ) test where appropriate for pairwise comparisons between DS+ and DS- for each training session . For the repeated relapse tests ( Figure 1D , n = 11–14 ) , we used mixed two-way factorial models with the within-subjects factors duration of forced abstinence ( 1 , 21 , 60 , 120 , 200 , 300 , and 400 days ) and DS type ( DS+ , DS- ) , followed by Dunnett’s test for pairwise comparisons between the day 1 relapse test and each of the following days’ relapse tests . We also used Tukey's HSD for pairwise comparisons between DS+ and DS- for each relapse-test day . For the tests of priming-induced reinstatement ( Figure 1E , n = 11 ) , we used 2-way models with the within-subjects factors cocaine priming dose ( 0 , 10 , 0 , 20 mg/kg ) and DS type ( DS+ , DS- ) . We used Tukey's HSD for pairwise comparisons between different cocaine priming doses within each DS trial type . We also used Tukey's HSD for pairwise comparisons between DS+ and DS- for each reinstatement-test day . In Experiment 2 , for the analysis of discrimination ( Figure 2C , n = 16 ) , we used a two-way factorial model with within-subject factors of discrimination training session ( sessions 3–13 ) and DS type ( DS+ , DS- ) , accompanied by Tukey's HSD test for pairwise comparisons between DS+ and DS- for each training session . For the repeated relapse tests ( Figure 2D , n = 15–16 ) , we used two-way factorial models with the within-subjects factors duration of forced abstinence ( 1 , 21 , 60 , 120 , and 200 days ) and DS type ( DS+ , DS- ) , followed by Dunnett’s tests for pairwise comparisons between the day 1 relapse test and each of the following days’ relapse testing . We also used Tukey's HSD for pairwise comparisons between DS+ and DS- on each relapse-test day . In all models , we used a spatial-power error structure to account for autocorrelation across unevenly spaced intervals; this is similar to the use of a Huynh-Feldt or Greenhouse-Geisser correction in a repeated-measures ANOVA . Alpha ( significance ) level was set at 0 . 05 , two-tailed .	More than 85% of people who give up an addictive drug begin using it again within a year . Relapse rates have changed little over the past five decades . Situations , places and objects associated with drug-taking can trigger relapse long after a person’s last exposure to a drug . We can study relapse by training animals to self-administer drugs such as cocaine . For example , rats can learn to press a lever to receive an infusion of a drug paired with a cue ( a conditioned stimulus ) , such as a tone or a light . After training , the rats continue to press the lever to seek the drug , even if this behavior no longer delivers it . In addition , their lever pressing in response to cues increases with time for several months after their last drug exposure . This phenomenon , known as incubation of drug craving , mirrors the increase in cravings reported by abstinent drug users . In drug users , cues such as the crack pipe or syringe used to take the drug can later contribute to drug relapse during abstinence . Most studies modeling this phenomenon have focused on how the rats respond to a conditioned stimulus that signaled the delivery of a drug during training . However , a second type of signal , known as the discriminative stimulus , can also influence relapse . Discriminative stimuli are sets of cues that signal whether drugs are about to become available or not; for example , the presence of people selling drugs on a street corner as opposed to the presence of police . Madangopal et al . now show that discriminative stimuli – in the absence of conditioned stimuli – can also control the incubation of drug craving . Rats learned to press a lever in response to a light signaling the availability of cocaine ( the positive discriminative stimulus ) , and to avoid responding to a different light indicating that cocaine was unavailable ( the negative discriminative stimulus ) . When tested during abstinence , the rats only increased their lever pressing to the first light over time , i . e . , they showed an incubation of drug craving controlled by the positive discriminative stimulus . Lever pressing peaked after 60 days of abstinence and persisted for up to 300 days ( almost half the rats' lifespan ) . By contrast , the same discriminative stimuli did not trigger increased lever pressing when used to signal the availability of a palatable food . Discriminative stimuli are thus powerful and persistent triggers of craving for addictive drugs . They signal the availability of a drug prior to both drug-taking and relapse , making them a critical target for intervention strategies . Understanding the mechanisms by which discriminative stimuli promote drug craving could lead to new treatments to prevent relapse .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2019	Discriminative stimuli are sufficient for incubation of cocaine craving
Lecithin:cholesterol acyltransferase ( LCAT ) and LCAT-activating compounds are being investigated as treatments for coronary heart disease ( CHD ) and familial LCAT deficiency ( FLD ) . Herein we report the crystal structure of human LCAT in complex with a potent piperidinylpyrazolopyridine activator and an acyl intermediate-like inhibitor , revealing LCAT in an active conformation . Unlike other LCAT activators , the piperidinylpyrazolopyridine activator binds exclusively to the membrane-binding domain ( MBD ) . Functional studies indicate that the compound does not modulate the affinity of LCAT for HDL , but instead stabilizes residues in the MBD and facilitates channeling of substrates into the active site . By demonstrating that these activators increase the activity of an FLD variant , we show that compounds targeting the MBD have therapeutic potential . Our data better define the substrate binding site of LCAT and pave the way for rational design of LCAT agonists and improved biotherapeutics for augmenting or restoring reverse cholesterol transport in CHD and FLD patients . Coronary heart disease ( CHD ) is the leading cause of death in the world and typically develops as the result of atherosclerotic plaque build-up in the arteries . Risk for CHD is inversely related to high-density lipoprotein ( HDL ) cholesterol ( HDL-C ) levels in plasma . In reverse cholesterol transport ( RCT ) , HDL receives cholesterol from cholesterol-enriched macrophages , which is then esterified by lecithin:cholesterol acyltransferase ( LCAT ) bound to HDL . LCAT preferentially catalyzes the transfer of the sn-2 acyl group from phosphatidylcholine ( lecithin ) to cholesterol , creating a cholesteryl ester ( CE ) that partitions to the hydrophobic core of the HDL particle ( Calabresi et al . , 2012 ) . This process drives the maturation of discoidal pre-β HDL to spherical α-HDL and promotes further cholesterol efflux from arterial plaques ( Glomset , 1968 ) . LCAT esterification of cholesterol in HDL is promoted by ApoA-I , the most abundant structural apolipoprotein in HDL ( Fielding et al . , 1972; Jonas , 2000 ) . The structural determinants that underlie ApoA-I activation of LCAT are poorly understood , but clues have been provided by a series of crystal structures of LCAT ( Gunawardane et al . , 2016; Manthei et al . , 2017; Piper et al . , 2015 ) and the closely-related lysosomal phospholipase A2 ( LPLA2 ) ( Glukhova et al . , 2015 ) . Both enzymes contain an α/β-hydrolase domain and two accessory domains referred to as the membrane-binding domain ( MBD ) and cap domain . The MBD contains hydrophobic residues important for LPLA2 to bind liposomes and for LCAT to bind HDLs . Protruding from the cap domain is an active site lid that has been observed in multiple conformations . In the case of LCAT , crystallographic and hydrogen/deuterium exchange mass spectrometry ( HDX MS ) studies suggest that the lid blocks the active site in its inactive state , and opens in response to the binding of substrates and , presumably , upon interaction with HDL ( Manthei et al . , 2017 ) . The lid region is also important for HDL-binding ( Cooke et al . , 2018; Glukhova et al . , 2015; Manthei et al . , 2017 ) , and thus we hypothesize that activation imposed by ApoA-I involves conformational changes in LCAT that stabilize its lid in an open state that is more competent to bind substrates . To date , over 90 genetic mutations in LCAT have been described and are responsible for two phenotypes of LCAT deficiency: fish-eye disease ( FED ) , wherein patients retain residual LCAT activity , particularly on apoB-containing lipoproteins , and familial LCAT deficiency ( FLD ) , wherein patients exhibit a total loss of LCAT activity ( Kuivenhoven et al . , 1997; Rousset et al . , 2009 ) . Both are characterized by low levels of HDL-C and corneal opacities , but FLD presents additional serious symptoms including anemia , proteinuria , and progressive renal disease , the main cause of morbidity and mortality in these patients ( Ahsan et al . , 2014; Ossoli et al . , 2016; Rousset et al . , 2011 ) . Novel treatments for raising HDL-C largely based on inhibition of cholesteryl ester transfer protein have failed to protect against CHD in clinical trials ( Kingwell et al . , 2014; Rader , 2016 ) . Therefore , there is currently great interest in investigating alternative pathways for modulating HDL metabolism . In particular , the focus has switched from raising HDL-C to developing drugs that increase the beneficial properties of HDL , such as cholesterol efflux , which is enhanced by LCAT ( Czarnecka and Yokoyama , 1996 ) . New treatments that increase LCAT activity could therefore be beneficial for both FLD and CHD patients . Recombinant human LCAT ( rhLCAT ) , which raises HDL-C and increases cholesterol efflux , was shown to be safe in a phase I study ( Shamburek et al . , 2016b ) and is now in phase II trials for CHD ( clinicaltrials . gov , NCT02601560 , NCT03578809 ) . This same rhLCAT has also been tested in enzyme replacement therapy for one patient with FLD with encouraging results ( Shamburek et al . , 2016a ) . However , small molecule activators would be less expensive and easier to administer than a biotherapeutic . Previously , Amgen identified Compound A ( 3- ( 5- ( ethylthio ) −1 , 3 , 4-thiadiazol-2-ylthio ) pyrazine-2-carbonitrile ) ) , which binds covalently to Cys31 in the active site of LCAT and increases plasma CE and HDL-C levels in mice and hamsters ( Chen et al . , 2012; Freeman et al . , 2017; Kayser et al . , 2013 ) . Other sulfhydryl-reactive compounds based on monocyclic β-lactams have also been shown to activate LCAT ( Freeman et al . , 2017 ) . Although highlighting the promise of LCAT-activating molecules , these compounds are expected to have many off-target effects . Recently , Daiichi Sankyo reported a new class of reversible small molecule activators that have demonstrated the ability to activate LCAT isolated from human plasma ( Kobayashi et al . , 2016; Kobayashi et al . , 2015a; Kobayashi et al . , 2015b; Onoda et al . , 2015 ) , and increased HDL-C up to 1000-fold when orally administered to cynomolgus monkeys ( Onoda et al . , 2015 ) . Here we determined the structure of LCAT bound to both a Daiichi Sankyo piperidinylpyrazolopyridine activator and isopropyl dodecyl fluorophosphonate ( IDFP ) , a covalent inhibitor that mimics an acylated reaction intermediate , in which the enzyme adopts an active conformation with an open lid . The activator binds in a pocket formed exclusively by the MBD but does not influence affinity of LCAT for HDL . The lid , which contains positions mutated in FLD , undergoes a large conformational change from that observed in inactive LCAT structures . We show that variants of Arg244 within the lid recover acyltransferase activity when treated with a piperidinylpyrazolopyridine activator , highlighting the promise of compounds that target the MBD for many missense FLD variants . Our results thereby provide a better understanding of the key conformational changes that LCAT undergoes during activation , insight into how the enzyme alters its conformation in response to acyl substrates , and a rational framework for the design of new small molecule LCAT modulators . We first synthesized and confirmed the ability of three recently reported piperidinylpyrazolopyridine and piperidinylimidazopyridine LCAT activators ( Kobayashi et al . , 2015a; Onoda et al . , 2015 ) ( compounds 1–3 , Figure 1a ) to activate hydrolysis of 4-methylumbelliferyl palmitate ( MUP ) by full-length LCAT ( Figure 1b ) . All three activated LCAT greater than 2-fold , with EC50 values of 160 , 280 and 320 nM for 1 , 2 , and 3 , respectively ( Tables 1 and 2 ) . We also examined the acyltransferase activity of LCAT with dehydroergosterol ( DHE ) incorporated in peptide-based HDLs in response to compound 2 , as it has lower background fluorescence in this assay . We observed that compound 2 activates LCAT 2 . 8-fold with an EC50 of 280 nM ( Table 1 , Figure 1c ) . To gain insight into the mechanism of activation , we determined the Vmax and Km values for the DHE assay with and without 5 μM compound 2 . The Vmax increased from 22 to 37 μM DHE-ester hr−1 , whereas the Km was not significantly changed ( 11 μM vs . 6 . 6 μM with compound 2 ) ( Figure 1d ) . We next examined the ability of compound 1 to modulate HDL-binding by pre-incubating the compound with LCAT and then monitoring the kinetics of LCAT binding to ApoA-I HDLs with bio-layer interferometry ( BLI ) . There was no change in the kon , koff , or overall Kd in BLI , and thus the compounds do not appear to act by increasing LCAT affinity for HDL ( Table 3 , Figure 1e , Figure 1—figure supplement 1a ) . The activators did however increase the melting temperature ( Tm ) of LCAT ( ∆Tm values of 2 . 7–5 . 0 ˚C ) , similar to that which occurs upon reaction of LCAT with isopropyl dodecyl fluorophosphonate ( IDFP ) ( ∆Tm = 7 ˚C ) ( Manthei et al . , 2017 ) ( Figure 1f–g ) . A Kd value of 100 ± 14 nM was determined for compound 1 binding to LCAT via microscale thermophoresis ( MST ) ( Figure 1—figure supplement 1b ) . With the goal of visualizing an active conformation of LCAT , we examined the combined ability of both compound 1 and IDFP to stabilize ∆N∆C-LCAT ( residues 21–397 ) , a truncation variant that lacks the dynamic N- and C-termini of the enzyme and thus is more readily crystallized ( Glukhova et al . , 2015; Gunawardane et al . , 2016; Manthei et al . , 2017; Piper et al . , 2015 ) . The ligands had an additive effect ( ∆Tm of 12 . 7 ˚C ) , suggesting that the two ligands have distinct , non-overlapping binding sites ( Figure 1f–g ) . Because increased protein stability improves the chances of obtaining crystals , ∆N∆C-LCAT incubated with both IDFP and 1 ( ∆N∆C-IDFP·1 ) was thus subjected to crystallization trials . The combined use of these ligands was expected to trap an active conformation of LCAT . The resulting structure was determined using diffraction data to 3 . 1 Å spacings ( Figure 2 , Figure 2—figure supplement 1 , Table 4 ) . Crystals could not be obtained without both ligands . There are two protomers of ∆N∆C-IDFP·1 in the asymmetric unit with a root mean square deviation ( RMSD ) of 0 . 35 Å for all Cα atoms , indicating nearly identical conformations ( Krissinel and Henrick , 2004 ) . Density was observed for residues spanning 21–397 of chain A and 21–395 of chain B , although in both chains a portion of the lid is disordered ( 239–240 in chain A and 236–242 in chain B ) . Strong omit map density is observed for both compound 1 and portions of IDFP ( Figure 2b–d ) . Compound 1 binds in a groove formed by the MBD of each subunit , burying 380 Å2 of accessible surface area of the protein ( Pettersen et al . , 2004 ) ( Figure 2b–c ) . The bicyclic head of 1 binds in a pocket chiefly formed by the b1-b2 loop and a1 and a2 helices ( nomenclature as in LPLA2 ( Glukhova et al . , 2015 ) ) , including the Cys50-Cys74 disulfide bond ( Figure 2a–c ) . Its pyrazole ring donates and accepts a hydrogen bond with the backbone carbonyl and amide of Met49 and Tyr51 , respectively , which mandates the hydrogen to be on the 2-position of the ring ( Figure 2—figure supplement 2a , Compound 1-b ) . The C4 hydroxyl donates a hydrogen bond to the side chain of Asp63 , and the C6 carbonyl accepts a hydrogen bond from the side chain of Asn78 . The C4 trifluoromethyl group is buried against the a1 and a2 helices . Thus , although compound 1 was synthesized as a racemic mixture at the C4 position , the binding site is only compatible with the R enantiomer ( Figure 2—figure supplement 2a , Compound 1-c ) . For simplicity , in future descriptions the compound observed in the structure is still referred to as compound 1 . The stereochemical preference is consistent with previous observations that one optical enantiomer of a given activator is typically at least ten-fold more potent than the other ( Kobayashi et al . , 2015a; Kobayashi et al . , 2015b ) . The pyrazole moiety packs between the side chain of Tyr51 and the Cys50-Cys74 disulfide . The central piperidine ring of 1 forms van der Waals contacts , but also positions the terminal pyrazine ring of 1 in a hydrophobic cleft formed by the side chains of Met49 , Leu68 , Pro69 , and Leu70 ( Figure 2b ) . One edge of the pyrazine moiety also participates in crystal lattice contacts with residues in the αA-αA′ loop ( residues 111–119 ) , a region proposed to be involved in cholesterol binding ( Glukhova et al . , 2015; Manthei et al . , 2017 ) , although these lattice contacts are distinct in each chain ( Figure 2—figure supplement 3a–b ) . This contact may explain why similar crystals could not be obtained with compounds 2 and 3 , which have bulky trifluoromethyl substitutions for the pyrazine cyano group . Notably , the binding site for compound 1 is also occupied in some prior LCAT and LPLA2 crystal structures ( Figure 3a–b ) , either by a Phe-Tyr dipeptide of an inhibitory Fab fragment ( Fab1 ) ( PDB entries 4XWG , 4XX1 , 5BV7 ) ( Gunawardane et al . , 2016; Piper et al . , 2015 ) or by a HEPES molecule in structures of LPLA2 ( Glukhova et al . , 2015 ) , indicating that the MBD in the LCAT/LPLA2 family serves as a robust binding site for diverse chemical matter . Because the 4XWG and 4XX1 structures ( referred to as LCAT–Fab1 ) of LCAT adopt what seems to be an inactive conformation ( Manthei et al . , 2017; Piper et al . , 2015 ) , a general occupation of the activator binding site however seems insufficient to trigger a global conformational transition in LCAT . The strongest omit density for IDFP corresponds to its phosphonate head group , which is covalently bound to Ser181 and occupies the oxyanion hole ( Figure 2d–e ) . The density is progressively weaker beyond the phosphonate , and the alkyl chain past the C2 carbon is not observed . However , the location of IDFP in our structure and the dynamic nature of the alkyl chain is consistent with results from the LPLA2-IDFP structure ( PDB entry 4X91 ) , wherein multiple conformations of bound IDFP suggested two hydrophobic tracks likely used for binding the acyl chains of phospholipid substrates ( Glukhova et al . , 2015 ) ( Figure 2—figure supplement 3c ) . Indeed , there is a similar hydrophobic track corresponding to track A that takes a straighter path to the back of the LCAT as compared the one observed for LPLA2 , which results from the different orientations of their lids ( Figure 2—figure supplement 3b–c ) . We previously used HDX MS to show that IDFP stabilizes elements in the MBD and the lid region of LCAT ( Manthei et al . , 2017 ) . This data is in agreement with what we observe in the crystal structure of ∆N∆C-IDFP·1 , in that residues 67–72 in the MBD and residues 226–236 in the lid have markedly lower temperature factors in the structure reported here as compared to LCAT structures without IDFP ( Figure 3—figure supplement 1 ) . Reported atomic structures of LCAT include that of full-length LCAT wherein its lid extends over and shields the active site ( PDB entry 5TXF , LCAT-closed ) , LCAT in complex with inhibitory Fab1 ( LCAT–Fab1 ) , and LCAT in complex with Fab1 and a second agonistic Fab fragment ( 27C3 ) ( entry 5BV7 , 27C3–LCAT–Fab1; Figure 3c ) . In these structures , the N- and C-termini are disordered except for an N-terminal pentapeptide in the 27C3–LCAT–Fab1 structure ( containing mutations L4F/N5D ) that docks in the active site of a neighboring symmetry mate . It is unclear which of these structures , if any , represent an activated conformation of LCAT , although the LCAT–Fab1 and LCAT-closed structures are more similar to each other and likely to be inactive , whereas 27C3–LCAT–Fab1 has a more exposed active site . The conformation of the active site lid is highly variable among these three structures . The ∆N∆C-IDFP·1 structure affords a high-resolution view of LCAT in what is expected to be a fully activated conformation unobstructed by conformational changes that might be induced by Fab binding . The structure of LCAT here is most similar to that in 27C3–LCAT–Fab1 ( RMSD 0 . 70 Å for all Cα atoms ) ( Gunawardane et al . , 2016; Krissinel and Henrick , 2004 ) , including in their active site lid regions and in the relative configuration of their three domains ( Figure 3—figure supplement 1a–b ) . The active site lid can be divided into two regions , with the C-terminal portion ( residues 233–249 ) being most consistent between the two structures . Both structures contain similar disordered segments ( residues 236–242 in 27C3–LCAT–Fab1 , chain A residues 239–240 and chain B residues 236–242 in ∆N∆C-IDFP·1 ) . The N-terminal portion of the lid ( residues 225–232 ) is most variable , although it is consistent between the two unique chains of the ∆N∆C-IDFP·1 structure and , given the substrate analog , more likely to adopt a physiological conformation . Indeed , the N-terminal pentapeptide of a symmetry mate in the 27C3–LCAT–Fab1 structure would clash with Asn228 in the lid region of ∆N∆C-IDFP·1 . Regardless , such differences highlight the high plasticity of the active site , which is likely required for LCAT to accommodate its various lipidic substrates . Comparison of the structure of LCAT-closed with ∆N∆C-IDFP·1 provides a unique glimpse of how LCAT transitions from inactive to active states ( Video 1 ) . Domain motion analysis ( Hayward and Berendsen , 1998 ) reveals two hinge regions: residues 219–229 and 251–255 ( Video 2 ) . The dihedral angles between Asn228-Gln229 and Gln229-Gly230 undergo a large rotation that flips the lid region away from the active site in the ΔNΔC-IDFP·1 complex . On the other end of the lid , the α5 helix of the cap domain unwinds in the lid open state , with the dihedral angles between Pro250-Trp251 undergoing the most change ( Figure 3c–d , Video 2 ) . The lid transition is accompanied by a 4˚ change in the orientation of the adjacent cap domain relative to both the α/β-hydrolase and MBD , which remain fixed with respect to each other ( Figure 3c , Video 1 ) . Interestingly , in all reported LCAT structures the binding site for compound 1 is accessible ( with obvious exception of those in complex with Fab1 , which takes advantage of the same site ) , regardless of the orientation of the cap domain . In other words , initial HDL-binding and subsequent occupation of the active site by a ligand are most likely responsible for triggering the lid opening and rearrangement of the cap domain we observe in the structure , and not the binding of 1 . As a consequence of these conformational changes in the lid and reorientation of the cap domain , there are alterations within the active site that likely facilitate binding to substrates . In LPLA2 , two distinct tracks for the acyl chains of lipid substrates were observed ( Figure 2—figure supplement 3c ) ( Glukhova et al . , 2015 ) . Track A is furthest from the lid loop and is only solvent-accessible when the lid is retracted , and the α5 helix , including hinge residue Trp251 , unwinds and moves inwards to block this track in the closed lid conformation of LCAT-closed ( Figure 3—figure supplement 2 ) . In the lid-open structures , Lys218 moves with the cap domain away from the MBD in the activated conformation , where it would be in better position to bind the phosphate in the substrate lipid head group ( Glukhova et al . , 2015 ) ( Figure 3—figure supplement 2a ) . The structure of the ΔNΔC-IDFP·1 complex confirms the structure-activity relationships we and others have observed for the pyrazolopyridine scaffold . The hydrogen bonds formed by the pyrazole ring with the backbone carbonyl of Met49 and amide of Tyr51 ( Figure 2b–c ) indicate that 1-b is the dominant tautomerized isoform in the co-crystallized structure ( Figure 2—figure supplement 2a ) . Although the exchange of pyrazole ( 2 ) to imidazole ( 3 ) eliminates the hydrogen bond with Met49 , this resulted in only a minimal change in EC50 ( 280 and 320 nM for 2 and 3 , respectively ) and no change in the maximum response ( Tables 1 and 2 ) . However , interruption of both of these hydrogen bonds by swapping the pyrazole ( 2 ) for isoxazole ( 9 , Figure 2—figure supplement 2b ) dramatically increased the EC50 to 7 . 7 μM and decreased the response to 1 . 6-fold ( Tables 2 and 5 ) . It was previously shown that removal of the C4 hydroxyl group ( 4 , Figure 2—figure supplement 2b ) , which interacts with Asp63 in the structure , caused a ~ 6 fold drop in potency compared with 2 ( Kobayashi et al . , 2015a; Kobayashi et al . , 2015b ) . This is consistent with elimination of the hydroxyl group of 3 to give the more planar structure of 8 ( Figure 2—figure supplement 2b ) which decreased the potency to 4 . 6 μM , yet interestingly it still activated LCAT with increased efficacy of 3 . 7-fold ( Tables 2 and 5 ) . Surprisingly , although the bicyclic head of these compounds is expected to play an important role in the retention of potency , the imidazole-containing head group of 3 has no activating effect at concentrations up to 10 μM ( 6 , Figure 2—figure supplement 2b ) , perhaps due to loss of favorable interactions with Met49 . Consistent with the above data , compounds 6 , 8 , and 9 could not thermal stabilize LCAT at 10 μM in DSF , although 8 could at 100 μM ( Figure 3—figure supplement 3a ) . MST further confirmed that compound 6 was unable to bind to LCAT ( Figure 3—figure supplement 3b ) . Thus , in this series of activators , potency and efficacy are therefore highly dependent on a hydroxyl and chirality at the C4 position , as well as maintenance of a pyrazine ring system that likely assists in interactions with hydrophobic substrates . To further validate the crystal structure and better understand the mechanistic role of the MBD , we exchanged residues in the activator binding site of LCAT with their equivalents in LPLA2 , which is not stabilized by 1 or related compounds ( Figure 1g ) . The Y51S , G71I , and Y51S/G71I ( Figure 3a ) variants were thus expected to be impaired in binding . These variants exhibited similar or higher Tm values than WT LCAT , and were able to hydrolyze both the soluble substrate p-nitrophenyl butyrate ( pNPB ) and the micellar substrate MUP ( Figure 4 , Figure 4—figure supplement 1 ) , indicating an intact fold . As expected , compound 1 was far less effective at increasing the Tm of the three variants compared to WT ( Figure 4a ) . The Y51S/G71I variant also exhibited a nearly 4-fold decrease in HDL binding affinity and a reduced ability to catalyze acyl transfer ( Figure 4b–c , Figure 4—figure supplement 2 ) . These results are consistent with recent studies probing nearby positions at Trp48 ( mutated to Ala ) and Leu70 ( mutated to Ser ) ( Manthei et al . , 2017 ) or the analogous positions in LPLA2 ( Glukhova et al . , 2015 ) . Conversely , the analogous LPLA2 chimeric variants ( S33Y , I53G , and S33Y/I53G ) had lower Tm values relative to WT ( Figure 4a ) . However , these variants remained unable to be stabilized by 1 . We were unable to express and test a triple mutant expected to fully restore binding ( S33Y/I53G/L48N ) . Compound 1 did not stimulate pNPB esterase activity for any variant of LCAT ( Figure 4—figure supplement 1b ) , and in fact seemed to inhibit the activity of WT . Perturbation of the activator binding site decreased this effect . Compound 1 and related compounds activated hydrolysis in the MUP assay ( Figure 4d , Tables 1 and 2 ) . The EC50 of Y51S with 1 was 4-fold higher than WT at 0 . 59 μM , G71I had an EC50 >5 μM , and Y51S/G71I had no response at concentrations up to 10 μM 1 . We confirmed these results in a DHE acyltransferase assay with the Y51S/G71I variant , wherein the mutation failed to increase activity in the presence of compound 2 ( Figure 4c , e , Table 1 ) . These results confirm that the binding site for compound 1 in the crystal structure is responsible for the biochemical effects observed in solution . Arg244 is a position commonly mutated in LCAT genetic disease ( R244G ( McLean , 1992; Vrabec et al . , 1988 ) , R244H ( Pisciotta et al . , 2005; Sampaio et al . , 2017; Strøm et al . , 2011 ) , R244C ( Charlton-Menys et al . , 2007 ) , and R244L ( Castro-Ferreira et al . , 2018 ) ) and its side chain forms unique interactions in the observed active and inactive states of LCAT . In data obtained from patient plasma , the amount of LCAT-R244G isolated from homozygotes was ~25% of the amount from WT LCAT plasma and there was ~15% of WT LCAT activity , whereas heterozygotes of the R244G and R244H mutations had ~80% and~50% of WT LCAT activity , respectively ( Pisciotta et al . , 2005; Vrabec et al . , 1988 ) , thus supporting an important role for this residue . Arg244 is found in the lid of LCAT and interacts with the backbone carbonyls of Leu223 and Leu285 in ΔNΔC-IDFP·1 , and with the side chain of Asp335 in the lid closed state of LCAT-closed ( Figure 5a , Video 1 ) . We hypothesized that molecules targeting the MBD could restore some stability and function of mutations at Arg244 because this residue does not participate in the binding site for 1 . The LCAT-R244A and -R244H variants were purified and shown to be less stable than WT with ∆Tm values of −2 . 3 and −2 . 4 ˚C , respectively , consistent with Arg244 playing an important structural role ( Figure 5b , Figure 4—figure supplement 1a ) . Both LCAT-R244A and -R244H exhibited WT levels of pNPB activity , but 44% and 78% of WT in the MUP hydrolysis assay ( Figure 4—figure supplement 1b–c ) . In HDL binding analyses , both variants had an increased koff ( 2-fold for R244A and 3 . 5-fold for R244H ) which led to an increase in their overall Kd values ( Figure 4—figure supplement 2 , Table 3 ) . For R244H , the kon was also decreased from 0 . 091 ( WT ) to 0 . 022 μM−1 s−1 . Thus , in the context of HDL binding , the histidine mutant is less tolerated , perhaps due to steric clashes in the lid open conformation . Neither variant had substantial activity in the acyltransferase assay ( Figure 5c ) , consistent with their contribution to FLD . R244A and R244H were both stabilized by the addition of compound 1 ( ∆Tm of 6 . 0˚C and 4 . 8 ˚C , respectively , Figure 5b ) . R244A , R244H , and WT LCAT all exhibited similar EC50 values in response to 1 in the MUP esterase assay ( ~150 nM ) , with all three variants being activated about 2-fold by compounds 1–3 ( Figure 5d , Tables 1 and 2 ) . In the DHE acyltransferase assay , the EC50 values in the presence of saturating compound 2 were 0 . 28 , 0 . 76 , and ≥4 . 6 μM for WT , R244A , and R244H , respectively ( Figure 5e , Table 1 ) . At the highest concentration tested ( 10 μM compound 2 ) , the acyltransferase rate was 18 and 26 μM h−1 for R244A and R244H , respectively , both greater than WT LCAT which had a rate of 11 μM h−1 at the lowest concentration of compoud 2 examined . The activator affected HDL binding of the two Arg244 variants differently . For R244A , compound 1 decreased the kon from 0 . 069 to 0 . 017 μM−1 s−1 , which increases the Kd from 3 . 2 to 11 μM . For R244H , compound 1 enhanced binding to HDL by reducing the koff from 0 . 40 to 0 . 15 s−1 , reducing the Kd from 18 to 4 . 3 μM ( Figure 4—figure supplement 2 , Table 3 ) . Thus , piperidinylpyrazolopyridine and piperidinylimidazopyridine activators like compound 1 can partially rescue defects in the activity of LCAT-Arg244 variants . Here we have defined a novel activator binding site in the MBD of LCAT as well as the active conformation of LCAT , and have demonstrated that these activators can restore the activity of some FLD variants . However , the mechanism of activation mediated by compound 1 and its analogs is not straightforward . The activators do not alter the binding constant of WT LCAT for HDL ( Figure 1e , Table 3 ) , suggesting that they do not contribute to HDL binding despite occupying a site in the MBD . Thus , one would expect that the residues that interact with compound 1 are not involved in HDL binding , or else these compounds would act as inhibitors . However , the site is closely juxtaposed with residues that are involved in HDL binding . HDL-binding residues such as Trp48 and Leu70 are adjacent to the activator binding site ( Manthei et al . , 2017 ) , and the double mutant Y51S/G71I was 4-fold decreased in its affinity for HDLs due to a defect in the koff , and lost acyltransferase activity ( Figure 4 , Table 3 ) . A G71R variant has also been reported in LCAT genetic disease ( Hörl et al . , 2006 ) . The compounds increase activity of WT LCAT up to 3 . 7-fold , specifically by increasing the Vmax , although by acting at a site remote from the catalytic triad and IDFP binding site ( Figure 1 , Table 2 ) . The typical mechanism for acting at a distance would be allostery , wherein ligand binding induces a conformational change that alters the active site . Indeed , ΔNΔC-IDFP·1 adopts what we believe is a more active state with alterations in the active site that should promote activity . However , the MBD of LCAT does not appreciably change its orientation with respect to the hydrolase domain in any reported structure thus far , and the activator binding site seems available regardless of LCAT conformation . Moreover , the increase in Tm caused by IDFP and compound 1 is additive , not synergistic ( Figure 1f–g ) , and our previous HDX MS data suggested that IDFP alone can stabilize LCAT in an active , lid open conformation that is likely represented by the current structure ( Manthei et al . , 2017 ) . Thus , IDFP is more likely to be the driver of the observed global conformation change observed in the crystal structure of ΔNΔC-IDFP·1 . Although both ligands stabilize , they do so via independent mechanisms and compound 1 may only do so locally . Thus , we hypothesize that the activators such as compund 1 act by stabilizing the MBD and facilitating substrate entry into the active site cleft of the enzyme ( Figure 6 ) . In support of such a model , we note that the two chains of LCAT in the asymmetric unit of the ΔNΔC-IDFP·1 crystals pack to form a pseudo-symmetric homodimer utilizing an interface with many of the hydrophobic residues from the MBD including Trp48 , Leu64 , Phe67 , Leu68 , Pro69 , Leu70 and Leu117 from the αA-αA′ loop ( Figure 2—figure supplement 3a , Figure 6—figure supplement 1 ) . The interface is centered on the side chains of Leu64 and Phe67 . The pyrazine ring of the activator is prominently featured in this hydrophobic surface . This hydrophobic ring packs next to residues in the MBD well-known to be important for membrane interactions , such as the conspicuously solvent exposed Trp48 side chain ( Figure 6—figure supplement 1a ) . This same interface was also proposed by a recent molecular dynamics study exploring the ability of LCAT to dock to a model membrane in both the closed and open conformations ( Casteleijn et al . , 2018 ) . In the closed conformation , the active site lid blocks Leu64 , Phe67 , and Leu117 from being able to access membranes , though the rest of the MBD and the hydrophobic N-terminus of LCAT , which is also key for HDL binding ( Manthei et al . , 2017 ) , would still be available ( Figure 6—figure supplement 1b ) . The simulations in this study also suggested that residues such as Phe67 were involved in promoting transfer of lipids into the active site tunnel of the enzyme . Mutation of Arg244 , unlike compound 1 , clearly affects binding to HDL , and thus this residue , or the lid region in which it resides , could be a major ApoA-I binding determinant ( Figure 4—figure supplement 2 , Table 3 ) . Indeed , a recent paper identified a crosslink between LCAT and ApoA-I at nearby residue Lys240 within the lid ( Cooke et al . , 2018 ) . A better understanding of how ligands fit within the activator pocket enables rational design to create more potent and effective LCAT activators . For example , our crystal structure revealed the preferred enantiomer of bound piperidinylpyrazolopyridines , thus one could expect at least two-fold higher potency could be achieved with an enantiopure preparation . A recent patent has improved the potency of these compounds 3-fold by using an optically pure compound , as well as adding a hydroxyl to the C5 position on the bicyclic head , which our structure indicates would add a second hydrogen bond with the side chain of Asp63 ( Kobayashi et al . , 2016 ) . Furthermore , we have shown that there is potential to increase the efficacy of the compounds , because compound 8 activated 3 . 7-fold compared to the parent compounds , which activated an average of 2 . 3-fold . However , 8 had lowerpotency , and so more modulations will be required to determine if potency and efficacy can be improved simultaneously . The ability to perform rational design is important because we also demonstrated here the therapeutic potential of using small molecule activators targeting the MBD in FLD patients . We focused on mutations at Arg244 ( Castro-Ferreira et al . , 2018; Charlton-Menys et al . , 2007; McLean , 1992; Pisciotta et al . , 2005; Sampaio et al . , 2017; Strøm et al . , 2011; Vrabec et al . , 1988 ) because of its apparent role in the switch mechanism of the active site lid , but in principle any patient harboring an alternative missense mutation that does not directly perturb the hydrolase active site may also benefit from this compound series . In this sense the ability of piperidinylpyrazolopyridine LCAT activators to rescue Arg244 mutants parallels the allosteric action of ivacaftor on the G551D mutant of the cystic fibrosis transmembrane conductance regulator , although their mechanisms of action are necessarily different due to the unique structure of the MBD ( McPhail and Clancy , 2013 ) . Even a relatively small increase in activity could potentially slow or reverse the progression of renal disease in some FLD patients because FED patients with only partial LCAT activity do not develop renal disease ( Ahsan et al . , 2014 ) . Certainly , treatment with a small molecule activator would be more cost effective and easier for patients comply with than rhLCAT enzyme replacement therapy . In future experiments , it will be important to examine the utility of activators like compound 1 for other FLD variants . Lastly , because these compounds were demonstrated to effectively increase HDL-C in monkeys with normal levels of LCAT ( Kobayashi et al . , 2016; Kobayashi et al . , 2015a; Kobayashi et al . , 2015b; Onoda et al . , 2015 ) , it will be important to continue to interrogate their mechanism and determine if they also increase cholesterol efflux and promote atherosclerotic plaque regression . If so , then activation of LCAT by a small molecule approach and improving HDL function could be widely used in the primary prevention of cardiovascular disease and would likely complement our existing drugs for lowering LDL-C , such as statins and PCSK9-inhibitors . To produce protein for crystallographic screens , a stable cell line expressing ∆N∆C-LCAT was created in HEK293F cells . A codon-optimized human ∆N∆C-LCAT construct with a C-terminal 6x histidine-tag in pcDNA4 was SspI digested and transfected into HEK293F cells . Cells were selected with zeocin and grown in adherent culture on 150 mm plates in Dulbecco's Modified Eagle Medium high glucose medium with GlutaMAX and 1 mM pyruvate , supplemented with 10% fetal bovine serum ( Sigma ) , 100 U/ml penicillin , 100 μg/ml streptomycin and 50 μg mL−1 zeocin . Kifunensine was added to 5 μM once the cells were confluent to prevent complex glycosylation . Conditioned media was harvested every 5 days , purified via Ni-NTA , dialyzed against reaction buffer ( 20 mM HEPES pH 7 . 5 , 150 mM NaCl ) , and then frozen . For crystallographic trials , samples were thawed and subsequently cleaved with a 1:3 endoglycosidase H:LCAT molar ratio in reaction buffer supplemented with 100 mM NaOAc pH 5 . 2 for 2 . 5 hr at room temperature , which reduces the heterogeneous N-glycans to single N-acetylglucosamines . HEPES pH 8 was then added to 100 mM prior to re-purification via Ni-NTA to remove the glycosidase , and finally LCAT was polished via tandem Superdex 75 size exclusion chromatography ( SEC ) in reaction buffer ( 20 mM HEPES pH 7 . 5 , 150 mM NaCl ) . The identity of the stable cell line expressing ∆N∆C-LCAT was initially verified by western blot with an anti-His antibody and abundant secretion into the conditioned media , followed by structural characterization of the correct protein . Protein for biochemical analysis was made using pcDNA4 containing the codon-optimized human LCAT gene with a C-terminal 6x histidine-tag , which was transiently transfected in HEK293F cells as previously described ( Glukhova et al . , 2015 ) . The cells were grown in suspension in FreeStyle medium supplemented with 100 U mL−1 penicillin and 100 μg mL−1 streptomycin , and conditioned media was harvested 5 d later . The secreted protein was purified via Ni-NTA and dialyzed against reaction buffer . The LCAT proteins used in pNPB , MUP , and DSF experiments were further polished via Superdex 75 s to remove any background contaminating reactivity . ∆N∆C-LCAT was derivatized with isopropyl dodecyl fluorophosphonate ( IDFP ) to give ∆N∆C-IDFP in reaction buffer as previously described ( Manthei et al . , 2017 ) . ∆N∆C-IDFP at 5 mg mL−1 was incubated with 1 mM compound 1 for 30 min at room temperature in reaction buffer with 1% DMSO . Sparse matrix screens were set with a Crystal Gryphon ( Art Robbins Instruments ) . Initial crystals of ∆N∆C-IDFP·1 were obtained via sitting drop vapor diffusion from the Index HT screen . Crystals formed at 20 ˚C in a 1 μL drop with a protein to mother liquor ratio of 1:1 . The crystals were optimized to a final condition of 0 . 25 M lithium sulfate , 0 . 1 M Tris pH 8 . 5 , and 16% PEG 3350 via hanging drop vapor diffusion , and cryoprotected by moving the crystals to buffer with 0 . 2 M lithium sulfate , 0 . 1 M Tris pH 8 . 5 , and 24% PEG 3350 , and 20% glycerol . Crystals were frozen in nylon cryoloops ( Hampton ) , and the data were collected at the Advanced Photon Source ( APS ) at Argonne National Laboratories on the LS-CAT 21-ID-G ( λ = 0 . 97857 ) beam line . The data were processed and scaled with HKL-2000 ( Otwinowski and Minor , 1997 ) . The closed LCAT structure ( PDB 5TXF ) with the lid removed ( residues 226–249 ) was used as a search model in molecular replacement with PHASER ( McCoy , 2007 ) to generate initial phases . Non-crystallographic symmetry ( NCS ) restraints were applied to the two copies of LCAT per asymmetric unit during refinement in REFMAC5 ( Murshudov et al . , 2011 ) and Phenix ( Adams et al . , 2010 ) but removed during the final rounds of refinement . Reciprocal space refinement alternated with manual model building in Coot ( Emsley et al . , 2010 ) . A Ni2+ was observed coordinated by a portion of the exogenous His-tag beginning at residue 398 of chain A and aided in crystal packing . The final model was validated for stereochemical correctness with MolProbity ( Chen et al . , 2010 ) . The esterase assay was performed as previously described ( Glukhova et al . , 2015 ) at least in triplicate . pNPB was diluted to 10 mM into reaction buffer containing 10% dimethylsulfoxide . The reaction was started by addition of 40 μL 1 μM LCAT containing either 3 . 2% DMSO or 11 . 1 μM compound 1 to 10 μL of pNPB . The increase in absorbance at 400 nm was monitored on a Spectramax plate reader for 15 min . Significance was determined using a one-way analysis of variance followed by Tukey’s multiple comparisons post-test in GraphPad Prism . The lipase activity of LCAT was measured using MUP as a substrate . The assay was performed at room temperature in 0 . 1 M sodium phosphate buffer , pH 7 . 4 containing 0 . 01% Triton X-100 . 4 μL of LCAT ( 6 nM final concentration ) were dispensed into a 1536-well Greiner solid black plate . The same volume of assay buffer was dispensed into column 1 and 2 for a no-enzyme control . Then 23 nL DMSO or compounds titrated at 11-point 1:3 dilution series starting at 10 mM were transferred using a pintool . After 15 min incubation , 2 μL MUP ( 16 μM final concentration ) was added to initiate the reaction . The hydrolysis of MUP was monitored using a ViewLux plate reader ( excitation 380 nm/emission 450 nm ) for 20 min . The fluorescence signal was normalized against no-activator and no-enzyme control after subtraction of background signal ( t = 0 min ) . To plot percent activation , in each assay 100% was set at the rate of LCAT or LCAT variant without compound . The resulting data were fitted to a sigmoidal dose response curve . Tm values were determined using an Applied Biosystems QuantStudio 7 Flex qPCR machine with two replicates performed at least in triplicate . LCAT at 0 . 05 mg mL−1 was diluted into reaction buffer containing 5X Sypro Orange in a final volume of 10 μL in 384-well PCR plates . DMSO or compound 1 was added so that all reactions contained 3% DMSO . The reactions were run from 25–95˚C with a ramp rate of 0 . 03 ˚C s−1 . Tm values were determined as the derivative using Protein Thermal Shift software . Significance was determined using a one-way analysis of variance followed by Tukey’s multiple comparisons post-test in GraphPad Prism . MST was used to determine the binding affinity of the compounds to LCAT . Recombinant proteins were labeled with a fluorophore using the Monolith His-tag labeling RED-Tris-NTA 2nd Generation kit following manufacturer’s protocol . Compounds were titrated in a two-fold dilution series starting at 20 μM and incubated with the same volume of 100 nM labeled recombinant protein for 5 min at room temperature . Measurements were carried out in PBS containing 0 . 05% Tween-20 and standard capillaries using a Monolith NT . 115 instrument ( Nanotemper Technologies ) with 50% LED excitation power , 60% MST power , MST on-time of 30 s and off-time of 5 s . Kd values were calculated by fitting the thermophoresis signal at 20 s of the thermograph using MO . AffinityAnalysis software . A FortéBio Octet RED system was used to measure the binding of LCAT to ApoA-I HDLs . HDLs were prepared with 1 , 2-dipalmitoyl-sn-glycero-3-phosphocholine ( DPPC ) , 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine ( POPC ) , and 16:0 biotinyl Cap PE in a ratio of 49 . 5:49 . 5:1 ( Manthei et al . , 2017 ) . HDLs were diluted 1/20 in assay buffer ( 1X PBS pH 7 . 4 , 1 mM EDTA , 60 µM fatty acid free bovine serum albumin ) and then immobilized on streptavidin tips for 600 s , followed by a wash in assay buffer for 600 s to remove unbound HDLs . The tips were then moved to buffer containing DMSO or compound 1 and allowed to equilibrate for 120 s before a baseline was established for 30 s . The tips were then moved into LCAT protein in assay buffer ( containing DMSO or 10 µM 1 ) or buffer alone ( with DMSO or 10 µM 1 ) as a control and allowed to associate for 200 s , and then dissociated in assay buffer for 480 s . All steps were performed at 25 ˚C with shaking at 1000 rpm . LCAT was titrated from 0 . 4 to 2 . 4 µM in triplicate . However , for some data sets ( R244H , R244H + 1 , and Y51S/G71I ) , the 0 . 4 µM point was excluded due to low signal . The appropriate control of buffer containing DMSO or compound 1 was used to subtract the baseline and correct for drift using FortéBio’s Data Analysis 7 . 0 . The association and dissociation curves were fit using GraphPad Prism with a two-phase model . In order to determine Kd values , the kobs ( from association ) were determined at each concentration for the fast phase and then plotted against LCAT concentration . The slope of the line was evaluated as kon using the equation kobs = kon[LCAT]+koff and the resultant Kd = koff/kon . For statistical analysis , the kon , koff , and Kd for each replicate was determined individually and the results were compared to WT using a one-way analysis of variance followed by Tukey’s multiple comparisons post-test in GraphPad Prism . Peptide-based HDLs were used in this assay as there is no difference between peptide HDLs and ApoA-I HDLs in both HDL binding and acyltransferase assays ( Manthei et al . , 2017 ) . The peptide HDLs were made using the ESP24218 peptide with the sequence PVLDLFRELLNELLEALKQKLK ( Dassuex et al . , 1999; Li et al . , 2015 ) with a DPPC:POPC:DHE ratio of 47:47:6 as previously described ( Manthei et al . , 2017 ) . The assay was performed in 384-well low volume black microplates ( Corning 4514 ) with a total assay volume of 16 μL . In each reaction , LCAT was diluted in assay buffer to 15 μg mL−1 in the presence of either 1% DMSO or 10 μM 2 with 1% DMSO . Compound 2 was used in this assay because it has lower background fluorescence than 1 . The DHE HDLs were diluted in 1X PBS with 1 mM EDTA and 5 mM β-mercaptoethanol . 8 μL of the HDLs were added to the plate , and the reactions were initiated with 8 μL of LCAT , so that LCAT was assayed at 7 . 5 μg mL−1 with and without 5 μM compound with a range of DHE concentrations from 0 to 50 μM . The reactions were stopped after 25 min at 37 ˚C with the addition of 4 μL of stop solution ( 1X PBS with 1 mM EDTA , 5 U mL−1 cholesterol oxidase ( COx ) , and 7% Triton X100 ) . Following the addition of stop solution , the plates were incubated for another 60 min at 37 ˚C to allow for the COx to react . After the plates were re-equilibrated at room temperature , fluorescence was determined on a SpectraMax plate reader with excitation at 325 nm and emission at 425 nm , with a 420 nm cutoff . Reactions without LCAT were used for background subtraction , and reactions without LCAT and stop solution lacking COx were used to generate a standard curve for DHE . Reactions were performed in triplicate with three independent experiments per LCAT variant . Data were processed via background subtraction to remove excess fluorescence that results from the higher concentrations of DHE . These values were divided by the slope of the line from the standard curve , which yields the amount of DHE-ester that resulted in each well , and then by time to determine the rate . Outliers were removed using automatic outlier elimination within Prism . For statistical analysis , the Vmax for each variant was compared to WT using a one-way analysis of variance followed by Tukey’s multiple comparisons post-test in GraphPad Prism . To determine EC50 values , compound 2 was titrated from 0 . 004 to 10 μM , and the DHE concentration was set at 50 μM . LCAT was diluted in assay buffer and compound 2 dilutions were made with assay buffer containing 5 . 3% DMSO . 1 . 5 μL compound was added , then 6 . 5 μL LCAT , followed by 8 μL DHE . Dilutions were adjusted so that LCAT was assayed at 7 . 5 μg mL−1 , as above . All values were background subtracted to buffer with the same concentration of compound 2 . A standard curve was included in one experiment with DHE from 0 to 50 μM in order to adjust the final fluorescence values to a rate by dividing by the slope of the line and time ( 25 min ) , as above . Outliers were removed using automatic outlier elimination within Prism . For statistical analysis , the EC50 for each variant was compared to WT using a one-way analysis of variance followed by Tukey’s multiple comparisons post-test in GraphPad Prism . In most cases and as indicated in the methods and figure legends , statistical analysis was performed a one-way analysis of variance followed by Tukey’s multiple comparisons post-test in GraphPad Prism . A paired t-test was used to compare the basal MUP hydrolysis levels . The statistical parameters , P value cutoffs , and number of replicates for each experiment are indicated in the table that corresponds to each experiment , the figure legends , and/or methods .	Cholesterol is a fatty substance found throughout the body that is essential to our health . However , if too much cholesterol builds up in our blood vessels , it can cause blockages that lead to heart and kidney problems . The body removes excess cholesterol by sending out high-density lipoproteins ( HDL ) that capture the fatty molecules and carry them to the liver where they are eliminated . The first step in this process requires an enzyme called LCAT , which converts cholesterol into a form that HDL particles can efficiently pack and transport . The enzyme acts by interacting with HDL particles , and chemically joining cholesterol with another compound . Finding ways to make LCAT perform better and produce more HDL could improve treatments for heart disease . This could be particularly helpful to people with genetic changes that make LCAT defective . Several small molecules that ‘dial up’ the activity of LCAT have been identified , but how they act on the enzyme is not always well understood . Manthei et al . therefore set out to determine precisely how one such small activator promotes LCAT function . The experiments involved using a method known as crystallography to look at the structure of LCAT when it is attached to the small molecule . They also evaluated the activity of the enzyme and other aspects of the protein in the presence of the small molecule and HDL particles . Taken together , the results led Manthei et al . to suggest that the small molecule works by more efficiently bringing into LCAT the materials that this enzyme needs to create the transport-ready form of cholesterol . The small molecule also partially restored the activity of mutant LCAT found in human disease . This knowledge may help to design more drug-like chemicals to ‘boost’ the activity of LCAT and prevent heart and kidney disease , especially in people who carry a defective version of the enzyme .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2018	Molecular basis for activation of lecithin:cholesterol acyltransferase by a compound that increases HDL cholesterol
Primary Ovarian Insufficiency ( POI ) affects ~1% of women under forty . Exome sequencing of two Finnish sisters with non-syndromic POI revealed a homozygous mutation in FANCM , leading to a truncated protein ( p . Gln1701* ) . FANCM is a DNA-damage response gene whose heterozygous mutations predispose to breast cancer . Compared to the mother's cells , the patients’ lymphocytes displayed higher levels of basal and mitomycin C ( MMC ) -induced chromosomal abnormalities . Their lymphoblasts were hypersensitive to MMC and MMC-induced monoubiquitination of FANCD2 was impaired . Genetic complementation of patient's cells with wild-type FANCM improved their resistance to MMC re-establishing FANCD2 monoubiquitination . FANCM was more strongly expressed in human fetal germ cells than in somatic cells . FANCM protein was preferentially expressed along the chromosomes in pachytene cells , which undergo meiotic recombination . This mutation may provoke meiotic defects leading to a depleted follicular stock , as in Fancm-/- mice . Our findings document the first Mendelian phenotype due to a biallelic FANCM mutation . Primary Ovarian Insufficiency ( POI ) affects about 1% of women under forty years . It is often diagnosed too late , thus generating infertility and significant morbidity and mortality due to steroid-deprivation associated symptoms . Infertility is usually definitive but resumption of ovarian function occurs in ~24% of cases ( Tucker et al . , 2016 ) . POI is etiologically heterogeneous ( OMIM: ODG1 # 233300 . ODG2 # 300510 ODG3 # 614324 , ODG4 # 616185 ) and remains idiopathic in ~70% of the cases , but a number of genetic variants have been identified , including mutations in meiotic and DNA repair genes ( Tucker et al . , 2016 ) . Consistently , several DNA repair and genomic instability disorders , such as Fanconi anemia ( FA ) , are known to be associated with hypogonadism , ovarian failure and/or infertility . FA is a bone marrow failure syndrome accompanied by developmental defects , predisposition to leukemia , chromosome fragility and hypersensitivity to DNA interstrand crosslinks ( ICL ) ( Bogliolo and Surrallés , 2015; Ceccaldi et al . , 2016 ) . The products of the 21 genes ( FANCA to FANCV ) whose loss-of-function has been associated to FA are subdivided into three functional groups ( Bogliolo and Surrallés , 2015; Ceccaldi et al . , 2016; Wang , 2007 ) . The first one , the FANCcore complex consisting of seven FA proteins ( A , B , C , E , F , G , and L ) and two FA-associated proteins ( FAAP20 and FAAP100 ) , is assembled in response to DNA damage and/or stalled replication forks to monoubiquitinate FANCD2 and FANCI ( the second group ) . This monoubiquitination allows the FANCD2-FANCI heterodimer to coordinate the DNA repair/replication rescue activities of the third group of FANC proteins , which includes nucleases and proteins involved in homologous recombination . The third group includes BRCA1 , BRCA2 , RAD51 , PALPB2 and BRIP1 , whose mutations predispose to breast and ovarian cancer ( BOC ) ( Bogliolo and Surrallés , 2015 ) . However , a more recent analysis reconsidered the role of some FA-associated genes in the establishment of bona fide FA clinical and cellular phenotypes and excluded FANCM from the group of FA genes ( Bogliolo and Surrallés , 2015 ) . The phenotypes associated to FANCM biallelic mutations thus far are cancer predisposition , in particular early-onset breast cancer in females , and chemosensitivity . ( Michl et al . , 2016; Bogliolo et al . , 2017; Catucci et al . , 2017 ) . Here , we have performed a whole-exome sequencing in a Finnish family with two patients presenting with non-syndromic POI and identified a homozygous truncating mutation ( c . 5101C>T; p . Gln1701* ) in FANCM , which explains POI in these patients . Both sisters with POI and their mother were studied by whole-exome sequencing , which generated ~45 millions of mapped read pairs per sample ( 91% of targeted exome at ≥10X ) . Exome capture , sequencing and data processing were performed as described in ( Fauchereau et al . , 2016 ) . Variants were filtered according to the following criteria: minimum depth of 5 reads and variant quality of 20 , with potential impact on transcript or protein , homozygosity in both affected sisters and heterozygosity in the mother and minor allele frequency ( MAF ) under 1% . The only variant fulfilling these criteria was the non-sense mutation chr14:45658326 C/T ( rs147021911 ) in exon 20 of FANCM ( Figure 1B , Tables 2–3 ) . According to the ExAC database , its MAF is 0 . 0013 , slightly higher in the Finnish population ( 0 . 0089 ) , with only one homozygous individual described in the ExAC database . The variant and its segregation in the family were verified by Sanger sequencing of the exon 20 of the FANCM ( Figure 1C ) . This variant leads to the p . Gln1701* truncation at the protein level , which removes the C-terminal endonuclease and the FA associated protein 24 ( FAAP24 ) -interaction domain ( Figure 1D ) . FAAP24 mediates chromatin association of FANCM . EBV-immortalized cells were obtained from patient 2 and the mother . Immunoblot with an anti-FANCM antibody confirmed the expression of a truncated FANCM protein of ~195 kDa in the cell line derived from the patient , whereas both the wild-type/WT ( MW ~235 kDa ) and the truncated proteins were present in the heterozygous mother ( Figure 1E ) . The expression level of the truncated protein was significantly reduced compared to the WT , suggesting that the mutation destabilizes either the protein or the corresponding mRNA ( i . e . , non-sense mediated decay ) . No variants were observed in other FANC genes . Indeed , we have exhaustively looked for the presence of variants in a list of FA genes ( Table 4 ) . The very good coverage of each gene ( 80-90X in the POI patients ) excludes the possibility of having missed a variant due to a lack of coverage . The presence of heterozygous variants , both in coding regions and elsewhere ( UTRs , the covered intronic sequences , etc . ) , in one or both POI patients , excludes the possibility of hemizygosity due to heterozygous deletions . The ratios of the number of reads of the alleles for these heterozygous variants were unbiased ( i . e . , close to 1 ) and argue against gene deletions . The detection of variants outside the coding portions of the exons also argues against the possibility of partial deletions . Exome-based CNV analysis showed that the detected CNVs were unrelated to the disease as they were either not common to the two sisters or were present in the mother . Finally , the total lack of correctly segregating variants ( i . e common to both POI patients and heterozygous in the mother ) in FA genes excludes the possibility of their involvement in the pathology . FANCM expression was studied in germ cells of human fetal and adult ovaries . Human fetal material was obtained from the Antoine Béclère Hospital ( Clamart , France ) following legally-induced abortions or therapeutic termination of pregnancy . The identification of meiotic stages was performed in histological pieces that we have previously characterized ( Frydman et al . , 2017 ) and based on the chromatin features . qRT-PCR in human fetal ovaries demonstrated that FANCM mRNAs were expressed throughout ovarian development ( 5–32 weeks post-fertilization , wpf ) ( Figure 2A ) . Of note , expression tended to be higher than average in 14 and 17 wpf ovaries that are stages containing the highest proportion of germ cells progressing into meiotic prophase I . Cell-sorting experiments conducted in 8–12 wpf ovaries indicated that FANCM transcripts were predominant in oogonial cells ( D2-40-positive ) compared to somatic cells ( Figure 2B ) . Immunohistochemical studies in human fetal ovaries show that FANCM protein was present in the nuclei of oogonia but staining was stronger in pachytene stage oocytes ( i . e . , at 8 and 14 wpf respectively in Figure 2C ) . Of note , staining localized along the chromosomes in pachytene cells that undergo meiotic recombination . FANCM was also observed in oocytes arrested at the diplotene stage of prophase I during the last trimester of pregnancy and in adults ( Figure 2C ) . The co-staining with Synaptonemal complex protein 3 ( SYCP3 ) or DEAD box protein 4 ( DDX4/VASA ) confirmed respectively the meiotic and germinal nature of the FANCM-positive cells ( Figure 2D ) . Next , we monitored chromosome breakage in primary lymphocytes from the two patients and their mother . Baseline and DNA-damage induced chromosome breakage and rearrangement were blindly scored on 50 metaphases without treatment or after exposure to Mitomycin C ( MMC ) for 72 hr . In line with a role of FANCM in the maintenance of genome stability , the occurrence of chromosome breakages and rearrangements was higher in both POI patients than in their mother ( Figure 3A and B ) . We also determined the impact of the FANCM truncating mutation on the sensitivity to Mitomycin C and on FANCD2 monoubiquitination in response to DNA interstrand crosslinks and replication inhibition . In response to MMC , primary lymphocytes from both patients had a reduced capability to monoubiquitinate FANCD2 ( Figure 3C ) . The reduced level of FANCD2 in lymphocytes from POI patients is likely due to a reduced proliferation of their cells in culture conditions . Next , we determined the growth inhibition response of the lymphoblasts from POI patient-2 and her mother in response to MMC ( Figure 3D ) , as previously described ( Ridet et al . , 1997 ) . The cells from the mother behaved like FANC-pathway proficient cells , while the response of the mutated cells was similar to that of FANCA- and FANCC-deficient lymphoblasts . Finally , we assessed FANCD2 monoubiquitination in lymphoblastoid cells in response to MMC , that arrests replication by inducing DNA lesions ( Figure 4A ) , or to two replication inhibitors , hydroxyurea ( HU ) and aphidicolin ( APH ) , that block replication forks by poisoning DNA polymerases ( Figure 4B ) . Surprisingly , whereas in response to MMC , FANCD2 monoubiquitination was clearly impaired in FANCMmut cells , those cells maintained a residual but detectable capability to monoubiquitinate FANCD2 after HU or APH treatments . Finally , consistent with a proficient DNA damage and stalled replication forks signaling in FANCM mutated cells , we failed to observe any major impairment in the MMC- phosphorylation of H2AX and CHK1 ( Figure 4A ) ( Durocher and Jackson , 2001 ) . To further validate that the identified bi-allelic mutation in FANCM was responsible for the MMC hypersensitivity observed in the patient's lymphoblasts , we transduced them with a FANCM-WT cDNA-expressing lentiviral vector ( see Materials and methods ) . Transiently genetically complemented cells recovered a significant resistance to MMC ( Figure 3D ) as well as an improved monoubiquitination of FANCD2 in response to MMC ( Figure 3E ) . The two patients with POI studied here belong to a consanguineous family and are thus homozygous for the FANCM mutation inherited from their parents . Our FANCM mRNA and protein expression studies in the developing human ovary suggest that the expression of FANCM starts in mitotic germ cells ( first trimester of pregnancy ) notably along chromosome axes and increases at the onset of meiosis ( second trimester ) that are maintained up to the diplotene stage in follicles . Given the known role of FANCM to sustain primordial germ cell proliferation in mouse ( Luo et al . , 2014 ) and its conserved function during meiotic recombination ( Lorenz et al . , 2012; Crismani et al . , 2012 ) , it is likely that both processes are sensitive to a FANCM mutation . FANCM is a nuclear partner of the FANCcore complex , belonging to the FANC pathway that promotes DNA repair and safeguards replication . Indeed , FANCM is targeted to damaged DNA and/or stalled replication forks by its partners FAAP24 , FANCM-interacting histone fold protein 1 ( MHF1 ) and 2 ( MHF2 ) , where it recruits the FANCcore complex to monoubiquitinate FANCD2 and FANCI and coordinates DNA repair/replication rescue ( Ceccaldi et al . , 2016; Michl et al . , 2016 ) allowing the faithful transmission of undamaged chromosome to the daughter cells . Consistently , the occurrence of chromosome breakages and rearrangements was higher in both POI patients than in their mother ( Figure 3A and B ) . Our biochemical studies show that lymphocytes from both patients have a reduced capability to monoubiquitinate FANCD2 in response to MMC pointing to a defect in the activation of the FANCcore complex in FANCMmut cells . However the cells maintained a detectable capability to monoubiquitinate FANCD2 in response to HU or APH treatments . A previous study links POI to other breast cancer genes such as BRCA1 ( Oktay et al . , 2010 ) . Although recent works established a link between the heterozygous c . 5101C>T FANCM truncating mutation and BOC predisposition in the Finnish population ( Kiiski et al . , 2016 , 2014; Neidhardt et al . , 2017 ) , there is no history of BOC in the family investigated here . Interestingly , a few apparently healthy ( i . e . , without cancer ) individuals homozygous for the c . 5101C>T FANCM mutation were identified in previous studies ( Kiiski et al . , 2016 , 2014; Neidhardt et al . , 2017 ) . Indeed , the authors stated , by direct observation or by looking at registered clinical data , that none of them presented FA stigmata ( although no data on age , sex or clinical findings was made available ) . This is in line with the fact that previously identified bi-allelic inactivating FANCM mutations in a few FA patients co-occurred with mutations in other FANC genes , which were indeed responsible for FA ( Chang et al . , 2014; Singh et al . , 2009 ) . This has led to the recent exclusion of FANCM from the group of the bona fide FA genes ( Bogliolo and Surrallés , 2015 ) . Very recently two studies report that the phenotypes associated to FANCM biallelic mutations thus far are cancer predisposition , in particular early-onset breast cancer in females , and chemosensitivity ( Catucci et al . , 2017; Bogliolo et al . , 2017 ) . In one study ( Catucci et al . , 2017 ) , two female patients out of five had premature menopause and one patient had an early reduction of the ovarian reserve according to her AMH levels . Our results suggest that the cells that harbor biallelic c . 5101C>T FANCM mutations maintain a FANCM residual activity allowing them to cope with the stalling of the replication forks during their normal proliferation in vivo , probably protecting individuals from the hematopoietic and developmental abnormalities that constitute the bona fide features of FA . However , the repair defects associated to the c . 5101C>T truncating mutation that we describe likely leads to meiotic defects and oocyte apoptosis . Oocytes deficient for DNA repair may accumulate double-strand DNA breaks over time , resulting in reduced oocyte viability ( Oktay et al . , 2015 ) . Along similar lines , in Fancm-/- female mice , the ovarian cortex is depleted of primary follicles and the number of developing follicles is reduced compared to WT ovaries . However , due to a bias against Fancm-/- females , their fertility was not thoroughly investigated ( Bakker et al . , 2009 ) . Antral follicles were present , albeit in reduced number , which is consistent with the observations in our patients . Some follicles might escape the DNA repair defects and achieve maturation , as epitomized by the presence of antral follicles and corpora lutea in Fancm-deficient mice and by the spontaneous resumption of ovarian function with pregnancy reported in one of our patients . In conclusion , in this report , we document the first case implicating FANCM mutations in non-syndromic POI . Our findings clearly support a genetic link between infertility and DNA-repair/cancer genes and show the necessity to perform an enhanced genetic counseling of POI patients with a long-term follow-up . The study was approved by all the institutions involved . All participants gave informed consent for the study and the study was approved by the agence de Biomedecine ( reference number PFS12-002 ) . Written informed consent was received from participants prior to inclusion in the study and the institutions involved . FSH , LH , Prolactin , TSH: Electrochemiluminescence immunoassay ECLIA , cobas kit insert Elecsys and cobas e analyzers ( 2013–10 , V 19 ) , Roche Diagnostics GmbH/Roche Diagnostics GmbH , Sandhofer Strasse 116 , D-68305 Mannheim , Germany . Equipment: cobas 8000 e 602 , Roche Diagnostics . Estradiol: radioimmunological assay , Spectra kits 125I Coated Tube Radioimmunoassay kit insert , Orion Diagnostica , Espoo , Finland . . Equipment: Gammamaster 1270 , Wallac Oy , Turku , Finland . AMH: AMH Gen II ELISA . Equipment: Beckman Coulter , Inc . 250 s . Kraemer Blvd . Brea , CA 92821 U . S . A . Blood samples were collected for hormonal and genetic studies and for the production of EBV-immortalized lymphoblastoid cell lines , which was performed at the Genopole , Evry , France , following a standard in-house protocol . EBV-immortalized cells were obtained for patient 2 and her mother . HEK293 ( Research Resource Identifier/RRID:CVCL_0045 ) cells and FANC pathway-proficient ( HSC-93 ( RRID:CVCL_G049 ) ) , and -deficient ( HSC-72 ( RRID:CVCL_G047 ) , HSC-536 ( RRID:CVCL_G045 ) and GM16756 ( RRID:CVCL_G041 ) ) cells have been previously described ( Bourseguin et al . , 2016 ) . HSC-72 , HSC-536 and GM16756 bring bi-allelic inactivating mutations in FANCA , FANCC and FANCD2 , respectively . Lymphoblastoid cells were cultured in RPMI 1640 medium , supplemented with 12% FCS and antibiotics . Wild-type FANCM cDNA-expressing lentiviral vectors were a gift of M . Bogliolo ( Dpt of Genetics and Microbiology , Universitat Autonoma de Barcelona , Spain ) . Production and titration of lentiviral particles were performed as described ( Hamelin et al . , 2006 ) . The infection was performed on retronectin-coated plates ( TaKaRa Bio , CA , USA ) and efficiency was assayed by testing GFP expression using flow cytometry . Library preparation , exome capture , sequencing and data processing were performed by IntegraGen SA ( Evry , France ) according to their in-house procedures . Data analysis was performed as described in ( Fauchereau et al . , 2016 ) . Briefly , genomic DNA libraries were prepared from 600 ng of genomic DNA from three individuals ( the mother and the two affected sisters ) with NEBNext Ultra kit ( New England Biolabs ) . Target capture , enrichment and elution were performed according to manufacturer’s instructions and protocols ( SureSelect Human All Exon Kits Version CRE , Agilent ) without modification . Libraries were sequenced on an Illumina HiSEQ 2500 as paired-end 75 bp reads . Image analysis and base calling was performed using Illumina Real Time Analysis ( RTA 1 . 18 . 64 ) with default parameters . The various metrics of the NGS sequencing are detailed in the Table 2 . The sequencing data was analyzed with the Illumina pipeline ( CASAVA 1 . 8 . 2 ) for read alignment and variant calling , and an IntegraGen in-house pipeline was used for variant annotation . Annotation of kown variants was performed according to the following databases: dbSNP144 ( RRID:SCR_002338 ) , 1000 Genomes Project ( release_v3 . 20101123 ) ( RRID:SCR_008801 ) , Exome Variant Server ( ESP6500SI-V2-SSA137 ) ( RRID:SCR_012761 ) , HapMap3 ( RRID:SCR_002846 ) , ExAC r3 . 0 ( RRID:SCR_004068 ) , COSMIC 71 ( RRID:SCR_002260 ) , ClinVar 201507 ( RRID:SCR_006169 ) and an Integragen internal database ( containing 201 control exomes for SNPs and 130 control exomes for Indels ) . The variants were filtered using IntegraGen ERISv3 platform . The number of called variants and the effect of the various variant filters on the number of remaining candidate variants are shown in Table 3 . The sequence of the exon 20 of the FANCM gene was verified by PCR amplification and Sanger sequencing from the genomic DNA of the available individuals , using the primers FANCM-Forward ( 5’-AAAACTCGACGTGCAGTAATG-3’ ) and FANCM-Reverse ( 5’-GAGGTTTGAAGTCTGAGACTT-3’ ) . Human fetal material was provided by the Department of Obstetrics and Gynecology at the Antoine Béclère Hospital ( Clamart , France ) following legally induced abortions ( first trimester ) or therapeutic termination of pregnancy ( second and third trimesters ) . The fetal age was evaluated by measuring the length of limbs and feet as previously described ( Evtouchenko et al . , 1996 ) . Adult human ovaries were provided by the Laboratory of Pathology of Gustave Roussy Institute ( Villejuif , France ) following prophylactic removal due to breast cancer . Adult ovaries did not present any pathological aspect , showing corpora lutea at various stages of regression compatible with at least three successive ovulatory cycles . All women provided an informed consent and this study was approved by the Biomedicine Agency ( reference number PFS12-002 ) . Tissue was either snap-frozen for RNA analysis of formalin-fixed for immunohistochemistry studies . Human fetal material was obtained from the Antoine Béclère Hospital ( Clamart , France ) following legally-induced abortions or therapeutic termination of pregnancy . The fetal age was evaluated as previously described ( Evtouchenko et al . , 1996 ) . All women provided an informed consent and this study was approved by the Biomedicine Agency ( reference number PFS12-002 ) . After collection , fetal ovaries were stored in RNA lysis buffer ( RLT ) ( Qiagen Courtaboeuf , France ) for gene expression profiling . For some experiments human gonads were dissociated and germ and somatic cells were sorted using M2A ( D240/PODOPLANIN ) as previously described ( Muczynski et al . , 2012 ) . The M2A antigen was first reported as being present in testicular germ cells and germ cell tumors ( Marks et al . , 1999 ) . It has recently been shown that the M2A antigen is also expressed in female germ cells in the human developing ovary ( Frydman et al . , 2017 ) . Sorting was performed using a BD-Influx biohazard system ( BD Biosciences; San Jose , CA; USA ) with the D2-40-positive cells being mostly germ cells . Total RNA from fetal ovaries and sorted cells was extracted using the RNeasy Mini Kit ( Qiagen ) . The 7900HT Fast Real-Time PCR System ( Applied Biosystems , Foster City , CA ) and SYBR-green labelling were used for quantitative RT-PCR ( RT-qPCR ) . The comparative ΔΔcycle threshold method was used to determine the relative quantities of mRNA using ACTB ( ß-actin ) mRNA as reference gene for normalization . Each RNA sample was analyzed in duplicate . The sequences of oligonucleotides used for amplification are: FANCM ForwardExp ( 5’-GAGGAGCTTGTCCCGCTG-3’ ) , ReverseExp: ( 5‘-TGACTAGTTCTCTTACAACCTGGCAATA-3’ ) ; B-ACT Forward: ( 5’-TGACCCAGATCATGTTTGAGA-3’ ) , Reverse ( 5’-TACGGCCAGAGGCGTACAGG-3’ ) . Immunohistochemistry was performed on paraffin-embedded human fetal and adult ovaries as previously described ( François et al . , 2017 ) . Briefly , human fetal ovaries were fixed in 10% neutral formalin , dehydrated , embedded in paraffin and sectioned ( 5 μm ) as described in by ( François et al . , 2017 ) . Slides were deparaffinized , rehydrated and heated to 98°C in 0 . 05% citraconic anhydride , pH 7 . 4 ( Sigma-Aldrich Corp . ) , for 45 min and then blocked for 1 hr in 10% fetal calf serum at room temperature . After washing in phosphate-buffered saline , slides were incubated overnight with anti-FANCM antibody ( CV5 . 1 , Novus biologicals , Abingdon , UK , dilution 1/500 ) ( RRID:AB_2716711 ) . After washing , slides were incubated with goat anti-mouse secondary antibody for 90 min at room temperature and rinsed in phosphate-buffered saline . Slides were then incubated in 3 , 3′-diaminobenzidene ( DAB substrate kit for peroxidase; Vector Laboratories , Burlingame , CA; SK-4100 ) , and after staining development , they were counterstained with hematoxylin and mounted in Eukitt ( Sigma ) . Similarly , for double immunostaining anti-SYCP3 ( NB300-232 , Novus biological ) ( RRID:AB_2087193 ) and anti-DDX4 ( ab13840 , Abcam ) ( RRID:AB_443012 ) antibodies were used and VIP ( Vector laboratories ) was used as a second substrate . Primary lymphocytes were cultured under standard conditions for karyotyping . Baseline and DNA damage induced chromosome breakage and rearrangement were scored blinded on 50 metaphases . DNA damage was induced with Mitomycin C ( MMC , Sigma ) added for 72 hr . Immunoblots were performed as described ( Bourseguin et al . , 2016 ) . The antibodies used were: mouse monoclonal anti-FANCM antibody ( CV5 . 1 , Novus biologicals , Abingdon , UK ) ( RRID:AB_2716711 ) , mouse monoclonal anti-FANCD2 , ( Santa-Cruz Biotechnology , Dallas , Texas , USA , SC-20022 ) ( RRID:AB_2278211 ) , rabbit anti-FANCA ( Abcam ) ( Bethyl Laboratories , Montgomeryn Texas , USA , Cat# A301-980A RRID:AB_1547945 ) , mouse anti-vinculin ( Abcam ) ( RRID:AB_11156698 ) , Rabbit anti-CHK1 antibody from Santa-Cruz Biotechnology ( SC-8408 ) ( RRID:AB_627257 ) , mouse monoclonal anti-Phospho-CHK1 ( Ser317 ) antibody from Cell Signaling Technology ( #2344 ) ( RRID:AB_331488 ) , monoclonal anti-phospho-H2AX ( Ser139 ) antibody from Millipore ( RRID:AB_309864 ) . To transiently deplete FANCM , HEK293 cells ( RRID:CVCL_0045 ) were transfected with 20 nmol/L of small interfering RNA ( siRNA ) targeting FANCM , 5'-GGC-UAC-GUC-CAG-GAG-CGC-3' with 8 μL INTERFERin ( Polyplus ) in Opti-MEM . The protein bands were visualized and recorded using an ImageQuant apparatus . Western blot quantifications were performed using densitometry measures and the ImageJ software . Measurement of MMC growth inhibition was performed as previously described ( Ridet et al . , 1997 ) . Measurement of Mitomycin C ( MMC , Sigma ) growth inhibition was performed as previously described ( Bourseguin et al . , 2016 ) . Briefly , 5−10 × 105 cells/wells were seeded in 24-wells plate in 1 ml of complete culture medium , and left untreated or exposed to various concentration of MMC . After a growth period of 3–5 days , cells were counted in a Coulter counter . Growth percentage was calculated as follows: % Growth = 100 Xfinal number MMC-treated cells/final number untreated cells . Web resources dbSNP144 , https://www . ncbi . nlm . nih . gov/projects/SNP/ ( RRID:SCR_002338 ) 1000 Genomes Project ( release_v3 . 20101123 ) , http://www . internationalgenome . org/ ( RRID:SCR_008801 ) Exome Variant Server ( ESP6500SI-V2-SSA137 ) , http://evs . gs . washington . edu/EVS/ ( RRID:SCR_012761 ) HapMap3 , ftp://ftp . ncbi . nlm . nih . gov/hapmap/ ( RRID:SCR_002846 ) ExAC r3 . 0 , http://exac . broadinstitute . org/ ( RRID:SCR_004068 ) COSMIC 71 , http://cancer . sanger . ac . uk/cosmic ( RRID:SCR_002260 ) ClinVar 201507 , https://www . ncbi . nlm . nih . gov/clinvar/ ( RRID:SCR_006169 ) Exac database: http://exac . broadinstitute . org/ ( RRID:SCR_004068 ) OMIM: http://omim . org/ ( RRID:SCR_006437 )	About one in 100 women under the age of 40 experience a condition known as primary ovarian insufficiency , which is sometimes known as premature menopause . Women with this condition may have fewer egg cells and are usually infertile . Women with primary ovarian insufficiency are also more at risk of other diseases , including the bone disorder osteoporosis and cardiovascular diseases . The condition is thought to have a genetic basis in part , although so far its causes are largely unknown . Fouquet , Pawlikowska et al . looked at all the genes in genomes of three women in one Finnish family: two sisters and their mother . Both of the sisters had primary ovarian insufficiency , but were otherwise healthy . Their mother did not have the condition . The genetic analysis identified a mutation in a gene called FANCM , which is involved in the cell’s repair response to DNA damage and has recently been linked to breast cancer . This gene is mostly active in egg cells within the ovary . The sisters’ protein made from this mutated copy of the gene was cut short compared with the protein produced by the mother’s FANCM gene . Due to the mutation , the sisters were more sensitive to chemicals that can damage the DNA , effectively making their genome less stable . The affected sisters also had higher levels of abnormalities in the chromosomes compared with their unaffected mother . Fouquet , Pawlikowska et al . then inserted a healthy version of the FANCM gene into the sisters’ cells . This reversed the sensitivity of the sisters’ cells to DNA-damaging chemicals . The findings confirm a genetic link between primary ovarian insufficiency and genes responsible for DNA repair . Mutations in these genes can also make people more at risk of certain cancers . The findings point towards offering some women who have primary ovarian insufficiency in-depth genetic counselling with a long-term follow-up , when alterations of cancer-susceptibility genes are responsible for their condition .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "medicine" ]	2017	A homozygous FANCM mutation underlies a familial case of non-syndromic primary ovarian insufficiency
Information in a computer is quantified by the number of bits that can be stored and recovered . An important question about the brain is how much information can be stored at a synapse through synaptic plasticity , which depends on the history of probabilistic synaptic activity . The strong correlation between size and efficacy of a synapse allowed us to estimate the variability of synaptic plasticity . In an EM reconstruction of hippocampal neuropil we found single axons making two or more synaptic contacts onto the same dendrites , having shared histories of presynaptic and postsynaptic activity . The spine heads and neck diameters , but not neck lengths , of these pairs were nearly identical in size . We found that there is a minimum of 26 distinguishable synaptic strengths , corresponding to storing 4 . 7 bits of information at each synapse . Because of stochastic variability of synaptic activation the observed precision requires averaging activity over several minutes . Synapses between neurons control the flow of information in the brain and their strengths are regulated by experience . Synapses in the hippocampus are involved in the formation of new declarative memories . Understanding how and why synaptic strengths undergo changes in the hippocampus is important for understanding how we remember facts about the world . A fundamental question is the degree of precision in the adjustment of synaptic strengths in view of the many sources of variability at synapses . In this study we provide an upper bound on the variability of synaptic plasticity and quantify a lower bound on the amount of information that can be stored at a single synapse . Excitatory synapses on dendritic spines of hippocampal pyramidal neurons have a wide range of sizes . Anatomical measurements of the spine size , the area of the postsynaptic density ( PSD ) , the number of AMPA receptors , the area of the presynaptic active zone and the number of docked vesicles in the presynaptic terminal are all highly correlated with each other and with physiological measurements of the release probability and the efficacy of the synapse ( Harris and Stevens , 1989; Lisman and Harris , 1994; Harris and Sultan , 1995; Schikorski and Stevens , 1997; Murthy et al . , 2001; Branco et al . , 2008; Bourne et al . , 2013 ) . Thus , each of these individual characteristics is a correlate of synaptic strength . The sizes and strengths of these synapses can increase or decrease according to the history of relative timing of presynaptic inputs and postsynaptic spikes ( Bi and Poo , 1998 ) . If experience regulates synaptic strength then one might expect that synapses having the same pre- and postsynaptic histories would be adjusted to have the same strength . But what would be the inherent variability , or conversely the precision , of this process ? Due to the high failure rate and other sources of stochastic variability at synapses one might expect that the precision of changes in the strengths of these synapses in vivo to be low . The failure rate at synapses depends inversely on the strength , and therefore the size , of the synapse . On this basis the strengths of weaker , and therefore smaller and less reliable synapses , would be expected to be less precisely controlled than the larger and stronger synapses , which have a lower failure rate . An ideal experiment to test for the precision of the changes in synaptic strength would be to stimulate in vivo the axonal inputs to two well-separated spines on the same dendrite to insure that they have the same presynaptic and postsynaptic history of stimulation . Nature has already done the experiment for us as pairs of spines on the same dendrite contacting the same axon satisfy this condition . Prior work suggests that such pairs of spines are more similar in size than those from the same axon on different dendrites ( Sorra and Harris , 1993 ) . Here we evaluated this axon-spine coupling in a complete nanoconnectomic three-dimensional reconstruction from serial electron microscopy ( 3DEM ) ( Harris et al . , 2015 ) of hippocampal neuropil . We determined the similarity of synapses among pairs of spines and set an upper bound on the variability and the time window over which pre- and postsynaptic histories would need to be averaged to achieve the observed precision . In a 6 x 6 x 5 μm3 complete 3DEM from the middle of stratum radiatum in hippocampal area CA1 ( Mishchenko et al . , 2010; Kinney et al . , 2013; see Materials and Methods ) . We identified 449 synapses , 446 axons and 149 dendrites , which except for one identified branch point , are likely to originate from different neurons based on the size of the reconstructed volume and the obtuse branching angles of dendrites from these neurons ( Ishizuka et al . , 1995; Megı́as et al . , 1997 ) . We measured spine head volume and surface area , surface area of the postsynaptic density ( PSD ) adjacent to the presynaptic active zone , and spine neck volume , neck length and neck diameter at the 287 spines that were fully contained within the volume . We also quantified the number of vesicles at the 236 spines and presynaptic boutons that were fully contained within the volume . The strong correlations between these metrics , the skewed shape of the frequency histograms , and the number of synapses per unit of volume ( Figures 1–3 ) , are consistent with previous observations ( Harris and Stevens , 1989; Schikorski and Stevens , 1997; Sorra et al . , 2006; Bourne and Harris , 2011; Bourne et al . , 2013; Bell et al . , 2014 ) . To reduce error , we averaged over multiple independent spine volume measurements for each spine ( Figure 3—figure supplement 1 ) . We determined that the relationship between PSD area and spine head volume did not differ significantly across different dendrites ( Figure 1—figure supplement 1 ) . The correlation between spine head area and spine head volume accounted for 99% of the variance , despite the wide range in spine head shapes and dimensions ( Figure 1A ) , which suggests that the accuracy of our measurements matched the precision of the spine . We also measured spine neck length , diameter , and volume and found a weak trend between the neck diameter ( Figure 1D ) and spine head volume but no correlation between neck length ( Figure 1E ) and spine head volume , consistent with previous studies ( Harris and Stevens , 1989; Tønnesen et al . , 2014 ) . 10 . 7554/eLife . 10778 . 003Figure 1 . Correlations among metrics of dendritic spine morphology . Strong correlations were found between ( A ) Spine head area and spine head volume , ( B ) PSD area and spine head volume , and ( C ) Spine head area and PSD area . ( D ) Weak correlation was found between spine neck diameter and spine head volume . No correlation was found between ( E ) spine neck length and spine head volume and ( F ) spine neck diameter and spine neck length . Regression lines in red and error bars for each data point represent SEM based on multiple tracers who also edited each spine . Equations are based on the log-log distributions , with r2 values indicated , and n=287 complete spines . DOI: http://dx . doi . org/10 . 7554/eLife . 10778 . 00310 . 7554/eLife . 10778 . 004Figure 1—figure supplement 1 . Area of postsynaptic density plotted against spine head volume . Nine individual dendrites all have similar slopes that are not significantly different showing the uniformity of this comparison across dendrites . Error bars , regression lines and equations as described in Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10778 . 00410 . 7554/eLife . 10778 . 005Figure 2 . Presynaptic docked vesicle numbers are correlated with PSD areas , spine head volumes , and neck diameter , but not with neck length . ( A ) All 31 , 377 presynaptic vesicles . ( B ) En face view of the 24 docked vesicles ( gray spheres ) viewed through an axon ( green ) onto the PSD ( red ) of example spine ( yellow ) . ( C ) Number of docked vesicles is correlated strongly with both PSD area and ( D ) spine head volume , weakly with ( E ) neck diameter , but is not correlated with ( F ) spine neck length . Regression lines , SEM ( from multiple tracers ) , and r2 are as in Figure 1 n = 236 complete axonal boutons , each associated with one of the 287 complete spines . One human tracer marked PSDs and vesicles , hence no SEM for these two metrics . DOI: http://dx . doi . org/10 . 7554/eLife . 10778 . 00510 . 7554/eLife . 10778 . 006Figure 3 . Morphometric analysis of 287 complete spines in reconstruction . Distributions of ( A ) spine head volumes , ( B ) PSD areas , ( C ) docked vesicles , ( D ) spine neck volumes , ( E ) spine neck diameters , and ( F ) spine neck lengths are highly skewed with a long tail . DOI: http://dx . doi . org/10 . 7554/eLife . 10778 . 00610 . 7554/eLife . 10778 . 007Figure 3—figure supplement 1 . Spine measurement and estimation of measurement error . ( A ) Example segmentation of spine head ( yellow ) , neck ( gray ) , and PSD area ( red ) . ( B ) Histogram of the measurement error across all spines measured . ( C ) Measurement error plotted against spine head volume . DOI: http://dx . doi . org/10 . 7554/eLife . 10778 . 007 Next , we analyzed spine volumes according to their axonal connectivity and dendrite origin . Pairs of spines on the same dendrite that received input from the same axon ( ‘axon-coupled’ ) , were of the same size and had nearly identical head volumes ( Figure 4 ) . We compared this sample of 10 axon-coupled pairs on the same dendrite ( Figure 4B , pairs a-j ) to those identified on dendrites from the two additional animals ( Bourne et al . , 2013 ) , for a total of 17 axon-coupled spine pairs . When plotted against one another , the paired head volumes were highly correlated with slope 0 . 91 , and despite the small sample size , were highly significantly different from random pairings of spines ( Figure 4C and Figure 4—figure supplement 1A , KS test p=0 . 0002 ) . Similarly , there was a strong positive correlation between their paired neck diameters ( Figure 4D ) , PSD areas ( Figure 4E ) , and number of presynaptic docked vesicles ( Figure 4F ) . These features of axon-coupled spines from the same dendrite spanned the distribution of the overall spine population ( Figure 3 ) . In contrast , the spine neck lengths ( Figure 4G ) , and neck volumes ( Figure 4—figure supplement 1B ) of the pairs were not well-correlated indicating that regulation of neck length and neck volume are not important for synaptic strength . 10 . 7554/eLife . 10778 . 008Figure 4 . Spine head volumes , PSD areas and neck diameters , but not neck lengths , are highly correlated between pairs of axon-coupled same-dendrite spines . ( A ) Visualization of a pair of spines ( gray necks ) from the same dendrite ( yellow ) with synapses ( red , indicated by white arrows ) on the same axon ( black stippling ) with presynaptic vesicles ( white spheres ) . ( B ) All axon-coupled same-dendrite spine pairs ( colors as in A , pair c is elaborated in A ) . Strong correlations with slopes near 1 ( dashed diagonal line ) occur between paired ( C ) spine head volumes ( slope = 0 . 91 ) , ( D ) neck diameters ( slope = 0 . 93 ) , ( E ) PSD areas ( slope = 0 . 74 ) , and ( F ) docked vesicles ( slope = 0 . 91 ) ; but not ( G ) spine neck lengths ( slope = 0 . 48 ) . Larger values from each pairing are plotted on the X axis . Regression lines ( red ) include the 10 a-j pairings ( blue points ) and 7 pairs from 2 additional animals ( green points in ( C ) ) , but do not include triplet bouton pairings ( k-m , gray points ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10778 . 00810 . 7554/eLife . 10778 . 009Figure 4—figure supplement 1 . Analysis of whole spine volume and spine neck volume of axon-coupled same dendrite spines . ( A ) Whole spine volumes of pairs of axon-coupled spines on the same dendrite are highly correlated and significantly different from random pairs ( KS test p = 0 . 018 ) . ( B ) Correlation of neck volumes of pairs of axon-coupled spines on the same dendrite are not significantly different from random pairs ( KS test p = 0 . 74 ) . Larger value in each pair is plotted on the X axis . Regression lines shown in red . Equations are based on regression of log-log distributions , with r2 values indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 10778 . 00910 . 7554/eLife . 10778 . 010Figure 4—figure supplement 2 . Analysis of spines paired randomly . Distributions represent random pairings of ( A ) spine head volumes , ( B ) PSD areas , ( C ) docked vesicles , ( D ) neck volumes , ( E ) neck diameters , and ( F ) neck lengths , from the population of complete spines in the reconstruction . Larger value in each pair is plotted on the X axis . Regression lines shown in red . Error bars for each data point are not shown for clarity . Equations are based on regression of log-log distributions , with r2 values indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 10778 . 01010 . 7554/eLife . 10778 . 011Figure 4—figure supplement 3 . Axon-coupled same dendrite pairs a–f . Large white arrows indicate the red PSDs of the spine pairs , the edited necks are dark gray , and the axons are stippled black with vesicles inside . These illustrate how the axon weaves through the neuropil , synapses with two spines yet passes by others . DOI: http://dx . doi . org/10 . 7554/eLife . 10778 . 01110 . 7554/eLife . 10778 . 012Figure 4—figure supplement 4 . Axon-coupled same-dendrite pairs g-m , illustrated in same way as in Figure 4—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 10778 . 012 The coupled triplet of synapses ( Figure 4C , gray points ‘k , l , m’ ) are on three different spines along a single dendrite and receive synaptic input from a single multi-synaptic bouton . A larger central spine between two similar in size ( Figure 4B , ‘k , l , m’ ) produces one same size pair ( ‘k’ ) and two different size pairs ( ‘ l’ , ‘m’ ) . This unusual configuration is probably driven by processes , such as competition for available resources , that differ from the other pairs ( Sorra and Harris , 1993; Sorra et al . , 1998 ) . As one possibility , perhaps the the size of the larger postsynaptic spine was influenced by the larger size of the available pool of presynaptic vesicles in close proximity to its active zone . Excluding this triple synapse , the median value of the coefficient of variation of volume differences between pairs was CV = 0 . 083 and was as precise for small synapses as it was for large ones ( Figure 5 ) . This precision ( i . e . low CV ) suggests that accurately maintaining the size of every synapse , regardless of size and strength , could be important for the function , flexibility and computational power of the hippocampus . 10 . 7554/eLife . 10778 . 013Figure 5 . CV of axon-coupled spines on the same dendrite does not vary with spine size . There is no significant correlation , which implies that paired small synapses are as precisely matched as paired large synapses . DOI: http://dx . doi . org/10 . 7554/eLife . 10778 . 013 This near-identical size relationship does not hold for axon-coupled spines on different dendrites ( Figure 6B , CV = 0 . 39 , n = 127 , example Figure 6A ) , nor for non-axon-coupled spines on the same or different dendrites ( Figures 6E , F , example Figure 6D ) -- all cases which would have had different activation histories . The volumes of axon-coupled different-dendrite spines are no different from the volumes of random pairs when plotted against one another ( KS test p=0 . 94 , Figure 4—figure supplement 2A , and Figures 6B , C ) and the distribution of their sizes was no different from the whole population ( KS test p=0 . 41 ) . The number of docked vesicles for pairs on different dendrites ( Figure 6—figure supplement 1B ) is not different from random pairings ( KS test p=0 . 08 ) , nor are the neck diameters ( Figure 6—figure supplement 1C , KS test p=0 . 06 ) , nor the neck lengths ( Figure 6—figure supplement 1D , KS test p=0 . 75 ) . The size difference of pairs of axon-coupled spines on the same or different dendrites shows a weak trend with separation distance along the axon or dendrite ( Figure 6—figure supplement 2 ) . The sizes of pairs of axon-coupled spines on the same or different dendrites is unaffected by proximity of glia processes to the synapses ( Figure 7 ) ( Ventura and Harris , 1999; Witcher et al . , 2007 ) , or location of mitochondria in the axon ( Billups and Forsythe , 2002 ) . 10 . 7554/eLife . 10778 . 014Figure 6 . Paired spine head volumes are not correlated when they are not both axon and dendrite coupled . ( A ) Representative visualization and ( B ) plot showing lack of correlation between spine head volumes of all pairs of axon-coupled spines on different dendrites ( n=127 ) . ( C ) Similarly , randomly associated pairs of spine head volumes were not correlated . ( D ) Representative visualization and plots show lack of correlation between spine head volumes from randomly selected pairs ( n=127 ) of non-axon-coupled spines ( E ) on the same or ( F ) different dendrites . Color scheme and regression analyses as in Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 10778 . 01410 . 7554/eLife . 10778 . 015Figure 6—figure supplement 1 . Morphologies of PSD , docked vesicles , and necks are not correlated when spines are not both axon and dendrite coupled . There is no correlation between ( A ) PSD areas , ( B ) docked vesicles , ( C ) neck diameters , and ( D ) neck lengths , in pairs of axon-coupled spines on different dendrites . DOI: http://dx . doi . org/10 . 7554/eLife . 10778 . 01510 . 7554/eLife . 10778 . 016Figure 6—figure supplement 2 . Difference in volume between pairs of axon-coupled spines exhibits a weak trend with separation distance . Differences in spine head volumes plotted against: A ) Distance along the axon for axon-coupled spines on the same dendrite; B ) Distance along the axon for axon-coupled spines on different dendrites; C ) Distance along the axon for randomly paired spines; D ) Distance along the dendrite for axon-coupled spines on the same dendrite; E ) Distance along the dendrite for axon-coupled spines on different dendrites; F ) Distance along the dendrite for randomly paired spines . DOI: http://dx . doi . org/10 . 7554/eLife . 10778 . 01610 . 7554/eLife . 10778 . 017Figure 7 . Proximity of the glial cell to axon-coupled dendritic spines on either the same or different dendrites . Proximity of astrocytic glial processes is not significantly correlated with spine head volumes of axon coupled pairs . ( A ) Histogram of spine head volume for spines that contain a spinule that is engulfed within the glial process ( ‘ spinule’ ) . ( B ) Representation of an engulfed spinule . ( C ) Histogram of spine head volume for spines that are surrounded by and making contact with a glial process ( ‘ensheathed’ ) . ( D ) Representation of ‘ensheathed’ spine . ( E ) Histogram of spine head volume for spines that are proximal but not contacting a glial process ( “‘ adjacent”’ ) . ( F ) Representation of “‘adjacent”’ spine . ( G ) Histogram of spine head volume for spines that are distant from any glial process . ( H ) Representation of a spine “‘distant”’ from the glial process . The KS p value is shown on each inset and indicates that none of these distributions differ from the distribution for the whole population of spines . DOI: http://dx . doi . org/10 . 7554/eLife . 10778 . 017 We found that spine head volumes ranged in size over a factor of 60 from smallest to largest while the CV of any given size was 0 . 083 and was constant across the range of sizes . Measurements of these of 20 pairs allowed us to estimate the number of distinct spine sizes , and by extension synaptic strengths , that can be reliably distinguished across this range . Signal detection theory holds that at a Signal-to-Noise Ratio ( SNR ) of 1 , a common detection threshold used in psychophysical experiments , an ideal observer can correctly detect whether a signal is higher or lower than some threshold 69% of the time ( Green and Swets , 1966; Schultz , 2007 ) . Put another way , if random samples are drawn from two Gaussian distributions whose areas overlap by 31% , an ideal observer will correctly assign a given sample to the correct distribution 69% of the time . Using this logic , we found that ~26 different mean synaptic strengths could span the entire range , assuming CV = 0 . 083 for each strength level , and a 69% discrimination threshold ( Figure 8 , see Materials and methods ) . These 26 distinct strength levels can be represented with 4 . 7 bits of information ( i . e . 24 . 7 ≈ 26 ) which means 4 . 7 bits of information that can be stored at each synapse as synaptic strength . At a discrimination threshold of 76% ( corresponding to SNR = 2 ) there would be ~23 distinct strengths and 4 . 5 bits of information . 10 . 7554/eLife . 10778 . 018Figure 8 . Distinguishable spine sizes . Over the factor of 60 range in spine head volumes from the data set there are 26 distinguishable intervals of spine sizes with a discrimination probability of 69% for each interval based on signal detection theory ( Green and Swets , 1966; Schultz , 2007 ) . The graph illustrates how distinct Gaussian distributions of spine sizes , each with a certain mean size and standard deviation , covers the entire range of spine head sizes on a log scale . The CV of each distribution is a constant value of 0 . 083 ( Figure 5 ) and the intervals are spaced to achieve a total of 31% overlap with adjacent intervals giving a 69% discrimination threshold ( see Materials and Methods ) . Note that the constant CV observed in the data set ( Figure 5 ) means that the intervals appear uniform in width and spacing on a logarithmic scale . This is a form of non-uniform quantization which efficiently encodes the dynamic range of synaptic strengths at constant precision . DOI: http://dx . doi . org/10 . 7554/eLife . 10778 . 018 To explain the high precision observed in spine head volumes , we propose that time-window averaging smooths out fluctuations due to plasticity and other sources of variability including differences in the age of the synapses . To set a lower bound on averaging time , we chose to examine neurotransmitter release probability as a single source of variability . Let us first consider release caused by single action potentials , ignoring short-term plasticity . Release of this type can be analyzed using a binomial model in which n presynaptic action potentials , each with a probability pr of releasing one or more vesicles , leads to a mean number of releases μ = npr having variance σ2 = npr* ( 1−pr ) . The coefficient of variation around the mean is CV = σ/μ = sqrt [ ( 1−pr ) / ( npr ) ] and can be compared with the measured values . Therefore , the number of spikes that are needed to reduce the variability to achieve a given CV is n= ( 1−pr ) / ( prCV2 ) . Table 1 gives averaging time windows T = n/R , where R is spiking rate of the presynaptic axon , for representative values of pr and a range of spiking rates . 10 . 7554/eLife . 10778 . 019Table 1 . Lower bounds on time window for averaging binomially distributed synaptic input to achieve CV = 0 . 083 . DOI: http://dx . doi . org/10 . 7554/eLife . 10778 . 019Release probability ( pr ) Presynaptic spikes ( n ) Averaging time ( R = 1 Hz ) Averaging time ( R = 25 Hz ) 0 . 1130621 . 8 min52 . 2 sec0 . 25819 . 68 min23 . 2 sec0 . 51452 . 42 min5 . 8 sec Accounting for other known sources of variability at dendritic spines would require even longer time windows . In particular , the impact of short-term plasticity during bursts of action potentials on the length of the time-window is complicated by the interplay of facilitation and depression . Synapses with a low initial pr ( and corresponding long time-window in Table 1 ) exhibit marked facilitation and slowly depress during bursts ( Kandaswamy et al . , 2010; Nadkarni et al . , 2010 ) which would shorten the time-window . But synapses with a high initial pr ( and short time-window ) only weakly facilitate , if at all , and quickly depress , which would lengthen the time-window . Previous upper bounds on the variability of spine volume in the hippocampus , based on the whole spine volume ( Sorra and Harris , 1993; O'Connor et al . , 2005 ) , underestimated the precision by including the spine neck volume ( Figure 4—figure supplement 1A ) , which was not correlated between pairs of spines in our volume ( Figure 4—figure supplement 1B ) . Our dense reconstruction included a complete inventory of every synapse in the reconstructed volume and in this respect was unbiased . Additional pairs of synapses from two other rats confirmed that this finding is not confined to a single brain . Of course , additional measurements in the hippocampus and other brain regions would be needed to confirm and extend this finding . The very high statistical significance of the finding ( Figure 4C , KS test p=0 . 0002 ) despite a relatively small number of pairs in our sample implies a large effect magnitude , which would be much smaller if many more samples were needed to reach the same level of significance . To make this p value concrete , if 17 random pairs were chosen from all 287 synapses in the reconstructed volume , there is only a one in 5000 chance that the spine heads would be as precisely matched as the 17 axon-coupled pairs discovered here . Previous studies have shown that there is a high correlation of the size of the spine head with the PSD area and numbers of docked vesicles ( Harris and Stevens , 1989; Lisman and Harris , 1994; Harris and Sultan , 1995; Schikorski and Stevens , 1997; Murthy et al . , 2001; Branco et al . , 2008; Bourne et al . , 2013 ) . Since the correlations between the head sizes of axon-coupled pairs of spines is high , the high correlation between the PSD areas and numbers of docked vesicles observed in axon-coupled spines is not surprising ( Figures 4E and 4F ) . However , it was unexpected to find that the spine neck diameters were also highly correlated between axon-coupled pairs of spines ( Figure 4D r2=0 . 70 ) , since the correlation between spine head volumes and spine neck diameters is not statistically significant ( Figure 1D ) . Thus , there are at least two geometric aspects of the spine geometry that are under tight control of synaptic plasticity , which may reflect different aspects of synaptic function . The diameter of the spine neck may reflect the need for trafficking of materials between the spine shaft and spine head , which is known to be regulated by LTP and LTD ( Araki et al . , 2015 ) . Complementing our observations and analysis in the hippocampus , highly correlated pr at multiple contacts in the neocortex between the axon of a given layer 2/3 pyramidal neuron and the same target cell has been reported ( Koester and Johnston , 2005 ) . Our estimate of synaptic variability , based on spine head volume , is an order of magnitude lower . In a recent connectomic reconstruction of the mouse cortex , the similarity in the volumes of axon-coupled pairs of dendritic spines were statistically significant ( Kasthuri et al . , 2015 ) . This observation is further evidence for the high precision of synaptic plasticity and suggests that the same may be true in other brain areas . The axon-coupled pairs of synapses that we studied were within a few microns of each other on the same dendrite , which raises the question of how far apart the two synapses can be and still converge to the same size . Related to this question , two synapses from the same axon on two different dendrites of the same neuron might not share the same postsynaptic history . These questions cannot be answered with our current data due to the small dimensions and the fact that the position in the neuropil from which our reconstruction was taken makes it highly unlikely that any of the dendrites , other than the one branch point captured in the volume , belong to the same neuron ( Ishizuka et al . , 1995 ) . Synaptic tagging and capture , in which inputs that are too weak to trigger LTP or LTD can be ‘rescued’ by a stronger input to neighboring synapses if it occurs within an hour ( Frey and Morris , 1997; O’Donnell and Sejnowski , 2014 ) , is much less effective when the synapses are on different branches ( Govindarajan et al . , 2011 ) , which would tend to make two synapses from the same axon on different dendritic branches less similar . Probing these questions will require reconstructing a larger extent of hippocampus when a single axon can contact multiple dendritic branches of the same neuron ( Sorra and Harris , 1993 ) or of other cells , such as layer 5 pyramidal cells , which can have 4–8 connections between pairs of neurons ( Markram et al . , 1997 ) . An unusual triple synapse from a single axon ( Figure 4B , ‘k , l , m’ ) was excluded from the analysis because the presynaptic terminal was a single large varicosity filled with vesicles ( i . e . an MSB ) shared by three synapses , unlike the other pairs that had isolated presynaptic specializations ( n=9 ) , or an MSB shared by two synapses ( n=8 ) . It is possible that the large , central spine had an effectively larger pool of vesicles by virtue of proximity , whereas the two synapses on the outside had a more limited population to draw from , and the size of the postsynaptic spine was influenced by the size of the available pool . More examples are needed before we can reach any conclusions . Regardless of the explanation , our estimate of the variability would not be greatly affected by including these 3 additional pairs of synapses in the analysis . How can the high precision in spine head volume be achieved despite the many sources of stochastic variability observed in synaptic responses ? These include: 1 ) The low probability of release from the presynaptic axon in response to an action potential ( Murthy et al . , 2001 ) ; 2 ) Short-term plasticity of release of neurotransmitter ( Dobrunz et al . , 1997 ) ; 3 ) Stochastic fluctuations in the opening of postsynaptic NMDA receptors , with only a few of the 2–20 conducting at any time ( Nimchinsky , 2004 ) ; 4 ) Location of release site relative to AMPA receptors ( Franks et al . , 2003; Ashby et al . , 2006; Kusters et al . , 2013 ) 5 ) Few voltage-dependent calcium channels ( VDCCs ) in spines that affect synaptic plasticity ( smallest spines contain none ) ( Mills et al . , 1994; Magee and Johnston , 1995 ) ; 6 ) VDCCs depress after back propagating action potentials ( Yasuda et al . , 2003 ) ; 7 ) Capacity for local ribosomal protein synthesis in some spines while others depend on transport of proteins from the dendrites ( Ostroff et al . , 2002; Sutton and Schuman , 2006; Bourne et al . , 2007; Bourne and Harris , 2011 ) ; 8 ) Homeostatic mechanisms for synaptic scaling may vary ( Turrigiano , 2008; Bourne and Harris , 2011 ) ; 9 ) Presence or absence of glia ( Ventura and Harris , 1999; Witcher et al . , 2007; Clarke and Barres , 2013 ) ; and 10 ) Frequency of axonal firing ( Callaway and Ross , 1995 ) . One way that high precision can be achieved is through time averaging . Long-term changes in the structure of the synapse and the efficacy of synaptic transmission are triggered by the entry of calcium into the spine . A strategy for identifying the time-averaging mechanism is to follow the calcium . Phosphorylation of calcium/calmodulin-dependent protein kinase II ( CaMKII ) , required for spike-timing dependent plasticity processes , integrates calcium signals over minutes to hours and is a critical step in enzyme cascades leading to structural changes induced by long-term potentiation ( LTP ) and long-term depression ( LTD ) ( Kennedy et al . , 2005 ) , including rearrangements of the cytoskeleton ( Kramár et al . , 2012 ) . The time window over which CaMKII integrates calcium signals is within the range of time windows we predict would be needed to achieve the observed precision ( Table 1 ) . Similar time windows occur in synaptic tagging and capture , which also requires CaMKII ( Redondo and Morris , 2011; de Carvalho Myskiw et al . , 2014 ) . These observations suggest that biochemical pathways within the postsynaptic spine have the long time scales required to record and maintain the history of activity patterns leading to structural changes in the size of the spine heads . The information stored at a single synapse is encoded in the form of the synaptic strength , which reflects the pre- and postsynaptic history experienced by the synapse . But due to the many sources of variability , this information cannot be read out with a single input spike . This apparent limitation may have several advantages . First , the stochastic variability might reflect a sampling strategy designed for energetic efficiency since it is the physical substrate that must be stable for long-term memory retention , not the read out of individual spikes ( Laughlin and Sejnowski , 2003 ) . Second , some algorithms depend on stochastic sampling , such as the Markov Chain Monte Carlo algorithm that achieves estimates by sampling from probability distributions , and can be used for Bayesian inference ( Gamerman and Lopes , 2006 ) . Each synapse in essence samples from a probability distribution with a highly accurate mean , which collectively produces a sample from the joint probability distribution across all synapses . A final advantage derives from the problem of overfitting , which occurs when the number of parameters in a model is very large . This problem can be ameliorated by ‘drop out’ , a procedure in which only a random fraction of the elements in the model are used on any given trial ( Wan et al . , 2013; Srivastava et al . , 2014 ) . Drop out regularizes the learning since a different network is being used on every learning trial , which reduces co-adaptation and overfitting . We are just beginning to appreciate the level of precision with which synapses are regulated and the wide range of time scales that govern the structural organization of synapses . The upper bound on the variability that we have found may be limited by errors in the reconstruction and could be even lower if a more accurate method could be devised to compute the volume of a spine head , neck diameter , PSD area , number of docked vesicles , or other salient features of dendritic spines . Much can be learned about the computational resources of synapses by exploring axon-coupled synaptic pairs in other brain regions and in other species . Three separate 3DEM data sets were used in this study . Each of these data sets has been used for other purposes in prior studies . Images were obtained from serial thin sections in the middle of stratum radiatum of hippocampal area CA1 from three adult male rats ( 55–65 days old ) ( Mishchenko et al . , 2010; Bourne et al . , 2013 ) . One set of images was used to make a dense model of 6 x 6 x 5 μm3 of hippocampal neuropil and processed as previously described in a study of the extracellular space ( Kinney et al . , 2013 ) . In this data set , we identified 13 axon-coupled synaptic pairs on 11 dendrites ( Figure 4—figure supplement 3 and Figure 4—figure supplement 4 ) . The other two sets of images were part of a prior study ( Bourne et al . , 2013 ) in which subsets of dendrites and axons had been reconstructed . In this data set , we identified 7 axon-coupled synaptic pairs on 4 dendrites for a total of 20 axon-coupled dendrite-coupled spine pairs . To perform an accurate and robust geometric analysis of the dendrites , dendritic spines , axons , and glial processes , it was necessary to correct the reconstructed surface meshes for artifacts and make them into computational-quality meshes as described elsewhere ( Kinney et al . , 2013; Edwards et al . , 2014 ) . The postsynaptic densities ( PSDs ) and presynaptic active zones ( AZs ) were identified in the serial section transmission electron microscopy ( ssTEM ) images by their electron density and presence of closely apposed presynaptic vesicles . We devised a method to segment the PSD-AZ features in the electron micrographs and mark their pre- and post-synaptic locations as subregions of the membrane in the final 3D mesh . To accomplish this , contours were hand-drawn on each serial section micrograph closely encompassing , as a single closed contour , the pre- and post-synaptic extent of the electron dense region . Taken together , the stack of contours for a given PSD-AZ forms a 3D capsule which encloses the entire feature . VolRoverN ( Edwards et al . , 2014 ) was used to reconstruct the 3D surface of the capsule enclosing each PSD-AZ pair in 3D . Because these capsules enclose the intracellular domain of both the PSD and AZ they also overlap with the pre- and post-synaptic membrane associated with these subcellular features . Each of these closed capsules was then used as a ‘3D lasso’ to tag mesh triangles of the pre- and post-synaptic membrane contained within the lasso , marking the enclosed membrane area as a synaptic contact region—PSD postsynaptically and AZ presynaptically . Figure 3—figure supplement 1A shows a postsynaptic contact area labeled in red on a dendritic spine . The reconstructed neuropil models were then visualized and analyzed using Blender , a free , open-source tool for 3D computer graphics modeling ( http://blender . org ) . A total of 449 synaptic contacts were found in the dense reconstructed volume of neuropil . We excluded a number of synapses from the analysis if they were partially clipped by the edge of the data set ( 142 ) , or were shaft synapses ( 20 ) leaving 287 valid synapses on dendritic spines in the dense model . An additional 70 spines were excluded from the analysis of axon-coupled spines as the axon which contacted these spines did not contact any other spines within the reconstructed volume . Example visualizations of the spines and axons , generated using Blender , are shown in Figure 2A , B , Figure 3—figure supplement 1A , Figure 4A , B , Figure 4—figure supplement 3 , Figure 4—figure supplement 4 , Figure 6A , D , Figure 7B , D , F , H . Blender’s functionality is user-extensible via a Python interface for creating add-ons . We created a Python add-on for Blender that enabled the selection of the mesh triangles of the dendrite corresponding to the spine head and whole spine of each individual spine . Our add-on tagged each selected set of triangles with metadata for the spine name and geometric attributes of the head , whole spine , and neck as described below . The selection of the spine head was made by hand based on a standardized procedure in which the junction between the head and neck was visually identified as half-way along the concave arc as the head narrows to form the neck ( see Figure 3—figure supplement 1A ) . To select the whole spine , a similar visual judgment was made to locate the junction where the neck widens as it joins the dendritic shaft . Once the appropriate area was selected , the tool was designed to automatically create the convex hull of the selected region . The closed mesh formed by the Boolean intersection of the convex hull and the dendrite was used to determine the measured volume of the spine head or whole spine . The volume of the neck was calculated by taking the difference between these two measurements . Areas were computed from the selected regions for spine head and whole spine . Active zone and postsynaptic density areas were calculated using regions that had been determined during the hand-drawn reconstruction phase described above . Distances between spine heads along the axon were calculated as the Euclidean distance between the centroids of the PSD/AZ regions . Distances between whole spines along the dendritic shaft were calculated as the Euclidean distance between the spine necks to shaft junctions . Glial classification , mitochondria classification and shape classification were performed by hand using set criteria . Some error in the measurement of spine head volume is expected to occur in the human judgment required to segment the dendritic spines into whole spine , head , and neck . To estimate this error , the valid spines in the dense model were segmented and measured a total of four times per spine ( twice each by two people ) . The standard error of the mean in spine head volume decreases with volume and is less than 5% for the majority of spines with a median error of about 1% ( Figure 3—figure supplement 1 ) . The head volumes in the other two data sets were only measured once . Synaptic vesicles in the presynaptic terminals , totaling 31 , 377 in number , were identified along with their 3D locations within the dense reconstruction . Of the 449 presynaptic terminals , we excluded 193 terminals from the analysis due to truncation at the edge of the volume , and 20 terminals at shaft synapses , leaving 236 valid terminals . A visualization of all the synaptic vesicles in the reconstruction is shown in Figure 2A . Positive identification of docked vesicles in these ssTEM data sets is problematic due to the thickness of the sections and density of the staining . To estimate docked vesicles , we counted the number of vesicles whose centers were located within 100 nm of the presynaptic membrane across from the postsynaptic density of a given spine . Of the 31 , 377 vesicles , 3437 were labeled as docked according to this criterion which yielded estimates in good agreement with previous estimates ( Harris and Sultan , 1995; Schikorski and Stevens , 1997; Figures 2B–D ) . An en face view of the docked vesicles at one synapse is shown in Figure 2B . All statistical analysis and plots were generated using Python 2 . 7 ( http://python . org ) with NumPy , SciPy , and Matplotlib . The distributions of spine head volume , spine head area , spine neck volume , PSD area , and AZ area were highly skewed with a long tail at larger values ( Figure 1 ) . Consequently , all regression analysis was performed using Pearson’s linear regression on the data after applying a log-normal transformation ( r2 values shown in Figures 1–6 ) . The coefficient of variation ( CV ) of the population of spine pairings ( Figures 4 and 6 ) was calculated as the median value of the CVs of each individual pair . The CV of each individual pair is simply the standard deviation of the volumes of the pair divided by the mean volume of the pair ( Figure 5 ) . Population distributions were highly skewed making it necessary to make comparisons of distributions using non-parametric methods . We used the two-sample Kolmogorov-Smirnoff ( KS ) test to make these comparisons in Figures 4 and 6 . To estimate the number of distinguishable spine sizes and corresponding bits of precision we calculated the number of distinct Gaussian distributions of spine sizes , each with a certain mean size and standard deviation that together would cover and span the entire range of spine head sizes seen in Figure 4A . Figure 5 demonstrates that it is reasonable to assume that the CV of each these sub-distributions is a constant value of 0 . 083 . From this CV , the spacing between the mean values of each sub-distribution can be chosen to achieve a total of 31% overlap with adjacent sub-distributions giving a 69% discrimination threshold . A 69% discrimination threshold is commonly used in the field of psychophysics and corresponds to a Signal-to-Noise Ratio ( SNR ) of 1 ( Green and Swets , 1966; Schultz , 2007 ) . The 69% confidence interval , z , of a Gaussian distribution is given by: z = sqrt ( 2 ) erf-1 ( 0 . 69 ) The spacing , s , of adjacent intervals of mean , μ , is given by: s = μ2CVz The number , N , of such distributions that would span the factor of 60 range of spine sizes is: N = log ( 60 ) /log ( 1+2CVz ) N = 26 . 3 The number of bits of precision implied by N distinguishable distributions is given by: bits = log2 ( N ) bits = 4 . 72 Figure 8 shows that ~26 distinguishable distributions can cover the entire range of spine sizes , implying that there are ~4 . 7 bits of precision in the spine size . All data and software tools described here are available at: http://www . mcell . cnl . salk . edu/models/hippocampus-spine-analysis-2015-1	What is the memory capacity of a human brain ? The storage capacity in a computer memory is measured in bits , each of which can have a value of 0 or 1 . In the brain , information is stored in the form of synaptic strength , a measure of how strongly activity in one neuron influences another neuron to which it is connected . The number of different strengths can be measured in bits . The total storage capacity of the brain therefore depends on both the number of synapses and the number of distinguishable synaptic strengths . Structurally , neurons consist of a cell body that influences other neurons through a cable-like axon . The cell body bears numerous short branches called dendrites , which are covered in tiny protrusions , or “spines” . Most excitatory synapses are formed between the axon of one neuron and a dendritic spine on another . When two neurons on either side of a synapse are active simultaneously , that synapse becomes stronger , a form of memory . The dendritic spine also becomes larger to accommodate the extra molecular machinery needed to support a stronger synapse . Some axons form two or more synapses with the same dendrite , but on different dendritic spines . These synapses should be the same strength because they will have experienced the same history of neural activity . Bartol et al . used a technique called serial section electron microscopy to create a 3D reconstruction of part of the brain that allowed the sizes of the dendritic spines these synapses form on to be compared . This revealed that the synaptic areas and volumes of the spine heads were nearly identical . This remarkable similarity can be used to estimate the number of bits of information that a single synapse can store , since the size of dendritic spines and their synapses can be used as proxies for synaptic strength . Measurements in a small cube of brain tissue revealed 26 different dendritic spine sizes , each associated with a distinct synaptic strength . This number translates into a storage capacity of roughly 4 . 7 bits of information per synapse . This estimate is markedly higher than previous suggestions . It implies that the total memory capacity of the brain – with its many trillions of synapses – may have been underestimated by an order of magnitude . Additional measurements in the same and other brain regions are needed to confirm this possibility .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Nanoconnectomic upper bound on the variability of synaptic plasticity
Motor neurons of the crustacean cardiac ganglion generate virtually identical , synchronized output despite the fact that each neuron uses distinct conductance magnitudes . As a result of this variability , manipulations that target ionic conductances have distinct effects on neurons within the same ganglion , disrupting synchronized motor neuron output that is necessary for proper cardiac function . We hypothesized that robustness in network output is accomplished via plasticity that counters such destabilizing influences . By blocking high-threshold K+ conductances in motor neurons within the ongoing cardiac network , we discovered that compensation both resynchronized the network and helped restore excitability . Using model findings to guide experimentation , we determined that compensatory increases of both GA and electrical coupling restored function in the network . This is one of the first direct demonstrations of the physiological regulation of coupling conductance in a compensatory context , and of synergistic plasticity across cell- and network-level mechanisms in the restoration of output . The hallmarks of robust central pattern generator ( CPG ) output are appropriately tuned excitability of individual neurons combined with circuit-level interactions that maintain appropriate temporal coordination ( i . e . , phasing ) of these neurons . Through both developmental and ongoing tuning processes , CPGs can maintain reliable network output for decades across the lifespan of an individual , despite constant feedback from the changing nature of both the organismal and natural environment . Yet underlying this constant reliability of network output exists a surprising amount of variability in the individual parameters necessary for producing activity . For instance , despite having nearly identical output across animals , networks can exhibit a five-fold or more range in intrinsic and synaptic conductance values ( Marder and Goaillard , 2006; Schulz et al . , 2006; Marder , 2011; Roffman et al . , 2012 ) . Such variability in intrinsic conductances is not limited to CPGs , but has been documented in several cell types of the mammalian brain , including cerebellar Purkinje cells ( Swensen and Bean , 2005 ) , and globus pallidus neurons ( Günay et al . , 2008 ) . Additionally , the synaptic strengths between mammalian central neurons have been shown to vary in several brain regions ( Nelson and Turrigiano , 2008; Turrigiano , 2008; Maffei et al . , 2012 ) . The origins and implications of this variability are still an intense area of investigation ( Krubitzer and Kahn , 2003; Turrigiano and Nelson , 2004; Ciarleglio et al . , 2015 ) . We hypothesize that this variability might be a result of ongoing compensatory changes required to maintain reliable output over time . This compensation , termed homeostatic plasticity , has been well documented both for plasticity of intrinsic excitability via changing ionic conductances ( Turrigiano et al . , 1994 ) , as well as for changes in chemical synaptic strength ( Desai , 2004; Turrigiano , 2012 ) . Variability in conductances may also be an adaptive trait in and of itself: variable solutions that produce convergent circuit output may provide a selective advantage , or perhaps be a substrate for adaptation and evolution ( Marder and Goaillard , 2006; Grashow et al . , 2009 ) . Regardless of whether such variability is the result of homeostatic compensation , differential tuning across networks , or a combination of these and other heretofore undiscovered causes , a potential cost to such variability has recently been identified . In the cardiac ganglion ( CG ) of the Jonah crab ( Cancer borealis ) , five Large Cell motor neurons ( LCs ) generate completely synchronous output , as a result of pacemaker inputs within the network , to drive simultaneous heart muscle contraction in the crab ( Tazaki , 1972 ) . Despite completely uniform and synchronous activity within the network , LCs show highly variable underlying maximal conductances ( Ransdell et al . , 2012 ) . These variable conductances render the neurons susceptible to perturbations that target a subset of ionic conductances: when high-threshold K+ currents were blocked with tetraethylammonium ( TEA ) , the motor neurons lost coordinated output and became divergent in their patterns of firing ( Ransdell et al . , 2013a ) . These CG neurons compensate for this change in excitability , presumably to homeostatically maintain a target level of excitability ( Ransdell et al . , 2012 ) . However , none of the network level impacts of this perturbation and plasticity have been investigated . Indeed , it is difficult to study homeostatic plasticity in intact networks and to simultaneously take into account both properties of individual cells as well as their network interactions . In the present study , we discovered that LC variability makes the network vulnerable to desynchronization as a result of TEA exposure , but that compensation resynchronizes the network within 30–60 min via both intrinsic cellular and circuit-level physiological mechanisms . To examine the underlying mechanisms , we developed a biophysical computational model of the entire cardiac network . The network model enabled a comprehensive search of the conductance space for potential compensatory mechanisms that preserved network synchrony , and we used these findings to guide further experimentation . Our study revealed cooperative homeostatic plasticity among intrinsic conductances and electrical coupling across multiple cells in the cardiac network . We interpret this as a novel homeostatic compensatory mechanism contributing to the overall robustness of CPG output . Ransdell et al . ( 2013a ) were able to repeatably reduce the magnitude of high-threshold K+ currents ( IKd + IBKKCA ) by ~92% in isolated LCs with 25 mM TEA . We used this experimental manipulation on the 3 anterior LCs in intact CGs ( Figure 1A ) causing LCs in the same ganglion to change from identical ( Figure 1B ) to divergent , asynchronous output ( Figure 2A , I , II ) after exposure to TEA . After application of TEA , motor neurons became noticeably more depolarized during burst potentials and LC spiking seen on the extracellular recordings increased substantially ( Figure 2A ) . Additionally , comparison of intracellular voltage waveforms revealed a loss of conserved output ( Figure 2A ) . These results are consistent with the hypothesis that variable underlying conductances of the LCs makes them vulnerable to a uniform perturbation of a subset of conductances such as the TEA blockade that targets high-threshold K+ conductances . Because our previous results demonstrated that the change in excitability that accompanies TEA exposure in LCs is accompanied by an increase in A-type K+ current , we hypothesized that compensation may also occur at the network level to restore synchrony among LCs subsequent to the TEA block . 10 . 7554/eLife . 16879 . 003Figure 1 . Experimental setup for the recording and superfusion of CG neurons . ( A ) Petroleum jelly wells ( gray ) allow the posterior LC1 and LC2 as well as the pacemaker small cells ( SCs ) to be pharmacologically isolated from the anterior large cells ( LC3 , LC4 , LC5 ) . Pacemaker cells can be maintained in physiological saline , or the network can be temporarily shut down by replacing saline with 750 mM sucrose . Extracellular recordings are performed with stainless steel pin electrodes from the 'trunk' nerve that contains the axons of all 5 LCs and the pacemaker cells . Intracellular recordings are taken from the anterior LCs . The area outside the petroleum jelly wells is superfused with pharmacological agents to target only the anterior large cells . ( B ) Simultaneous intracellular recordings from the three anterior LCs and extracellular recording of the network output via the trunk nerve , demonstrating synchrony among LCs in the control ongoing rhythm . Scale bars = 10 mV , recording duration = 9 s . DOI: http://dx . doi . org/10 . 7554/eLife . 16879 . 00310 . 7554/eLife . 16879 . 004Figure 2 . Restored excitability and waveform synchrony among LCs after 1 hr of TEA exposure . ( A ) Representative recordings of LC3 and LC5 and of network activity over one hour of TEA exposure . Roman numerals ( I , II , III , IV ) designate time points of reference throughout the remainder of the figure , as follows: I – control saline , II – acute TEA exposure identified as the maximum effect on loss of synchrony across LCs ( i . e . lowest R2 value ) , III – 30 min of TEA exposure , IV – 60 min of TEA exposure . ( B ) Scatterplots show pairwise correlation of time-matched voltages ( sampled at 10 kHz ) of the waveforms shown in the representative traces . R2 values are calculated from Pearson’s correlation tests for these two cells . Loss and restoration of conserved output is demonstrated by changes in coherence in the scatterplot as well as in R2 value . ( C ) Synchrony of waveforms of the two cells seen in panels A and B plotted as R2 values over the entire time course of the experiment . Roman numerals and large gray circles represent the values that were obtained from the scatterplots as each time point shown in panel B . TEA perfusion persists from time zero through 60 min . Box plots show distributions of R2 values from cross-correlation analyses of LC voltage waveforms for pairs of LCs from N = 11 preparations . Lines within boxes mark the median , box boundaries represent 25th and 75th percentiles , whiskers represent 5th and 95th percentiles , and points represent outlying observations . Groups with significant differences in median synchrony ( p<0 . 05; Wilcoxon signed rank tests ) are denoted with different letters . ( D ) Excitability of LCs was quantified by five measurements ( mean ± SD ) . Analysis of each preparation used the average of 10 consecutive bursts at each time point ( N = 8 preparations ) . Number of spikes per burst , spike frequency within each burst , and the latency between pacemaker firing and first motor neuron spike ( SC-LC Phase Delay ) were calculated from extracellular traces . Total depolarization and amplitude of each burst are based on intracellular recordings . Significant differences across groups ( p<0 . 05; paired t-tests ) are denoted with different letters , such that any two bars with a letter in common are not significantly different . DOI: http://dx . doi . org/10 . 7554/eLife . 16879 . 004 To determine whether compensatory responses can restore both excitability and synchrony of LC output following TEA block , anterior LCs were superfused with TEA for at least 1 hr while the activity of the individual LCs and the network were continuously recorded . Our data demonstrate that both synchrony and excitability are restored towards baseline levels over a period of 30–60 min following TEA exposure . Figure 2A and B illustrate a typical progression through the loss and subsequent restoration of synchrony among LCs during one hour of continuous exposure to TEA . Acutely after application of TEA , we saw a significant reduction in synchrony as measured by R2 ( see Materials and methods; Ransdell et al . , 2013a ) across LC voltage waveforms ( Figure 2B , C , time point II ) . Following this reduction , waveform synchrony values consistently recovered towards baseline levels , often by 30 min , and were no longer statistically different from baseline by 1 hr of treatment with TEA ( Figure 2C , right ) . To monitor compensatory changes in excitability , five different measures of excitability were calculated using both extracellular and intracellular recordings ( Figure 2D; see Experimental Methods ) . The spikes per burst , average spike frequency , total burst depolarization and burst amplitude all were significantly increased immediately after exposure to TEA ( 'acute' ) , while the small cell pacemaker ( SC ) -to-LC phase delay was significantly decreased ( Figure 2D ) . All 5 measures also then showed a significant change back towards their baseline levels between the 30 min and 60 min time points . While all measures showed clear shifts towards restoration of baseline excitability , the number of spikes per burst , spike frequency within each burst , burst amplitude , and total burst depolarization were not completely restored to control levels; the exception was the SC-LC phase delay which was fully restored ( Figure 2D ) . Preparations exposed to TEA for 2–3 hr showed no further change in excitability ( data not shown ) . For this reason , time scales longer than 1 hr were not included in our analyses . TEA exposure reduces LC synchrony and induces hyperexcitability . Our model development and selection criteria resulted in a population of 27 model CG networks with variable underlying conductances of the constituent neurons that successfully recapitulated the biological data observed in TEA ( see Methods , Supplemental Information ) . Our previous results identified an approximate 2 . 2 ± 0 . 8-fold change in IA in LCs as a result of 60 min of TEA exposure ( Ransdell et al . , 2012 ) . Therefore , we used the model networks to explore potential mechanisms of compensation by first increasing and decreasing each individual maximal membrane conductance by a similar factor of 2 . We searched for changes that would increase LC spike synchrony while countering the hyperexcitability induced by TEA . To easily visualize the trends , each network was normalized to its initial value for spike synchrony . These data are shown for all conductances in Figure 3 . 10 . 7554/eLife . 16879 . 005Figure 3 . Effects of increasing and decreasing individual ionic conductances on excitability and synchrony in model CG networks . ( A ) Schematic representation of model network organization and connectivity . Five large cell ( LC ) motor neurons are innervated via excitatory synapses from a common small cell pacemaker input ( SCs ) . LC model neurons consist of two compartments - soma and axon - of which only the somata are pictured . Somata contain 9 conductances: GCaS , GCaT , GLEAK , GCAN , GA , GBKKCa , GSKKCa , GKd , and GNaP . Paired LCs ( 1+2 , 4+5 ) have stronger local coupling ( black resistor symbols ) , and all 5 LCs are reciprocally electrically coupled via weaker gap junctions ( gray resistor symbols ) . An example of LC3 and LC4 model output within a network burst activity is shown in the red and blue traces under both control and TEA ( 90% reduction in both GKd and GBKKCa ) conditions . Graphical representations of spike synchrony ( raster plots ) and waveform synchrony ( scatterplots; as in Figure 2 ) are shown for the model neurons , demonstrating that both measures reflect the loss of LC synchrony as a result of TEA . ( B ) Measurements of both output variables ( # of spikes and spike synchrony ) were made under three model conditions: control , TEA , and TEA + either a 2x increase ( G↑ ) or 2x decrease ( G↓ ) in a given conductance . N = 27 distinct model networks . All output measurements are normalized to their initial ( control ) conditions . Red lines represent the mean for a given group . Dashed line represents the 1 . 0 value ( baseline ) for a given measure . Compensatory responses that restore excitability and synchrony will tend to move the mean towards baseline . DOI: http://dx . doi . org/10 . 7554/eLife . 16879 . 005 Our initial goal with the model was to determine whether changes in single conductances were sufficient to elicit compensatory changes in output that help restore both excitability and synchrony . While it is not difficult to conceive of a change in multiple aspects of the parameter set that could achieve restoration of output , it is perhaps not as intuitive – but presumably the most parsimonious solution – for a single conductance to have such an impact . True to this expectation , while various manipulations of Gmax values improved either excitability or synchrony , very few conductance changes improved both . The optimal solution of significantly improving spike synchrony and also significantly decreasing the total number of action potentials was achieved in only one case: two-fold increase in GA resulted in a mean synchrony score that was significantly different from the TEA case ( p<0 . 05 , paired t-test ) but not significantly different from control ( p=0 . 157 ) . No other change in a given conductance resulted in this combination of statistical outcomes . Not every model cell or networked improved uniformly with this conductance change . Therefore , while these results do not rule out a contribution for other conductances , they do suggest that an increase in GA , as seen in previous experimental studies on isolated LCs ( Ransdell et al . , 2012 ) , may be the most likely candidate for a change in intrinsic conductance promoting synchrony at the network level . These data suggest that while a single conductance change ( increased GA ) can help restore both excitability and synchrony , variations in a single voltage-dependent conductance may not be sufficient to account for the full compensation response . In addition to perturbing only individual conductances , we also varied current kinetics and activation parameters ( half-activation voltage V1/2 , ± 10 mV , and slope factor k , by 0 . 5x and 2x ( Ballo et al . , 2010 ) and time constant by ± 10 ms ) for all the cell currents individually , and found that no changes in parameters for a single current could simultaneously restore excitability at the single cell level , and synchrony at the network level ( data not shown ) . While the analysis has focused only on the parameters of a single current , simultaneous changes in parameters of multiple currents could also potentially provide similar compensation , and that remains to be explored . However , our analysis does reveal the substantial contribution of changing a single parameter – GA – on multiple aspects of network compensation , to an extent that is beyond simple intuition . Importantly , the model also extends the biological data by demonstrating that waveform synchrony can translate into spike synchrony . Because of the electrotonic distance between the somata and axons of LCs , we cannot measure spike synchrony directly in this preparation . The model allows us to infer that waveform synchrony ( and loss of synchrony ) can indeed translate to the level of the most proximal cellular output – spiking . Model runs predicted that increases in IA help restore LC excitability and synchrony . To test this experimentally , we silenced pacemaker activity with isotonic sucrose solution ( see Methods , Supplemental; Figure 1A ) , and tested the similarity of responses of each individual LC to a biologically realistic current stimulus ( Ransdell et al . , 2013a ) . We compared LC3 and LC5 to the same current injection at three time points: control , 5 min post-TEA , and 1 hr post-TEA . Between current injections , pacemaker activity was restarted by removal of the sucrose block . This allows us to test each cell in isolation , but compensation occurs in the intact network . The initial voltage responses to our stimulation protocol in LC3 and LC5 in control conditions are highly similar to one another , and their level of waveform synchrony was not significantly different from the synchronous activity across these LCs during intact control network activity ( Figure 4A , B ) . Immediately following TEA application , LC3 and LC5 show disparate output when driven with a common stimulus protocol ( Figure 4A ) . Finally , our data show significant increases in R2 of voltage activity within 1 hr across isolated LCs ( Figure 4B ) , demonstrating that intrinsic compensation does improve network synchrony . However , after 1 hr the synchrony values were significantly lower than control values ( Figure 4B , C ) , suggesting that intrinsic compensation alone is insufficient to restore synchrony . To determine whether compensatory changes in IA occur in the intact network , we measured IA with two-electrode voltage clamp in LCs before and after 1 hr of TEA exposure . Measurements were made while the network activity was temporarily halted with sucrose block , and compensation occurred with ongoing network activity . In all cases peak IA current increased ( Figure 4D ) , with a mean increase of 56% ( p<0 . 05 , n = 6 , Wilcoxon signed rank test ) . These data are consistent with the hypothesis that a compensatory increase in IA can help promote synchrony in these networks . 10 . 7554/eLife . 16879 . 006Figure 4 . Intrinsic compensation involving GA partially restores synchrony after TEA block . A reversible sucrose block was used to temporarily stop network activity and a current stimulus protocol was delivered to individual LCs at three time points: control , after 5 min of TEA perfusion ( acute ) , and after 1 hr of TEA perfusion ( after compensation ) . ( A ) Representative traces from 2 different preparations compare the responses of LC3 ( orange ) and LC5 ( blue ) to the same current injection ( Stimulus Protocol ) at each of the three time points . ( B ) R2 values for N = 6 preparations at each time point are plotted with the same preparation connected across time points . All 3 conditions were significantly different from one another ( p<0 . 01; paired t-tests ) . ( C ) When isolated cell output was compared with the output in the network , there was no difference in R2 at the control state , but following 1 hr of compensation in the network , there was a significant ( p<0 . 002 – t-test ) difference in synchrony scores ( mean ± SD ) between cells when isolated vs . when they are in the intact network . ( D ) IA was measured by two-electrode voltage clamp before and after compensation in N = 6 LCs in the intact CG . Voltage clamp data were obtained by temporarily silencing network activity with 750 mM sucrose . There was a significant increase in IA after 1 hr in TEA ( mean 56 ± 65% increase , p<0 . 05 –Wilcoxon signed rank test ) . ( E ) Similarity of waveform in the same neurons before and after 1 hr TEA exposure . R2 values were calculated for the output of the same cells before and after TEA exposure , and are shown as before-and-after values in the same cell connected by a line . Box plots show distributions of R2 values from cross-correlation analysis of LC voltage waveforms N = 8 cells . Lines within boxes mark the median , box boundaries represent 25th and 75th percentiles . 'Control' is the comparison of voltage waveforms to 2 separate rounds of current injection in the absence of TEA . Although there was improvement in similarity of waveforms after 1 hr in TEA ( panel A ) , the newly compensated output after 1 hr of TEA exposure does not recapitulate the original response to the stimulus protocol ( significantly different from control; p<0 . 01 – Wilcoxon signed rank test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16879 . 006 Although the waveforms of LC3 and LC5 were not different from one another after 1 hr of compensation , anterior LC burst potentials did not reproduce their original waveform after compensation ( see Figure 2; Panels I and IV ) . We also used the LCs from data shown in Figure 4A to compare the voltage responses of individual LCs to a fixed stimulus before and after compensation . Repeatable voltage responses under control conditions indicate that trial-to-trial variability is negligible ( mean R2 = 0 . 997; Figure 4E , control ) . However , the voltage response after compensation was significantly different from the control voltage response ( p<0 . 01 , n = 8 , Wilcoxon signed rank test ) , indicating that intrinsic compensation does not restore the original cellular output ( Figure 4E; control vs . TEA [1 Hour] ) . If intrinsic compensation does not fully restore synchrony , another mechanism must be present to explain the results observed during network compensation . LCs receive common excitatory inputs from the pacemakers and one hypothesis is that changing the strength of these chemical synapses might help to restore LC firing to appropriate levels . LCs in the network are also electrically coupled to one another via gap junctions which presumably promotes synchrony , although clearly the native coupling is not able to maintain LC synchrony in TEA ( Hagiwara et al . , 1959; Tazaki and Cooke , 1983; Cooke , 2002 ) . A second hypothesis is that increased electrotonic coupling between LCs could buffer against disparate output and help to restore synchrony . Using our set of model networks , we increased and decreased the strength of chemical synapses in 10% increments to test the effects on excitability and synchrony . We then did the same with model electrical coupling conductance . We found that increasing the strength of either chemical synapses or electrical coupling increased both synchrony and excitability ( Figure 5 left ) . However , increasing the chemical synaptic conductance in conjunction with TEA blockade also increases spiking of the LCs ~25–30% in contrast to the biological decrease in excitability relative to the acute TEA exposure seen with compensation . Conversely , only a small change in LC spiking occurs with an increase in electrical coupling ( ~9% , Figure 5 left ) . Reducing the strength of either chemical synapses or electrical coupling decreased overall spike synchrony ( Figure 5 right ) , violating the assumptions of compensation based on the biological data . Reducing chemical synaptic strength eventually ceased LC firing altogether ( data not shown ) . 10 . 7554/eLife . 16879 . 007Figure 5 . Effects of increased or decreased strength of chemical synapses and electrical coupling on excitability and synchrony in model CG networks . Measurements of two output variables ( # of spikes and synchrony ) were made under three model conditions: control , TEA ( 90% reduction in both GKd and GBKKCa ) , and TEA + an incremental increase or decrease ( up to 100% by 10% increments ) for both chemical ( pacemaker to LC ) or electrical ( LC to LC ) connections . N = 27 distinct model networks . All output measurements are normalized to their initial ( control ) conditions to visualize trends . Dashed line represents the 1 . 0 value ( baseline ) for a given measure . Red lines represent the mean for each group . Each different colour and shape for points corresponds to one model network , and the same networks are shown across conductance levels . P-values in each plot refer to the results of a one-way ANOVA across all groups . Asterisks ( * ) denote groups in each plot that were significantly different from the TEA group via Holm-Sidak post-hoc tests . DOI: http://dx . doi . org/10 . 7554/eLife . 16879 . 007 Increased electrical coupling restored synchrony in model LCs with only a modest effect on excitability . To investigate this relationship in experiments , we isolated LCs and used a dynamic clamp to add an artificial coupling conductance . Pairs of LCs from the same network were physically isolated by thread ligature , exposed to TEA , and simultaneously received the same stimulus protocol , while the dynamic clamp added a non-rectifying artificial coupling conductance ( from 0 to 0 . 2 μS ) between the cells . The driving force was equivalent to the voltage difference in membrane potential between the coupled cells . Increasing the artificial coupling conductance significantly increased the correlation coefficient of the waveform between the two cells ( Figure 6A ) , with a synaptic conductance value of 0 . 2 μS able to rescue synchrony of LCs to levels observed in intact networks ( Figure 6A ) . These results provided proof of principle that increasing electrical coupling could be responsible for resynchronization in the network . 10 . 7554/eLife . 16879 . 008Figure 6 . Changes in electrical coupling associated with compensation in biological CG networks . ( A ) Artificial electrical coupling restores synchrony in isolated LCs . With acute exposure to TEA , isolated LCs produce disparate output in response to an identical stimulus . The stimulus protocol consists of current injections that mimic biological synaptic currents and back propagating action potentials ( see Ransdell et al . , 2013a ) from four consecutive bursts of network activity . The current was injected simultaneously into isolated cells , while the dynamic clamp was used to provide an artificial coupling conductance . Representative traces of the same two cells shown in TEA with different levels of synaptic current applied . Only the final burst of the four-burst input stimulus is shown for clarity . N = 7 different preparations ( bottom panel ) show an increase in synchrony with increasing coupling conductance . ( B ) Biological coupling increases during network compensation . Hyperpolarizing current injections were used to measure coupling coefficients between LC3 and LC5 during control conditions , with acute TEA exposure ( 5 min ) and after 1 hr compensation in TEA . The input resistance of LCs was measured and showed no significant change at any time point . No significant differences were observed for coupling coefficient or coupling conductance between control and acute TEA conditions . Coupling coefficient significantly increased after 1 hr in TEA ( Mean increase 85 ± 82% from control , N = 11 , p<0 . 01 , Wilcoxon signed rank test ) . Measurements from LC4 and LC5 show that coupling conductance increased significantly as a result of 1 hr TEA exposure ( mean increase 49 . 5 ± 36% from control , N = 13 , p<0 . 001 , Wilcoxon signed rank test ) . Significant differences across groups are denoted by different letters . Plots show mean ± SD . ( C ) Representative traces from two different preparations of changes in coupling observed before ( Control ) and after 1 hr of TEA exposure . Top traces are between LC3-LC5 and the bottom traces are between LC4-LC5 . Measurements were made in two-electrode current clamp , and the current was injected into LC5 . Because current injections were manually timed to occur between bursts of network activity , slightly different durations of current pulses occurred in the two recordings in the top recordings . Recordings from LC4-LC5 were used for coupling conductance measurements seen in panel B , as their close proximity allows for much less influence of electrotonic distance on calculations of conductance . DOI: http://dx . doi . org/10 . 7554/eLife . 16879 . 008 We then measured coupling coefficients between LCs during compensation in the intact network . Coupling coefficients between cells increased significantly after 1 hr in TEA ( mean increase 85% , p<0 . 01; Figure 6B , C ) . The coupling coefficient is a useful description of the functional relationship in coupling , but does not identify the electrophysiological mechanism . Plasticity of coupling properties can ultimately be influenced by two fundamentally different mechanisms: altered resistance of the non-junctional membrane of the coupled cells , or modification of gap junctional conductance . Using two-electrode current clamp , we saw no significant differences in the apparent input resistance of LCs in control physiological saline , after acute TEA exposure , or after compensation ( 1 hr TEA exposure; Figure 6B , C ) . These results indicate that changes in passive membrane conductance ( GLeak ) are not responsible for increased coupling coefficients . To directly test whether coupling conductance between LCs increases as a result of TEA-induced compensation , we focused on LC4 and LC5 pairs within the same network . In the crab CG , anterior LCs exhibit strong local electrical and dye coupling ( Tazaki and Cooke , 1979 , 1983 ) . The branch containing LC4 and LC5 somata can be separated and electrotonically 'sealed' from the network by thread ligature to create ideal conditions for measuring coupling conductance . With two electrodes in each cell , we used hyperpolarizing current injections to measure resistance and calculate junctional conductance independent of membrane resistance ( as in Bennett , 1966; see Materials and methods ) . Coupling conductance ( GC ) between LC4 and LC5 significantly increased during 1 hr of TEA exposure ( mean increase 49 . 5% , p<0 . 01 , N = 8; Figure 6B , C ) . Taken together , our experimental and modelling results suggest that an increase in GA is able to counter the increase in excitability of LCs in TEA in a compensatory fashion , as well as promote restoration of synchrony , but was insufficient to restore synchrony fully . Additionally , our results suggest that an increase in coupling among LCs can greatly promote synchrony with only a modest effect on excitability . Therefore , we next used our model networks to investigate how GA and GC might interact to promote synchrony by calculating synchrony scores as conductances of all 27 model networks were adjusted . First , we increased GA alone in 10% increments up to a 100% increase ( Figure 7 ) . Increasing GA up to +40% promoted greater synchrony after TEA blockade , but was unable to fully restore synchrony even with increasing conductance levels , consistent with our biological data ( Figure 4 ) . Increasing GA beyond +40% did not further improve synchrony ( Figure 7 ) , and ultimately caused LCs to cease firing altogether . We also increased model GC incrementally ( from +10% to +150% ) , and found that electrical coupling alone was capable of restoring synchrony fully , but this required a 140% increase in its value ( Figure 7 ) . Finally , we increased both GA and GC together in 10% increments , revealing a potentially synergistic relationship: a smaller increase of 70% in each conductance was able to produce spike synchrony that was indistinguishable from control ( Figure 7 ) . 10 . 7554/eLife . 16879 . 009Figure 7 . Increased GA and coupling conductance ( GC ) among LCs act in concert to help restore synchrony across LCs in model networks . ( A ) Voltage response of a typical network LC3 ( gray ) and LC4 ( black ) cells under three model conditions: control , TEA ( 90% reduction in both GKd and GBKKCa ) , TEA + 70% increase in GA , and TEA + 70% increase in GA and GC . ( B ) Effects of increasing GA alone , GC alone , or both GA and GC on synchrony in model networks . Dashed lines represent simple linear regression fits to the points for each condition . Black line represents the change in synchrony from control to TEA . Points shown are the average values for N = 27 networks . ( C ) Summary of Synchrony Score shown for all 27 model networks . Stepwise changes are shown from Control , TEA , TEA + increasing GA by 70% , and TEA + increasing both GA and GC by 70% ( the point at which maximal synchrony is restored as per the analysis in panel B ) . Individual points correspond to those used to generate averages in panel A at the 70% level to give an idea of the variability in the data set . Individual preparations are connected with lines to show trends across networks . DOI: http://dx . doi . org/10 . 7554/eLife . 16879 . 009 Previous modelling studies found that K+ currents can increase or help restore synchrony between electrically coupled neurons ( Pfeuty et al . , 2003 ) , so we first hypothesized that a compensatory increase in A-Type K+ membrane conductance could be a mechanism underlying both restored excitability and resynchronization . Over the course of 30–60 min , increased IA was associated with decreasing cellular excitability [see also ( Golowasch et al . , 1999 ) ] and improvement of coordinated motor neuron firing . However , intrinsic compensation alone was insufficient to fully restore synchrony across LCs . A concomitant increase in electrotonic coupling ensured virtually complete resynchronization . Our modelling results suggest that although a sufficient increase in electrical coupling alone could restore full synchrony ( 140% increase ) , it could not simultaneously restore the original level of excitability . Only a 70% increase was necessary when accompanied by a concomitant increase in GA . Therefore , we conclude that multi-component mechanisms are not only necessary for full compensation , but also that their synergistic action is potentially more efficient than either mechanism operating in isolation . While our results occurred in a compensatory context , the underlying mechanisms bear striking similarity to motor output plasticity induced by operant conditioning in Aplysia . In a series of elegant experiments , it has been shown that chaotic exploratory and consummatory radula biting movements of Aplysia during food searching behaviour can be stably modified by operant conditioning , leading to prolonged bouts of radula movements with increased frequency and more stereotyped rhythmic organization ( Nargeot et al . , 2007; Nargeot and Simmers , 2011 ) . Chaotic biting patterns result from inherently variable and unsynchronized bursting of CPG neurons that are each randomly capable of triggering bites ( Nargeot et al . , 2009 ) . Following operant conditioning , induction of regular and synchronized bursting of pattern-initiating cells can be attributed to changes in both intrinsic excitability and electrical coupling strength . Specifically , changes in intrinsic excitability attributed to changes in leak conductance underlie the increase in frequency of motor output , while increases in coupling strength allow for the synchronization and regularization of bursting ( Nargeot et al . , 2009; Sieling et al . , 2014 ) . The full shift in behavioural and circuit output is therefore the additive influence of both intrinsic and electrical synaptic conductances . The striking similarity in these underlying mechanisms suggests that these kinds of circuit-level mechanisms may be a conserved strategy for stabilization of synchrony within network output , be it compensatory or in the context of behavioural plasticity . The speed ( within 30 min ) and magnitude ( up to a doubling of effective coupling ) of physiological changes seen in electrical coupling was remarkable . Although electrical coupling has long been known to promote synchrony in many systems , including the CG ( Tazaki , 1972; Bennett and Zukin , 2004 ) , the physiological interaction of electrical coupling with intrinsic conductances to affect a compensatory output has not been examined . Previous work in the crustacean STG has demonstrated how the synchronized activity of pacemaker cells is dependent on an interaction of intrinsic conductances and electrical coupling ( Szücs et al . , 2000 , 2001; Soto-Treviño et al . , 2005 ) , and that distinct circuits can be brought into synchrony via manipulations of electrical and chemical synapses ( Elson et al . , 1998; Szücs et al . , 2000 , 2009 ) . But none of these studies have addressed the interaction of membrane conductance and electrical coupling in a compensatory context . Similarly , plasticity of electrical synapses has drawn considerable attention after being discovered in the mammalian central nervous system , including the thalamic reticular nucleus ( Landisman and Connors , 2005; Haas et al . , 2011 ) , inferior olive ( Lefler et al . , 2014; Mathy et al . , 2014 ) , and retina ( Kothmann et al . , 2009; Völgyi et al . , 2013 ) . Studies in the thalamic reticular nucleus have suggested that potentially compensatory changes in coupling are important to maintain network stability as large changes in intrinsic excitability occur across development ( Parker et al . , 2009 ) . These discoveries increased awareness of the complex functional roles and plasticity of coupling ( Pereda et al . , 2013; O’Brien , 2014; Haas , 2015 ) , and also spurred research to identify molecular mechanisms that underlie plasticity and maintenance of these structures ( Flores et al . , 2012; Li et al . , 2012; Turecek et al . , 2014 ) . Our study adds to this growing appreciation for plasticity of electrical synaptic connections in the context of homeostatic plasticity . Stable levels of intrinsic neuronal excitability and temporal coordination within networks are critical features across all nervous systems . Underlying both neuronal and network outputs are complex , and often highly variable intrinsic and synaptic properties of constituent neurons ( Marder , 2011; Norris et al . , 2011 ) . Our data demonstrate that the intrinsic variability between cells of the same type can make networks vulnerable to loss of temporal coordination , in this case desynchronization of motor neuron output . Although LC activity was fully resynchronized within 1 hr , recovered LCs never recapitulated their original voltage waveforms . While the intrinsic conductances involved in our manipulation and compensation ( GKd , GBKKCa , GA ) have overlapping functions and characteristics ( Ransdell et al . , 2012 ) , our findings demonstrate that individual conductances are not truly redundant . Degeneracy of ion channel properties leading to this type of relationship has been put forth as a mechanism underlying robustness and adaptability in neural networks ( Tononi et al . , 1999; Marder and Goaillard , 2006 ) , but our study suggests physiological limits to neural network compensation and robustness . These limits may themselves be a contributing factor to the nature and progression of pathology in neurodegenerative diseases ( Trasande and Ramirez , 2007; Beck and Yaari , 2008; Small , 2008 ) . We induce desynchronization and compensation in our studies through pharmacological block of a subset of K+ conductances with TEA . However , the precise role these mechanisms play in fully intact biological networks is unclear . Intrinsic conductances can be differentially affected by ubiquitous natural mechanisms such as neuromodulation ( Marder , 2011 , 2012 ) or temperature changes ( Tang et al . , 2012; Marder et al . , 2015 ) . Further , the effectiveness of electrical coupling can be affected by the modulation of intrinsic cellular conductances ( Szabo et al . , 2010 ) . Maintaining reliable synchronization of the output under changing conditions is not trivial , and understanding the robustness and the constraints of homeostatic systems that cope with such perturbations remains an important area for future investigation ( Marder et al . , 2014 ) . Adult male Jonah crabs , Cancer borealis , were shipped overnight from The Fresh Lobster Company ( Gloucester , MA ) . Crabs were maintained in artificial seawater at 12°C until used . Crabs were anaesthetized by keeping them on ice for 30 min prior to dissection . The complete CG was dissected from the animal and pinned out in a Sylgard-lined petri dish in chilled physiological saline ( 440 mM NaCl , 26 mM MgCl2 , 13 mM CaCl2 , 11 mM KCl , and 10 mM HEPES , pH 7 . 4–7 . 5 , 12°C ) . Chemicals were obtained from Fisher Scientific unless otherwise noted . The CG network is comprised of 9 cells: 4 Small Cell ( SC ) pacemaker interneurons which give simultaneous excitatory input to 5 Large Cell ( LC ) motor neurons . Superfusate of SCs can be separated from the anterior LC somata using petroleum jelly wells ( Figure 1A ) . Intact network activity was monitored with intracellular recordings from anterior LCs along with extracellular recording of the network output . For most experiments , the posterior end of the ganglion was maintained in normal physiological saline and protected from TEA superfusate with a barrier of petroleum jelly . The anterior end of the preparation was superfused at a rate of approximately 2 ml/min . A schematic of this experimental setup is shown in Figure 1A . All experiments were performed at 12°C . Extracellular recordings using a Model 1700 Differential AC Amplifier ( A-M Systems , Carlsborg , WA ) were taken with stainless steel pin electrodes from a petroleum jelly well on the ganglionic trunk containing axons of all 9 neurons in the CG . LC spikes on the extracellular traces are easily distinguishable by their large amplitude . The LC somata were desheathed for sharp electrode recordings . Intracellular recordings were made using glass electrodes containing 3 M KCl ( 8–25 MΩ ) and AxoClamp 900A and AxoClamp 2B amplifiers ( Molecular Devices , Sunnyvale , CA ) . Two-electrode voltage clamp ( TEVC ) and two-electrode current clamp ( TECC ) protocols were created and run using Clampex 10 . 3 software ( Molecular Devices ) . Somata were isolated for dynamic clamp experiments by tightening a thread ligature past the anterior branch point on the nerve containing the LC soma . Isolated cells were simultaneously driven with the same current stimulus , described previously ( Ransdell et al . , 2013a ) . Briefly , a stimulus protocol was generated by recording the voltage waveform from a LC somata during intact network activity . This consisted of a 20 s recording from a LC3 soma which included four burst potentials with both pacemaker cell EPSPs and LC back-propagating APs present . In addition to the stimulus current , dynamic clamp artificial coupling current was applied with NetClamp software ( developed in the Fishberg Department of Neuroscience of the Mount Sinai School of Medicine and available at http://gothamsci . com/NetClamp/ ) at a sampling rate of 50 kHz according to the equation: Igap = g* ( Vm1 - Vm2 ) where g is a non-rectifying coupling conductance under experimental control , and the voltage difference between the cells ( Vm1 - Vm2 ) determines the driving force . IA was measured before and after compensation using voltage clamp protocols as described previously ( Ransdell et al . , 2012 ) . Briefly , outward currents were measured from a holding potential of −30 mV and stepped from −50 mV to +5 mV in 5 mV increments in order to measure the high threshold K+ current IHTK which is blocked by TEA . In LCs , IHTK is predominantly a mix of BKKCa and delayed rectifier currents ( Ransdell et al . , 2012 ) . A-Type K+ current ( IA ) was measured by performing an identical voltage clamp steps from a holding potential of -80mV and subtracting IHTK . P/N leak subtraction was used for all TEVC . Coupling in the intact network was measured using TECC in both cells during the sucrose block . Negative current steps ranging from 1–6 nA were injected into one cell at a time while measuring voltage changes in both cells . Coupling coefficients were calculated as the ratio: ( ΔVcoupled cell / ΔV Injected Cell ) . To determine whether intrinsic compensation may be contributing to restoration of synchrony in the biological network , we used a current stimulus protocol simulating realistic network inputs to LCs in order to deliver the same biologically relevant stimulus at 3 time points ( Ransdell et al . , 2013a ) . Using this reversible sucrose block to suspend pacemaking activity , we were able to compare the similarity of responses of LC3 and LC5 to the same current injection at three time points . This allows us to test each cell in isolation with respect to its output waveform , but compensation occurs with full network activity after removal of the sucrose over the course of 1 hr . We measured individual LC responses to current injection in control saline , repeated after 5 min of TEA exposure ( acute ) , and again after 1 hr of TEA exposure . Our modelling results suggested that an increase in IA may act in a manner to accomplish both the decrease in excitability and the restoration of synchrony seen in our biological data . In order to determine whether compensatory changes in IA occur in the intact network , we performed two-electrode voltage clamp in the same LC before and after 1 hr of TEA exposure . In order to track changes in IA in individual LCs we again used reversible sucrose block to perform voltage clamp immediately after acute TEA application and again after 1 hr of TEA perfusion . After voltage clamping anterior LCs , the sucrose was washed out and replaced with physiological saline , allowing the network to resume normal activity . This process was repeated after 1 hr of compensation . Intracellular burst waveforms were considered to begin with the first EPSP from pacemaker activity and ended upon return of the waveform to resting membrane potential . Recordings were analyzed using Clampfit 10 . 3 ( Molecular Devices ) and Spike 2 version 7 ( CED , Cambridge , UK ) software . Statistical analyses were performed using SigmaPlot 11 . 0 . Correlation coefficients ( R-values ) were obtained by a Pearson correlation , and squared to calculate the coefficient of determination ( R2 ) . Most data are 'before and after' effects within the same ganglion or cell , and therefore any two groups were compared with paired t-tests when the data were normally distributed , or Wilcoxon signed rank tests in the case of non-normality . The sample sizes for comparison of waveform synchrony were calculated with power analyses based on projected means and standard deviations from data reported in our previous study with very similar experimental manipulations of TEA exposure of LCs ( Ransdell et al . , 2013a ) , which yielded target sample size of N=6–10 to yield a power of 0 . 8 to 0 . 97 . Sample sizes for changes in network output and changes in IA after TEA exposure were based on similar data in our previous work ( Ransdell et al . , 2012 ) , and yielded target sample sizes of N=5 to achieve a power of 0 . 909 . Power analyses were conducted based on the use of paired t-tests to analyze the data . However , when data were not normally distributed we ended up using a Wilcoxon signed rank test , which was not utilized in our initial power analyses . All sample sizes used in our studies are reported in Figure Legends and/or in the Results section when significance values are reported . To quantify synchrony of LC voltage waveforms , we performed a cross-correlation of the digitized voltage signal from LC pairs , as shown in Figure 2B . The first pacemaker spike was used to define the start of each LC burst , and bursts were considered to have terminated upon return to VRest in the LCs . The coefficient of determination from this cross-correlation ( R2 ) was used to examine how accurately one burst waveform could predict the waveform in another LC ( see Ransdell et al . , 2013a ) . This cross-correlation was performed for every burst across the full time-course of the experiment ( Figure 2C , left ) . A decrease in R2 therefore indicates a loss or reduction in waveform synchrony . R2 values from 10 consecutive bursts were averaged for all data points presented as waveform synchrony data , save for the individual points found in Figure 2B and C . Five different measures of excitability of LCs were calculated using both extracellular and intracellular recordings ( Figure 2D ) . Extracellular recordings from the ganglionic trunk were used to calculate the number of spikes per burst and spike frequency . Because our hypothesis predicts both increased LC excitability and desynchronization with TEA exposure , it should be noted that there is a potential confound in distinguishing these effects based on extracellular analysis alone . Axons of all LCs run through the ganglionic trunk , thus the increased spike count observed could result from an increase in the total number of action potentials , desynchronization of action potentials across LCs , or both . We therefore included three additional measures of excitability that helped clarify the effect . We measured the latency between the onset of SC pacemaker bursting to the first LC spike in each burst ( Figure 2D; SC-LC phase delay ) . Two other measures of excitability were calculated from intracellular voltage changes: burst amplitude and total burst depolarization . Burst amplitude was defined as the maximal voltage change from VRest to the highest peak of the burst , and total depolarization is the area under the curve above VRest , measured in mVsec . For all measures of excitability , values from 10 consecutive bursts were averaged at each time-point in each preparation ( N=8 ) . A detailed model was created with eight voltage-dependent conductances ( GA , GKd , GNaP , GCaS , GCaT , GCAN , GSK ( Ca ) , GBK ( Ca ) ) and passive leak channels . These conductances are defined as: A-type potassium ( GA ) , delayed rectifier ( GKd ) , persistent sodium ( GNaP ) , transient calcium ( GCaT ) , slow persistent calcium ( GCaS ) , calcium-dependent non-selective cation ( GCAN ) , two calcium-dependent potassium currents ( GSKKCa and GBKKCa ) , and leak ( GLeak ) . A single compartment model with biological dimensions for soma ( Ransdell et al . , 2010 , 2013b ) was created in NEURON and its capacitance ( Cm ) and leak conductance ( Gleak ) were tuned to match the observed biological membrane time constant ( τ ) and input resistance ( Rin ) . This resulted in a soma with a length of 284 . 87 µm and a diameter of 125 µm , with a capacitance of 2 . 719 µF/cm2 . Channels were then added and their maximal conductances were tuned to match three biological properties observed , i . e . , a ) Total outward current b ) Response to synaptic drive and c ) Response to synaptic drive in the presence of TEA . These results were obtained from experiments performed on ligatured somata of C . borealis LCs ( Ransdell et al . , 2012 , 2013a , 2013b ) . For the network studies another compartment termed Spike Initiation Zone ( SIZ ) was added to the model . This compartment was modelled as a cylinder with a capacitance of 1 µF/cm2 , a length of 400 µm , and a diameter of 8 µm , and contained only sodium , potassium and passive leak channels . This compartment allowed us to obtain spiking activity in the model cells for spike synchrony analysis . The resulting model equations were as follows:CdVdt=−IA−IKd−INaP−ICaS−ICaT−ICAN−ISKKCa−IBKKCa−ILeak ( Soma ) CdVdt=−INa−IKdr−ILeak ( SIZ ) The individual currents were modelled as Ic=gmax , cmphq ( V−Ec ) , where gmax , c is its maximal conductance , m its activation variable ( with exponent p ) , h its inactivation variable ( with exponent q ) , and Ec its reversal potential ( a similar equation is used for the synaptic current but without m and h ) . The kinetic equation for each of the gating functions x ( m or h ) takes the formdxdt=x∞ ( V , [ Ca2+ ]i ) −xτx ( V , [ Ca2+ ]i ) where x∞ is the steady state gating voltage- and/or Ca2+- dependent gating variable and τx is the voltage- and/or Ca2+- dependent time constant . The equations for the active channels in the soma compartment were fit using biological recordings for these currents from the cardiac ganglion of Cancer borealis . These currents were fit as follows: Voltage clamp data obtained with Clampfit were imported into MATLAB ( Mathworks , Natick , MA ) and fit using the MATLAB curve-fitting toolbox . Current data were converted to conductance data by dividing by ( Vm – ERev ) , where ERev was as follows: ENa = +55 mV , EK = −80 mV , ECa = +45 mV , ELeak = −50 mV , and ECAN = −30 mV . The time axis was adjusted to start from 0 for the beginning of the clamp . The following parameterization was used:g ( t ) =∑i=1nAi ( 1−exp ( −tτm , i ) ) ( hi− ( hi−1 ) exp ( −tτh , i ) ) In this equation , Ai=Gi , max×mi was the maximal conductance of the current i multiplied by its voltage-dependent steady-state activation ( mi ) , hi was the steady-state inactivation value , and τm , i and τh , i were the time constants with which activation and inactivation reached steady-state , respectively . This fitting procedure assumed that ionic currents were completely deactivated ( m = 0 ) and de-inactivated ( h = 1 ) prior to the onset of the voltage clamp . This was fit to each trace in a voltage clamp experiment , giving the values of each of the four parameters for each test clamp voltage ( Vc ) . These values were then fit for each current as functions of Vc using the general forms as stated below . This procedure yielded equations for the currents recorded in voltage clamp that could be used in simulations according to the Hodgkin-Huxley mathematical formalism . A ( Vc ) =Gmax×m ( Vc ) =Gmax× ( 1+exp ( ( Vc−Vm , 1/2 ) /km ) ) −1h ( Vc ) = ( 1+exp ( ( Vc−Vh , 1/2 ) /kh ) ) −1τm ( Vc ) =τbase , m+τamp , m ( exp ( ( Vc−Vτ1 , m ) /kτ1 , m ) +exp ( ( Vc−Vτ2 , m ) /kτ2 , m ) ) −1τh ( Vc ) =τbase , h+τamp , h ( exp ( ( Vc−Vτ1 , h ) /kτ1 , h ) +exp ( ( Vc−Vτ2 , h ) /kτ2 , h ) ) −1 All the maximal conductances ( Gi , max ) were in μS , time constants in ms and voltages in mV . Intracellular calcium modulates the conductance of the calcium-activated potassium currents ( BKKCa and SKKCa ) , calcium-activated nonselective cation current ( CAN ) , and influences the magnitude of the inward calcium current in the LC ( Tazaki and Cooke , 1990 ) . A calcium pool was modelled in the LC with its concentration governed by the first-order dynamics ( Prinz et al . , 2003; Soto-Treviño et al . , 2005 ) below:τCad[Ca2+]dt= −F ×ICa− ( [ Ca2+ ]− [ Ca2+ ]rest ) where F=0 . 256 μM/nA is the constant specifying the amount of calcium influx that results per unit ( nanoampere ) inward calcium current; τCa represents the calcium removal rate from the pool; and [Ca2+]rest=0 . 5 μM . Voltage-clamp experiments of the calcium current ( Ransdell et al . , 2013b ) showed the calcium buffering time constant to be around 690 ms ( τCa ) . After creating a nominal LC model ( Tables 1 , 2 ) , we wanted to search the conductance space for other possible conductance combinations that might exhibit appropriate LC output . The properties that had to be maintained were; a ) Input Resistance ( Rin ) and Resting Membrane Potential , b ) Pre-TEA and Post TEA response to current injection c ) Response to Synaptic drive obtained from biological cell . 10 . 7554/eLife . 16879 . 010Table 1 . Nominal model conductance values:DOI: http://dx . doi . org/10 . 7554/eLife . 16879 . 010ConductanceValue ( S/cm2 ) Leak2e-4A6e-4BKKCa7 . 3e-3g1_Kd3e-4g2_Kd3 . 5e-5CaS1 . 7e-4CAN1 . 06e-4SKKCa8 . 79e-5CaT1 . 5e-4NaP3 . 06 e-410 . 7554/eLife . 16879 . 011Table 2 . Model current parameters . DOI: http://dx . doi . org/10 . 7554/eLife . 16879 . 011Iionxpx∞τx ( msec ) IAm311+exp ( ( V+21 . 46 ) /−17 . 96 ) 3 . 002+4 . 0731+exp ( ( V+24 . 18 ) /2 . 592 ) h11+exp ( ( V+21 . 14 ) /25 . 99 ) 9 . 434+11 . 71+exp ( ( V+1 ) /5 . 317 ) ICaSm211+exp ( ( V+24 . 75 ) /−5 ) 20+50 . 2exp ( ( V+20 . 25 ) /1 ) h4540+[Ca2+]10 . 02ICaTm11+exp ( ( V+20 ) /−1 . 898 ) 18 . 51−3 . 388exp ( ( V−6 . 53 ) /9 . 736 ) +exp ( ( V+12 . 39 ) /−2 . 525 ) h11+exp ( ( V+55 . 27 ) /6 . 11 ) 20 . 23+40exp ( ( V+23 . 48 ) /−9 . 976 ) +exp ( ( V+5 . 196 ) /10 . 84 ) Ikdm1411+exp ( ( V+24 . 19 ) /−10 . 77 ) 25 . 049+251+exp ( ( V+25 . 84 ) /6 . 252 ) h10 . 3+1−0 . 31+exp ( ( V+15 . 87 ) /5 . 916 ) 550+954 . 9exp ( ( V+10 . 8 ) /−15 ) m2411+exp ( ( V+23 . 32 ) /−10 ) 100+550exp ( ( V+15 ) /12 . 46 ) INaPm311+exp ( ( V+32 . 7 ) /−18 . 81 ) ) 3 . 15+0 . 8464exp ( ( V+0 . 8703 ) /−6 . 106 ) ) ICANw ( 0 . 0002∗[Ca2+]∧ 2 / ( 0 . 0002∗[Ca2+]∧2+0 . 05 ) ) ( 40/ ( 0 . 0002∗[Ca2+]∧2+0 . 05 ) ) ISKKCaw ( 0 . 0001∗[Ca2+]∧2 / ( 0 . 0001∗[Ca2+]∧2+0 . 1 ) ) ( 4/ ( 0 . 0001∗[Ca2+]∧2+0 . 1 ) ) IBKKCaa[Ca2+]1+exp ( ( V−15+0 . 08∗[Ca2+] ) /−15 ) ∗ ( 1+exp ( ( V+5+0 . 08∗[Ca2+] ) /−9 ) ∗ ( 2+[Ca2+]10 . 4b75+[Ca2+]10 . 2F = Faradays constantR = Gas constantV = Membrane voltage[Ca2+] = Calcium concentration The rules used to select the potential parameters were as follows ( based on biological recordings ) : Synaptic Drive response should have an R2 value of at least 0 . 8 or higher when compared to biological Synaptic Drive response . The duration of the pre-TEA response to a 6 nA , 50 ms current injection should be less than 120 ms . Also the peak should be less than −22 mV . The duration of the post-TEA ( GBKKCa and GKd reduced by 90% ) response should be between 255–667 ms and its peak should be greater than −15 mV . A 9-D max conductance parameter space ( 5-fold variation over each conductance except GLeak ) was searched randomly for sets that satisfied the constraints above . We searched 20 , 000 different combinations of parameter sets with these criteria , and most of those which passed did not have a proper termination of activity following current injection ( i . e . , did not return to Vrest ) . We concluded this was due to an inappropriate relationship between ICAN and ISKKCa . Subsequent trials revealed that a given ratio range ( ~1:0 . 83 respectively ) of these two currents was necessary for proper termination of the activity . Larger ratios cause Vrest to be higher due to the reversal potential ( −30 mV ) of CAN current . A higher fraction of ISKKCa ( reversal potential −80mV ) caused a large AHP after termination and reduced the duration of the post-TEA response . Using the updated selection criteria with a ratio ICAN to ISKKCa , we found 180 parameter sets that passed . Of these 180 potential model sets , we selected only the ones that had Synaptic Drive response R2 value > 0 . 9 compared to the biological Synaptic Drive response . This resulted in 49 potential parameter sets . Biological data showed that IA and IBKKCa had a negative correlation in their magnitudes in LCs ( Ransdell et al . , 2012 ) . We added this to our criteria for screening potential parameter sets for the network studies . We converted biological IA-IBKKCa current data into factor data by dividing IA and IBKKCa by their respective factor average . GA and GBKKCa values of passed parameters were similarly divided by its average to get its factor data . The biological data were fitted using a linear polynomial from 95% to 70% confidence intervals , in steps of 10% . For network studies we used a 70% confidence interval , which left us with 14 potential parameter sets that represented LC model neurons for use in modeling studies . Our results demonstrate that intact cardiac ganglia are able to compensate for the loss of high-threshold K+ currents and restore both excitability and synchrony within one hour of TEA blockade . We next set out to explore the mechanisms by which excitability and synchrony could be restored in this network . To maximize our ability to interrogate multiple parameters that may be responsible for compensation in this system , we constructed a population of conductance-based biophysical models of the CG network . This allowed us to simulate the TEA conductance blockade and then manipulate individual conductances , both voltage-gated and synaptic , to examine their effects on network excitability and synchrony . Our 14 parameter sets for LCs were used to create 50 random 5-cell networks of LCs , ensuring that the same model LC never appeared twice in the same network . The five cells within a network were then electrically coupled using conductance values tuned to reflect experimental observations of coupling coefficients . Small cell ( SC ) pacemaker drive was simulated as excitatory synapses via the NetStim function in NEURON . Parameters for the model of the synaptic drive onto LCs were tuned to get 6 to 9 spikes in the nominal LC model . It was observed biologically that frequency of SC firing increases within the slow wave oscillation cycle of LCs . Based on these recordings , the model SC burst initially fired at 18 Hz for first 440 ms and then increased to 25 Hz for 560 ms , with the burst terminating at 1000 ms . Our experimental TEA block was simulated in these networks by reducing GBKKCa and GKd conductances by 90% in the 3 anterior LCs based on biological data in LCs ( Ransdell et al . , 2013a ) . We imposed a final set of selection criteria on the randomly generated model networks , rejecting networks that increased synchrony or decreased the total number of spikes after the simulated TEA block , as this was never observed in biological networks . This left 27 networks that reproduced the biological trends and these were used in subsequent analyses to explore potential conductance changes that could restore network synchrony . Somatic burst potentials drive action potentials in LC axons , so divergent burst waveforms would be expected to cause desynchronized spiking . Our biological data qualitatively agreed with this , but a precise quantification of synchrony for all spikes within a burst is subject to many ambiguities . Our model networks easily provided precise spike times for each cell in the network , so we chose to examine actual spike synchrony in the model to complement the burst waveform analysis in the biological preparation . Our analysis considered synchrony for paired anterior LCs with a nominal coupling conductance of 0 . 0182 S using a 25 ms bin width for spike-times ( Wang and Buzsáki , 1996 ) . Spikes occurring in both cells during the same bin were considered synchronized , while spikes that did not bin together were tallied as desynchronized . Using the definition of synchrony listed in the next section , these randomly generated model networks exhibited 'control' synchrony scores ranging from 0 . 642 to 1 . 0 with a median value of 0 . 915 ( matching data in biology ) , where 1 . 0 represents perfect spike synchrony . In the models , spike synchrony between two cells was calculated based on spike times ( Wang and Buzsáki , 1996 ) . Spike times were recorded from each LC’s Spike Initiation Zone ( SIZ ) . The simulation time was divided into 25 ms bins . After initializing all bins to zero , each cell spike was added to the corresponding bin . Synchrony ( SY ) between two cells A and B was calculated using the following equation:SYAB= ∑l=1kA ( l ) B ( l ) ∑l=1kA ( l ) * ∑l=1kB ( l ) where l is the current bin and k is the maximum number of bins . Spikes occurring in both cells during the same bin were considered synchronized , while spikes that did not bin together were tallied as desynchronized . To compare the measures of spike and waveform synchrony in model networks , model waveform synchrony measures ( Figure 3 ) were performed as described above on filtered voltage traces ( Gaussian lowpass filter , 15 Hz cutoff frequency; Clampfit 10 . 3 ) that remove the influence of the axonal spikes on the voltage waveforms . Statistical analyses were performed in SigmaPlot v11 . 0 . The effects of changing Gmax on the number of spikes and synchrony among TEA-'treated' model neurons were tested with paired t-tests . Analyses changes in spike number and spike synchrony with changing coupling and synaptic strengths ( Figure 5 ) were analyzed with One-Way ANOVAs with post-hoc pairwise comparisons between a given percent change and the TEA case conducted via Holm-Sidak tests .	Neurons can communicate with each other by releasing chemicals called neurotransmitters , or by forming direct connections with each other known as gap junctions . These direct connections allow electrical impulses to flow from one neuron to another via pores in the membranes between the cells . Unlike communication via neurotransmitters , gap junctions are usually thought to be hard-wired and unchanging over the life of the animal . Lane et al . recorded electrical activity in a network of neurons that generates rhythmic heart contractions in the Jonah crab . Neurons in this network usually all fire an electrical impulse at the same time , which is crucial to make sure that the whole heart contracts at the same time . The experiments show that drugs that block potassium channel pores in the membrane cause the neurons to fire too much and at different times to each other . However , the network of neurons soon adapted to the changes caused by the drugs and returned to working as normal . Mimicking these changes in a computer model of the neuron network , together with experimental data , showed that changes to the gap junctions play a major role in restoring normal activity to the network . The next step following on from this research is to understand how a network of neurons ‘senses’ that it is not working normally and changes its electrical activity .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Synergistic plasticity of intrinsic conductance and electrical coupling restores synchrony in an intact motor network
In multicellular organisms , proteins of the extracellular matrix ( ECM ) play structural and functional roles in essentially all organs , so understanding ECM protein organization in health and disease remains an important goal . Here , we used sub-diffraction resolution stochastic optical reconstruction microscopy ( STORM ) to resolve the in situ molecular organization of proteins within the kidney glomerular basement membrane ( GBM ) , an essential mediator of glomerular ultrafiltration . Using multichannel STORM and STORM-electron microscopy correlation , we constructed a molecular reference frame that revealed a laminar organization of ECM proteins within the GBM . Separate analyses of domains near the N- and C-termini of agrin , laminin , and collagen IV in mouse and human GBM revealed a highly oriented macromolecular organization . Our analysis also revealed disruptions in this GBM architecture in a mouse model of Alport syndrome . These results provide the first nanoscopic glimpse into the organization of a complex ECM . Kidney glomeruli are specialized capillary tufts responsible for filtration . They contain the glomerular filtration barrier ( GFB ) , an intricate structure that constitutes the permselective barrier between the bloodstream and the urine . The GFB is organized into three major layers: ( 1 ) endothelial cells with fenestrations that line the glomerular capillaries; ( 2 ) unique epithelial cells called podocytes that extend many ‘foot’ processes that interdigitate with those of adjacent podocytes and are linked to each other via a specialized cell/cell junction called the slit diaphragm; and ( 3 ) the glomerular basement membrane ( GBM ) , which is composed of extracellular matrix ( ECM ) proteins secreted by the endothelial cells and podocytes that flank it ( Miner , 2011 ) . The GFB’s major physiological role is to allow the unimpeded passage of water and small solutes ( including metabolic waste products ) from the bloodstream into the urinary space , while restricting the passage of plasma proteins such as albumin and immunoglobulins . Although the exact mechanism whereby the GFB functions in this manner is the subject of much controversy , it is clear that injuries or genetic defects in any one of the three layers can cause GFB malfunction that results in increased levels of protein in the urine , a hallmark of diverse kidney diseases . In this study we focus on the ultrastructure and molecular organization of the GBM and its connections to podocytes and endothelial cells . The GBM contains specific isoforms of laminin ( laminin α5β2γ1 , or LM-521 ) , type IV collagen ( primarily the collagen α3α4α5 ( IV ) network ) , heparan sulfate proteoglycan ( HSPG ) ( primarily agrin ) , and nidogen ( Miner , 2012 ) . Laminin and type IV collagen self-polymerize into networks that are connected to each other and/or to cell surface receptors by HSPGs and nidogen; in addition , some receptors can bind directly to laminin or type IV collagen ( Yurchenco et al . , 2004 ) . A great number of biochemical studies of ECM proteins have revealed the molecular basis for their self-assembly and interactions with each other and with cellular receptors ( reviewed in Yurchenco , 2011 ) , and detailed structures are available for specific domains of several ECM proteins ( Khoshnoodi et al . , 2006; Carafoli et al . , 2012 ) . However , a systematic analysis of the actual spatial arrangement of ECM components within basement membranes , with respect to each other and to their receptors , has not yet been performed in vivo . Studying the molecular architecture of ECM protein arrangement poses particular challenges in the context of the GBM . First , the GFB with the GBM cannot yet be reconstituted in vitro . Second , the ∼200-nm thickness of a typical mouse GBM is close to the diffraction limited resolution of conventional light microscopy , thus preventing its use to analyze GBM ultrastructural and macromolecular organization . Transmission electron microscopy ( EM ) studies show electron dense and translucent layers ( lamina densa and lamina lucida ) , but it is unclear whether these reflect an organization of proteins in the GBM ( Chan et al . , 1993 ) . The ability to decipher ECM protein organization thus will require methods with nanometer resolution and high efficiency molecular labeling . Recently , the invention of ‘super-resolution’ light microscopic approaches have broken the diffraction limit of light microscopy , combining a nanoscopic image resolution with several advantages of fluorescence microscopy , such as high efficiency and multi-color labeling ( reviewed in Huang et al . , 2009; Ji et al . , 2008; Hell , 2009 ) . In particular , single molecule localization-based imaging methods such as ( fluorescence ) photoactivable localization microscopy ( ( F ) PALM ) and stochastic optical reconstruction microscopy ( STORM ) ( Betzig et al . , 2006; Hess et al . , 2006; Rust et al . , 2006 ) achieve a molecular scale resolution that can bridge our understanding of supramolecular complexes from the molecular/biochemical to the cellular scales . The strength of various sub-diffraction microscopy methods to dissect the nanoscale architecture of multi-protein complexes has been exemplified recently in the case of neuronal synapses ( Dani et al . , 2010 ) , adhesion complexes ( Kanchanawong et al . , 2010 ) , centrosomes ( Lau et al . , 2012; Lawo et al . , 2012; Mennella et al . , 2012 ) and the nuclear pore complex ( Szymborska et al . , 2013 ) . Given the importance of ECM proteins in tissue morphogenesis and in diverse human diseases ( Miner and Yurchenco , 2004 ) , in this study we investigated the ECM protein organization within the GBM . Using a panel of antibodies , we performed 3D , multichannel STORM in kidney sections and determined the positions of well-defined protein epitopes with nanometer precision . Our data reveal that the GBM is composed of two layers of laminin and agrin molecules , with their N terminal domains oriented towards the center of the GBM . Using transgenic mice expressing a human laminin chain we observed a striking difference in laminin dynamics within the GBM . Results from our imaging approach provide an in situ model of GBM supramolecular organization and align remarkably well with biochemical interactions gleaned from in vitro studies . Finally , we observed distinct distributions of collagen IV molecules and their disruption in a mouse model of Alport syndrome , a hereditary disease of the GBM caused by collagen IV mutations . To reveal supramolecular organization within the GBM , we performed STORM on glomeruli in kidney sections using antibodies conjugated to Alexa 647 , a bright fluorescent photoswitchable dye ( Bates et al . , 2007; Dani et al . , 2010 ) . The dense protein network within the GBM results in a higher fluorescence background and significant light scattering . This poses a major hurdle in obtaining high resolution STORM data from kidney sections . We therefore optimized the fixation and tissue sectioning methods extensively . These preliminary studies established that a modification of the Tokuyasu method of cryo-embedding and cryosectioning at 200 nanometers ( nm ) thickness was optimal ( Tokuyasu , 1973 ) . We began with studies of agrin , a ∼200 kDa protein that is the major HSPG component of the GBM ( Groffen et al . , 1998 ) . Agrin’s N-terminus ( agrinN ) binds to laminin via the laminin γ1 chain’s coiled-coil domain ( Kammerer et al . , 1999 ) , whereas agrin’s C-terminus ( agrinC ) binds to integrins and dystroglycan ( Singhal and Martin , 2011 ) , receptors present on both podocytes and endothelial cells . To investigate agrin’s organization in the GBM , we first labeled kidney sections with an antibody to agrinC . In contrast to conventional immunofluorescence images of agrin , STORM resolved two distinct layers of agrinC within the GBM ( Figure 1A , B ) . To quantitatively document agrinC’s distribution , we digitally selected multiple regions across glomeruli . AgrinC localizations in each region were fitted with a double Gaussian distribution , and the mid-position between the peaks was identified from each image and used as a reference to align subsequent regions . The cumulative localizations obtained from 80 regions are displayed as a histogram ( Figure 1C ) . The gap between agrinC layers , measured as the peak-to-peak distance , showed a mean of 137 . 9 nm ± 1 . 9 nm standard error of mean ( SEM ) for the 80 regions; the distance was 132 . 7 nm ( ±0 . 9 nm ) across more than 600 GBM regions ( Figure 1D ) . 10 . 7554/eLife . 01149 . 003Figure 1 . STORM and STORM-EM image correlation of the mouse GBM . Conventional fluorescence ( A ) and STORM ( B ) images of a kidney glomerular capillary loop labeled with an antibody to agrin . ( C ) Projection histogram of agrin STORM localizations accumulated from 80 different capillary regions . ( D ) AgrinC peak-to-peak distance measured across 604 regions in this study . ( E and F ) 2-color STORM and quantification of agrinC and podocaylxin labeling along a capillary region . ( G ) Platinum deep etch replica prepared from the section shown in ( E ) and imaged by EM . ( H ) Overlay of the STORM and EM images shows ultrastructural features such as podocyte foot processes ( fp ) , endothelial cells ( en ) and the GBM . Podocalyxin labeling is seen along the foot process periphery and agrin localization within the GBM . Figure 1—figure supplement 1 shows a wide field podocalyxin-agrin STORM image overlayed with its EM correlation and Figure 1—figure supplement 2 shows a schematic overview of the STORM–EM correlation procedure . ( I and J ) Immuno-gold labeling and platinum replica EM from a kidney section confirm podocalyxin localization to the fp and en sides ( I ) and agrinC localization in two layers in the GBM ( J ) . Figure 1—figure supplement 3 shows a STORM image and histogram of two separate antibodies labeling the C-terminus end of agrin . DOI: http://dx . doi . org/10 . 7554/eLife . 01149 . 00310 . 7554/eLife . 01149 . 004Figure 1—figure supplement 1 . Low magnification image of EM/STORM correlation . Wide-field STORM image of a kidney section labeled with podocalyxin and agrinC , overlaid with a platinum deep etch replica EM image obtained from the same section . Podocyte cell bodies and capillary loops are marked and the blue boxed region is zoomed in Figure 1E–H . Yellow boxes are examples of capillary loop regions used to quantify localizations . DOI: http://dx . doi . org/10 . 7554/eLife . 01149 . 00410 . 7554/eLife . 01149 . 005Figure 1—figure supplement 2 . Schematic showing steps involved in processing of samples . Schematic of the STORM–EM correlation procedure consisting of the following steps: ( 1 ) glass coverslip is coated with carbon and glow discharged , ( 2 ) Tokuyasu cryo-sections are collected on the carbon coated coverslip surface , ( 3 ) immunolabeled sections are inverted onto a glass slide containing STORM imaging buffer and coverslip edges sealed with nail polish . After STORM image acquisition on an inverted microscope ( 4 ) , nail polish and coverslips are removed , fixed with 2% glutaraldehyde ( 5 ) . ( 6 ) Coverslips are quick frozen and coated with platinum/carbon followed by dissolving the glass and tissue underneath the carbon . ( 7 ) Carbon films with platinum replicas are imaged by EM after transferring to a grid . ( 8 ) STORM and EM images are superimposed . DOI: http://dx . doi . org/10 . 7554/eLife . 01149 . 00510 . 7554/eLife . 01149 . 006Figure 1—figure supplement 3 . Similar pattern of staining from two different agrin antibodies . STORM image and projection histogram of two separate antibodies labeling the C-terminus end of agrin . Scale bar: 200 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 01149 . 006 We next imaged the relationship between the GBM and the flanking podocytes and endothelial cells by double labeling with antibodies to agrinC and to the sialoprotein podocalyxin , which is expressed on the surfaces of both endothelial cells and podocytes ( Kerjaschki et al . , 1984 ) . Dual channel STORM imaging showed that the two layers of agrinC were indeed localized between two layers of podocalyxin ( Figure 1E , F and Figure 1—figure supplement 1 ) . To further correlate the STORM molecular localizations , we developed a hybrid STORM-electron microscopy ( EM ) approach ( see schematic in Figure 1—figure supplement 2 ) . After performing STORM , tissue sections attached to the cover glass were subjected to quick-freezing followed by platinum deep-etching ( Heuser , 1981 ) . The tissue and cover glass underlying the platinum replicas were dissolved , and the replicas were transferred to a grid for examination by EM ( Figure 1G ) . Superimposition of the podocalyxin-agrinC STORM image with deep-etch EM confirmed that podocalyxin labeled endothelial cells as well as podocytes , while agrin was localized to the GBM ( Figure 1H ) . The STORM–EM correlation also confirmed that ultrastructural features of the GFB such as podocyte foot processes and the GBM were well-preserved . Lastly , we confirmed these results using immunogold labeling and electron microscopy ( EM ) . Kidney Tokuyasu sections labeled with anti-podocalyxin and anti-agrinC were initially detected using gold-conjugated secondary antibodies followed by deep-etch EM . Similar to its STORM localizations , anti-podocalyxin labeled both podocyte foot process and endothelial membranes ( Figure 1I ) , but initial attempts to detect agrinC resulted in very sparse labeling . We postulated that the gold-labeled secondary antibody might not efficiently penetrate the GBM . To circumvent this problem , we used a third antibody ( goat anti-rabbit ) to bridge the primary antibody ( rabbit anti-agrinC ) and the gold-conjugated antibody ( anti-goat ) . Images generated using this method confirmed the two agrin layers in the GBM ( Figure 1J ) . These data validate the STORM localizations obtained on kidney sections and also confirm the challenges associated with achieving high efficiency immunogold labeling on a highly cross-linked structure such as the GBM . Having established methods for tissue preparation , STORM imaging , and EM correlation , we decided to analyze the molecular organization of the GBM by comparing the positions of various ECM components to each other . Taking advantage of the robust and bimodal distribution of agrinC , we used a localization scheme in which the center position between the two agrinC layers was set as the origin , and the position of a second protein , imaged along with agrinC by 2-color STORM , was plotted on a reference frame that was marked by the agrinC positions at −68 . 9 and +68 . 9 ± 0 . 9 nm towards the endothelial and podocyte sides , respectively . By iterating this procedure across multiple regions and glomeruli , we could determine the position of various protein epitopes across the GBM with a high degree of precision . The various ECM proteins and their epitopes mapped in this study are illustrated in Figure 2—figure supplement 1 . The human GBM seems to be composed of components that are orthologous to those in mouse , but it is at least two times thicker than the mouse GBM . We therefore attempted to map ECM protein epitopes within the human GBM to better understand its architecture and compare with the mouse GBM . We first labeled the integrin β1 ectodomain along with agrinC and observed that , similar to mouse GBM , both protein epitopes localized along the endothelial and podocyte surfaces of the human GBM ( Figure 5A , B ) . It is notable that significantly more agrinC was detected at the podocyte vs the endothelial aspect ( Figure 5B ) ; this could be due to either epitope masking or differences in expression . The peak-to-peak distance of integrin β1 was measured to be 428 ± 14 . 4 nm , consistent with the known thickness of the human GBM . Given these data , and since two separate anti-integrin β1 antibodies showed similar distributions ( Figure 5—figure supplement 1 ) , we selected integrin β1 as a reference and mapped laminin and collagen IV epitopes . We observed that the anti-LMα5-LG epitope labeled two layers close to the endothelial and podocyte sides of the GBM in a location near the integrin β1 epitope ( Figure 5C ) . However , unlike the mouse , where LMα5-LG was restricted to the two sides of the GBM , in human we also observed LMα5-LG towards the central aspect of the GBM . STORM–EM correlation confirmed that the human GBM was intact and the GBM central LMα5-LG localizations could not arise due a poorly preserved GBM or sectioning artifact ( Figure 5—figure supplement 2 ) . Given the inability of cell surface receptors to interact with LMα5-LG domains in these internal sites , it is not clear how internal LM-521 molecules would be organized . We speculate that LN domain interactions with the subepithelial and subendothelial LM-521 networks could be involved . 10 . 7554/eLife . 01149 . 013Figure 5 . Molecular organization of the human GBM . ( A ) Single channel STORM of integrin β1-labeled human GBM sections . Figure 5—figure supplement 1 shows two channel STORM using two different integrin β1 antibodies . Human kidney sections were labeled with antibodies to integrin β1 ectodomain and: agrinC ( B ) , laminin α5 ( LG ) ( C ) , collagen α1α1α2 ( IV ) ( D ) , collagen α3α4α5 ( IV ) -NC1 ( E ) and collagen α3α4α5 ( IV ) -periN ( F ) . Figure 5—figure supplement 2 shows anti-laminin α5-LG labeling of the human GBM and a replica EM image of the same section . Figure 5—figure supplement 3 shows anti-collagen α1α1α2 ( IV ) and collagen α3α4α5 ( IV ) -NC1 labeling of a human GBM region . Scale bar: 200 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 01149 . 01310 . 7554/eLife . 01149 . 014Figure 5—source data 1 . Calculated relative position of molecules in the human GBM . DOI: http://dx . doi . org/10 . 7554/eLife . 01149 . 01410 . 7554/eLife . 01149 . 015Figure 5—figure supplement 1 . Similar pattern of staining obtained with two different human integrin β1 antibodies . Double channel STORM using two different integrin β1 antibodies to label human GBM sections ( MAB13: blue , TS2/16: red ) . The projection histogram reveals a peak-to-peak distance of ∼400 nm . Scale bar: 200 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 01149 . 01510 . 7554/eLife . 01149 . 016Figure 5—figure supplement 2 . Laminin a5 stains multiple layers in normal human GBM . ( A and B ) STORM image of laminin α5-LG labeling of the human GBM and a replica EM image of the same section . Scale bar: 200 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 01149 . 01610 . 7554/eLife . 01149 . 017Figure 5—figure supplement 3 . Distinct localization of Collagen α1α1α2 ( IV ) and Collagen α3α4α5 ( IV ) in human GBM . Collagen α1α1α2 ( IV ) and collagen α3α4α5 ( IV ) -NC1 labeling of a human GBM region . Since agrin or integrin were not used for orientation , the localizations from the region depicted in the image were quantified . Scale bar: 200 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 01149 . 017 We next imaged collagen α3α4α5 ( IV ) NC1 and peri-N epitopes along with integrin β1 in the human GBM ( Figure 5E , F ) . Both collagen α3α4α5 ( IV ) epitopes were localized at the center of the GBM , with a clear separation from integrin β1 . Staining for collagen α1α1α2 ( IV ) using an antibody to the N-terminal part of the collagenous domain ( Foellmer et al . , 1983 ) showed that it was localized to the endothelial side of the GBM ( Figure 5D ) , consistent with previous results using conventional immunofluorescence microscopy ( Butkowski et al . , 1989 ) . Further , co-labeling for α3α4α5 ( IV ) -NC1 and α1α1α2 ( IV ) epitopes showed that the two collagen networks had distinct adjacent distributions ( Figure 5—figure supplement 3 ) , with the α1α1α2 ( IV ) network occupying the space between the central α3α4α5 ( IV ) layer and the endothelial surface of the GBM . These data indicate that the increased thickness of the human GBM can be attributed to an expanded collagen α3α4α5 ( IV ) network that is significantly wider than that in the mouse GBM together with an additional layer of LM-521 . The distinct distributions of the components of the GBM prompted us to assess whether the organization of the GBM is disrupted in a specific disorder of the basement membrane . We therefore analyzed a mouse model of autosomal recessive Alport syndrome that lacks the collagen α3α4α5 ( IV ) network due to a Col4a3 null mutation ( Miner and Sanes , 1996 ) . In both humans and mice , the lack of this network results in a compensatory increase in the expression of collagen α1α1α2 ( IV ) , which is normally found at low levels in adult GBM ( Kashtan and Kim , 1992; Miner and Sanes , 1996 ) . Despite this compensation , the GBM becomes segmentally split and thickened , and this is associated with hematuria , proteinuria , and progressive renal failure ( Hudson et al . , 2003 ) . To investigate the architecture of the GBM in this disease model , we examined the organization of agrinC and collagen α1α1α2 ( IV ) . In contrast to the wild-type GBM , the Alport GBM showed segments of capillary loops where the two layered organization of agrinC was disrupted , while in some segments it was intact ( Figure 6A , B ) . The localization of collagen α1α1α2 ( IV ) was also altered , with its distribution spread across the width of the GBM and no longer restricted to the endothelial side . This altered collagen α1α1α2 ( IV ) distribution was evident even in capillary loop segments that showed a relatively preserved agrinC organization ( Figure 6C , D ) . Thus , we conclude that an intact collagen α3α4α5 ( IV ) network helps to maintain agrin and collagen α1α1α2 ( IV ) organization in the healthy GBM . 10 . 7554/eLife . 01149 . 018Figure 6 . Breakdown of the GBM molecular architecture in a mouse model of Alport syndrome . ( A and B ) STORM and EM images of a capillary loop from a collagen α3 ( IV ) knockout ( KO ) mouse kidney labeled with agrinC shows a thin GBM with 2-layered agrin ( single arrow ) as well as a breakdown of the 2-layered agrin labeling pattern ( double arrow ) and a thick irregular GBM stretch showing disorganized , diffuse agrinC labeling ( arrowheads ) . ( C and D ) Images and quantification of capillary loops selected from collagen α3 ( IV ) KO and heterozygous ( HET ) kidney that show two layers of agrinC . Despite the intact agrin layers , collagen α1α1α2 ( IV ) shows an atypical distribution spread across the GBM in the KO vs a single peak in the HET littermate control . DOI: http://dx . doi . org/10 . 7554/eLife . 01149 . 018 Light microscopic imaging shows the components of the GBM as diffusely distributed , suggesting either an amorphous structure or an organization below the resolution of conventional light microscopy . While transmission EM imaging of the glomerulus with OsO4 demonstrates an electron dense layer in the GBM , the significance of this finding has been unclear . Immunogold-EM has been deployed to study ECM protein organization ( Miosge et al . , 1999 ) , but constraints of sample preparation , antibody accessibility , and quantitation prevent its extensive application . In our study , we performed STORM as well as EM and immunogold-EM using the same tissue and antibody labeling procedures . This allowed us to validate data using both methods and also illustrated the challenges associated with immunogold EM . Super resolution fluorescence microscopic methods have now been widely used to resolve subcellular structures in cultured cells , but the use of these methods to analyze tissue has been challenging . Single molecule probe based methods ( F ) PALM and STORM were originally developed using photoswitchable fluorescent proteins ( Betzig et al . , 2006; Hess et al . , 2006 ) or photoswitching properties of fluorescent dyes ( Rust et al . , 2006 ) . Exogenous expression of multiple photoswitchable fusion proteins and their functional validation is not yet feasible for studying multi-protein complexes in mammalian tissues . Using a number of well characterized antibodies provides a practical alternative for tissue imaging . Thus , we focused on using photoswitchable dye-conjugated antibody labeling for our experiments . Fluorescence imaging of tissue also presents special challenges , mainly regarding the methods of fixation , labeling , sectioning and the method of electron microscopic correlation . Sub-diffraction microscopy methods typically map the positions of labeled proteins; however , putting these in the context of membranes and other ultrastructures requires EM . Recently , a few studies have described super-resolution imaging of photoswitchable fusion proteins expressed in cultured cells and in C . elegans , followed by plastic embedding and EM correlation ( Betzig et al . , 2006; Watanabe et al . , 2011; Kopek et al . , 2012 ) . Conventional fluorescence microscopy of resin embedded tissue sections labeled with antibodies has also been described ( Micheva and Smith , 2007 ) . Here , we tested a variety of fixation methods , sectioning thicknesses and electron microscopic approaches . Our best results were obtained when we collected 200-nm thickness Tokuyasu cryosections directly onto a coverglass . Since the Tokuyasu method was originally developed for the preparation of tissue for EM imaging , we were able to use freeze etch EM after STORM , to produce correlated images . We believe that the method described here is a breakthrough as it should be useful for the imaging of a wide variety of other tissues . The results from our study demonstrated that the ECM within the GBM , far from being amorphous , is a highly structured and laminated amalgam of interacting protein networks ( Figure 4 ) . We showed that agrin and laminin-521 each form two layers within the mouse GBM . In each layer , both proteins were oriented with their C-terminal LG domains close to the adjacent cell membranes , whereas the more N-terminal laminin short arm epitopes and the agrin N-terminus were localized towards the interior of the GBM . Cell culture and biochemical data support these observations: the LG domains of laminin α5 and agrin are known to interact with cell surface receptors ( Martin and Sanes , 1997; Yurchenco et al . , 2004; Kikkawa et al . , 2007 ) , and the agrin N-terminal domain interacts with laminin γ1 near the center of the coiled-coil domain ( Kammerer et al . , 1999 ) . But the ∼140 nm vertical axis of laminin must be reconciled with our definitive demonstration in mouse that there are two separate networks that meet in the middle of the GBM; the ∼200 nm wide mouse GBM cannot contain two fully extended laminin trimers , which would approach a 300 nm span . Models predicting either a bent laminin conformation ( Yurchenco et al . , 2004 ) or an extended laminin situated at an oblique angle relative to the adjacent cells would be consistent with our data . One of our most important and unexpected findings was the mapping of the collagen α3α4α5 ( IV ) N- and C-terminal domains to near the center of the GBM , a position that in human GBM is too distant from the positions of the integrin β1 extracellular domain ( extending from both podocytes and endothelial cells ) for a meaningful ligand–receptor interaction . In contrast , both agrinC and LMα5LG localized near integrin β1 , implicating them as the likely biologically relevant ligands . At the endothelial aspect of the GBM , collagen α1α1α2 ( IV ) was near enough to the cell to be capable of binding integrin β1 . In our Alport syndrome mouse model , which lacks the collagen α3α4α5 ( IV ) network , collagen α1α1α2 ( IV ) was found at increased levels throughout the width of the GBM . One implication of this is that podocytes should be exposed to unfamiliar collagen IV ligands , which might then lead to podocyte injury and the glomerulosclerosis that is observed in the disease . The GBM’s collagen α3α4α5 ( IV ) is secreted solely by podocytes ( Abrahamson et al . , 2009 ) , and our results show that it eventually becomes concentrated near the center of the GBM , away from the podocytes . The secreted collagen α3α4α5 ( IV ) protomers are thus likely able to permeate through the LM-521 and agrin networks adjacent to the podocyte , against the flow of filtrate , to reach the center of the GBM . Given the greater thickness of the human vs the mouse GBM , this process should either be faster or persist for a longer period in human glomeruli . In contrast , the collagen α1α1α2 ( IV ) protomers made by endothelial cells form a network closely juxtaposed to the endothelium . These different collagen IV protomer behaviors may be fundamental to establishing and maintaining the GBM and could be related to its splitting and thickening in the setting of Alport syndrome . Kidneys were isolated from mice after transcardial perfusion with phosphate buffered saline ( PBS ) containing 4% ( wt/vol ) Paraformaldeyde ( PFA; EM Sciences , Hatfield , PA ) . After dissection , kidneys were cut into smaller pieces and fixed overnight with 4% PFA , followed by washing off excess PFA with PBS . Kidney pieces were immersed overnight at 4°C in a cryoprotectant solution of 2 . 3 M sucrose + 10% polyvinylpyrrolidone ( PVP ) in 0 . 1 M PIPES ( pH = 7 . 2 ) . Cryoprotected tissues were mounted on a metal sectioning pin and frozen by immersion in liquid nitrogen . All transgenic and knockout mice used have been previously described . These included tetO7-regulated human Laminin α5 cDNA ( Goldberg et al . , 2010 ) , Nphs2-rtTA ( Shigehara et al . , 2003 ) , Tie2-Cre ( Koni et al . , 2001 ) , Rosa26-LoxP-neo-LoxP-rtTA ( Belteki et al . , 2005 ) , and Col4a3 null ( Miner and Sanes , 1996 ) mice . Induction of human laminin α5 expression was achieved by feeding pregnant females doxycycline chow ( 0 . 15% ) beginning when pregnancy was apparent and continuing after birth and after weaning . De-identified human kidney samples from individuals with no known history of kidney disease were obtained through the Washington University George O’Brien Center for Kidney Disease Research . Samples were fixed overnight in 4% paraformaldehyde and cryoprotected as described above . To capture tissue sections , acid cleaned and air-dried No . 1 . 0 coverglass were carbon coated for 1 min at a pressure of 2 × 10−6 mbar . Carbon coated coverglass were glow discharged at 2 × 10−2 mbar immediately before collecting tissue sections . Frozen tissues were sectioned at ∼200-nm thickness on a Leica EM-FC6 ultracryomicrotome equipped with a diamond knife and sections were collected on the carbon coated coverslips . Sections were re-fixed for 20 min at room temperature ( RT ) with 4% PFA , followed by three washes in PBS and excess PFA was quenched using 50 mM glycine in PBS . Sections were further processed for immunolabeling in the following manner: ( 1 ) blocked overnight at 4°C using 2% bovine serum albumin ( BSA ) in PBS , ( 2 ) primary antibodies diluted in 2% BSA-PBS were applied overnight at 4°C followed by 3 × 20 min PBS washes at RT , ( 3 ) secondary antibodies diluted 3% BSA-PBS were applied at RT for 2–3 hr followed by PBS washes . ( 4 ) Immunolabeled sections were post-fixed using 3% PFA+ 0 . 05% Glutaraldehyde ( EM Sciences ) in PBS , washed in PBS in used for STORM . The primary antibodies used in this study and their approximate concentrations/dilutions used are shown in Supplementary file 1 . Secondary antibodies for STORM were purchased from Jackson Immunoresearch and were custom conjugated to Alexa647 reporter dye and either Alexa405 , Cy2 or Cy3 activator dyes as described before ( Bates et al . , 2007 ) . STORM setup and image acquisition scheme were similar to that described before ( Dani et al . , 2010 ) , with the following modifications . Briefly , the STORM rig was constructed around a Nikon Eclipse TiE inverted microscope fitted with the Nikon perfect focus system for focus stabilization , a motorized stage ( Marzhauser ) and a 100X 1 . 4NA objective ( Olympus UPLSAPO ) . Illumination lasers , 642 nm ( Vortran ) , 561 nm , 488 nm ( Coherent , Sapphire ) and 405 nm ( Coherent , Cube ) were shuttered using an acousto-optical tunable filter ( AOTF , Crystal technologies ) . Laser beams were combined , expanded , collimated and focused at the back focal plane of the 100X objective . Total internal reflection fluorescence ( TIRF ) illumination was achieved with an objective type TIRF geometry using a custom-built translation stage . Sections immunolabeled on carbon coverglass were inverted onto a slide containing a drop of imaging buffer containing mercaptoethylamine along with an oxygen scavenger system , and coverglass edges were sealed with nail polish . In all experiments described here , we performed single reporter ( Alexa647 ) -multiple activator STORM as described before ( Bates et al . , 2007; Dani et al . , 2010 ) . Alexa647 images ( ∼10 , 000 images per channel ) acquired by the same objective were separated using a quad band dichroic ZT405/488/561/640rpc and filtered using ET705/72m emission filter ( Chroma ) . Images were captured on an EM-CCD camera ( iXon+ DU897 , Andor ) and analyzed using custom software . Image stacks were fitted with an elliptical Gaussian function to determine the centroid positions of fluorescent pixel intensity peaks . These positions termed ‘STORM localizations’ were rendered as STORM images or analyzed further quantitatively . To quantify STORM localizations and to estimate molecule positions with high precision , multiple GBM regions , each as a ∼800 nm wide window , were selected towards the most peripheral aspect of circular capillary loops , to avoid the mesangial regions where the mesangial matrix meets the GBM . Selected regions were rotated to uniformly orient the podocyte and endothelial sides . STORM localizations from agrinC and a co-labeled molecule , were projected onto the perpendicular axis for each region and fitted with Gaussian functions to identify the centroid positions of the STORM localizations . The midpoint between the two agrinC centroid positions was set as zero and the position of the second molecule of interest was identified by the distance of its centroid position with respect to the zero . This procedure was iterated over multiple regions to estimate the mean position , standard error of the mean and standard deviation for each molecule . To view the quantitative profile of STORM localizations , a histogram was generated from each region by projecting localization points onto a line perpendicular to the long axis of the capillary loop . Each histogram was shifted to align them by their zero position and an accumulated histogram from several regions was constructed by adding up all localizations . The number of regions used to generate the histograms , the mean position of each molecule , standard error of the mean and standard deviation are reported as Figure 4—source data 1 and Figure 5—source data 1 . All quantifications were performed using custom scripts in Matlab and the data were plotted in Origin software . Quick-freeze deep-etch EM was performed according to published protocol , with minor modifications ( Heuser , 1980 ) . After STORM , nail polish from the coverglass was carefully removed by immersing in PBS , and the tissue sections were fixed in 2% glutaraldehyde in100 mM NaCl , 30 mM HEPES and 2 mM CaCl , pH 7 . 2 ( NaHCaCl ) at room temperature . 3 × 3 mm areas of the coverglass containing STORM imaged sections were cut , rinsed in dH2O and frozen by abrupt application of the sample against a liquid helium cooled copper block with a Cryopress freezing machine . Frozen samples were transferred to a liquid nitrogen cooled Balzers 400 vacuum evaporator , etched for 20 min at −80°C and rotary replicated with ∼ 2 nm platinum deposited from a 20° angle above the horizontal , followed by an immediate ∼10 nm stabilization film of pure carbon deposited from an 85° angle . Replicas were floated onto a dish of concentrated hydrofluoric acid and transferred through several rinses of dH20 with a loopful of Photo-flo , picked up on Luxel grids ( Luxel , Friday Harbor , WA ) , and photographed on a JEOL 1400 microscope with attached AMT digital camera . The glomeruli imaged by EM were matched with the corresponding STORM images , and the two images were superimposed using Adobe Photoshop . The STORM–EM correlation procedure is illustrated in Figure 1—figure supplement 2 . Immunogold labeling was done on cryosections collected on a 3 × 3 mm coverglass followed by detection with 12 nm colloidal gold affinity purified secondary antibodies ( Jackson Immuno Research ) diluted 1: 15 in PBS/2% BSA over 2 hr at RT , rinsed with three 10 min washes of PBS , followed by fixation in 2% glutaraldehyde in NaHCaCl . Prior to freezing , coverglass was rinsed in dH20 , frozen and platinum replicas made as described above .	The blood that flows through the body must be continually filtered to remove waste products and to ensure that it contains optimal levels of water and salts . Filtration is performed inside the kidneys by tufts of small blood vessels called glomeruli . These glomerular capillaries allow water and waste products to pass from the blood into the urine , while holding back proteins and blood cells . The wall of a glomerular capillary consists of two layers of cells flanking a third layer called the glomerular basement membrane . If any of these layers malfunctions , it becomes possible for proteins to pass into the urine . This is a clear sign of kidney disease . The basement membrane is composed of proteins secreted by the two layers of cells , but little was known about how these proteins are organized . Now , Suleiman et al . have adapted a new form of high-resolution optical microscopy called STORM to study the structure of the glomerular basement membrane in both mouse and human kidney tissue . By combining data from STORM and electron microscopy , Suleiman et al . showed that the proteins in the glomerular basement membranes of both species are arranged similarly to form a distinctive layered structure . This suggests that the organization of the basement membrane plays a critical role in its function . The technique was used to demonstrate that proteins were not organized in the glomerular basement membrane in tissue samples taken from mice suffering from Alport syndrome , a genetic disorder of the kidneys . In addition to suggesting that the disorganization of basement membranes may play an important role in disease , this work also provides a method for investigating the structure of the basement membrane in diverse types of tissue .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2013	Nanoscale protein architecture of the kidney glomerular basement membrane
Blood is arguably the most important bodily fluid and its analysis provides crucial health status information . A first routine measure to narrow down diagnosis in clinical practice is the differential blood count , determining the frequency of all major blood cells . What is lacking to advance initial blood diagnostics is an unbiased and quick functional assessment of blood that can narrow down the diagnosis and generate specific hypotheses . To address this need , we introduce the continuous , cell-by-cell morpho-rheological ( MORE ) analysis of diluted whole blood , without labeling , enrichment or separation , at rates of 1000 cells/sec . In a drop of blood we can identify all major blood cells and characterize their pathological changes in several disease conditions in vitro and in patient samples . This approach takes previous results of mechanical studies on specifically isolated blood cells to the level of application directly in blood and adds a functional dimension to conventional blood analysis . Blood is responsible for the distribution of oxygen and nutrients , and centrally involved in the immune response . Consequently , its analysis yields crucial information about the health status of patients . The complete blood count , the analysis of presence and frequency of all major blood cells , constitutes a basic , routine measure in clinical practice . It is often accompanied by analysis of blood biochemistry and molecular markers reflecting the current focus on molecular considerations in biology and biomedicine . An orthogonal approach could be seen in the study of the overall rheological properties of blood . It is evident that the flow of blood throughout the body will be determined by its physical properties in the vasculature , and their alterations could cause or reflect pathological conditions ( Lichtman , 1973; Baskurt and Meiselman , 2003; Popel and Johnson , 2005 ) . In this context , blood is a poly-disperse suspension of colloids with different deformability and the flow properties of such non-Newtonian fluids have been the center of study in hydrodynamics and colloidal physics ( Lan and Khismatullin , 2012 ) . Probably due to the dominant importance of erythrocytes , at the expense of sensitivity to leukocyte properties , whole blood rheology has not resulted in wide-spread diagnostic application . Focusing on the physical properties of individual blood cells has suggested a third possibility to glean maximum diagnostic information from blood . Various cell mechanics measurement techniques , such as atomic force microscopy ( Worthen et al . , 1989; Rosenbluth et al . , 2006; Lam et al . , 2008 ) , micropipette aspiration ( Lichtman , 1973; Lichtman , 1970; Dombret et al . , 1995; Ravetto et al . , 2014 ) or optical traps ( Lautenschläger et al . , 2009; Ekpenyong et al . , 2012; Paul et al . , 2013 ) , have been used to show that leukocyte activation , leukemia , and malaria infection , amongst many other physiological and pathological changes , lead to readily quantifiable mechanical alterations in the major blood cells ( Worthen et al . , 1989; Lautenschläger et al . , 2009; Schmid-Schönbein et al . , 1973; Suresh et al . , 2005; Rosenbluth et al . , 2008; Bow et al . , 2011; Gossett et al . , 2012 ) . These proof-of-concept studies have so far been done on few tens of specifically isolated cells . This line of research has not progressed towards clinical application for lack of an appropriate measurement technique that can assess single-cell properties of sufficient number directly in blood . This report aims to close this gap by presenting a novel approach for high-throughput single-cell morpho-rheological ( MORE ) characterization of all major blood cells in continuous flow . Mimicking capillary flow , we analyze human blood without any labeling or separation at rates of 1000 cells/sec . We show that we can sensitively detect morphological and rheological changes of erythrocytes in spherocytosis and malaria infection , of leukocytes in viral and bacterial infection , and of malignant transformed cells in myeloid and lymphatic leukemias . Readily available quantitative morphological parameters such as cell shape , size , aggregation , and brightness , as well as rheological information of each blood cell type with excellent statistics might not only inform further investigation of blood as a complex fluid . It also connects many previous reports of mechanical changes of specifically isolated cells to a measurement now done directly in blood . As such , it adds a new functional dimension to conventional blood analysis — a MORE complete blood count — and , thus , opens the door to a new era of exploration in investigating and diagnosing hematological and systemic disorders . In order to establish the normal MORE phenotype of cells found in blood , we obtained venous , citrate-anticoagulated blood of healthy donors , of which 50 µl was diluted in 950 µl of measurement buffer with a controlled elevated viscosity , but without any additional labeling , sorting , or enrichment . The cell suspension was then pumped through a micro-channel not unlike micro-capillaries in the blood vasculature ( Figure 1A ) . Brightfield images of the cells , deformed by hydrodynamic shear stresses in the channel ( Mietke et al . , 2015 ) , were obtained continuously by RT-DC ( Otto et al . , 2015 ) ( see Materials and methods; Video 1 ) . These images revealed distinct differences in overall morphology , brightness , and amount of deformation between all major cell types found in blood ( Figure 1B ) . RT-DC further enabled the continuous , real-time quantification of the cross-sectional area and of the deformed shape ( see detailed description in Materials and methods and Figure 1—figure supplement 1 ) of an , in principle , unlimited number of cells at measurement rates of 100–1000 cells/sec ( Figure 1C ) . For each cell detected and analyzed , an image was saved and the average pixel brightness within the cell determined ( Figure 1D , Figure 1—figure supplement 1 ) . This single-cell MORE analysis of blood revealed distinct and well-separated cell populations in the space spanned by the three parameters ( Video 2 ) . Notably , size and brightness alone — parameters not unlike those accessible by light scattering analysis in standard flow cytometers — were sufficient for the identification of the cell types ( Figure 1D ) , so that deformation , as an additional and independent parameter , was available for assessing their functional changes . The identity of the individual cell populations by size and brightness was established by magnetic cell sorting , controlled by fluorescence immunophenotyping , and subsequent MORE analysis ( Figure 1—figure supplement 2 ) . A key feature is the very clear separation of the abundant erythrocytes ( red blood cells; RBCs ) from other cells as a result of their much greater deformation and lower brightness . This feature gives access to leukocyte properties directly in diluted whole blood , without the potentially detrimental effects of hemolysis ( see Figure 1—figure supplement 3 ) or other separation steps , which are required for analysis with cell mechanics techniques with lower specificity and throughput , or non-continuous measurement . This aspect contributes to the well-established field of hemorheology the possibility to interrogate mechanical properties of all individual blood cells and to specifically investigate their contribution to the overall blood rheological properties . In extensive tests of the variability of this approach , MORE phenotyping yielded identical results in repeated measurements of blood from the same donor , with sodium citrate added as an anti-coagulant and for different storage times ( Figure 1—figure supplement 4 ) , between different donors of both sexes ( Figure 1—figure supplement 5 ) , and blood samples taken at different times during the day ( Figure 1—figure supplement 6 ) . This robustness served to establish a norm for the different cell types ( Figure 1E ) . MORE analysis provided the identity and frequency of all major white blood cells as with a conventional differential blood count ( Figure 1F; Supplementary file 1 ) — obtained from a single drop of blood , with minimal preparation , and within 15 min . Going beyond this current gold-standard of routine blood cell analysis , and importantly also beyond all other single-blood-cell mechanical analysis studies to date , MORE phenotyping allowed the sensitive characterization of pathophysiological changes of individual cells directly in merely diluted whole blood . In the following , we exemplarily demonstrate , in turn for each of the blood cell types , the new possibilities of gaining MORE information from an initial blood test as a time-critical step in generating specific hypotheses and steering further investigation enabled by this approach . Spherocytosis is a prototypical hereditary disease in humans in which genetic changes ( here ankyrin and spectrin mutations ) cause abnormal shape and mechanical properties of erythrocytes . Current diagnosis is based on the detection of abnormal cell shapes in a blood smear , followed up by assessment of the osmotic fragility quantified by Acidified Glycerol Lysis Time ( AGLT ) or by osmotic gradient ektacytometry . These manual assays take time and do not lend themselves to quick , initial screening . MORE analysis of the blood of patients with spherocytosis directly revealed significantly less deformed and smaller erythrocytes than normal ( Figure 2A–C ) as the functional correlate of the cytoskeletal mutation . The differences are so clear ( Figure 2—figure supplement 1 ) that this analysis can serve as a fast primary and cheap screening test for spherocytosis . Detection of such RBC changes would then warrant confirmation by more specific analysis using flow-cytometric detection of Eosin-5-Maleimide staining ( EMA test ) or the direct detection of the mutation by PCR , which require specific preparation , are more expensive , and thus benefit from a strong and clear initial hypothesis . A change in RBC deformability has also been implicated in malaria pathogenesis , since single cells infected by parasites have been reported to be stiffer ( Bow et al . , 2011 ) . This insight has not progressed towards clinical application and the gold standard in malaria diagnosis is still a manual thick blood smear analysis . To evaluate whether MORE analysis could provide a sensitive , automated alternative , we analyzed populations of RBCs exposed in vitro to Plasmodium falciparum ( P . f . ) with a parasitemia ( percentage of actually infected cells ) of 7–8% at time points over the 2 day parasite life cycle . We found a clear , significant , and increasing reduction in the deformation of the entire exposed RBC population detectable after 4 hr ( Figure 2D–F; Figure 2—figure supplement 2 ) . Inspection of the individual cell images revealed the appearance of characteristic features likely associated with the maturation of parasites inside a subset of RBCs ( Figure 2D , E insets; Figure 2—figure supplement 2 ) . These features permitted the direct identification of positively infected cells , whose relative frequency peaked at 36 hr ( Figure 2—figure supplement 2 ) . The separate assessment of overtly infected cells showed an even greater reduction in deformation than observed in the entire exposed population ( Figure 2F; Figure 2—figure supplement 2 ) , which — extrapolating our in vitro results to the situation in vivo — relates to the possibility of clearance of stiff , infected cells from the circulation by the spleen ( Cranston et al . , 1984; Shelby et al . , 2003 ) . However , this small fraction of stiffer cells alone cannot account for the reduced deformation of the whole population , so that a bystander stiffening of exposed but non-infected cells seems involved ( Paul et al . , 2013 ) . Reduced membrane-cytoskeleton interactions have previously been correlated with elliptocytic RBCs and resistance to P . f . infection ( Chishti et al . , 1996 ) . The characteristic biconcave morphology of RBCs can be chemically altered by the use of 2-bromo-palmitate ( 2 BP ) , an efficient inhibitor of palmitoyl acyltransferases ( Biernatowska et al . , 2013 ) . Here , 2 BP-treated RBC samples showed changes in deformation ( Figure 2G ) with a concurrent reduction in P . f . infectivity ( Figure 2H ) , compared to buffer control or RBCs treated with palmitic acid ( PA ) . PA is an analogue of 2 BP that does not inhibit palmitoylation ( Biernatowska et al . , 2013 ) . Since both , 2 BP and PA readily accumulate in the membranes , but only 2 BP causes a reduction in infectivity of P . f . , we suggest that palmitoylation of RBC proteins is important for RBC morphology and infectivity of P . f . While a previous report had found no change in infectability of RBCs treated with 2 BP ( Jones et al . , 2012 ) , the difference could stem from the different RBC receptors involved in invasion by the different parasite clones ( 3D7 vs . HB3 ) , which in turn are differentially affected by palmitoylation . Thus , MORE analysis has the potential not only to simplify , automate , and speed up malaria diagnosis , but also to provide additional quantitative information aiding research on the pathogenesis of the disease ( Koch et al . , 2017 ) . While RBC mechanics has already been used for clinical diagnostics using rheoscopes and ektacytometers for over 40 years ( Schmid-Schönbein et al . , 1973; Cranston et al . , 1984; Reid et al . , 1976 ) , leukocyte mechanics has not been utilized for diagnostic purposes . This is likely due to their increased stiffness compared to RBCs and a lack of convenient techniques capable of sufficiently deforming them in suspension — their physiological state . Until recently , techniques with sufficient throughput , obviating the need for specifically isolating the relevant cells of interest , which always bears the potential of inadvertent cell change ( see Figure 1—figure supplement 3 ) , did not exist . In this sense , the mechanical phenotyping of diagnostic changes of leukocytes directly in diluted whole blood is the most transformative application area of MORE analysis . For example , there have been proof-of-concept studies on the mechanical changes associated with activation of isolated neutrophils showing a stiffening , in line with the pronounced actin cortex that is a hallmark of neutrophil activation ( Worthen et al . , 1989; Rosenbluth et al . , 2008 ) . MORE analysis of the in vitro neutrophil activation in blood with the bacterial wall-derived tripeptide fMLP confirmed that neutrophils were indeed less deformed and smaller within the first 15 min post fMLP treatment . Interestingly , the subsequent time-course showed a reversal to more deformed and larger cells ( Figure 3A , B; Figure 3—figure supplement 1 ) . These observations by themselves do not permit a conclusion about a change in cell stiffness , since a smaller size also leads to less stress acting on the cells in the channel , and less deformation ( Mietke et al . , 2015; Mokbel et al . , 2017 ) . Thus , we also calculated the apparent Young’s modulus of the cells , which increased from E = 742 ± 12 Pa to E = 853 ± 20 Pa ( mean ±SEM . p=0 . 009 , n = 5 ) during the first 15 min , and then subsequently reverted to values statistically indistinguishable but slightly lower than before stimulation ( 15–30 min: E = 717 ± 9 Pa , p=0 . 347; 30–45 min: E = 719 ± 7 Pa , p=0 . 117; 45–60 min: E = 731 ± 11 Pa , p=0 . 465 ) . Such mechanical activation kinetics of neutrophils has not been reported before as the lower measurement rate of previous techniques yielded only cumulative data over the time period investigated . We also found a similar increase in size and greater deformation of the neutrophils at the later time points in an experimental medicine trial , where healthy human volunteers inhaled lipopolysaccharide ( LPS; from E . coli ) ( Figure 3A , B; Figure 3—figure supplement 1 ) . Also , infecting blood in vitro with Staphylococcus aureus ( S . aureus ) , a Gram-positive bacterium and one of the major pathogens responsible for life-threatening infections world-wide , resulted in larger and more deformed neutrophils , measured between 30–60 min after blood stimulation ( Figure 3A , B; Figure 3—figure supplement 2 ) . Congruently , blood taken from patients with an acute lung injury ( ALI ) of most likely bacterial origin had larger and more deformed neutrophils compared to healthy controls ( Figure 3C , E , H ) . The same neutrophil response was found in blood samples from patients hospitalized with viral respiratory tract infections ( RTI; Figure 3D , E , H ) . Also monocytes responded by a size increase in both RTI and ALI patients and after in vitro stimulation with S . aureus , but showed a significantly increased deformation only in viral RTI , while blood lymphocytes did not show any consistent response ( Figure 3F–H; Figure 3—figure supplements 2 and 3 ) . The lymphocyte response changed when analyzing blood of patients with acute Epstein-Barr-virus ( EBV ) infection , which is known to also stimulate the lymphatic system , where both monocytes and lymphocytes showed an increase in cell size and deformation , while neutrophils showed less of a response ( Figure 3I–L , Figure 3—figure supplement 3 ) . These results suggest that MORE blood analysis might be sufficiently sensitive to distinguish bacterial from viral infections , and potentially other inflammatory diseases , by the differential response of selective blood leukocyte populations . This possibility will be followed up in future specific trials . Importantly , MORE blood analysis is of special interest for blood tests in neonatology with patients at high risk of infections but only minute amounts of blood available for diagnostics , or to characterize neutrophils in neutropenic patients , as it merely requires longer data acquisition periods . Blood cancers , or leukemias , affecting both myeloid and lymphoid cell lineages , are a further large area where MORE analysis could potentially contribute fundamental insight , aid diagnosis , and improve therapy monitoring . While solid cancer cell mechanics has been a focus of cell mechanics research and extensively documented ( Suresh , 2007; Kumar and Weaver , 2009; Guck and Chilvers , 2013 ) , the mechanical properties of blood cancers are comparatively understudied . The available research on mechanics of leukemic cells has been undertaken either on cell lines or fully purified cells ( Lichtman , 1973; Rosenbluth et al . , 2006; Lam et al . , 2008; Lichtman , 1970; Dombret et al . , 1995; Lautenschläger et al . , 2009; Ekpenyong et al . , 2012; Rosenbluth et al . , 2008; Zheng et al . , 2015 ) but so far not directly in blood . MORE analysis of the blood of patients with acute myeloid ( AML ) and lymphatic leukemias ( ALL ) revealed the new presence of atypical cell populations — the characteristic immature blasts not normally present in healthy donors ( Figure 4A–C ) . Cell populations gated for AML revealed less deformed cells but of about the same size compared to healthy and fully differentiated myeloid cells ( Figure 4D , Figure 4—figure supplement 1 ) , in line with previous results ( Rosenbluth et al . , 2006; Lichtman , 1970; Dombret et al . , 1995; Lautenschläger et al . , 2009; Ekpenyong et al . , 2012 ) . ALL blast cells were larger in size compared to mature lymphocytes , but did not show any consistent trend in deformation ( Figure 4D; Figure 4—figure supplement 1 ) . Since cell size and deformation in the channel are interrelated ( Mietke et al . , 2015; Otto et al . , 2015; Mokbel et al . , 2017 ) , which can be seen by the isoelasticity lines parameterizing the deformation-size space ( Figure 4D ) , we also calculated the apparent Young’s modulus of these cells ( Figure 4—figure supplement 1 ) . Together these results show that mature lymphocytes , ALL blasts , mature myeloid cells , and AML blasts have increasing levels of stiffness , consistent with the composite findings of previous reports ( Lichtman , 1973; Rosenbluth et al . , 2006; Lam et al . , 2008; Lichtman , 1970; Dombret et al . , 1995; Lautenschläger et al . , 2009; Ekpenyong et al . , 2012; Rosenbluth et al . , 2008 ) . This is quite different than the general trend in solid tumors , where cancer cells are found to be more deformable than their healthy counterparts ( Suresh , 2007; Kumar and Weaver , 2009; Guck and Chilvers , 2013 ) . Sensibly , the differential stiffness of AML and ALL blasts , and its potential further increase with chemotherapy , has been implicated in the occurrence of leukostasis ( Lam et al . , 2008; Rosenbluth et al . , 2008; Lam et al . , 2007 ) . MORE analysis might not only permit screening for novel therapeutic targets to soften cells ( Gossett et al . , 2012; Di Carlo , 2012; Surcel et al . , 2015 ) , but also assessing the risk of leukostasis directly in each patient . Finally , by following the ALL blast population in a patient over 12 days of methylprednisolone treatment we could monitor the return to the normal morpho-rheological fingerprint of blood ( Figure 4E–H ) . The evolution of this fingerprint likely comprises multiple contributions with blast cells undergoing apoptosis over a time course of 2–7 days ( Ito et al . , 1996 ) , which is associated with an increase in stiffness ( Lam et al . , 2007 ) . Blast cells are sequestered by the spleen and new , but immature and likely stiffer , blast cells are being added to the circulation from the bone marrow . There could also be ALL subclones with different morpho-rheological characteristics that respond differently and at different times to treatment . And the final increase in deformation from day 9 to 12 coincides with the addition of cytostatic drugs ( vincristine , daunorubicin ) to the methylprednisolone treatment . Dissecting this multifaceted response will be aided by adding simultaneous fluorescence identification of the cells in the future ( Rosendahl , 2017 ) . Of note , of the conventional biomarkers and techniques that are used in the diagnosis of leukemia ( see Supplementary file 2 ) , only morphological analysis of air-dried Romanowsky-stained blood ( or bone marrow ) smears is traditionally applied to monitor treatment success in ALL . The response to treatment is one of the most powerful prognostic in vivo markers of leukemia survival . In pediatric ALL the number of blasts at day eight after start of methylprednisolone treatment is predictive of the relapse rate ( <1000 blasts/µl of blood: relapse rate 20–30%; >1000 blasts/µl of blood: relapse rate 50–80% ) . MORE analysis provides at least the same information as conventional morphological analysis , but in a shorter amount of time and with smaller sample sizes required ( for a comparison between MORE analysis and conventional biomarkers , see Supplementary file 2 ) . In summary , MORE blood analysis can be used to monitor morpho-rheological effects of chemotherapy and the successful replacement of lymphoblasts with mature lymphocytes in a quantitative manner . This last finding also touches upon the study of hematopoietic differentiation of cells in the bone marrow , which is an obvious further potential area of application of this approach . Morpho-rheological phenotyping allows individual blood cell mechanics to be studied in a range of human diseases and takes cell mechanical phenotyping to an entirely new level . While established techniques such as micropipette aspiration ( Lichtman , 1973; Lichtman , 1970; Dombret et al . , 1995; Ravetto et al . , 2014 ) , indentation by cell pokers and atomic force microscopes ( Worthen et al . , 1989; Rosenbluth et al . , 2006; Lam et al . , 2008 ) , or optical trapping ( Lautenschläger et al . , 2009; Ekpenyong et al . , 2012; Paul et al . , 2013 ) have provided important proof-of-concept insight over the last decades , the recent advent of microfluidic techniques approaching the throughput of conventional flow cytometers ( Gossett et al . , 2012; Otto et al . , 2015; Zheng et al . , 2015; Byun et al . , 2013; Lange et al . , 2015 ) has finally brought mechanical phenotyping close to real-world applications ( Guck and Chilvers , 2013; Tse et al . , 2013 ) . Amongst the latter techniques , RT-DC stands out because it can continuously monitor an , in principle , unlimited number of cells , which enables the direct sensitive assessment of the state of all major cell types found in blood . A volume as small as 10 µl can be analyzed cell-by-cell , with only dilution in measurement buffer to adjust cell density and prevent sedimentation , but no labeling , enrichment or separation , which could otherwise cause inadvertent activation of blood cells . The conventional blood count is extended by information about characteristic , and diagnostic , morpho-rheological changes of the major cell types . Cell mechanics and morphology are inherent and sensitive markers intimately linked to functional changes associated with the cytoskeleton ( Chimini and Chavrier , 2000; Fletcher and Mullins , 2010; Kasza et al . , 2007; Patel et al . , 2012; Salbreux et al . , 2012 ) and other intracellular shape-determining and load-bearing entities ( Rowat et al . , 2006; Munder et al . , 2016 ) . Thus , label-free , disease-specific morpho-rheological blood signatures are a novel resource for generating hypotheses about the underlying molecular mechanisms . The availability of such parameters in real-time , easily combined with conventional fluorescence detection ( Rosendahl , 2017 ) , are the necessary prerequisite for future sorting of morpho-rheologically distinct subpopulations , which then provides a novel opportunity for further molecular biological analysis . Of course , at present , MORE phenotyping provides a sensitive , but not a very specific marker . For example , neutrophil softening could be a signature of different underlying pathological changes . In the future , fuller exploration of the large combinatorial space afforded by the multi-parametric response of the various blood cells , exploiting many additional morpho-rheological parameters in conjunction with machine learning , and inclusion of conventional fluorescence-based marker analysis ( Rosendahl , 2017 ) will further increase the specificity of this approach . Apart from now enabling realistic blood cell research ex vivo close to physiological conditions , delivering for example previously unavailable information about leukocyte activation kinetics , and after further in-depth studies of the phenomena reported here , MORE phenotyping could have a tangible impact on diagnosis , prognosis , and monitoring of treatment success of many hematological diseases as well as inflammatory , infectious , and metabolic disorders . Beyond blood analysis , MORE phenotyping has the potential to become a standard approach in flow cytometry with many applications in biology , biophysics , biotechnology , and medicine . Real-time deformability cytometry ( RT-DC ) was carried out as described previously ( Otto et al . , 2015 ) . For RT-DC measurements , cells were suspended in a viscosity-adjusted measurement buffer ( MB ) based on 1x phosphate buffered saline ( PBS ) containing methylcellulose . The viscosity was adjusted to 15 mPa s at room ( and measurement ) temperature , determined using a falling ball viscometer ( Haake , Thermo Scientific ) . Cells in the MB were taken up into a 1 ml syringe , placed on a syringe pump ( neMESYS , Cetoni GmbH ) and connected via tubing to the sample inlet of the microfluidic chip with a square measurement channel cross section of 20 × 20 µm2 . The microfluidic chip was made from cured polydimethylsiloxane bonded to a thickness #2 cover glass . Another syringe containing MB without cells was connected to the sheath flow inlet of the chip . Measurements were carried out at a total flow rate of 0 . 12 µl/s with a sample flow rate of 0 . 03 µl/s and a sheath flow rate of 0 . 09 µl/s unless stated otherwise . Different gating settings for cell dimensions could be employed during the measurement ( Figure 1—figure supplement 1 ) . Images of the cells in the channel were acquired in a region of interest of 250 × 80 pixels at a frame rate of 2000 fps . Real-time analysis of the images was performed during the measurement and the parameters necessary for MORE analysis were stored for all detected cells . The raw data obtained from RT-DC measurements consisted of the following information of every detected cell: a bright field image of the cell , the contour of the cell , its deformation value , and the cell size as the cross-sectional area of the cell in the image ( Figure 1—figure supplement 1 ) . The deformation was calculated from the convex hull contour of the cell — a processed contour , where all points contributing to concave curvature were removed:deformation=1-2πAl , where A is the area enclosed by the convex hull contour and l is the length of the convex hull contour . Therefore , deformation is the deviation from a perfectly circular cell image . It describes the change of the cell’s shape by the hydrodynamic forces in the measurement channel but may also contain pre-existing shape deviations from a sphere , for example for the biconcave , disk-like shapes of healthy red blood cells or strongly activated and polarized neutrophils . Image brightness analysis was carried out using the contour information and the image of the cell . The mean brightness of the cell was determined from all pixel values within the cell’s contour ( Figure 1D ) . With this information the distinction of leukocyte subpopulations was possible in the space spanned by cell size and mean brightness ( Figure 1D and Figure 1—figure supplement 2 ) . It is worth noting that the absolute value of the resulting brightness was influenced by several experimental conditions such as focus of the image and the thickness of the microfluidic chip . However , this did not affect the quality of the distinction of cells by their brightness . Special care had to be taken when comparing the brightness of different purified leukocyte subpopulations of similar size ( like neutrophils , eosinophils , and monocytes ) . In order to achieve a situation similar to the diluted whole blood measurement , we used the same microfluidic chip repeatedly after thorough flushing . All brightness values reported were normalized to 100 by the background brightness of the channel . Apart from the initial brightness distinction , in a second step , the root mean square of pixel brightness values was calculated in an area of 9 × 5 pixels ( nine in the flow direction , five perpendicular to the flow direction ) around the geometrical center of the cell . This information was used to distinguish the relevant leukocyte subpopulations from eventual erythrocyte doublets present ( Figure 1D ) . To ensure best validity of the deformation measure based on the area within the cell’s contour and the length of the contour , only cells without prominent protrusions were considered for comparisons based on deformation . A reliable criterion to select those cells was found by comparing the area within the originally detected cell contour and within the convex hull contour . For erythrocytes , the difference of these two areas was limited to 15% . For leukocytes , a suitable limit was found at 5% . For the identification of malaria-infected erythrocytes we used a semi-automated procedure designed to obtain only clearly positive results and to avoid false negatives . The defining property of infected cells was the presence of bright spots within the cells . In a first step , all pixel values outside the cell’s contour were set to 0 . In a twice-repeated procedure , the image of the erythrocyte was further reduced by setting all pixel values of the contour pixels to 0 and finding the new contour . This measure was used to eliminate possible bright spots due to fringes at the border of the cell . From this image , the brightness of every pixel of the remaining cell was calculated by taking the mean of the pixel itself and its eight nearest neighbors . The user was then able to set the minimal threshold for this brightness in order to identify a cell as potentially infected . Since higher pixel values are frequently obtained at the rear of the cell ( in flow direction ) only bright spots within 70% of the cell’s length counted from the front of the cell were considered . As a last criterion , the calculated brightness was compared to the brightness of the cell directly surrounding the bright spot in order to eliminate cases of generally bright cells . For this a mean brightness value was formed from all pixels located within the two rectangular areas spanned from [k-3 , l-1] to [k-2 , l + 1] as well as [k + 2 , l-1] to [k + 3 , l + 1] , where k is the pixel position of the bright spot in the flow direction and l is the pixel position of the bright spot orthogonal to the flow direction . Most of this analysis can be performed with ShapeOut , except for the last aspect of considering details of internal brightness , for which a custom-written Python script was used . All studies complied with the Declaration of Helsinki and involved written informed consent from all participants or their legal guardians . Ethics for experiments with human blood were approved by the ethics committee of the Technische Universität Dresden ( EK89032013 , EK458102015 ) , and for human blood and LPS inhalation in healthy volunteers by the East of England , Cambridge Central ethics committee ( Study No . 06/Q0108/281 and ClinicalTrialReference NCT02551614 ) . Study participants were enrolled according to good clinical practice and recruited at the University Medical Centre Carl Gustav Carus Dresden , Germany , the Biotechnology Center , Technische Universität Dresden , Germany , or Cambridge University Hospitals , Cambridge , UK . Human blood and serum used to culture the malaria parasites was obtained from the Glasgow and West of Scotland Blood Transfusion Service; the provision of the material was approved by the Scottish National Blood Transfusion Service Committee For The Governance Of Blood And Tissue Samples For Non-Therapeutic Use . Venous blood was drawn from donors with a 20-gauge multifly needle into a sodium citrate S-monovette ( Sarstedt ) by vacuum aspiration . In case of blood volumes above 9 ml , blood was manually drawn via a 19-gauge multifly needle into a 40 ml syringe and transferred to 50 ml Falcon polypropylene tubes ( BD ) containing 4 ml 3 . 8% sodium citrate ( Martindale Pharmaceuticals ) . For RT-DC measurements of blood , 50 µl of anti-coagulated blood were diluted in 950 µl MB and mixed gently by manual rotation of the sample tube . This fixed dilution of 1:20 was the result of optimization series to dilute as little possible , while still enabling the reliable detection of single cell events for both erythrocytes and leukocytes at typical cell densities found in blood . Measurements were typically carried out within 2 hr past blood donation unless stated otherwise . Two different gating settings were employed in the measurement software for erythrocyte and leukocyte acquisition , respectively ( Figure 1—figure supplement 1A ) . For erythrocytes , gates were essentially open allowing cell dimensions in flow direction from 0 µm to 30 µm . The leukocyte gate was set to a size of 5–16 µm in flow direction and >5 µm perpendicular to it . This setting allowed filtering out single erythrocytes and almost all erythrocyte multiples . The leukocyte populations remained unaltered as confirmed in experiments with purified leukocytes at open gate settings . Using the leukocyte gate , the majority of thrombocytes was also ignored as they possess typical diameters of 2–3 µm . A small fraction of very large thrombocytes and microerythrocytes were still found within this gate as seen in Figure 1C and D . Mechanical analysis of these events constitutes an interesting challenge in that they can be detected and counted , but at present not tested for activation via their deformation given their very small size compared to the channel size , which was chosen to accommodate all cells found in blood . Measurements in the leukocyte gate were carried out over a fixed timespan of 15 min ( to acquire typically 500 to 3000 leukocytes , depending on donor and disease state ) , followed by a separate measurement in the erythrocyte gate for a few seconds until data of 5 , 000–10 , 000 cells were acquired . Measurements for establishing the normal MORE blood phenotype in healthy human volunteers ( Figure 1E ) , and all measurements directly compared to this norm , e . g . , blood samples derived from patients , were carried out at a temperature of 30°C . The remaining measurements — fMLP stimulation , LPS stimulation , purified leukocyte subpopulations , malaria infection , and erythrocyte palmitoylation — were carried out at a temperature of 23°C . The viscosity of the MB was always adjusted to 15 mPa s at the different temperatures to keep the acting hydrodynamic stress and , thus , the resulting deformation regimes the same . An MB with the viscosity of 25 mPa s ( to slow blood cell sedimentation in the tubing ) was used in experiments for comparing the relative cell count results of leukocyte subpopulation by MORE analysis and conventional blood count ( Figure 1F; Supplementary file 1 ) . Here , the total flow rate was 0 . 06 µl/s ( sample flow 0 . 015 µl/s , sheath flow 0 . 045 µl/s ) and images were acquired at 4000 fps . Leukocyte subpopulations were purified by negative and/or positive magnetic-activated cell sorting ( MACS ) following the instructions provided by the manufacturer . Reagents for cell isolation with magnetic beads purchased from Miltenyi Biotec were MACSxpress Neutrophil Isolation Kit human ( 130-104-434 ) , Monocyte Isolation Kit human ( 130-091-153 ) , Basophil Isolation Kit II human ( 130-092-662 ) , Pan T Cell Isolation Kit human ( 130-096-535 ) and CD3 MicroBeads ( 130-050-101 ) , as well as Pan B Cell Isolation Kit human ( 130-101-638 ) and CD19 MicroBeads ( 130-050-301 ) . EasySep Human Eosinophil Enrichment Kit ( 19256 ) was obtained from StemCell Technologies . The purity of the derived cell isolates was controlled twice by staining with 7-Color-Immunophenotyping Kit ( Miltenyi Biotec , 5140627058 ) , as well as additional single staining of each cell subset for fluorescence-activated cell sorting ( FACS ) . Individual cell type staining antibodies from BioLegend were used for granulocytes ( target: CD66ACE , staining: PE , order no . : 342304 , RRID:AB_2077337 ) , eosinophils ( Siglec-8 , APC , 347105 , RRID:AB_2561401 ) , B lymphocytes ( CD19 , FITC , 302205 , RRID:AB_314235 ) , NK cells ( CD56 , PE , 318305 , RRID:AB_604093 ) , T helper cells ( CD4 , PE-Cy7 , 300511 , RRID:AB_314079 ) , T lymphocytes ( CD3 , APC , 300411 , RRID:AB_314065 ) , cytotoxic T cells ( CD8 , PacificBlue , 301026 , RRID:AB_493111 ) , monocytes ( CD14 , FITC , 325603 , RRID:AB_830676 ) , as well as eosinophils , basophils , mast cells , and mononuclear phagocytes ( CD193 , PE , 310705 , RRID:AB_345395 ) . For RT-DC measurements , purified cells were pelleted by centrifugation ( 200 g , 5 min ) and re-suspended in MB at concentrations of about 5 · 106 cells/ml by repeated , gentle shaking . Plasmodium falciparum ( P . falciparum , HB3 clone , NCBI Taxonomy ID: 137071 ) cultures were grown accordingly to standard protocols ( Trager and Jensen , 1976 ) . Two P . falciparum cultures were grown independently for 3 weeks , treated with Plasmion ( Lelièvre et al . , 2005 ) to enrich for the schizont stages , and then allowed to reinvade fresh red blood cells in a shaking incubator for 3 hr . The cultures were then treated with sorbitol ( Lambros and Vanderberg , 1979 ) , to remove all schizonts that had not reached full maturity; only ring stage parasites survive sorbitol treatment . The highly synchronized culture used for the RT-DC measurements therefore consisted of erythrocytes exposed to P . falciparum , into some of which parasites had invaded within a 3 hr window . Samples were removed at 4 , 12 , 16 , 20 , 24 , 36 , 42 and 46 hr post invasion for the RT-DC measurements . At the time of each measurement a thin blood smear was taken and stained with Giemsa’s stain to assess the parasitemia and the stage of the parasites ( Figure 2—figure supplement 2A ) . A control sample of the same blood without the parasites underwent the identical treatment as the P . falciparum exposed samples . For RT-DC measurements , at each time point , 10 µl of the blood culture were diluted in 990 µl of the MB to a final concentration of 2 . 5 · 105 cells/μl . The total flow rate through the channel was 0 . 04 µl/s for all malaria infection experiments ( sample flow rate 0 . 01 µl/s , sheath flow rate 0 . 03 µl/s ) . For experiments on growth and invasion depending on erythrocyte palmitoylation status , blood , treated as described in the palmitoylation section below , was shipped from Germany to Scotland in PBS buffer containing 15 mM glucose , 5 mM sodium pyruvate , 5 µM Coenzyme A , 5 mM MgCl2 , 5 mM KCl , 130 mM NaCl . Parasites were synchronized by collecting P . falciparum mature stages ( trophozoites and schizonts ) from P . falciparum clone HB3 using MACS columns ( Ribaut et al . , 2008 ) . The trophozoite and schizont enriched cultures were mixed with erythrocytes to achieve a starting parasitemia of 0 . 5–1 . 0% . Each erythrocyte type was set up in a separate culture flask at 3 ml volume and 5% hematocrit . The parasites were incubated in a shaking incubator at 37°C under standard culture conditions of gas and medium . Parasitemia was monitored on day 2 ( post invasion ) and day 4 ( second round of invasion ) . For all experimental conditions , a minimum of 500 RBCs were counted . Experiments were repeated on three different days with erythrocytes of 3 different donors yielding the same results . Red blood cells were pelleted by blood centrifugation ( 800 g , 5 min ) , plasma was removed , and the RBCs were pretreated with one volume of 1% fatty acid-free bovine serum albumin ( BSA ) in PBS-glucose ( 10 mM phosphate , 140 mM NaCl , 5 mM KCl , 0 . 5 mM EDTA , 5 mM glucose , pH 7 . 4 ) at 37°C for 15 min , in order to lower the endogenous content of free fatty acids in their membrane pools , and washed three times with PBS-glucose . Cells were re-suspended in 3 volumes of incubation buffer , containing 40 mM imidazole , 90 mM NaCl , 5 mM KCl , 5 mM MgCl2 , 15 mM D-glucose , 0 . 5 mM EGTA , 30 mM sucrose , 5 mM sodium pyruvate , 5 mM Coenzyme A , 50 mg PMSF/ml and 200 U penicillin/streptomycin ( 320 mOsm , pH 7 . 6 ) . For inhibition of palmitoylation , 100 µM final concentration of 2-bromopalmitate ( 2 BP ) was used . 100 µM palmitic acid ( PA ) was added as a control . The RBCs were incubated in a humidified incubator with 5% CO2 for 24 hr at 37°C . Prior to measurement , RBCs were pelleted , re-suspended in 1% BSA , incubated for 15 min at 37°C and washed two times with PBS-glucose . Glucose , sucrose , 2-bromopalmitate , palmitic acid , fatty acid free BSA , Coenzyme A , and PMSF were purchased from Sigma-Aldrich; Penicillin/streptomycin and sodium pyruvate from Gibco . RT-DC measurements were carried out at a room temperature of 23°C and with a total flow rate of 0 . 032 µl/s ( sample flow 0 . 008 µl/s , sheath flow 0 . 024 µl/s ) after adding 10 µl of the RBC suspension to 990 µl of MB . Experiments were carried out on two different days with erythrocytes of 4 different donors . For in vitro fMLP stimulation , blood was stimulated with 100 nM N-Formylmethionyl-leucyl-phenylalanine ( fMLP; Sigma-Aldrich , 47729 , 10 mg-F ) . Separate samples were analyzed in time intervals of 0–15 min , 15–30 min , 30–45 min , and 45–60 min after activation . During incubation all samples were stored in 2 ml Eppendorf tubes at 37°C at 450 rpm in a ThermoMicer C ( Eppendorf ) . All experiments were performed within 2 hr maximum after blood drawing . Experiments were repeated with blood samples of 5 different donors on five different days . Due to experimental feasibility PBS controls of these donors were measured before fMLP stimulation and after the 60 min fMLP sample . Additionally , three control samples of different donors were treated similarly adding 10 µl 1 x PBS instead of fMLP and were analyzed in time intervals of 0–15 min , 15–30 min , 30–45 min , and 45–60 min after bleeding to exclude kinetic effects due to blood alteration with storage . Blood stimulation was performed with Staphylococcus aureus Newmann strain ( S . aureus; ATCC 25904; NCBI Taxonomy ID: 426430 ) . For reproducible repetitive testing with competent bacterial strains cryo-aliquots of S . aureus were prepared as follows . Bacterial cells were pre-cultured to the log phase for synchronization of growth in BHI broth ( Bacto Brain Heart Infusion , Becton Dickinson ) at 37°C and transferred to a second culture . Aiming at a high bacterial virulence factor expression , the cells were grown to an early stationary phase in a 96-well-plate ( 100 µl , OD600nm 0 . 1837 , Infinite 200 reader , TECAN ) , pelleted by centrifugation ( 2671 g for 5 min at 4°C ) , washed two times in PBS and re-suspended in cell-freezing media ( Iscove Basal Medium , Biochrom ) with 40% endotoxin-free FBS ( FBS Superior , Biochrom ) at a final concentration of 2 . 54⋅109 CFU/ml . Aliquots were immediately frozen at –80°C and only thawed once for a single experiment . Blood stimulation and measurement were carried out at 30°C temperature for 15 min with one multiplicity of infection ( MOI ) in 1:20 RT-DC measurement buffer . MOI ( 0 . 9–1 . 09 ) was controlled retrospectively by granulocyte count and 5% sheep blood agar culture ( Columbia agar , bioMérieux ) at 37°C and bacterial colony counting on the following day . PBS blood controls were conducted before and after S . aureus blood stimulation . The experiment was repeated with blood of 4 different donors on four different days . All experiments were performed within 2 hr after blood drawing . E . coli lipopolysaccharide ( LPS ) 50 µg ( GSK ) was administered to healthy , never-smoker volunteers via a specialized dosimeter ( MB3 Markos Mefar ) 90 min prior to injection of autologous 99mTechnetium-Hexamethylpropleneamine-oxime labeled neutrophils . Temperature , forced expiratory volume in 1 s , forced ventilator capacity and triplicate blood pressures were recorded prior to , and at 30 min post LPS administration . RT-DC measurements were obtained at baseline , 90 , 135 , 210 , 330 , and 450 min post LPS . Patient inclusion criteria for RTI: Patients with clinical signs of lower RTI , a core temperature >38 . 5°C and the need for supplemental oxygen were recruited on the day of hospitalization . Only patients without treatment prior to hospitalization were included . None of the included patients received antibiotic treatment for reconstitution . Patient inclusion criteria for ALI: Patients diagnosed with ALI according to the criteria of the North American European Consensus Conference ( NAECC ) ( Bernard et al . , 1994 ) and without underlying diseases prior to ALI were included . All blood samples were analyzed within 30 min of venipuncture . Size and deformation of blood leukocytes was characterized for all blood cells in which the area within the original cell contour differed less than 5% from the area within the convex hull contour . Samples from patients diagnosed with ALL or AML based on cytogenetic , molecular-genetic and morphological criteria according to WHO classification from 2008 ( Vardiman et al . , 2009 ) were assessed by MORE blood analysis on the day of diagnosis . In order to evaluate mechanical properties of AML and ALL blast cells in diluted whole blood , several brightness and size gates had to be combined as shown in Figure 4A–C . The AML gate spanned the regions normally used for basophils and monocytes . The ALL gate spanned the regions used for lymphocytes , basophils and monocytes . In all AML cases , blasts made up >80% of all leukocytes , and up to 99% of events in the AML gate . In all ALL cases , blasts made up >60% of leukocytes , and up to 85% of events in the ALL gate . The blast cell fraction was obtained from the standard differential blood count , by comparing the number of blast cells with the number of normal cells that would also populate the respective blasts gate in MORE analysis . RT-DC data of cell size and deformation can be converted into apparent Young’s moduli using theoretical models ( Mietke et al . , 2015 ) and numerical simulations ( Mokbel et al . , 2017 ) . To ensure a correct conversion , effects of shear thinning of the MC medium and a deformation offset due to pixelation were taken into account as described in Herold , 2017 . The calculation of apparent Young’s moduli for AML and ALL blasts and isoelasticity lines are based on the assumption that cells can be approximated as purely elastic , homogeneous isotropic spheres . This assumption is equivalent to using the Hertz model to extract an apparent Young’s modulus of cells in atomic force microscopy-enabled nano-indentation experiments . The conversion of deformation and size into Young’s modulus for every cell measured is included in the analysis software ShapeOut . Throughout , the number of cells in a single measurement is denoted as N , while the number of independently repeated experiments — typically the number of donor or patient samples measured , as stated — is denoted as n . For comparison of different donors or treatment conditions the median of deformation and cell size of a specific cell population was used . In order to evaluate effects of a disease we calculated a 2D confidence ellipse at 68 . 3% ( or one sigma ) as well as 95 . 5% ( or two sigma ) for the control group/norm norm of healthy human blood donors in the space of cell size and deformation . The confidence ellipse was calculated from the covariance matrix of the data and the calculation was carried out with OriginPro 2015 ( Originlab ) . Statistically significant differences between two sets of experiments were checked to the significance level of p<0 . 05 by comparing the groups of individual median values of an experiment using a Kruskal-Wallis one-way ANOVA as implemented in OriginPro 2015 ( Originlab ) . In erythrocyte MORE analysis in malaria infection and palmitoylation , statistically significant differences were checked using linear mixed models in combination with a likelihood ratio test to obtain significance levels for the comparison of the complete populations ( Herbig et al . , 2017 ) . This analysis can be performed in the software ShapeOut . One , two , or three asterisks were awarded for significance levels p<0 . 05 , p<0 . 01 and p<0 . 001 , respectively . In manual counts of malaria infection in RBCs , statistical analyses were performed using a χ2 test with Bonferroni correction ( adjusted statistical significance for p<0 . 0125 ) to compare the numbers of infected and non-infected erythrocytes between erythrocyte samples , except where number of parasite infected cells was zero , in which case Fisher`s exact test was used . The standard deviation for the parasitemia was calculated assuming a binomial random variable as SD= N∙p ( 1-p ) , where N is the number of cells counted and p is the fraction of infected cells . The raw data of all measurements are available from the Dryad Digital Repository ( Toepfner et al . , 2017 ) . The TDMS files can be read , processed , and analyzed using ShapeOut , a custom written , open source software . RT-DC measurement software is commercially available . The analysis software ShapeOut is available as an open source application on GitHub ( https://github . com/ZELLMECHANIK-DRESDEN/ShapeOut/releases; Müller , 2017 ) . A copy is archived at https://github . com/elifesciences-publications/ShapeOut .	When you are sick and go to the doctor , it is often not obvious what exactly is wrong — what is causing fever , nausea , shortness of breath or other symptoms . It is important to find this out quickly so that the right action can be taken . One of the first steps is to obtain a blood sample and to count how many of the different blood cells are present in it . This is called a complete blood count , and the information it provides has turned out to be surprisingly useful . A large number of certain white blood cells , for example , can show that the body is fighting an infection . But there might be several reasons why the number of white blood cells has increased , so this information alone is often not enough for a specific diagnosis . There are many hundreds of possible tests that can supplement the results of a complete blood count . These might identify bacteria or measure the concentrations of certain molecules in the blood , for example . But which test will give the important clue that reveals the source of the illness ? This can be difficult to predict . Although each test helps to narrow down the final diagnosis they become increasingly expensive and time-consuming to perform , and rapid action is often important when treating a disease . Can we get more decisive information from the initial blood test by measuring other properties of the blood cells ? Toepfner et al . now show that this is possible using a technique called real-time deformability cytometry . This method forces the blood cells in a small drop of blood to flow extremely rapidly through a narrow microfluidic channel while they are imaged by a fast camera . A computer algorithm can then analyze the size and stiffness of the blood cells in real-time . Toepfner et al . show that this approach can detect characteristic changes that affect blood cells as a result of malaria , spherocytosis , bacterial and viral infections , and leukemia . Furthermore , many thousands of blood cells can be measured in a few minutes — fast enough to be suitable as a diagnostic test . These proof-of-concept findings can now be used to develop actual diagnostic tests for a wide range of blood-related diseases . The approach could also be used to test which of several drugs is working to treat a certain medical condition , and to monitor whether the treatment is progressing as planned .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "tools", "and", "resources" ]	2018	Detection of human disease conditions by single-cell morpho-rheological phenotyping of blood
Most reinforcement learning models assume that the reward signal arrives after the activity that led to the reward , placing constraints on the possible underlying cellular mechanisms . Here we show that dopamine , a positive reinforcement signal , can retroactively convert hippocampal timing-dependent synaptic depression into potentiation . This effect requires functional NMDA receptors and is mediated in part through the activation of the cAMP/PKA cascade . Collectively , our results support the idea that reward-related signaling can act on a pre-established synaptic eligibility trace , thereby associating specific experiences with behaviorally distant , rewarding outcomes . This finding identifies a biologically plausible mechanism for solving the ‘distal reward problem’ . Spike timing-dependent plasticity ( STDP ) is a physiologically relevant form of Hebbian learning ( Caporale and Dan , 2008 ) . In its classic form , STDP depends on the order and precise timing of presynaptic and postsynaptic spikes: pre-before-post spike pairings induce timing-dependent long-term potentiation ( t-LTP ) , whereas post-before-pre pairings induce timing-dependent long-term depression ( t-LTD ) ( Markram et al . , 1997; Bi and Poo , 1998 ) . However , the quantitative rules of STDP are profoundly influenced by neuromodulators ( Seol et al . , 2007; Pawlak et al . , 2010 ) , including dopamine ( DA ) ( Zhang et al . , 2009; Edelmann and Lessmann , 2011; Yang and Dani , 2014 ) . Although it is well established that reward-motivated behavior depends on the activity of DA neurons ( Schultz et al . , 1997; Suri and Schultz , 1999; Pan et al . , 2005 ) , the mechanisms that associate specific experiences with rewarding outcomes , which typically occur after a delay , are not well understood . This is referred to as the distal reward problem ( Hull , 1943 ) . To address this problem , here , we examined whether DA modulates STDP not only when applied during , but also—more importantly—when applied after the pairing event . We first sought to corroborate the shape of the STDP induction curve by varying the time interval between the presynaptic and postsynaptic activity ( Δt; Figure 1A , B , G ) . To this end , we monitored excitatory postsynaptic potentials ( EPSPs ) that were evoked by extracellular stimulation of the Schaffer-collateral-CA1 pathway during whole-cell recordings of CA1 pyramidal cells in mouse horizontal slices ( postnatal days 12–18; ‘Materials and methods’ ) . Plasticity was induced in current clamp mode using an induction protocol that involved 100 pairings of a single EPSP followed by a single postsynaptic spike ( t-LTP; Figure 1A ) or a single postsynaptic spike followed by a single EPSP ( t-LTD; Figure 1B ) at 0 . 2 Hz . Consistent with previous studies ( Bi and Poo , 1998; Zhang et al . , 2009; Edelmann and Lessmann , 2011 ) , the pre-before-post pairing protocol with Δt = +10 ms induced t-LTP ( 182 ± 14%; t ( 4 ) = 5 . 6 , p = 0 . 0049 vs 100% , n = 5; Figure 1A ) and the post-before-pre pairing protocol with Δt = −20 ms induced t-LTD ( 61 ± 9%; t ( 4 ) = 4 . 4 , p = 0 . 0121 vs 100% , n = 5; Figure 1B ) . Surprisingly , however , we found that the post-before-pre pairing protocol with Δt = −10 ms instead of inducing t-LTD , elicited robust t-LTP ( 202 ± 21%; t ( 6 ) = 5 . 0 , p = 0 . 0025 vs 100% , n = 7; Figure 1C , D , G ) . This conflicts with previous reports from hippocampal cultures ( Bi and Poo , 1998; Zhang et al . , 2009 ) and acute slices ( Edelmann and Lessmann , 2011; Yang and Dani , 2014 ) , where post-before-pre pairing protocols never elicited synaptic potentiation in baseline conditions . Given that DA has been found to widen the time window for the induction of t-LTP ( Zhang et al . , 2009; Yang and Dani , 2014 ) , we wanted to assess whether endogenous DA could be responsible for the potentiation observed with the post-before-pre pairing under our experimental conditions . Therefore , we repeated this set of experiments using the post-before-pre pairing protocol with Δt = −10 ms in the presence of DA receptor ( DAR ) antagonists . Indeed , combined application of the D1/D5 receptor antagonist SCH23390 ( 10 μM ) and D2-like receptor antagonist sulpiride ( 50 μM ) from the start of the recordings prevented t-LTP and enabled t-LTD instead ( 72 ± 8%; t ( 5 ) = 3 . 6 , p = 0 . 0160 vs 100% , n = 6; Figure 1C , D , G ) , rendering the STDP induction curve similar to that observed in hippocampal cultures ( Bi and Poo , 1998; Zhang et al . , 2009 ) . These results suggest a modulatory action of endogenous DA , presumably released during the pairing protocol ( Frey et al . , 1990; Yang and Dani , 2014 ) , which resulted in the changed polarity of plasticity at narrow negative spike-timing intervals . 10 . 7554/eLife . 09685 . 003Figure 1 . Dopamine widens the time window for induction of t-LTP . Example plots of normalized EPSP slopes demonstrating that ( A ) pre-before-post pairing protocol with Δt = +10 ms induced t-LTP , whereas ( B ) post-before-pre pairing protocol with Δt = −20 ms induced t-LTD . Insets , pairing protocols . Traces show an excitatory postsynaptic potential ( EPSP ) before ( 1 ) and 40 min after ( 2 ) pairing . ( C ) Endogenous DA widens the spike time window for induction of t-LTP . In the absence of DA receptor antagonists , the post-before-pre pairing protocol with Δt = −10 ms induced t-LTP ( red ) , whereas application of SCH23390 and sulpiride at the start of the recordings prevented this t-LTP and enabled t-LTD instead ( black ) . Traces are presented as in A . ( D ) Summary of results . ( E ) Exogenous DA widens the spike time window for induction of t-LTP . In the presence of 20 μM DA , the post-before-pre pairing protocol with Δt = −20 ms induced t-LTP ( red ) , whereas in control condition the same pairing protocol induced t-LTD ( black ) . Traces are presented as in A . ( F ) Summary of results . ( G ) Summary of the spike timing-dependent plasticity ( STDP ) induction with various spike-timing intervals ( Δt in ms ) in control condition ( green; data point at −20 ms represents data combined from B and E ) , in the presence of DA receptor antagonists ( SCH23390 and sulpiride; black ) , or DA ( red ) . Each data point is the group average EPSP slope percentage change from baseline . Error bars represent s . e . m . Significant difference ( p < 0 . 05 , p < 0 . 01 , p < 0 . 001 ) compared with the baseline ( one-sample two-tailed Student's t-test ) or between the indicated two groups ( paired two-tailed Student's t-test ) . The numbers of cells are shown in parentheses . DOI: http://dx . doi . org/10 . 7554/eLife . 09685 . 00310 . 7554/eLife . 09685 . 004Figure 1—source data 1 . Source data for Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 09685 . 004 Next , we wanted to examine whether the application of exogenous DA during pairing at negative spike-timing intervals facilitates synaptic potentiation . Indeed , the post-before-pre pairing protocol with Δt = −20 ms , which elicited robust t-LTD in control condition ( 74 ± 9% , t ( 8 ) = 3 . 0 , p = 0 . 0165 vs 100% , n = 9; Figure 1E–G ) , induced significant t-LTP when exogenous DA ( 20 μM ) was bath-applied for 10–12 min from 2 min before and during the post-before-pre pairings in interleaved experiments ( 144 ± 12%; t ( 5 ) = 3 . 7 , p = 0 . 0148 vs 100% , n = 6; Figure 1E–G ) . Therefore , in accordance with previous findings ( Zhang et al . , 2009; Yang and Dani , 2014 ) , the presence of DA during the coordinated spiking activity widens the spike time interval for induction of t-LTP . A crucial aspect of reinforcement learning models is the ability of the reinforcing signal ( DA ) to strengthen active synapses , even when it arrives after the activity ( Sutton and Barto , 1981; Izhikevich , 2007 ) . To test this hypothesis experimentally , we applied DA after the t-LTD induction protocol . Exogenous DA ( 100 μM ) added to the perfusion system for 10–12 min starting within 1 min after the post-before-pre pairing protocol with Δt = −20 ms converted t-LTD into t-LTP ( 169 ± 16% , t ( 5 ) = 4 . 3 , p = 0 . 0078 vs 100% , n = 6; Figure 2A , B ) . This implies that DA can have a retroactive effect allowing negative spike pairings to induce t-LTP . The specificity of this DAergic conversion of STDP was assessed using DAR antagonists . Indeed , combined application of DAR antagonists , SCH23390 ( 10 μM ) and sulpiride ( 50 μM ) , prevented DA-induced conversion of t-LTD into t-LTP , resulting in significant t-LTD instead ( 63 ± 12% , t ( 5 ) = 3 . 0 , p = 0 . 0289 vs 100% , n = 6 , Δt = −20 ms ) . Thus , the conversion of t-LTD into t-LTP was due to specific DAR activation . Importantly , when the test pathway was not stimulated following the pairing protocol until after DA washout ( stimulation resumed 15 min after pairing ) , robust t-LTD was induced ( 61 ± 5% , t ( 5 ) = 8 . 1 , p = 0 . 0005 vs 100% , n = 6; Figure 2A , B ) . The effect of DA was , therefore , activity dependent , demonstrating that the reinforcing signal is capable of acting specifically on the active inputs . To exclude the possibility that DA by itself could potentiate the test pathway , control experiments with ongoing synaptic stimulation over 60 min at 0 . 2 Hz , but without pairing with postsynaptic action potentials , were performed . Consistent with earlier reports ( Otmakhova and Lisman , 1999 ) , DA had no significant effect on the basal Schaffer-collateral transmission ( 110 ± 7% , t ( 7 ) = 1 . 4 , p = 0 . 2169 vs 100% , n = 8; Figure 2A , B ) . 10 . 7554/eLife . 09685 . 005Figure 2 . Dopamine retroactively converts t-LTD into t-LTP . ( A ) DA applied immediately after the pairing protocol with Δt = −20 ms converted t-LTD into t-LTP ( +Pairing +Stim; red ) . If the pathway was not stimulated following the pairing protocol until after DA washout , t-LTD was induced ( +Pairing −Stim; black ) . In the absence of the pairing protocol , DA had no effect on baseline EPSPs ( −Pairing +Stim; blue ) . Traces show an EPSP before ( 1 ) and 40 min after ( 2 ) pairing . ( B ) Summary of results . Error bars represent s . e . m . Significant difference ( p < 0 . 01 , **p < 0 . 001 ) compared with the baseline ( one-sample two-tailed Student's t-test ) . The number of cells is shown in parentheses . DOI: http://dx . doi . org/10 . 7554/eLife . 09685 . 00510 . 7554/eLife . 09685 . 006Figure 2—source data 1 . Source data for Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 09685 . 006 Subsequently , we wanted to determine whether the observed DA-induced conversion of t-LTD into t-LTP depends on the timing of DA application following the post-before-pre protocol ( Figure 3A–D ) . We found that delayed application of DA ( 10 or 30 min after t-LTD pairing protocol ) failed to convert t-LTD into t-LTP . Application of DA 10 min after the post-before-pre pairing caused a reversal of t-LTD back to baseline ( 94 ± 9% , t ( 11 ) = 0 . 7 , p = 0 . 5230 vs 100% , n = 12; Figure 3B , D ) , whereas application of DA 30 min after the post-before-pre pairing failed to influence t-LTD altogether ( 59 ± 12% , t ( 5 ) = 3 . 4 , p = 0 . 0200 vs 100% , n = 6; Figure 3C , D ) . Whilst it has previously been reported that DA applied after the induction of low frequency stimulation-induced LTD can reduce the magnitude of synaptic depression ( Mockett et al . , 2007 ) , our data demonstrate , for the first time to our knowledge , that DA can change the polarity of STDP when acting within a short time window following the induction protocol . 10 . 7554/eLife . 09685 . 007Figure 3 . Time dependence of the DA-induced conversion of t-LTD into t-LTP . ( A ) DA applied immediately after the pairing protocol with Δt = −20 ms converted t-LTD into t-LTP , whereas delayed application of DA failed to convert t-LTD into t-LTP and either ( B ) resulted in a reversal of t-LTD back to baseline ( 10 min after pairing ) or ( C ) failed to affect t-LTD altogether ( 30 min after pairing ) . Traces show an EPSP before ( 1 ) and 40 min ( A , B ) or 60 min ( C ) after pairing ( 2 ) . ( D ) Summary of results . Error bars represent s . e . m . Significant difference ( p < 0 . 05 , *p < 0 . 01 ) compared with the baseline ( one-sample two-tailed Student's t-test ) . The number of cells is shown in parentheses . DOI: http://dx . doi . org/10 . 7554/eLife . 09685 . 00710 . 7554/eLife . 09685 . 008Figure 3—source data 1 . Source data for Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 09685 . 008 Finally , we aimed to explore the possible mechanisms underlying the DA-induced conversion of t-LTD into t-LTP . Both hippocampal t-LTP and t-LTD ( Edelmann and Lessmann , 2011; Yang and Dani , 2014 ) , as well as the modulation of STDP by DA ( Zhang et al . , 2009 ) , require functional NMDA receptors . We , therefore , asked whether the DA-induced conversion of t-LTD into t-LTP is also NMDA receptor dependent . Application of the NMDA receptor antagonist d-2-amino-5-phosphonopentanoic acid ( d-AP5 , 50 μM ) after the post-before-pre pairing protocol did not by itself affect the development of t-LTD ( 57 ± 12% , t ( 5 ) = 3 . 6 , p = 0 . 0156 vs 100% , n = 6 , Δt = −20 ms; Figure 4A , D ) . Nevertheless , DA in the presence of d-AP5 reversed t-LTD back to baseline albeit failing to convert t-LTD into t-LTP ( 105 ± 12% , t ( 5 ) = 0 . 4 , p = 0 . 6898 vs 100% , n = 6 , Δt = −20 ms; Figure 4A , D ) . This suggests an important dissociation between two mechanisms involved in the DA-induced conversion of t-LTD into t-LTD , namely a reversal of synaptic depression ( de-depression ) and synaptic potentiation , one of which is NMDA receptor dependent . 10 . 7554/eLife . 09685 . 009Figure 4 . Cellular mechanisms involved in the DA-induced conversion of t-LTD into t-LTP . ( A ) DA applied immediately after the pairing protocol ( Δt = −20 ms ) converted t-LTD into t-LTP ( red; data combined from Figures 2A , 3A ) . This effect requires NMDA receptors as application of d-AP5 1 min before the DA application partially blocked the conversion of t-LTD into t-LTP ( blue ) , whereas application of d-AP5 alone failed to influence the development of t-LTD ( black ) . Traces show an EPSP before ( 1 ) and 40 min after pairing ( 2 ) . ( B ) The DA-induced conversion of t-LTD into t-LTP involves the activation of cAMP/PKA signaling cascade , which closely mimics the effect of DA . Forskolin , an AC activator , applied immediately after the pairing protocol ( Δt = −20 ms ) converted t-LTD into t-LTP ( red ) . This effect requires NMDA receptors as application of d-AP5 1 min before forskolin application partially blocked the forskolin-induced conversion of t-LTD into t-LTP ( blue ) , whereas application of d-AP5 alone failed to influence the development of t-LTD ( black; data same as in A ) . ( C ) Downstream of cAMP , PKA is involved in the conversion of t-LTD into t-LTP as application of the PKA inhibitor , H-89 , completely prevented the conversion of t-LTD into t-LTP . Traces are presented as in A . ( D ) Summary of results . Error bars represent s . e . m . Significant difference ( p < 0 . 05 , p < 0 . 01 , p < 0 . 001 ) compared with the baseline ( one-sample two-tailed Student's t-test ) or between the indicated two groups ( paired two-tailed Student's t-test ) . The number of cells is shown in parentheses . DOI: http://dx . doi . org/10 . 7554/eLife . 09685 . 00910 . 7554/eLife . 09685 . 010Figure 4—source data 1 . Source data for Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 09685 . 01010 . 7554/eLife . 09685 . 011Figure 4—source data 2 . Source data for Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 09685 . 01110 . 7554/eLife . 09685 . 012Figure 4—source data 3 . Source data for Figure 4—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 09685 . 01210 . 7554/eLife . 09685 . 013Figure 4—figure supplement 1 . The retroactive conversion of t-LTP into t-LTD is due to specific DA receptor activation . Plot of normalized EPSP slopes demonstrating that ( A ) application of 10 μM SCH23390 and 50 μM sulpiride prevented DA-induced conversion of t-LTD into t-LTP , which resulted in significant t-LTD , whereas ( B ) neither SCH23390 nor ( C ) sulpiride alone was able to completely block the DA effect . Traces show an EPSP before ( 1 ) and 40 min after ( 2 ) pairing . ( D ) Summary of results . Error bars represent s . e . m . Significant difference ( p < 0 . 05 , p < 0 . 01 , p < 0 . 001 ) compared with the baseline ( one-sample two-tailed Student's t-test ) or between the indicated two groups ( paired two-tailed Student's t-test ) . The number of cells is shown in parentheses . DOI: http://dx . doi . org/10 . 7554/eLife . 09685 . 01310 . 7554/eLife . 09685 . 014Figure 4—figure supplement 2 . Forskolin had no effect on baseline EPSPs . Plot of normalized EPSP slopes showing that in the absence of a pairing protocol , application of 50 μM forskolin does not result in significant synaptic modification ( n = 7 ) . Traces show an EPSP at 10 min ( 1 ) and 58 min ( 2 ) of a recording . Error bars represent s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 09685 . 014 To investigate the intracellular signaling mechanisms involved , we initially set out to establish the DAR subtype associated with the DA-induced conversion of t-LTD into t-LTP . Even though combined application of D1-like and D2-like receptor antagonist completely blocked the DA effect ( Figure 4—figure supplement 1A , D ) , application of either D1-like or D2-like receptor antagonist alone only partially prevented the conversion of t-LTD into t-LTP ( SCH 23390: 131 ± 16%; t ( 6 ) = 1 . 4 , p = 0 . 0113 vs DA ( t-LTP ) ; t ( 6 ) = 3 . 6 , p = 0 . 2277 vs control ( t-LTD ) ; t ( 6 ) = 1 . 9 , p = 0 . 0994 vs 100%; n = 7 . Sulpiride: 94 ± 13%; t ( 6 ) = 3 . 2 , p = 0 . 0228 vs DA ( t-LTP ) ; t ( 6 ) = 1 . 5 , p = 0 . 1932 vs control ( t-LTD ) ; t ( 6 ) = 0 . 5 , p = 0 . 6435 vs 100%; n = 7; Figure 4—figure supplement 1B–D ) . This suggests that both receptor subtypes might mediate the retroactive effect of DA on STDP . Given the astounding complexity of D2-like receptor pharmacology ( Neve et al . , 2004 ) , we first wanted to evaluate the possible involvement of D1-like receptor-activated cAMP/PKA signaling cascade in the DA-induced conversion of t-LTD into t-LTP . D1/D5 receptor stimulation leads to the activation of adenylyl cyclase ( AC ) and subsequent increase in cyclic adenosine monophosphate ( cAMP ) and protein kinase A ( PKA ) activation ( Greengard et al . , 1999; Neve et al . , 2004 ) . We found that the general AC activator , forskolin ( 50 μM ) , applied for 10–12 min immediately after the pairing protocol with Δt = −20 ms resulted in robust conversion of t-LTD into t-LTP ( 167 ± 17% , t ( 6 ) = 3 . 9 , p = 0 . 0078 vs 100% , n = 7; Figure 4B , D ) . Notably , forskolin ( 50 μM ) applied in the absence of the pairing protocol had no significant effect on baseline EPSPs ( 83 ± 12% , t ( 6 ) = 1 . 4 , p = 0 . 2046 vs 100% , n = 7; Figure 4—figure supplement 2 ) . The forskolin-induced conversion of t-LTD into t-LTP was also NMDA receptor dependent since bath application of d-AP5 ( 50 μM ) 1 min before forskolin treatment prevented synaptic potentiation from developing , although synaptic depression was reversed ( 102 ± 11% , t ( 5 ) = 0 . 2 , p = 0 . 8616 vs 100% , n = 6; Figure 4B , D ) . This result suggests that the cAMP cascade works either upstream of or in parallel to NMDA receptor activation for the development of potentiation to occur . Downstream of cAMP , PKA is involved because the PKA inhibitor , H-89 ( 20 µM ) , blocked the DA-induced conversion of t-LTD into t-LTP , revealing significant t-LTD ( 76 ± 5% , t ( 6 ) = 5 . 0 , p = 0 . 0025 vs 100% , n = 7; Figure 4C , D ) . Taken together , these results imply that the DA-induced conversion of t-LTD into t-LTP involves activation of the cAMP/PKA signaling cascade , which closely mimics the effects of DA ( Figure 4A vs Figure 4B ) . Although , D2-like receptors are typically associated with the inhibition of AC ( Neve et al . , 2004 ) , interestingly , there is also evidence that D2-like receptor stimulation can potentiate AC activity ( Glass and Felder , 1997; Watts and Neve , 1997 ) . Therefore , while the possibility that D2-like receptors contribute to the conversion of t-LTD into t-LTP via a different signaling cascade cannot be excluded , it is tempting to suggest that the DA-induced conversion of t-LTD into t-LTP is mediated primarily via the cAMP/PKA pathway . Hence , based on our results , we propose that the stimulation of postsynaptic ( Figure 5ai ) or presynaptic ( Figure 5aii ) DARs activates the cAMP/PKA pathway , which—via two cellular mechanisms ( de-depression and potentiation ) —leads to the retroactive conversion of t-LTD into t-LTP . 10 . 7554/eLife . 09685 . 015Figure 5 . Proposed mechanisms underlying the DA-induced conversion of t-LTD into t-LTP . ( A ) Schematic diagram depicting core components of the proposed cellular mechanisms underlying the DA-induced conversion of t-LTD into t-LTP ( de-depression and potentiation ) . ( Ai ) Model based on postsynaptic NMDAR-dependent potentiation ( Bi and Poo , 1998; Caporale and Dan , 2008; Zhang et al . , 2009; Edelmann and Lessmann , 2011; Yang and Dani , 2014 ) and metabotropic glutamate receptor-dependent ( mGluRs ) depression ( Otani and Connor , 1998; Kemp and Bashir , 1999; Huber et al . , 2000 ) . De-depression ( red , left ) : Activation of G protein-coupled D1/D5 receptors stimulates AC , increasing cAMP and activating PKA ( Greengard et al . , 1999; Neve et al . , 2004 ) , which , via phosphorylation of I-1 ( Ingebritsen and Cohen , 1983 ) , reverses the PP1-induced dephosphorylation of synaptic AMPARs ( Lee et al . , 2000; Mockett et al . , 2007 ) . Potentiation ( red , right ) : PKA activation enhances NMDAR function ( Westphal et al . , 1999; Chen and Roche , 2007 ) . ( Aii ) Model based on presynaptic depression ( Bolshakov and Siegelbaum , 1994; Siegel et al . , 1994; Oliet et al . , 1997; Charton et al . , 1999; Watabe et al . , 2002; Jourdain et al . , 2007 ) . De-depression ( red , left ) : Activation of presynaptic DA receptors stimulates AC , increasing cAMP and activating PKA ( Greengard et al . , 1999; Neve et al . , 2004 ) , which reverses the calcineurin-dependent presynaptic depression . Potentiation ( red , right ) : as in Ai . Arrow indicates activation/phosphorylation , blunt-ended line indicates inhibition/dephosphorylation . Abbreviations: AMPAR , AMPA-type glutamate receptor; NMDAR , NMDA-type glutamate receptor; mGluR1/5 , group I metabotropic glutamate receptor; DAR , dopamine receptor; AC , adenylate cyclase; cAMP , cyclic adenosine monophosphate; PKA , protein kinase A; I-1 , inhibitor 1; PP1 , protein phosphatase 1; PLC , phospholipase C; IP3 , inositol 1 , 4 , 5-trisphosphate; ER , endoplasmic reticulum; DAG , diacylglycerol; eCB , endocannabinoid; CB1R , cannabinoid receptor type 1; CN , calcineurin; CaMKII , calcium-calmodulin-dependent protein kinase II . ( B ) Schematic diagram of synaptic and behavioral timescales in reward learning . During Exploration , the activity-dependent modification of synaptic strength due to spike timing-dependent plasticity ( STDP ) depends on the coordinated spiking between presynaptic and postsynaptic neurons on a millisecond time scale . Post-before-pre pairing leads to synaptic depression that develops gradually on a scale of minutes . When Reward , signaled via dopamine , follows Exploration with a Delay of seconds to minutes , synaptic depression is converted into potentiation . DOI: http://dx . doi . org/10 . 7554/eLife . 09685 . 015 The functional implications of our finding depend on the presynaptic source of DA in the hippocampus , which remains controversial because of the apparent discrepancy between DAergic terminals and receptors ( Scatton et al . , 1980; Gasbarri et al . , 1997; Lisman and Grace , 2005 ) . On one hand , it has been argued that noradrenergic terminals from the locus coeruleus , which mediates arousal and the optimization of behavioral performance ( Aston-Jones and Cohen , 2005; Chamberlain and Robbins , 2013 ) , may provide a major source of DA release ( Smith and Greene , 2012 ) . On the other hand , hippocampal pyramidal cell assemblies are directly affected by concurrent activity in midbrain DAergic neurons ( McNamara et al . , 2014 ) , which have been linked to reward-seeking behavior ( Schultz et al . , 1997 ) and appetitive stimuli ( Mirenowicz and Schultz , 1996; Fiorillo et al . , 2013 ) . This latter finding ( McNamara et al . , 2014 ) not only supports the hypothesis that hippocampal DA is relevant for reward processing , but is also consistent with our result that the activation of DAergic receptors during coordinated spiking activity changes the functional outcome of STDP ( Figure 1C–G ) . Previous studies examining reinforcement learning at the level of synaptic plasticity have showed that neuromodulators can affect timing-dependent plasticity in locust ( Cassenaer and Laurent , 2012 ) and spine structural plasticity in striatal medium spiny neurons in mice ( Yagishita et al . , 2014 ) when acting within a delay time window of 1 s . While this narrow temporal detection window may be important in the striatum , it cannot account for the experimental evidence from behavioral studies of response acquisition with an extended reinforcement delay in rats and pigeons ( Lattal and Gleeson , 1990; Sutphin et al . , 1998 ) , rhesus monkeys ( Galuska and Woods , 2005 ) , and humans ( Okouchi , 2009 ) . Meanwhile , our finding demonstrates that DA can modulate STDP in the CA1 with a reinforcement delay of at least 1 min ( Figure 5B ) . Such extended reinforcement delay is likely to be particularly important in hippocampus-dependent learning during spatial exploration . In conclusion , our work demonstrates a retroactive effect of DA on STDP—converting t-LTD into t-LTP . This effect is mediated at least in part through the activation of the cAMP/PKA cascade and requires the activation of synaptic NMDA receptors . This in turn suggests that the conversion can only occur at synapses that are re-activated following the initial pairing event . Interestingly , it has been reported that hippocampal reactivation events ( sharp wave ripples ) increase in frequency following reward ( Singer and Frank , 2009; Atherton et al . , 2015 ) . Thus , in behaving animals , the conditions for the conversion of depression into potentiation might occur during reward-related sharp wave ripple activity . Together , these findings support the concept of a slowly decaying synaptic eligibility trace that is committed to memory by the occurrence of reward and provide a possible mechanism for associating specific experiences with behaviorally distant , rewarding outcomes in animals ( Sutton and Barto , 1981; Suri and Schultz , 1999; Pan et al . , 2005; Izhikevich , 2007; Harnett et al . , 2009 ) , including humans ( Dunsmoor et al . , 2015 ) . Wild-type mice ( C57BL/6; postnatal days 12–18; from Harlan , Bicester , UK or Central Animal Facility , Physiological Laboratory , Cambridge University ) of both sexes were housed on a 12-hr light/dark cycle at 19–23 °C , with water and food ad libitum . Experimental procedures and animal use were in accordance with the animal care guidelines of the UK Animals ( Scientific Procedures ) Act 1986 under personal and project licenses held by the authors . Caution was taken to minimize stress and the number of animals used in the experiments . Mice were anesthetized with isoflurane and decapitated . The brain was rapidly removed , glued to the stage of a vibrating microtome ( Leica VT 1200S , Leica Biosystems , Wetzlar , Germany ) and immersed in ice-cold artificial cerebrospinal fluid ( ACSF ) containing the following ( mM ) : 126 NaCl , 3 KCl , 26 . 4 NaH2CO3 , 1 . 25 NaH2PO4 , 2 MgSO4 , 2 CaCl2 , and 10 glucose . The ACSF solution , with pH adjusted to 7 . 2 and osmolarity to 270–290 mOsm l−1 , was continuously bubbled with carbogen gas ( 95% O2/5% CO2 ) . The brain was sectioned into 350-μm-thick horizontal slices . The slices were incubated in ACSF at room temperature in a submerged-style storage chamber for at least 1 hr . For recordings ( 1–7 hr after slicing ) , individual slices were transferred to an immersion-type recording chamber , perfused with ACSF ( 2 ml min−1 ) at 24–26 °C . Voltage signals were low-pass filtered at 2 kHz using an Axon Multiclamp 700B amplifier ( Molecular Devices , Sunnyvale , California , USA ) . Data were acquired at 5 kHz via an ITC18 interface board ( Instrutech , Port Washington , New York , USA ) , transmitting to a Dell computer running the Igor Pro software ( WaveMetrics , Lake Oswego , Oregon , USA ) . All experiments were carried out in the current clamp ( ‘bridge’ ) mode . Series resistance was monitored ( 10–15 MΩ ) and compensated for by adjusting the bridge balance . Data were discarded if series resistance changed by more than 30% . Data were analyzed using Igor Pro . EPSP slopes were measured on the rising phase of the EPSP as a linear fit between the time points corresponding to 25–30% and 70–75% of the peak amplitude . For statistical analysis , the mean EPSP slope per minute of the recording was calculated from 12 consecutive sweeps and normalized to the baseline . Normalized ESPS slopes from the last 5 min of the baseline ( immediately before pairing ) and from the last 5 min of the recording ( 35–40 min or 55–60 min after pairing ) were averaged . The magnitude of plasticity , as an indicator of synaptic change , was defined as the average EPSP slope after pairing expressed as a percentage of the average EPSP slope during baseline . The following drugs were used: dopamine hydrochloride 20 μM , forskolin 50 μM , d-AP5 50 μM , SCH23390 hydrochloride 10 μM , sulpiride 50 μM , H-89 20 µM . All drugs ( purchased from Sigma–Aldrich , Dorset , United Kingdom; Tocris Bioscience , Bristol , United Kindgom; or Abcam , Cambridge , United Kingdom ) were bath-applied through the perfusion system by dilution of concentrated stock solutions ( prepared in water or DMSO ) in ACSF . Statistical comparisons were made using one-sample two-tailed or paired two-tailed Student's t-test , with a significance level of α = 0 . 05 . Data are presented as mean ± s . e . m . Significance levels are indicated by p < 0 . 05 , p < 0 . 01 , *p < 0 . 001 .	To help someone learn a new task , we might give them a reward after they have performed well . However , these rewards tend to be given several seconds or minutes after the behavior they are supposed to promote . Therefore , it is unclear how the rewards affect the brain and help accelerate the learning process . Information is processed and sent around the brain by networks of cells called neurons . These networks are constantly remodeled because learning changes the connections—called synapses—that neighboring neurons signal across . Synapses can be strengthened so that signals are sent across them more easily in the future . Synapses can also be weakened , making it harder for the neurons to subsequently communicate . A chemical called dopamine is often produced in the brain when a reward is received . If dopamine is present in a synapse whilst a neuron is signaling to its neighbor , it can affect how effectively this communication occurs . Brzosko et al . have now investigated whether dopamine can also change the synapses if it is applied after signaling has already happened . The strengthening or weakening of synapses can be triggered by electrically stimulating the neurons on either side of a synapse at particular times . Brzosko et al . did this to neurons in slices of mouse brain , and then applied dopamine to the neurons . The results suggest that dopamine can reverse synaptic weakening and can even cause the synapses to strengthen . However , the dopamine had to be applied immediately after stimulation to be able to strengthen the synapse . The next challenge is to establish whether this change in synaptic strength is responsible for the change in behavior .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "short", "report", "neuroscience" ]	2015	Retroactive modulation of spike timing-dependent plasticity by dopamine
Synaptic target specificity , whereby neurons make distinct types of synapses with different target cells , is critical for brain function , yet the mechanisms driving it are poorly understood . In this study , we demonstrate Kirrel3 regulates target-specific synapse formation at hippocampal mossy fiber ( MF ) synapses , which connect dentate granule ( DG ) neurons to both CA3 and GABAergic neurons . Here , we show Kirrel3 is required for formation of MF filopodia; the structures that give rise to DG-GABA synapses and that regulate feed-forward inhibition of CA3 neurons . Consequently , loss of Kirrel3 robustly increases CA3 neuron activity in developing mice . Alterations in the Kirrel3 gene are repeatedly associated with intellectual disabilities , but the role of Kirrel3 at synapses remained largely unknown . Our findings demonstrate that subtle synaptic changes during development impact circuit function and provide the first insight toward understanding the cellular basis of Kirrel3-dependent neurodevelopmental disorders . Executing cognitive tasks requires coordination among neural circuits . Therefore , neurons usually send and receive neural information with different synaptic partners . One way neurons differentially regulate activity among partners is by forming different types of synapses with each partner ( Williams et al . , 2010; Emes and Grant , 2012 ) . This kind of synaptic target specificity is exquisitely exemplified by hippocampal mossy fiber ( MF ) synapses . MF synapses connect glutamatergic dentate granule ( DG ) neurons to glutamatergic CA3 neurons and GABAergic interneurons ( GABA neurons ) . The main DG-CA3 synapse consists of a giant presynaptic bouton apposed to a multi-headed CA3 spine called a thorny excrescence ( TE ) . In addition , filopodia project from the main bouton and synapse with nearby GABA neurons ( Frotscher , 1989; Acsády et al . , 1998 ) . Filopodial MF synapses mediate feed-forward inhibition of CA3 neurons and are essential for hippocampal function during learning and memory tasks ( Torborg et al . , 2010; Ruediger et al . , 2011 ) . Although main bouton and filopodial MF synapses are physically linked , they have different molecular and functional properties ( Toth et al . , 2000; McBain , 2008 ) . This suggests DG neurons utilize specific cues to construct different types of synapses with CA3 and GABA neurons , but the identity of the target-specific cues is unknown . Kirrel1 , 2 , and 3 ( also known as Neph1 , 3 , and 2 , respectively ) are transmembrane immunoglobulin superfamily members ( Figure 1D ) . In mice , Kirrels have been studied during formation of the kidney slit diaphragm , an adhesive cell junction for filtering blood ( Donoviel et al . , 2001; George and Holzman , 2012 ) and in axon targeting of the olfactory and vomeronasal systems ( Serizawa et al . , 2006; Prince et al . , 2013 ) . In humans , copy number variations and exonic point mutations in the Kirrel3 gene are associated with intellectual disability , autism , and Jacobsen's syndrome , a rare developmental disorder that often includes intellectual disabilities ( Bhalla et al . , 2008; Guerin et al . , 2012; Michaelson et al . , 2012; Neale et al . , 2012 ) . The Kirrel ortholog SYG-1 regulates synapse formation and axon branching in Caenorhabditis elegans ( Shen and Bargmann , 2003; Chia et al . , 2014 ) , but the role of Kirrel3 in mammalian synapse development is unknown . Here , we demonstrate Kirrel3 is a target-specific cue at MF synapses . Kirrel3 specifically regulates development of DG-GABA MF filopodia , which are necessary to constrain excitatory drive to CA3 neurons after DG stimulation . 10 . 7554/eLife . 09395 . 003Figure 1 . Kirrel3 is a synaptic molecule that mediates homophilic , trans-cellular adhesion . ( A ) Synaptosomes from mouse hippocampi from P9 , P21 , and adult ( P55 ) were immunoblotted for indicated proteins . lys; lysate . syn; synaptosome . 2 μg protein per lane . ( B , C ) 14 days in vitro ( DIV ) cultured hippocampal neurons immunostained with antibodies against Kirrel3 ( red ) , vGlut1 ( green ) , and MAGUK proteins ( blue ) . Boxed regions in B are magnified below in B′ , B′′ , and B′′′ . Neurons from Kirrel3 knockout mice have no Kirrel3 signal ( C ) . ( D ) Diagram of Kirrel3 protein and location of inserted FLAG tag . Ig; immunoglobulin . ( E–H ) Cultured hippocampal neurons were co-transfected with FLAG-Kirrel3 and GFP and immunostained for indicated proteins . Anti-FLAG antibodies were added prior to fixation to label only surface Kirrel3 . Note surface Kirrel3 is seen as puncta on axons and dendrites after synapse formation in 14DIV neurons ( E–G ) and prior to synapse formation in 4DIV neurons ( H ) . H shows that surface Kirrel3 also clusters at axon–dendrite crossings . Boxed regions in E are shown magnified in F and G . FLAG signal alone is shown in lower panels F′ , G′ , and H′ . ( I , J ) Kirrel3 clusters at cell junctions . 293HEK cells were co-transfected with GFP and FLAG-Kirrel3 , immunostained for GFP and FLAG , and nuclei labeled with Hoechst . ( K–M ) CHO cells transfected with either GFP control or GFP and Kirrel3 were tested for adhesion . Only cells expressing Kirrel3 formed aggregates . Aggregation index was calculated by dividing the total GFP fluorescence in cell aggregates by the total GFP fluorescence in the well . Mean ± SEM are shown , n = 3 , * indicates p = 0 . 001 by two-tailed t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 09395 . 00310 . 7554/eLife . 09395 . 004Figure 1—figure supplement 1 . Kirrel1 and 2 are not expressed in the hippocampus . ( A , B ) In situ hybridizations for Kirrel1 and Kirrel2 mRNA indicate little to no hippocampal expression . This is in agreement with the Allen Brain Atlas . A , B are tiled images . DOI: http://dx . doi . org/10 . 7554/eLife . 09395 . 00410 . 7554/eLife . 09395 . 005Figure 1—figure supplement 2 . Kirrel3 undergoes homophilic binding in cells . ( A ) HEK293 cells were co-transfected with GFP ( cyan ) and full-length FLAG-tagged Kirrel cDNAs . Cells were then live labeled with soluble Fc constructs ( red ) . This shows the Kirrel3 extracellular domain does not interact with Kirrel1 or Kirrel2 and the FLAG tag does not interfere with homophilic binding . ( B ) Fc binding was quantified by analyzing the percent area of GFP co-stained with Fc . The number of cells per condition from three independent experiments is K3 ( 96 ) , K1 ( 53 ) , K2 ( 70 ) , vec ( 84 ) , K3 + Fc ( 48 ) . indicates that K3-Fc binds K3 significantly ( p < 0 . 0001 ) more than any other condition as determined by ANOVA and pairwise post-tests . Mean ± SEM are shown . Kirrel1 , 2 , 3 is abbreviated K1 , 2 , 3 . ( C ) HEK293 cells were co-transfected with GFP ( cyan ) plus full-length FLAG-tagged Kirrel cDNAs as done in A and B . Cells were live labeled with anti-FLAG antibodies to confirm each Kirrel protein is properly expressed on the cell surface . DOI: http://dx . doi . org/10 . 7554/eLife . 09395 . 005 Given the association between Kirrel3 mutations and intellectual disabilities , we investigated the role of Kirrel3 in hippocampal circuits , which are critical for learning and memory , and may be impaired in patients with intellectual disabilities . Kirrel3 protein is enriched in synaptosomes prepared from hippocampal lysates with greatest enrichment at postnatal day ( P ) 21 ( Figure 1A ) . Next , we obtained Kirrel3 knockout mice , which were recently described ( Prince et al . , 2013 ) and prepared hippocampal neuron cultures from newborn wild-type and knockout mice . In hippocampal neurons cultured for 14 days in vitro ( DIV ) , Kirrel3 localizes to puncta adjacent to the pre- and post-synaptic markers vGlut1 and MAGUK in wild-type but not knockout neurons ( Figure 1B , C ) . This suggests that , like cadherins , Kirrel3 localizes to perisynaptic adhesion zones rather than the synaptic cleft . To determine if Kirrel3 is axonal , dendritic , or both , we analyzed the distribution of surface-expressed FLAG-Kirrel3 ( Figure 1D ) in sparsely transfected neurons using live labeling . Surface FLAG-Kirrel3 is seen as puncta on axons and dendrites of 14DIV neurons ( Figure 1E–G ) . Moreover , even prior to synaptogenesis in 4DIV neurons , FLAG-Kirrel3 already has a punctate distribution in axons and dendrites and clusters at axon–dendrite contact points ( Figure 1H ) . Kirrels can function via homophilic binding or heterophilic binding to nephrin , another Ig superfamily member ( Gerke et al . , 2003; Serizawa et al . , 2006 ) . However , neither nephrin nor the other Kirrel family members , Kirrel1 and Kirrel2 , have appreciable expression in the hippocampus ( Putaala et al . , 2001 ) ( Figure 1—figure supplement 1 ) . We noticed Kirrel3 clusters at cell junctions ( Figure 1I , J ) and therefore we directly tested the adhesive ability of Kirrel3 homophilic interactions using a cell aggregation assay . We demonstrate Kirrel3 mediates trans-cellular homophilic binding ( Figure 1K–M and Figure 1—figure supplement 2 ) . Taken together , our data indicate Kirrel3 is present at early axon–dendrite contacts , localizes at or near synapses , and is a bona fide homophilic adhesion molecule , all of which implicate Kirrel3 in synapse development . Next , we determined which hippocampal neurons express Kirrel3 . In developing P14 and adult hippocampi , Kirrel3 mRNA is present in two cell types: ( 1 ) DG neurons and ( 2 ) scattered cells of the hilus and area CA3 ( Figure 2A–D and Figure 2—figure supplement 1A ) . Correspondingly , Kirrel3 protein is present in the molecular and stratum lucidum layers of the hippocampus , containing DG dendrites and axons , respectively ( Figure 2E ) . It is also present in scattered cells of the hilus and area CA3 ( Figure 2F ) and faintly in the stratum lacunosum-moleculare , which contains axons from entorhinal cortex . No Kirrel3 signal was detected in knockout mice ( Figure 2G , H and Figure 2—figure supplement 1B ) . Instead , Kirrel3 knockout mice have farnesylated GFP in frame after exon 1 so they express membrane-associated GFP instead of Kirrel3 ( Prince et al . , 2013 ) . Examination of GFP expression in knockout mice indicates that again , Kirrel3 is selectively expressed by DG neurons and scattered cells of area CA3 ( Figure 2I–O ) . Notably , we never observe GFP expression in CA3 neurons ( Figure 2—figure supplement 1C–E ) . 10 . 7554/eLife . 09395 . 006Figure 2 . Hippocampal DG and GABA neurons express Kirrel3 . ( A–D ) In situ hybridizations for Kirrel3 mRNA on adult P60–P70 hippocampal sections from WT ( A–C ) and KO ( D ) mice . A negative control sense probe on WT tissue is shown in C . Red arrows in boxed region of A point to scattered Kirrel3-expressing cells shown magnified in ( B ) . ( E–H ) Hippocampal sections from Kirrel3 WT ( E , F ) and KO ( G , H ) mice were immunostained with anti-Kirrel3 antibodies ( yellow ) and Hoechst ( blue ) . F and H are magnified images of boxed regions in E and G . ( I–O ) P14 Kirrel3 KO mice with farnesylated GFP inserted in the Kirrel3 locus were immunostained with anti-GFP antibodies to identify Kirrel3-expressing cells ( green ) . Dentate granule ( DG ) dendrites and their mossy fiber ( MF ) axons are brightly labeled ( I ) as well as GABA-expressing cells ( magenta ) in area CA3 . ( P–R ) P14 Kirrel3 KO mice were immunostained for GFP ( green ) and calbindin ( Calb , magenta ) . ( S ) Analysis of Kirrel3-positive cells in P14 Kirrel3 KO mice co-expressing interneuron markers . Abbreviations: wild-type , WT; knockout , KO; molecular layer , ML; stratum lucidum , SL . Stratum lacunosum-moleculare , SLM . Images in A–D , E , G , and I are tiled . DOI: http://dx . doi . org/10 . 7554/eLife . 09395 . 00610 . 7554/eLife . 09395 . 007Figure 2—figure supplement 1 . Kirrel3 is not expressed by CA3 neurons . ( A ) In situ hybridization for Kirrel3 mRNA at P14 shows the same expression pattern as the adult brain shown in Figure 2A . ( B ) Western blot of hippocampal lysates from adult Kirrel3 wild-type ( WT ) , heterozygous ( HET ) , and knockout ( KO ) mice indicates Kirrel3 protein is absent in knockout mice as expected . ( C–E ) P25 Kirrel3 KO mice with farnesylated GFP inserted in the Kirrel3 locus were immunostained with anti-GFP antibodies to identify Kirrel3-expressing cells . Shown here are tiled confocal images of coronal sections through more ventral hippocampal sections compared to Figure 2E . Boxed region in C is magnified in D . Note that GFP is not present in CA3 neuron cell bodies or their axons that reside in the stratum radiatum ( SR ) layer . Thus , throughout the hippocampus , Kirrel3 is not expressed at detectable levels in CA3 neurons . EC; entorhinal cortex . Images A , C , and E are tiled . DOI: http://dx . doi . org/10 . 7554/eLife . 09395 . 00710 . 7554/eLife . 09395 . 008Figure 2—figure supplement 2 . Kirrel3 is expressed by mainly calbindin-positive GABA neurons . ( A ) Diagram indicating hippocampal region imaged and analyzed is represented inside the blue line . ( B ) Interneuron analysis of animals heterozygous for Kirrel3 shows little difference from knockout analysis shown in Figure 2S . ( C–F ) Sample images of interneuron staining to show each interneuron antibody worked for IHC despite having little overlap with Kirrel3 . Mice are immunostained for GFP ( green ) to mark K3-expressing neurons and interneuron markers ( magenta ) in tissue . DOI: http://dx . doi . org/10 . 7554/eLife . 09395 . 008 The scattered Kirrel3-positive cells reside mainly outside the pyramidal layer and have GFP-labeled arbors . This suggests they may be GABAergic interneurons . To test this , we co-stained P14 Kirrel3 heterozygous and knockout mice with antibodies against GFP to identify Kirrel3-expressing cells and GABA to identify GABAergic interneurons ( Figure 2I–S and Figure 2—figure supplement 2 ) . We find that nearly all Kirrel3-expressing cells express GABA ( Figure 2S ) . Conversely , about 20% of all GABA neurons in area CA3 express Kirrel3 ( Figure 2S ) . We also co-stained with several common GABA neuron markers and find that two thirds ( about 67% ) of Kirrel3/GABA neurons express calbindin ( Figure 2P–S ) and these Kirrel3-positive neurons make up about half of all calbindin-positive interneurons . Notably , glutamatergic DG neurons also highly express Kirrel3 and calbindin , providing a shared molecular profile between these neuronal populations . Kirrel3 cell populations are similar in P14 Kirrel3 heterozygous and knockout mice ( Figure 2—figure supplement 2 ) , suggesting that complete loss of Kirrel3 does not change cell fate or induce cell death of the neurons examined . Taken together , we find that , in the hippocampus , DG neurons and a subset of calbindin-positive GABA neurons selectively express Kirrel3 . Because Kirrel3 is a homophilic molecule expressed by DG and GABA neurons , we hypothesized Kirrel3 homophilic interactions may specifically regulate formation of MF filopodia connecting DG axons to GABA dendrites during development ( Figure 3A ) . To test this , we analyzed MF presynaptic morphology in Kirrel3 wild-type and knockout mice at P14 , the peak of MF synaptogenesis . DiI crystals were placed in the DG cell body layer of fixed brains . After 1 week , dye diffuses along DG axons and labels MF presynaptic terminals ( Figure 3B–G ) . We discovered Kirrel3 knockout mice have significantly fewer and shorter filopodia than wild-type ( Figure 3B–M and Figure 3—figure supplement 1 ) . In contrast , the main bouton area and perimeter are not affected by loss of Kirrel3 ( Figure 3J , K ) . To examine the main synapse in more detail , we analyzed postsynaptic CA3 TE spines by filling CA3 neurons with Alexa568 dye ( Figure 3N , O ) . TE spine head density and length are similar between P21 Kirrel3 wild-type and knockout mice ( Figure 3P , Q ) . Thus , Kirrel3 knockout mice have significant morphological defects in MF filopodia but not in pre- or post-synaptic structures of main MF synapses , suggesting that the number of DG synapses onto GABA neurons is reduced in Kirrel3 knockout mice . 10 . 7554/eLife . 09395 . 009Figure 3 . Kirrel3 regulates MF synapse form and function during development . ( A ) MF synapse diagram . TE; thorny excrescence . ( B , E ) DiI-labeled MF synapses from P14 Kirrel3 WT and KO mice . ( C , F ) 3D renderings of synapses in ( B , E ) . ( D , G ) Tracings of representative DiI-labeled MF synapses . ( H–M ) MF synapse morphology quantification . The number of filopodia per MF bouton ( H , L ) and filopodia length ( I , M ) are reduced in Kirrel3 KO mice . L and M are cumulative histograms of data shown in H and I , respectively . Area ( J ) and perimeter ( K ) of the main MF bouton are unaffected by genotype . n = 74 WT and 97 KO MF synapses from four mice of each genotype . Two-tailed t-tests: in H , = p < 0 . 001 and in I , * = p = 0 . 0001 . ( N , O ) Examples of P21 CA3 TE spines labeled by iontophoresis . ( P , Q ) Quantification of P21 spine morphology . No significant differences as determined by two-tailed t-tests . n = 16 WT neurons from four animals and 20 KO neurons from three animals . All bar graphs show mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 09395 . 00910 . 7554/eLife . 09395 . 010Figure 3—figure supplement 1 . Kirrel3 is required for normal development of MF filopodia . ( A , B ) Histograms for the number ( A ) and length ( B ) of MF filopodia in each genotype . This is the same P14 data plotted in main Figure 3H–K but here the entire range of values for all data is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 09395 . 010 If Kirrel3 knockout mice have fewer DG-GABA synapses , there should be reduced excitation to GABA neurons , particularly to Kirrel3-positive GABA neurons . Unfortunately , Kirrel3-positive GABA neurons cannot be identified in wild-type mice making it impossible to target them for direct electrophysiological studies at this time . However , reducing DG-GABA synapses in area CA3 is expected to decrease feed-forward inhibition and thereby increase excitation of CA3 neurons after DG stimulation . To test this , we recorded excitatory and inhibitory currents evoked in CA3 neurons after DG stimulation in acute slices ( Figure 4A ) . As predicted by our model , P14–P16 Kirrel3 knockout mice have a significantly increased excitatory/inhibitory ( E/I ) ratio compared to wild-type mice ( Figure 4B ) . We confirmed evoked responses were due to MF stimulation by applying DCG-IV , a metabotropic glutamate receptor agonist that selectively inhibits MF release ( Figure 4A ) ( Kamiya et al . , 1996; Yoshino et al . , 1996; Torborg et al . , 2010 ) . 10 . 7554/eLife . 09395 . 011Figure 4 . Kirrel3 regulates the activity of CA3 neurons during development . ( A ) Evoked responses at −70 and 0 mV of a CA3 neuron after stimulation of the MF pathway in Kirrel3 WT and KO mice . Lower traces show responses of the same cells after perfusion of 0 . 5 μm DCG-IV . ( B ) Average excitatory/inhibitory ( E/I ) ratio for CA3 neurons recorded from P14–P16 WT and KO mice . n = 15 cells from five WT animals and 24 cells from five KO animals . p = 0 . 02 with unpaired t-test . ( C ) Examples of anti-cFos staining in CA3 neurons from P14 mice . Note increased cell staining in Kirrel3 KO mice after 25 min stimulation ( stim ) in a novel , enriched environment . ( D ) Quantification of cFos-positive CA3 neurons at P14 . n = 14 ( WT no stim ) , 15 ( WT stim ) , 15 ( KO no stim ) , and 15 ( KO stim ) sections from three mice per condition . Two-way ANOVA indicates there is a significant difference among condition ( no stim vs stim ) and genotype . p values from post-tests are 0 . 0001 ( KO no stim vs KO stim ) and 0 . 0004 ( WT stim vs KO stim ) . ( E–H ) Quantification of MF synapse structure in adult mice . n = 115 WT and 131 KO synapses from three mice per genotype . Two-tailed t-tests indicate p = 0 . 01 ( E ) , p < 0 . 0001 ( G ) , and p = 0 . 002 ( H ) . ( I , J ) Tracings of representative DiI-labeled MF synapses from adult ( P60–P75 ) Kirrel3 WT and KO mice . ( K ) Quantification of cFos-positive cells in area CA3 of adult mice . n = 14 ( WT no stim ) , 14 ( WT stim ) , 16 ( KO no stim ) , and 16 ( KO stim ) sections from three different mice per condition . Two-way ANOVA indicates there is a significant difference among condition ( no stim vs stim ) but not genotype . p values from post-tests are 0 . 0005 ( WT no stim vs WT stim ) and 0 . 003 ( KO no stim vs KO stim ) . All graphs show mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 09395 . 01110 . 7554/eLife . 09395 . 012Figure 4—figure supplement 1 . Spontaneous mEPSC activity of CA3 and DG neurons is normal in the absence of Kirrel3 . ( A , B ) Graphs depicting time spent in motion ( gray ) and stationary ( orange ) of P14 ( A ) and adult P60–P75 ( B ) Kirrel3 WT and KO mice during cFos environmental stimulation . No significant differences by two-tailed t-test . Mean ± SEM is shown . n = 3 mice of each age and genotype . ( C , D ) Miniature excitatory postsynaptic current ( mEPSC ) amplitudes and frequencies from P17–P21 CA3 neurons . n = 12 cells from two WT animals and 14 cells from five KO animals . No significant differences by unpaired t-tests . ( E , F ) mEPSC amplitudes and frequencies from P17–P21 DG neurons . n = 16 cells from three WT animals and 21 cells from three KO animals . No significant differences by unpaired t-tests . All bar graphs show mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 09395 . 012 To test if Kirrel3 knockout mice have increased CA3 activity in an awake and behaving animal , we analyzed cFos , an immediate early gene marking recently activated neurons ( Kawashima et al . , 2014 ) , in P14 Kirrel3 wild-type and knockout mice . Mice were either removed from the home cage for immediate fixation ( no stim ) or allowed to explore an enriched environment for 25 min prior to fixation ( stim ) . At this age , basal , unstimulated cFos expression is low regardless of genotype , but with stimulation , P14 knockout mice have a twofold increase in cFos-positive CA3 neurons compared to wild-type ( Figure 4C , D ) . As a control , mice were recorded and exploration was comparable between genotypes ( Figure 4—figure supplement 1 ) . Interestingly , the number of cFos-positive neurons did not significantly increase in wild-type mice after stimulation but this is consistent with previous reports ( Waters et al . , 1997 ) and likely reflects that at P14 the CA3 is dominated by inhibition ( Figure 4B , note wild-type E/I ratio is <1 ) ( Torborg et al . , 2010 ) . Because Kirrel3 is expressed by entorhinal cortex axons that innervate the hippocampus , we also generally examined inputs to CA3 and DG neurons by recording miniature excitatory postsynaptic currents ( mEPSCs ) . No significant differences between wild-type and Kirrel3 knockout neurons in mEPSC amplitude or frequency were found for either cell type ( Figure 4—figure supplement 1 ) . Together , our functional data support the hypothesis that Kirrel3 selectively regulates formation of filopodial DG-GABA MF synapses . MF synapse complexity peaks at P14 and is refined such that adult synapses have fewer filopodia and a larger main bouton ( Wilke et al . , 2013 ) . To investigate MF synapse maturation under sustained loss of Kirrel3 , we analyzed MF morphology in adult ( P60–P75 ) wild-type and knockout mice . Both genotypes have fewer filopodia as adults than at P14 ( compare Figure 4E and Figure 3H ) , suggesting age-dependent filopodia refinement is Kirrel3-independent . However , while P14 knockout mice have fewer filopodia with a normal main bouton , adult knockout mice have fewer filopodia and a smaller main bouton ( Figure 4E–J ) . Thus , MF synapse defects worsen with age in Kirrel3 knockout mice . It is possible Kirrel3 is directly required for main bouton maturation or maintenance , but Kirrel3 is not expressed by CA3 neurons . Another possibility is that defects in the main bouton in adult knockout mice result from compensatory mechanisms enacted to dampen the dramatic increase in CA3 neuron excitability during development . To test this , we analyzed cFos expression in adult mice to determine if knockout CA3 neuron activity returns to normal . In support , unstimulated and stimulated cFos expression in CA3 neurons is similar between P60 adult wild-type and knockout mice ( Figure 4K and Figure 4—figure supplement 1 ) . Thus , our data suggest a model in which a reduction in MF filopodia during development leads to hyper-activation of CA3 neurons that , with time , causes homeostatic mechanisms to decrease the size of the main DG-CA3 MF synapse and return overall CA3 activity levels to a set point . This model remains to be tested in more detail in future studies . Interestingly , a battery of behavioral tests was recently performed on adult Kirrel3 knockout mice ( Choi et al . , 2015 ) . Kirrel3 knockout mice have only mild hyperactivity and a moderate impairment in novel object recognition compared to wild-type ( Choi et al . , 2015 ) . Our results indicate CA3 neuron activity is significantly impaired in 14-day-old animals but returns to normal by 2 months of age . This may explain why adult Kirrel3 knockout mice have only moderate behavioral impairments and suggests that younger animals may have more severe behavioral problems . In summary , we demonstrate Kirrel3 is required for normal development of MF filopodia , the synaptic structures connecting DG and GABA neurons . Because Kirrel3 is a homophilic adhesion molecule expressed by DG neurons and calbindin-positive GABA neurons in hippocampal area CA3 , our work suggests trans-cellular Kirrel3 interactions may stabilize MF filopodia contact and subsequent synapse formation between Kirrel3-expressing cells . Functionally , we demonstrate Kirrel3 is required to maintain feed-forward inhibition and constrain CA3 neuron activity in young animals . Although a few molecules have been shown to generally affect MF presynapse formation ( Danzer et al . , 2008; Williams et al . , 2011; Lanore et al . , 2012; Wilke et al . , 2012 ) , Kirrel3 is among the first shown to selectively act on MF filopodia and our results provide evidence that main and filopodial MF synapse formation can be uncoupled . The role of calbindin-positive GABA neurons in the hippocampus is little studied , but our findings suggest they may receive a substantial fraction of DG MF input and be a critical component of feed-forward inhibitory circuits regulating CA3 activity . Given that alterations in the Kirrel3 gene are associated with autism and intellectual disabilities , this work provides the first insight into cellular mechanisms that may underlie Kirrel3-dependent neurological disorders . Our work suggests Kirrel3 loss selectively reduces excitatory synapses made onto inhibitory neurons in hippocampal area CA3 . Consequently , CA3 neurons are over-active in young Kirrel3 knockout animals . Altered E/I ratios are hypothesized to underlie neurodevelopmental disorders ( Rubenstein and Merzenich , 2003; Baroncelli et al . , 2011; Fakhoury , 2015 ) , and our work suggests this may be an important circuit defect of Kirrel3-associated diseases . Interestingly , such a strong imbalance probably would not occur if all synapses were equally impaired . In this way , loss of synaptic specificity molecules can cause widespread circuit dysfunction . There is much evidence that hippocampal circuits are altered in autism and intellectual disabilities , but it is likely not the only brain region affected in these complex neurological disorders ( Philip et al . , 2012; Zoghbi and Bear , 2012 ) . Kirrel3 is also expressed by specific populations of neurons outside the hippocampus ( Lein et al . , 2007; Choi et al . , 2015 ) . Thus , Kirrel3 may also regulate formation of other specific types of synapses throughout the brain , which may further contribute to its association with neurodevelopmental disorders . Generation of the Kirrel3 knockout mouse line was recently reported ( Prince et al . , 2013 ) . All animals and experiments were maintained and conducted in accordance with the NIH guidelines on the care and use of animals and approved by the University of Utah IACUC committee . Kirrel3 cDNA was kindly provided by Dr Hitoshi Sakano and Kirrel1 and Kirrel2 cDNAs were obtained from OpenBiosystems . Standard PCR cloning was used to move cDNAs to the pBos vector and add an extracellular FLAG tag after the signal sequence . Full-length Kirrel1 , 2 , and 3 cDNAs were used to generate sense and anti-sense probes . Standard DIG-labeled , non-radioactive in situ hybridization protocol was carried out using the Roche DIG-labeling kit on coronal cryosections of brain tissue . Neurons: P0 rat cortical glia were cultured on PDL/collagen-coated coverslips to form a monolayer . 1 week later , P0 mouse or rat hippocampi were dissected in cold 4- ( 2-hydroxyethyl ) -1-piperazineethanesulfonic acid ( HEPES ) -buffered saline solution , incubated in papain for 30 min , dissociated , and plated to glial monolayers at 4–5 × 104 cells/ml . All media was from Life Technologies ( Carlsbad , CA , United States ) . Glia media: DMEM , 10% Fetal Bovine Serum ( FBS ) , 75 mM glucose , and penicillin/streptomycin . Neuron-plating media: MEM , 10% horse serum , 50 mM glucose , 0 . 250 mM pyruvic acid , 2 mM Glutamax , 100 U/ml penicillin , 100 μg/ml streptomycin . Neuron-feeding media: Neurobasal A , B27 , 30 mM glucose , 0 . 5 mM Glutamax , 20 U/ml penicillin , 20 μg/ml streptomycin . Neuron transfections were done using the calcium-phosphate method ( Dudek et al . , 2001 ) . Cell lines: 293HEK media: DMEM , 10% FBS , and penicillin/streptomycin . CHO media: F12K media , 10% FBS , and penicillin/streptomycin . Cell line transfections were done using polyethylenimine ( PEI , Polysciences , Warrington , PA , United States ) at a ratio of 5 μg PEI/1 μg DNA . CHO cells were co-transfected with FLAG-Kirrel3 pBOS ( 4 μg ) and GFP pBOS ( 2 μg ) using PEI . 48 hr later , cells were washed with HEPES Mg2+ free ( HMF ) buffer ( 137 mM NaCl , 5 . 4 mM KCl , 1 mM CaCl2 , 0 . 34 mM Na2HPO4 , 5 . 5 mM glucose , 10 mM HEPES , pH 7 . 4 adjusted with NaOH ) and detached from the dishes using 0 . 01% trypsin in HMF . Detached cells were spun down , resuspended in HMF , counted , and 100 , 000 cells were pipetted into single wells of 24-well plates precoated with 1% BSA in HMF . Subsequently , the plates were placed on a nutator for 90 min at 37°C . The cells were then fixed with paraformaldehyde ( PFA ) ( 4% final concentration ) , transferred to a 96-well glass bottom plate , and imaged in a Zeiss LSM 710 confocal microscope . The aggregation index was calculated by dividing the total GFP fluorescence in cell aggregates by the total GFP fluorescence in the well . Analysis was done using ImageJ . Cultured cells were fixed in 4% PFA for 10 min , washed with phosphate-buffered saline ( PBS ) , and incubated in blocking solution ( PBS with 3% bovine albumin and 0 . 1% Triton-X100 ) for 30 min . Primary antibody was diluted in blocking solution and incubated on cells for 1–2 hr . After three washes , secondary antibody was incubated for 45 min , washed , and cells were mounted for imaging using Fluoromount-G ( Southern Biotech , Birmingham , AL , United States ) . For live labeling , cells were incubated with anti-FLAG antibody diluted 1:250 in serum-free media for 20 min in the culture incubator . Cells were washed , fixed with PFA , and immunostained as above . For tissue sections , mice were transcardially perfused with 4% PFA . Brains were post-fixed in PFA overnight and 50–100 μm vibratome sections were cut . Sections were incubated in blocking solution ( PBS , 3% BSA , 0 . 2% Triton-X100 ) for more than 1 hr and incubated in primary antibody at 4°C overnight with gentle shaking . For VIP immunostaining , the blocking solution contained 0 . 3% triton +0 . 1% saponin . Secondary antibody incubation was done at room temperature for 2 hr . Sections were mounted in Fluoromount-G for imaging . Primary antibodies were used as follows: rabbit anti-Kirrel3 1:2000 ( this study , generated against C-terminal peptide ) , rabbit anti-GABA 1:5000 ( Sigma , St . Louis , MO , United States ) , goat anti-GFP 1:5000 ( Abcam , Cambridge , MA , United States ) , guinea pig anti-VGLUT1 1:10 , 000 ( Millipore , Billerica , MA , United States ) , mouse anti-MAGUK 1:1000 ( NeuroMab , UC Davis/NIH NeuroMAB Facility , Davis , CA , United States ) , mouse anti-FLAG M2 1:5000 ( Sigma ) , rabbit anti-synapsin 1:1000 ( Millipore ) , rabbit anti-cFos 1:500 ( Santa Cruz Biotech , Dallas , TX ) , rabbit anti-GFP 1:1000 ( Invitrogen , Waltham , MA , United States ) , chick anti-FLAG 1:1000 ( Gallus Immunotech , Cary , NC , United States ) , mouse anti-PSD95 1:2000 ( Thermo Scientific ) , mouse anti-GAPDH 1:5000 ( Millipore ) , chick anti-MAP2 1:10 , 000 ( Abcam ) , rabbit anti-calretinin 1:2000 ( Swant , Switzerland ) , mouse anti-parvalbumin 1:5000 ( Swant ) , mouse anti-CamKII 1:5000 ( Millipore ) , rat anti-Somatostatin 1:500 ( Chemicon ) , rabbit anti-calbindin d28k 1:2000 ( Swant ) , rabbit anti-VIP 1:500 ( Immunostar , Hudson , WI , United States ) . All secondary antibodies were obtained from Jackson ImmunoResearch ( West Grove , PA , United States ) and used at 1:1000 . The extracellular domain of Kirrel3 was cloned in frame with human Fc protein . Kirrel3-Fc was transfected into HEK293 cells using PEI . Cells were incubated in OptiMEM ( Life Technologies ) media for 5 days . Then , the Kirrel3-Fc-conditioned media was harvested and concentrated using Amicon Ultra filter units ( Millipore ) . Kirrel3-Fc concentration was estimated by Western blot using known concentrations of purified Fc ( Jackson Immuno ) as a standard . Kirrel3-Fc was pre-clustered by incubating it in OptiMEM plus anti-human Cy3 secondary antibodies at 1:100 . Kirrel3-Fc and control Fc ( used at ∼1 μg/ml ) were then tested for binding to transfected HEK293 cells using the live label immunostaining method described above . Synaptosomes were prepared according to methods described by Jones and Matus with minor modifications ( Jones and Matus , 1974 ) . Briefly , hippocampi were dissected from mice . Tissue was homogenized with a Dounce homogenizer ( 20% wt/vol ) in ice-cold 0 . 32 M sucrose +20 mM HEPES , pH 7 . 4 supplemented with protease inhibitors . Homogenates were cleared by spinning at 1000×g for 10 min at 4°C . The supernatant was spun at 17 , 000×g for 15 min . The pellet containing crude synaptosomes was resuspended in 0 . 32 M sucrose and 20 mM HEPES . Protein concentration was quantified with a BCA assay ( Thermo Scientific ) . 2 μg of synaptosomal or cleared lysate proteins was loaded per lane for Western blot analysis . Proteins were run on Bis-Tris gradient acrylamide gels and transferred to nitrocellulose membranes using the iBlot system ( Life Technologies ) . Membranes were incubated in blocking solution ( 50 mM Tris pH 7 . 5 , 300 mM NaCl , 3% wt/vol dry milk powder , and 0 . 05% Tween-20 ) for 10–60 min , primary antibody overnight at 4°C , washed , incubated in HRP-conjugated secondary antibodies ( Jackson Immuno ) for 1 hr at room temperature and then detected using the Bio-Rad Clarity ECL kit on a Bio-Rad ChemiDoc XRS+ imaging system . To prepare hippocampal lysates , 100 mg of hippocampal tissue was homogenized in 1 ml of reducing sample buffer . DiI crystals ( Life Technologies ) were placed in the DG of perfused hippocampi and incubated at 37°C in 2% PFA for 1 week . MF synapses from the suprapyramidal bundle in area CA3a/b were imaged , deconvolved in AutoQuant3 ( Bitplane ) , and analyzed in ImageJ . Filopodia length was analyzed in 3D using the Simple Neurite Tracer plug-in , while area and perimeter of the main bouton were analyzed in 2D . Neurons were microinjected with fluorescent dye as described ( Dumitriu et al . , 2011 ) . Briefly , P21 pups were perfused with 1% PFA in phosphate buffer ( PB ) for 1 min , followed by 4% PFA with 0 . 125% glutaraldehyde in PB for 9 min . The brains were extracted and post-fixed in 4% PFA for 30 min . 200 μm-thick coronal hippocampal slices were cut on a vibratome . Slices were submerged in 0 . 1 M PB and viewed through an Olympus BX51WI microscope coupled to a light source and fluorescent filters . High-resistance ( 150–250 MΩ ) glass pipettes were pulled on a Flaming/Brown P-97 Sutter pipette puller and backfilled with 10 mM Alexa568 ( dissolved in 200 mM KCl , Life Technologies ) . The pipettes were mounted on a micromanipulator connected to an S44 Grass square pulse stimulator . The pipette tip was gently advanced in tissue towards the cell of interest . On contact and penetration , a step stimulus of 1–5 V was used to inject the dye in the cell . Filled neurons were imaged on a Zeiss LSM710 confocal microscope . P14 and adult P60 wild-type and knockout mice were either removed from the home cage for immediate fixation by transcardial perfusion ( unstimulated ) or allowed to explore an enriched environment for 25 min prior to fixation ( stimulated ) . The enriched environment consisted of mice placed individually into a 40 × 40 clear plastic box containing five novel objects spaced 10 cm apart . Mice were allowed to explore freely for 25 min . Immunostaining was conducted as described in the above methods . When possible , experiments were conducted by an experimenter blind to condition or genotype . Sample sizes were based on previous experiments or power analysis . Statistics were calculated in Prism ( GraphPad ) . Intensity levels of some images were adjusted for visibility in publication but if so , the entire field of view and all comparable conditions were adjusted similarly . All images and conditions from the same experiment were collected and analyzed using the same confocal and analysis settings . Mice were rapidly decapitated and their brains carefully removed and kept in iced , artificial cerebrospinal fluid ( aCSF ) with sucrose ( in mM—sucrose 200 , KCl 3 , Na2PO4 1 . 4 , MgSO4 3 , NaHCO3 26 , glucose 10 , and CaCl2 0 . 5 ) . 300-μm thick transverse slices were cut on a Leica vibratome ( Leica VT1200 ) and the slices were left at room temperature in the holding chamber , until recording . P14–P16 mice were used for E/I ratio experiments and P17–P21 mice were used for mEPSC experiments . Neurons were visualized by differential interference contrast using a bright light source and an infrared filter on an Olympus BX51WI microscope with attached Hitachi color CCD camera ( KP-D20BU ) . For mEPSC recordings: slices were continuously superfused with aCSF containing ( in mM ) NaCl 126 , NaHCO3 26 , KCl 3 , NaH2PO4 1 . 4 , CaCl2 2 . 5 , MgSO4 1 , D-glucose 10 , and TTX 1 bubbled with 95% O2–5% CO2 . The intracellular pipette solution for mEPSC recordings contained ( in mM ) cesium methylsulfonate 80 , CsCl 60 , HEPES 10 , EGTA 1 ( adjusted with CsOH ) , CaCl2 0 . 5 , glucose 10 , and QX-314 5 , adjusted to 290–300 mOsm/Lt at pH 7 . 3 . For E/I ratio experiments , the aCSF was as above but without TTX and the intracellular solution was ( in mM ) cesium methylsulfonate 132 , CsCl 8 , HEPES 10 , EGTA 1 ( adjusted with CsOH ) , CaCl2 0 . 5 , glucose 10 , and QX-314 5 , adjusted to 290–300 mOsm/Lt at pH 7 . 3 . DCG-IV 0 . 5 μM ( CAS no . 147782-19-2 , TOCRIS ) was perfused in the ACSF in some E/I ratio experiments to verify the specificity of stimulation . Unless noted , chemicals were sourced from Fisher Scientific ( Pittsburgh , PA , United States ) . Somatic whole-cell recordings were performed with Axon Multiclamp 700B amplifier ( Molecular Devices , CA , United States ) in voltage clamp mode at 34 ± 1°C bath temperature for mEPSC experiments and at room temperature ( ∼22°C ) for E/I ratio experiments . Data acquisition was performed via an Axon Digidata 1550 ( Molecular Devices , CA , United States ) with pClamp ( Version 10 , Molecular Devices , CA , United States ) . Current signals were sampled at 1 kHz and filtered with a 2-kHz Bessel filter . Patch pipettes with a tip resistance of 5–10 MΩ were pulled with a Flaming/Brown micropipette puller P-97 ( Sutter Instruments and Co . ) using borosilicate glass capillaries with filaments ( 1B150F-4 , World Precision Instruments ) . Grass stimulator ( S88 , Grass Instruments ) and bipolar tungsten electrodes ( Harvard Apparatus , MA , United States , Ref . no . 72-0375 ) were used to deliver extracellular stimulation to the MF pathway at the hilus of the DG .	Nerve cells in the brain connect to each other via junctions called synapses to form vast networks that process information . Much like streets can be joined with stop signs , traffic lights , or exit ramps depending on the flow of traffic , different types of synapses control the flow of information along nerves in distinct ways . In a region of the brain called the hippocampus , nerve cells called DG neurons are connected to other neurons by two different types of synapses . One type of synapse allows the DG neurons to activate CA3 neurons , while the second type allows the DG neurons to activate GABAergic neurons . These same GABAergic neurons can then inhibit the activity of the CA3 neurons . Therefore , through these two different types of synapses , DG neurons can both increase and decrease the activity of the CA3 neurons . This delicate balance of activity across the two types of DG synapses is very important for the hippocampus to work properly , which is critical for our ability to learn and remember . Mutations in the gene that encodes a protein called Kirrel3 are associated with autism , Jacobsen's syndrome , and other disorders that affect intellectual ability in humans . Kirrel3 is similar to a protein found in roundworms that regulates the formation of synapses , but it is not known if it plays the same role in humans and other mammals . Now , Martin , Muralidhar et al . studied the role of Kirrel3 in mice . The experiments show that Kirrel3 is produced in both the DG neurons and the GABAergic neurons , but not the CA3 neurons . Young mutant mice that lacked Kirrel3 made fewer synapse-forming structures between DG neurons and GABAergic neurons than normal mice , but the synapses that connect DG neurons to CA3 neurons formed normally . This disrupted the balance of activity across the two types of DG synapses and the CA3 neurons in the mutant mice were over-active . Together , Martin , Muralidhar et al . 's findings show that altering the levels of Kirrel3 can alter the balance of synapses in the hippocampus . This may explain how even very small changes in synapse formation during brain development can have a big impact on nerve cell activity . The next challenge is to understand exactly how Kirrel3 helps build synapses , which may lead to the development of new drugs that help to rebalance brain activity in people that lack Kirrel3 .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "short", "report", "neuroscience" ]	2015	The intellectual disability gene Kirrel3 regulates target-specific mossy fiber synapse development in the hippocampus
Gene regulation relies on transcription factors ( TFs ) exploring the nucleus searching their targets . So far , most studies have focused on how fast TFs diffuse , underestimating the role of nuclear architecture . We implemented a single-molecule tracking assay to determine TFs dynamics . We found that c-Myc is a global explorer of the nucleus . In contrast , the positive transcription elongation factor P-TEFb is a local explorer that oversamples its environment . Consequently , each c-Myc molecule is equally available for all nuclear sites while P-TEFb reaches its targets in a position-dependent manner . Our observations are consistent with a model in which the exploration geometry of TFs is restrained by their interactions with nuclear structures and not by exclusion . The geometry-controlled kinetics of TFs target-search illustrates the influence of nuclear architecture on gene regulation , and has strong implications on how proteins react in the nucleus and how their function can be regulated in space and time . The nucleus is a complex environment where biochemical reactions are spatially organized in an interaction network devoted to transcription , replication , or repair of the genome ( Misteli , 2001 ) . Molecular interactions relevant to gene regulation involve transcription factors ( TFs ) that bind to specific DNA regulatory sequences or other components of the transcriptional machinery . In order to find their targets , TFs diffuse within the seemingly non-compartmentalized yet highly organized nuclear volume . Since the kinetics of a reaction can be largely determined by the mobility characteristics of the reactants ( Rice , 1985; Shlesinger and Zaslavsky , 1993 ) , the target-search strategy of TFs is a key element to understand the dynamics of transcriptional activity and regulation . Over the past decade , the nuclear dynamics of TFs has become an important topic of research and has been investigated with a variety of imaging and biochemical approaches . Overall , these studies have emphasized the high mobility of nuclear factors , which results from a combination of diffusive motion and transient specific and non-specific interactions with chromatin ( Darzacq et al . , 2009; Mueller et al . , 2010; Normanno et al . , 2012 ) . These transient interactions are essential to ensure a fine regulation of binding site occupancy—by competition or by altering the TF concentration—but must also be persistent enough to enable the assembly of multicomponent complexes ( Dundr , 2002; Darzacq and Singer , 2008; Gorski et al . , 2008; Cisse et al . , 2013 ) . In parallel to the experimental evidence of the fast diffusive motion of nuclear factors , our understanding of the intranuclear space has evolved from a homogeneous environment to an organelle where spatial arrangement among genes and regulatory sequences play an important role in transcriptional control ( Heard and Bickmore , 2007 ) . The nucleus of eukaryotes displays a hierarchy of organized structures ( Gibcus and Dekker , 2013 ) and is often referred to as a crowded environment . How crowding influences transport properties of macromolecules and organelles in the cell is a fundamental question in quantitative molecular biology . While a restriction of the available space for diffusion can slow down transport processes , it can also channel molecules towards their targets increasing their chance to meet interacting partners . A widespread observation in quantitative cell biology is that the diffusion of molecules is anomalous , often attributed to crowding in the nucleoplasm , cytoplasm , or in the membranes of the cell ( Höfling and Franosch , 2013 ) . An open debate remains on how to determine whether diffusion is anomalous or normal ( Malchus and Weiss , 2009; Saxton , 2012 ) , and the mechanisms behind anomalous diffusion ( Saxton , 2007 ) . The answer to these questions bears important consequences for the understanding of the biochemical reactions of the cell . The problem of diffusing molecules in non-homogenous media has been investigated in different fields . Following the seminal work of de Gennes ( 1982a ) , ( 1982b ) in polymer physics , the study of diffusivity of particles and their reactivity has been generalized to random or disordered media ( Kopelman , 1986; Lindenberg et al . , 1991 ) . These works have set a framework to interpret the mobility of macromolecular complexes in the cell , and recently in terms of kinetics of biochemical reactions ( Condamin et al . , 2007 ) . Experimental evidence has also been found , showing the influence of the glass-like properties of the bacterial cytoplasm in the molecular dynamics of intracellular processes ( Parry et al . , 2014 ) . These studies demonstrate that the geometry of the medium in which diffusion takes place has important repercussions for the search kinetics of molecules . The notion of compact and non-compact exploration was introduced by de Gennes ( 1982a ) in the context of dense polymers and describes two fundamental types of diffusive behavior . While a non-compact explorer leaves a significant number of available sites unvisited , a compact explorer performs a redundant exploration of the space . In chemistry , the influence of compactness is well established to describe dimensional effects on reaction rates ( Kopelman , 1986 ) . In this study , we aim to elucidate the existence of different types of mobility of TFs in the eukaryotic nucleus , as well as the principles governing nuclear exploration of factors relevant to transcriptional control . To this end , we used single-molecule ( SM ) imaging to address the relationship between the nuclear geometry and the search dynamics of two nuclear factors having distinct functional roles: the proto-oncogene c-Myc and the positive transcription elongation factor ( P-TEFb ) . c-Myc is a basic helix-loop-helix DNA-binding transcription factor that binds to E-Boxes; 18 , 000 E-boxes are found in the genome , and c-Myc affects the transcription of numerous genes ( Gallant and Steiger , 2009 ) . Recently , c-Myc has been demonstrated to be a general transcriptional activator upregulating transcription of nearly all genes ( Lin et al . , 2012; Nie et al . , 2012 ) . P-TEFb is an essential actor in the transcription regulation driven by RNA Polymerase II . P-TEFb is a cyclin-dependent kinase , comprising a CDK9 and a Cyclin T subunit . It phosphorylates the elongation control factors SPT5 and NELF to allow productive elongation of class II gene transcription ( Wada et al . , 1998 ) . The carboxy-terminal domain ( CTD ) of the catalytic subunit RPB1 of polymerase II is also a major target of P-TEFb ( Zhou et al . , 2012 ) . c-Myc and P-TEFb are therefore two good examples of transcriptional regulators binding to numerous sites in the nucleus; the latter binds to the transcription machinery itself and the former directly to DNA . Single particle tracking ( SPT ) constitutes a powerful method to probe the mobility of molecules in living cells ( Lord et al . , 2010 ) . In the nucleus , SPT has been first employed to investigate the dynamics of mRNAs ( Fusco et al . , 2003; Shav-Tal et al . , 2004 ) or for rheological measurements of the nucleoplasm using inert probes ( Bancaud et al . , 2009 ) . Recently , the tracking of single nuclear factors has been facilitated by the advent of efficient in situ tagging methods such as Halo tags ( Mazza et al . , 2012 ) . An alternative approach takes advantage of photoconvertible tags ( Lippincott-Schwartz and Patterson , 2009 ) and photoactivated localization microscopy ( PALM ) ( Betzig et al . , 2006; Hess et al . , 2006 ) . Single particle tracking PALM ( sptPALM ) was first used to achieve high-density diffusion maps of membrane proteins ( Manley et al . , 2008 ) . However , sptPALM experiments have typically been limited to proteins with slow mobility ( Manley et al . , 2008 ) or those that undergo restricted motions ( Frost et al . , 2010; English et al . , 2011 ) . Recently , by inclusion of light-sheet illumination , it has been used to determine the binding characteristics of TFs to DNA ( Gebhardt et al . , 2013 ) . In this study , we developed a new sptPALM procedure adapted for the recording of individual proteins rapidly diffusing in the nucleus of mammalian cells . We used the photoconvertible fluorophore Dendra2 ( Gurskaya et al . , 2006 ) and took advantage of tilted illumination ( Tokunaga et al . , 2008 ) . A careful control of the photoconversion rate minimized the background signal due to out-of-focus activated molecules , and we could thus follow the motion of individual proteins freely diffusing within the nuclear volume . With this sptPALM technique , we recorded large data sets ( on the order of 104 single translocations in a single imaging session ) , which were essential for a proper statistical analysis of the search dynamics . We applied our technique to several nuclear proteins and found that diffusing factors do not sense a unique nucleoplasmic architecture: c-Myc and P-TEFb adopt different nuclear space-exploration strategies , which drastically change the way they reach their specific targets . The differences observed between the two factors were not due to their diffusive kinetic parameters but to the geometry of their exploration path . c-Myc and our control protein , ‘free’ Dendra2 , showed free diffusion in a three-dimensional nuclear space . In contrast , P-TEFb explored the nuclear volume by sampling a space of reduced dimensionality , displaying characteristics of exploration constrained in fractal structures . The role of the space-sampling mode in the search strategy has long been discussed from a theoretical point of view ( de Gennes , 1982a; Kopelman , 1986; Lindenberg et al . , 1991 ) . Our experimental results support the notion that it could indeed be a key parameter for diffusion-limited chemical reactions in the closed environment of the nucleus ( Bénichou et al . , 2010 ) . We discuss the implications of our observations in terms of gene expression control , and its relation to the spatial organization of genes within the nucleus . We developed a simple and versatile approach based on photoconvertible protein tags that extends the use of sptPALM to any protein expressed in mammalian cells . Proteins of interest were fused to the photoconvertible protein Dendra2 ( Gurskaya et al . , 2006; Figure 1A ) . A standard wide-field configuration of the microscope allowed fast and sensitive acquisition with an EMCCD camera ( ‘Materials and methods—Single-molecule imaging and Detection and tracking of single molecules’ , Figure 1—figure supplement 1 ) . We used low activation intensity and tilted illumination ( Figure 1B ) in order to reach the regime of SM detection , characterized by single-step activation and photobleaching ( Figure 1C ) . Due to activation of out-of-focus fluorophores , a decreasing density of detected particles was correlated with an increasing average signal-to-noise ratio ( SNR ) ( Figure 1D ) . We found that activation intensity around 0 . 01 kW/cm2 offered the best trade-off between the number of detected particles ( ∼1 ) and SNR . 10 . 7554/eLife . 02230 . 003Figure 1 . From bulk to single molecule fluorescence imaging . ( A ) Images of the 525 nm bulk emission of the pre-converted form of Dendra2 in the cellular nucleus for the ‘free’ fluorophore Dendra2 and Dendra2 fused to H2B , c-Myc , and P-TEFb . ( B ) Schematics of the intracellular sptPALM; wide-field illumination is necessary in order to reach the nucleus of mammalian cells . A signature of single molecule detection is the on/off single-step fluorescence shown in panel ( C ) . To achieve single molecule detection , 405 nm laser photoactivation needs to be reduced to a level where no background noise is produced by out-of-focus fluorophores . Graphic in panel ( D ) shows the number of detected single molecules ( blue data , right axis ) and the mean SNR of the single molecule signal ( red data , left axis ) as a function of 405 nm photoactivation photon flux per pulse ( 10 ms pulses every 1 s ) . The signal-to-noise ratio ( SNR ) of the molecules within the image depth of focus indeed increases as the total number of detected particles decreases . In panel ( E ) , the trace of a single Dendra2 molecule freely diffusing in the nucleus of a living cell is depicted , imaged at a rate of 95 Hz ( 10 ms acquisition time and 0 . 5 ms interval between frames ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 00310 . 7554/eLife . 02230 . 004Figure 1—figure supplement 1 . Motion blur and detection algorithm . In panel A , schematics of the experimental set-up . The activation ( 405 nm ) laser and the excitation ( 561 nm ) laser were aligned on a single beam using a dichroic beam splitter . Their intensity and on/off switching ratio were independently controlled with an acousto-optic tunable filter ( AOTF ) . The combined laser beam was expanded through a beam expander and focused on the rear plane of the objective in an inverted microscope , with the help of a long-pass dichroic . The emission from the sample was imaged through the tube lens with an EMCCD . Panel B shows the transition of a fast diffusing particle to a bound state . The image of the moving particle results in a motion blur whereas the image of the bound molecule is a well-defined PSF . In panel C , schematics of the steps followed by the detection algorithm for both a well-defined PSF and a motion blur . The initial image ( i ) was smoothed by a Gaussian mask and the threshold value set as the 80% percentile of the raw image ( ii ) . Finally , the image was binarized according to the threshold and aggregates of sufficient size set as a positive detection . Panel D shows the estimation of the experimental localization accuracy computed for an immobile-like H2B protein . The position of the centroid of 29 consecutive detections represents the experimental error of the detection . Computation of the standard deviation of the mean along the X and Y axes is a measure of the pointing accuracy . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 00410 . 7554/eLife . 02230 . 005Figure 1—figure supplement 2 . Tracking algorithm . In A , decay of fluorescence intensity of the ensemble of Dendra2 fluorophores in the nucleus of a cell , after strong pulsed activation . The decay was fitted to a single exponential of lifetime τ = 600 ms . After 5 s of illumination , 99 . 9999% of the particles have bleached , allowing us to compute the misconnection probability . For each detected molecule , detections located at a time distance of at least 5 s were taken into account . Those detections have a negligible probability of 10−6 to arise from the same protein . For a given radius R , we could therefore compute the expected number of misconnections . In panel B , the number of connections , misconnections , and their difference as a function of the cut-off tracking radius R , for each protein under study are shown . Gradual increase of the cut-off distance increases the number of both connections and misconnections . The optimal cut-off radius R is reached when the number of connections reaches a plateau and only the number of misconnections increases . For all the proteins , we found an optimal value of ∼2 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 00510 . 7554/eLife . 02230 . 006Figure 1—figure supplement 3 . Localization accuracy and detection efficiency as a function of diffusion coefficient . In panels A and B , step translocation histograms as retrieved by the detection and tracking algorithms on simulated videos of particles following pure Brownian diffusion , with different diffusion coefficients ( D = 1 μm2/s and D = 10 μm2/s for A and B , respectively ) . In panel C , the localization accuracy of the detection is plotted as a function of the diffusion coefficient of the particle , for a series of simulated videos with increasing diffusion coefficient ( 0 . 001 , 0 . 01 , 0 . 1 , 1 , 10 , and 20 μm2/s ) . In panel D , the percentage of detected particles is plotted as a function their distance to the focal plane , averaged over the acquisition time , for two diffusion coefficients: 1 μm2/s and 10 μm2/s . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 006 Compared to membrane proteins or other proteins with constrained mobility , diffusion dynamics of intracellular molecules is much higher and can exceed 10 μm2/s . Images recorded for such fast moving objects depart from the well-defined point spread function ( PSF ) of the microscope and exhibit a motion blur that cannot be characterized with standard Gaussian localization algorithms ( Thompson et al . , 2002 ) . Therefore , we developed new localization and tracking algorithms ( ‘Materials and methods—Detection and tracking of single molecules’ and Figure 1—figure supplements 1 and 2 ) and validated them with simulations ( ‘Materials and methods—Numerical simulations’ and Figure 1—figure supplement 3 ) . We could thus obtain single trajectories formed by individual translocations recorded every 10 ms . 50% of the traces were reconstructed with more than four time points , and some of them were as long as 60 consecutive translocations . The step size of single translocations ranged between tens of nanometers ( limited by our localization accuracy of ∼70 nm ) and ∼2 μm ( Figure 1E and Videos 1–5 ) . Hence , it became possible to track molecules with diffusion coefficients exceeding 10 μm2/s . 10 . 7554/eLife . 02230 . 007Video 1 . Raw video of a single Dendra2 molecule diffusing in the nucleoplasm of a U2OS cell . Running parallel to the raw image , reconstruction of the trace by the localization and tracking algorithms . Exposure time was 10 ms , with 0 . 5 ms dead time between frames . Running time and scale bars are stamped on the video . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 00710 . 7554/eLife . 02230 . 008Video 2 . Raw video of a single H2B molecule in the nucleoplasm of a U2OS cell . Running parallel to the raw image , reconstruction of the trace by the localization and tracking algorithms . Exposure time was 10 ms , with 0 . 5 ms dead time between frames . Running time and scale bars are stamped on the video . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 00810 . 7554/eLife . 02230 . 009Video 3 . Raw video of a single c-Myc molecule displaying slow diffusion ( D2 ≈ 0 . 5 µm2/s ) in the nucleoplasm of a U2OS cell . Running parallel to the raw image , reconstruction of the trace by the localization and tracking algorithms . Exposure time was 10 ms , with 0 . 5 ms dead time between frames . Running time and scale bars are stamped on the video . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 00910 . 7554/eLife . 02230 . 010Video 4 . Raw video of a single c-Myc molecule displaying fast diffusion ( D1 ≈ 13 . 5 µm2/s ) in the nucleoplasm of a U2OS cell . Running parallel to the raw image , reconstruction of the trace by the localization and tracking algorithms . Exposure time was 10 ms , with 0 . 5 ms dead time between frames . Running time and scale bars are stamped on the video . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 01010 . 7554/eLife . 02230 . 011Video 5 . Raw video of a single P-TEFb molecule diffusing in the nucleoplasm of a U2OS cell . Running parallel to the raw image , reconstruction of the trace by the localization and tracking algorithms . Exposure time was 10 ms , with 0 . 5 ms dead time between frames . Running time and scale bars are stamped on the video . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 011 We first investigated two limit cases relevant to protein dynamics in the nucleoplasm: Dendra2 and DNA-associated histone H2B . Dendra2 is the fluorescent label that we fused to all other proteins used in our analysis . Green fluorescent protein ( GFP ) has no detectable interacting partners in mammalian cells ( Trinkle-Mulcahy et al . , 2008 ) , and we therefore considered ‘free’ Dendra2 as a model for freely diffusing particles due to its structural similarity with GFP . In contrast , Dendra2 fused to histone H2B ( Dendra2-H2B ) was expected to insert into chromatin and thus to display restricted motion . Indeed , from a visual inspection , ‘free’ Dendra2 and Dendra2-H2B trajectories ( Figure 2A , B , respectively ) exhibited obvious differences . Notably , translocation histograms for ‘free’ Dendra2 and for Dendra2-H2B were not consistent with a single diffusing species ( Figure 2—figure supplement 1 , ‘Materials and methods–Cumulative histogram analysis and mean square displacement’ ) , thus suggesting that displacements of these molecules were more complex than anticipated . Three distinct populations were needed to fit the translocation histograms at all time intervals ( Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 02230 . 012Figure 2 . Diffusion properties of ‘free’ Dendra2 and H2B . Examples of single molecule traces of the free fluorophore Dendra2 ( A ) and DNA-associated histone H2B ( B ) . In ( C ) , the averaged mean square displacement ( MSD ) as a function of time is represented for both proteins , with an interval of confidence of 95% . The averaged MSD curves were computed from a total of 18 , 364 trajectories ( from 39 cells ) for Dendra2 , and 40 , 546 trajectories ( from 32 cells ) for H2B . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 01210 . 7554/eLife . 02230 . 013Figure 2—figure supplement 1 . Translocation histograms of Dendra2 and H2B . Translocation histograms of Dendra2 ( A ) and H2B ( B ) , plotted for 1Δt ( 10 ms ) , 3Δt , and 6Δt . Fits of the step size distribution with one Brownian diffusive population ( dotted line ) , two populations ( dashed lines ) , and three diffusive populations ( red solid line ) are represented in the graphs . Three distinct populations were thus needed to fit the translocation histograms at all time intervals . For ‘free’ Dendra2 , ∼4% of the molecules were within the experimental localization accuracy ( ∼70 nm ) . The other two populations could be distinguished by their diffusion coefficients: 24% moved with a slow diffusion coefficient ( D2 = 2 . 6 μm2/s ) and 72% moved faster ( D3 = 13 μm2/s ) . For Dendra2-H2B , 35% of the molecules appeared immobile and might correspond to molecules engaged in chromatin-bound nucleosomes . The two populations of mobile Dendra2-H2B molecules were D2 = 0 . 5 μm2/s , 25% , and D3 = 13 μm2/s , 40% , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 01310 . 7554/eLife . 02230 . 014Figure 2—figure supplement 2 . Population exclusion . In panel A , the average single step translocations ( Δt = 10 . 5 ms ) plotted as a function of the length of the trace for ‘free’ Dendra2 . In panel B , the number of detected translocations is shown for increasing diffusion coefficient ( 0 . 001 , 0 . 01 , 0 . 1 , 1 , 10 , and 20 μm2/s ) for simulated videos of the same duration , volume , and particle concentration ( one single diffusing particle ) . Simulations with a mixture of two diffusive species with different diffusion coefficients of 1 μm2/s and 10 μm2/s were also performed . The translocation histogram in panel C shows the effect of the bias due to the lower percentage of detections of fast particles . The same effect can be observed at larger lag times in the difference for the theoretical and measured averaged MSD shown in panel D . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 014 To complement our analysis of the translocation histograms , we plotted the mean square displacement ( MSD ) of the molecules as a function of time ( ‘Materials and methods—Cumulative histogram analysis and mean square displacement’ ) . For Dendra-H2B , the MSD reached a plateau after ∼20 ms at ∼ 0 . 5 μm2 ( Figure 2C ) , consistent with a confined motion of individual histone molecules inserted into chromatin . The MSD of ‘free’ Dendra2 increased regularly with time . However , it slightly deviated from the linear behavior expected for molecules undergoing normal diffusion . This was attributed to a ‘population exclusion effect’ due to the different defocusing rates of the various diffusive subpopulations of Dendra2 . Because of their three-dimensional motion in the nucleus , slow moving particles remained within the focal depth of observation ( ∼0 . 5–1 μm ) for a longer time than fast moving ones . As a result , fast diffusing molecules contributed comparatively less than the slow ones to the MSD at longer time lags . Note that this effect is inevitable for any single-molecule experiment involving more than one diffusive population and in which the three-dimensional movement of particles is recorded in two dimensions ( ‘Materials and methods—Numerical simulations’ and Figure 2—figure supplement 2 ) . We therefore adjusted the rates of the different diffusive populations for each molecule , and have used the corrected values through the text and for our analysis . The deviation from linearity of the MSD curve produced by such an exclusion effect clearly illustrates the need to complement the analysis of molecular mobility with other observables , ideally independent of population heterogeneity . Finally , to carefully establish the range of application of our experimental and analytical methods , we performed numerical simulations ( ‘Materials and methods—Numerical simulations’ ) . On the one hand , the particle localization precision sets the lower bound to a reliable estimation of the diffusion parameters , that is ∼ 0 . 04 μm2/s for a pointing accuracy of ∼70 nm . On the other hand , fast moving particles can be tracked with a mobility up to ∼20 μm2/s , beyond the experimental values determined for ‘free’ Dendra2 . Altogether , our experimental and numerical results provide a benchmark for studying nuclear factors with a mobility ranging between that of chromatin-bound H2B molecules and of ‘free’ proteins such as Dendra2 . We next probed the mobility of transcription factors . Dendra2 was fused to the proto-oncogene c-Myc and to the Cyclin T1 subunit of P-TEFb . It has recently been shown that , rather than activating new sets of genes in the cell , the role of c-Myc is that of an amplifier of transcription of already active genes ( Lin et al . , 2012; Nie et al . , 2012 ) . We thus tested the functionality of c-Myc-Dendra2 by performing RT-qPCR on a set of active genes in our U2OS cell line . When comparing the wild-type cells and those expressing c-Myc-Dendra2 , we measured an increase of RNA expression levels in 10 out of 12 tested genes ( ‘Materials and methods—mRNA expression and c-Myc expression amplification analysis’ ) . Translocation histograms for c-Myc were well fit with three diffusive populations ( Figure 3—figure supplement 1 ) . The most abundant corresponded to rapidly diffusing particles ( 13 . 5 μm2/s , 70% of the molecules ) ( Figure 3A , black trajectories ) . In addition , a significant fraction of c-Myc was immobile ( 9 . 5% ) ( Figure 3A , green trajectory ) or displayed slow diffusion ( D2 = 0 . 5 μm2/s , 20 . 5% ) ( Figure 3A , blue trajectories ) . For P-TEFb , the typical translocation length and the translocation histograms were comparable to those obtained for c-Myc ( Figure 3—figure supplement 2 ) . 10 . 7554/eLife . 02230 . 015Figure 3 . Diffusion properties of c-Myc and P-TEFb . For c-Myc ( A ) and P-TEFb ( B ) , examples of single molecule traces . From these , we plotted the averaged mean square displacement ( MSD ) as a function of the lag time with intervals of confidence of 95% ( panel C ) , from a total of 33 , 645 trajectories ( from 42 cells ) for c-Myc and 16 , 852 trajectories ( from 38 cells ) for P-TEFb . In panel D , the MSD over time was represented as a function of time in logarithmic scale for ‘free’ Dendra2 , c-Myc and P-TEFb . The fit in the inset follows the time rescaling law MSD ( t ) = D tα , where α = 1 for normal diffusion , and 0 < α < 1 for subdiffusive behavior . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 01510 . 7554/eLife . 02230 . 016Figure 3—figure supplement 1 . Analysis of the cumulative distribution function of step translocations for c-Myc . Cumulative distribution function ( CDF ) of the step translocations is plotted in red for three different time intervals ( 1Δt , 3Δt and 6Δt in panels A , B , and C , respectively; Δt = 10 . 5 ms ) . The fit of the CDF is shown for 1 ( dotted line ) 2 ( dashed line ) and 3 ( solid line ) Brownian diffusive populations . In the inset , the residuals of the fits are shown . On the right hand side , the step translocation histograms from which the CDF were calculated , shown with the results of the CDF fit for 1 , 2 , and 3 populations . At increasing lag times , three diffusive species were needed to retrieve a good fit of the data . In panel D , the temporal evolution of the exponential coefficient µ for each population is shown . The diffusion coefficient for each population was calculated by a linear regression of the first four points of the MSD . In panel E , a simulation was performed using the results of the measurement of the diffusing coefficients and corresponding rescaled proportions ( D1 = 0 μm2/s , 9%; D2 = 1 . 4 μm2/s , 20%; D3 = 14 . 4 μm2/s , 70% ) and the resulting MSD is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 01610 . 7554/eLife . 02230 . 017Figure 3—figure supplement 2 . Histogram of single translocations for c-Myc and P-TEFb . Histograms of single step translocation lengths for c-Myc ( A ) and P-TEFb ( B ) , for 1Δt = 10 . 5 ms . Black curves are the result of the fit of the cumulative distribution function with three populations . Note that both histograms span within the same range of lengths . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 01710 . 7554/eLife . 02230 . 018Figure 3—figure supplement 3 . Cumulative histogram of square displacements rescaled in time . For ‘free’ Dendra2 ( A ) , c-Myc ( B ) , and P-TEFb ( C ) , cumulative histograms of the square displacements rescaled in time to Δtα , for increasing lag times . The α value for each molecule was obtained from the fit in Figure 3 . Only the data obtained from P-TEFb show a collapse of the curves , indicating the goodness of the fit to the anomalous diffusion model . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 018 When plotting the MSD as a function of time for c-Myc and P-TEFb , we observed a deviation from linearity for both factors ( Figure 3C ) . Such deviation could be due to the ‘population exclusion effect’ described above ( ‘Materials and methods—Numerical simulations’ , Figure 2—figure supplement 2 ) , but , alternatively , it could also be the signature of an anomalous diffusion process . When a particle undergoes anomalous diffusion , the MSD vs time scales as a power law tα , where α < 1 is characteristic of a subdiffusion process ( Saxton , 2007 ) . However , neither the ‘free’ Dendra2 nor the c-Myc MSD data could be properly fit by such a law ( Figure 3D ) . Similarly to ‘free’ Dendra2 , c-Myc molecules were distributed between populations of very distinct diffusion coefficients . In contrast , for P-TEFb , the MSD variations were remarkably fit by a tα power law with the anomalous coefficient α = 0 . 6 ( Figure 3D ) . The subdiffusion of P-TEFb was also apparent when we plotted the cumulative histograms of the square displacement for multiples of the time interval ( Δt ) between two frames and rescaled them by the factor tα , with α determined from the fit in Figure 3C . All the rescaled histograms curves collapsed remarkably well for P-TEFb but not for c-Myc or ‘free’ Dendra2 ( Figure 3—figure supplement 3 ) . We therefore concluded that the characteristics of single P-TEFb trajectories are consistent with an anomalous diffusive behavior whereas the deviation from linearity of the c-Myc MSD curve reflects the heterogeneity of its diffusion dynamics . Subdiffusion in cells is commonly attributed to one of the following two microscopic processes: a broad distribution of trapping times or an obstructed movement resulting from a reduction of the accessible space ( Condamin et al . , 2008 ) ( for a discussion about subdiffusion causes , see ‘Materials and methods—Numerical simulations of anomalous diffusion models’ ) . In other words , the subdiffusive behavior , evidenced by the sublinear MSD , is due to either temporal or spatial restrictions . In order to probe the spatial characteristics of the exploration independently of temporal considerations , we analyzed the distribution of angles Θ between two consecutive translocations , an observable that is predominantly sensitive to the geometry of the exploration space ( Liao et al . , 2012 ) and able to elucidate complex dynamics of molecules ( Burov et al . , 2013 ) . For ‘free’ Dendra2 and c-Myc , we found a quasi-uniform angular distribution ( Figure 4A ) , as expected for Brownian diffusion . In a three-dimensional space , there is no privileged direction and all angles Θ are equiprobable . In contrast , the angular distribution for P-TEFb was significantly biased toward 180° , reflecting an anti-correlation between two successive displacements . Such anisotropic angular distribution is consistent with diffusion in a space of reduced dimensionality such as a fractal network ( ben-Avraham and Havlin , 2005 ) . A particle that diffuses in such a structure encounters dead ends , in which case it cannot but return back to previously visited locations ( Θ = 180° ) . Noteworthy , the diffusing subpopulation of H2B molecules also showed a non-uniform angular distribution ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 02230 . 019Figure 4 . Angle distribution between consecutive steps . ( A ) Distribution histograms , in polar coordinates , of the angle θ formed between the vectors of two consecutive translocation steps ( vectors formed by positions at time 0 and 10 ms , and between 10 ms and 20 ms ) , for Dendra2 ( 23 , 883 total number of angles ) , H2B ( 54 , 820 angles ) , c-Myc ( 46 , 540 angles ) , and P-TEFb ( 13 , 820 angles ) . The asymmetry coefficient ( AC ) was calculated as the logarithm to the base 2 of the ratio between the frequency of forward angles ( between 0° and 30° ) and the backward angles ( 150°–180° ) ( B ) . In panel ( C ) , the temporal evolution of AC at increasing lag times has been plotted ( i . e . , the angle between the vectors formed by the positions at 0 to 10 ms and 10 ms to 20 ms , first data point at 10 ms; angle between the vectors formed at positions 0 to 20 ms and 20 ms to 40 ms , second data point at 20 ms , etc ) . In ( D ) , dependence of the AC with the average translocation value , calculated between the two consecutive steps forming the angle θ and binned at 150 nm . Error bars in ( C ) and ( D ) were calculated as the standard deviation of 50 resamplings using 50% of the data randomly chosen from the radial histograms . Note that the error bars increase as fewer angles are available at increasing lag times and large translocations . Also , how the limited localization accuracy is reflected in the first data point of the spatial dependence of AC in ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 01910 . 7554/eLife . 02230 . 020Figure 4—figure supplement 1 . Temporal and spatial dependence of the angular distribution of angles and their asymmetry coefficient ( AC ) . Radial distribution of angles and their temporal and spatial dependence for the four proteins under study ( Dendra2 , H2B , c-Myc , and P-TEFb ) ( A ) . The temporal evolution was calculated as a function of lag time ( angle between the vectors formed by the positions at 0 to 10 ms and 10 ms to 20 ms , 2Δt; angle between the vectors formed at positions 0 to 20 ms and 20 ms to 40 ms , 4Δt , etc ) . For each protein , the histograms were normalized to the total number of angles . The spatial dependence was calculated as a function of the average translocation length between the two steps forming the angle θ , with a bin of 150 nm . In ( B ) and ( C ) , the asymmetry coefficient was computed from the angular histograms plotted in panel ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 020 An alternative scenario in which such an asymmetric angular distribution of SM traces may arise is that of confined diffusion . If a diffusing particle is confined in a volume of size comparable to the translocation steps , its repetitive bouncing against the trap walls will produce a relative increase of angles larger than 90° . In such a case , the temporal evolution of the anti-persistence reflects the length ratio between the displacement steps and the size of the confining volume . On the other hand , a defining property of fractal structures is their scale invariance ( ben-Avraham and Havlin , 2005 ) , namely the repetition of structural motifs at different length scales . For a particle diffusing in such a fractal structure , we expected the scale invariance to be apparent in the characteristics of the movement of the particle . We therefore examined the temporal and spatial dependences of the angular distribution in order to further investigate the origin of the antipersistence of the trajectories and the underlying geometry of the space available for exploration . We defined the asymmetry coefficient ( AC ) as the logarithm to the base 2 of the ratio between the frequency of forward angles ( between 0° and 30° ) and backward angles ( 150°–180° ) ( Figure 4B ) . The AC is thus negative for angular distributions with a dominant number of backward angles , and it measures the deviation from a homogenous distribution . We calculated the AC for the angles formed at increasing lag times ( Figure 4C , Figure 4—figure supplement 1 ) as well as a function of the average length of the consecutive translocations forming the angle θ ( Figure 4D , Figure 4—figure supplement 1 ) . It is important to note that with this analysis , the experimental localization accuracy is reflected in the first data point of the spatial dependence of the AC , and not for the data above the 0–150 nm bin . Also , fewer particles are contributing to the AC at larger times , as can be observed in the angular distribution histograms in Figure 4—figure supplement 1 . We found out that the angular distribution of c-Myc deviates from homogeneity at increasing lag times with increasing negative AC ( Figure 4C ) , potentially reflecting a hindrance to the free diffusion of c-Myc and its confinement to domains significantly smaller than the nucleus . However , the angular distribution became isotropic ( AC = 0 ) at translocations larger than 300 nm ( Figure 4D ) . This transition can be interpreted as an indication of the upper limit size of the confining volume . Hence , the temporal and spatial evolution of the AC suggest two subpopulations of c-Myc , one confined into regions smaller than ∼300 nm and a non-confined fraction of c-Myc molecules . P-TEFb , on the other hand , displayed a remarkable constant value of AC for both , time and space ( Figure 4C , D ) , possibly reflecting a length-invariant property of the medium in which diffusion takes place . In order to gain insight about the different scenarios giving rise to the observed angular distributions , we performed numerical simulations of models with increased levels of complexity ( see ‘Materials and methods—Numerical simulations’ for details about the numerical simulations ) . In line with the observation of different diffusing populations even for free Dendra2 , we first considered an intermittent diffusion model . Here , particles had a probability to switch from a fast to a slow diffusion coefficient and vice versa . We also considered an intermittent trap model , where diffusing particles with fixed diffusion coefficient have a probability to be confined in a spherical trap . We adjusted the parameters of the models in order to obtain similar translocation histograms to those of c-Myc and P-TEFb ( Figure 5—figure supplement 1 , ‘Materials and methods—Numerical simulations’ ) . However , none of these simple intermittent models reproduced the antipersistent characteristics of the experimentally measured trajectories ( Figure 5—figure supplement 2 , ‘Materials and methods—Numerical simulations’ ) . We then considered a model that results from a combination of intermittent diffusion and intermittent trap . We performed simulations of fast diffusing particles ( diffusion coefficient D1 ) with a probability Kon to engage into a slower diffusion ( D2 ) confined in a trap of radius R ( Figure 5A ) . Here , the AC decreased with increasing lag times ( Figure 5B ) , reproducing the trend observed in c-Myc . Likewise , the AC displayed the same behavior as c-Myc , tending to zero for larger values of the translocation steps ( Figure 5C ) . Following this model , c-Myc performs thus a free exploration of the nuclear space , combined with slower yet still normal diffusion of confined domains , reflecting its interactions with a multiplicity of partners . 10 . 7554/eLife . 02230 . 021Figure 5 . Simulated trajectories and distribution of angles . ( A ) Distribution of angles between consecutive translocations for the intermittent diffusion plus confinement model . In this model , a fast diffusing particle with diffusion coefficient D1 has an association rate probability Kon to enter into a confined volume ( of radius Rtrap ) with slower diffusion coefficient D2 , and dissociation rate Koff . ( The values of the parameters were D1 = 14 μm2/s , D2 = 1 μm2/s , Rtrap = 500 nm , Kon = 0 . 0015 , Koff = 0 . 02 . ) . In panels ( B ) and ( C ) , the dependence of AC with the lag time and the average translocation step . In ( D ) , angular distributions of the intermittent trap simulations with a distribution of trap sizes given by a power law , as well as the simulations of random walks on a percolation cluster . In panel ( E ) , temporal dependence of the asymmetry coefficient ( AC ) for three types of simulations: power law distribution of trap sizes , random walks on a percolation cluster , and random walks on a 2D Sierpinski carpet . In panel ( F ) , dependence of the AC with the average translocation size . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 02110 . 7554/eLife . 02230 . 022Figure 5—figure supplement 1 . Simple models of intermittent diffusion and intermittent confinement . Simulation parameters that best fit the experimental step translocation histograms of c-Myc ( A ) and P-TEFb ( B ) . In the fast/slow diffusion model , D1 and D2 were determined by a two-population fit of the cumulative translocation histograms . Likewise , the rate Kon/Koff was determined by the population rate obtained in such a fit . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 02210 . 7554/eLife . 02230 . 023Figure 5—figure supplement 2 . Temporal and spatial dependence of the asymmetry coefficient for the simple intermittent models . Temporal and spatial dependence of the asymmetry coefficient ( Figure 4B ) for the intermittent diffusion and intermittent confinement models . In panel ( A ) , the parameters that matched the translocation histogram of c-Myc were used . In panel ( B ) , those that match P-TEFb . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 02310 . 7554/eLife . 02230 . 024Figure 5—figure supplement 3 . Continuous-time random walk . In A , ensemble averaged mean square displacement ( MSD ) for a continuous-time random walk ( CTRW ) on a cubic lattice , with a heavy-tailed probability distribution of power −1 . 6 and a position recorded every 1000 steps . The MSD exhibits an anomalous curvature . The corresponding asymmetry coefficient is shown in panel B . The distribution of angles is symmetric and uniform at all time scales . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 024 Finally , in order to reproduce the invariant properties of the angular asymmetry observed for P-TEFb , we needed to invoke a hierarchical organization of the space . We considered the intermittent trap model , this time with a distribution of trap sizes governed by a Pareto power law ( exponent 0 . 1 ) . With this model , we obtained an antipersistent angular distribution ( Figure 5D ) and a closer reproduction of the AC behavior observed for P-TEFb . Although the AC was not strictly constant with time , it did not show a tendency towards zero ( Figure 5E ) . Moreover , the spatial dependence of the angular asymmetry formed a plateau for translocations larger than 600 nm ( Figure 5F ) . Such a hierarchy of confining sizes led us to consider a fractal network as an underlying structure on which to simulate the diffusion of particles , also motivated by recent works on the geometry of the nuclear space ( Bancaud et al . , 2012 ) . We considered a 3D percolation cluster as well as a 2D Sierpinski carpet . The Sierpinski carpet is an exact fractal lattice with multi-scale self-similarities . Random walks on a Sierpinski lattice are anomalous because its structure induces spatial correlations between successive displacements . The percolation cluster at the critical percolation threshold possesses the property of statistical internal self-similarity . As a consequence , the percolation cluster exhibits fractal properties without a defined geometric shape ( ben-Avraham and Havlin , 2005 ) . For both fractal structures , the angular anisotropy was constant with time ( Figure 5E ) , illustrating the scale-invariant features of fractal structures , as observed in the experimental data of P-TEFb . Surprisingly , the AC decreased for larger translocations in the case of the percolation cluster , while the Sierpinski carpet yielded an invariant asymmetry in space . This was interesting because it indicates that the underlying network needs to reproduce a certain degree of geometrical self-similarity , as it is the case of the Sierpinski carpet . The percolation cluster , on the other hand , does not conserve its geometry at different scales but rather other features like the local density obey a power law . We have determined that while c-Myc undergoes normal diffusion ( with a subpopulation seemingly confined in domains smaller than the nucleus ) , the dynamics of P-TEFb is well described by a subdiffusive behavior . In the case of P-TEFb , our simulations support the notion that anomalous diffusion is compatible with an obstructed mobility of the proteins , as obtained on a fractal structure ( we have ruled out other models of subdiffusion , see Figure 5—figure supplement 3 and ‘Materials and methods—Numerical simulations of anomalous diffusion models’ for a more detailed discussion ) . As previously described , the exponent α = 0 . 6 of anomalous diffusion obtained for P-TEFb ( Figure 3D ) is a direct measure of the dimension of the walk Dw = 2/α = 3 . 3 . Since the fractal dimension Df has an upper limit at Df = 3 , we can therefore conclude that Dw > Df , and thus that P-TEFb is engaged in a compact exploration of the nucleoplasm . In contrast , the isotropic sampling of space of c-Myc excludes a compact mode of exploration; it undergoes normal 3D diffusion irrespective of its confinement , and hence the dimension of the walk is Dw = 2 , sampling the nucleoplasm in a non-compact manner . These results imply that different factors sense a protein-dependent nuclear environment , which can be determinant for their exploration strategy . The distinctive properties of compact and non-compact trajectories have potentially important functional consequences on the ability of searchers to find and react with molecular partners . As noted above , a striking difference is the distance-dependence of the mean first passage time ( MFPT ) of the searcher to the target site . The MFPT of non-compact explorers is essentially constant , depending solely on the total volume and not on the distance r to the target . Conversely , in the compact case , the MFPT still scales with the volume but also increases with the distance as r ( Dw−Df ) . As an illustration , we computed the MFPT as a function of the distance ( see analytical expressions of MFPT in Condamin et al . , 2005; Bénichou et al . , 2010 ) , using the experimental data for c-Myc and P-TEFb , two examples of non-compact and compact explorers . For c-Myc , which behaves as an ordinary Brownian walker , the fractal dimension is Df = 3 , and the dimension of the walk is Dw = 2 . We used a diffusion coefficient D = 9 . 8 μm2/s , the value obtained by a weighted average of the diffusion coefficients of the three subpopulations . ( It is important to note that the value used for the diffusion coefficient does not affect the dependence of the MFPT on the initial distance to the target . ) To calculate the MFPT , we used a nuclear volume of 600 μm3 and considered a target in its center . For P-TEFb , we did not have direct access to the value of Df and used several values previously reported as estimations in the nucleoplasm ( Bancaud et al . , 2012 ) . In Figure 6 , we used Df = 2 . 6 and the results were qualitatively similar for values of Df = 2 . 2 , and Df = 3 ( Figure 6—figure supplement 1 ) . For both proteins , we also varied the size a of the target between 1 nm ( i . e . , corresponding to a couple of base pairs ) , 10 nm ( the size of a protein complex ) , and 100 nm ( the size of a large multimolecular complex ) . 10 . 7554/eLife . 02230 . 025Figure 6 . Compact vs non-compact exploration . ( A ) Mean first passage time ( MFPT ) as a function of the initial distance to the target for both c-Myc ( non-compact exploration; Df = 3 , Dw = 2 , and diffusion coefficient D = 9 . 8 μm2/s ) and P-TEFb ( compact exploration; Df = 2 . 6 , Dw = 3 . 3 , and scale factor of the MSD fit D = 7 . 8 ) . The MFPT was calculated for three different target sizes: 1 nm , 10 nm , and 100 nm . Also , two-dimensional representation of the plots for a = 100 nm are depicted in the lower part of the panel . ( B ) Probability of interaction with target 1 before interacting with target 2 , placed at a distance of 20 μm from each other , as a function to the relative distance between the searcher and the targets; two-dimensional plots in the lower side of the panel . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 02510 . 7554/eLife . 02230 . 026Figure 6—figure supplement 1 . Mean first passage times with Df = 2 . 2 and Df = 3 . Mean first passage time ( MFPT ) as a function of the initial distance to the target for c-Myc ( non-compact exploration; Df = 3 , Dw = 2 , and diffusion coefficient D = 9 . 8 μm2/s ) and P-TEFb calculated with two fractal dimensions: Df = 2 . 2 in ( A ) , and Df = 3 in ( B ) ( Dw = 3 . 3 , and scale factor of the MSD fit D = 7 . 8 ) . The MFPT was calculated for three different target sizes: 1 nm , 10 nm , and 100 nm ( in A , the three curves overlap in the case of P-TEFb ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02230 . 026 For c-Myc , the MFPT was constant , irrespective of the distance r ( Figure 6A ) . However , it was inversely proportional to the size of the target , similar to what is predicted from the diffusion-limited rate of bimolecular reactions ( Nelson et al . , 2008 ) . In contrast , the MFPT of P-TEFb increased with the distance r but did not depend on the target size . The lack of size dependence can be simply viewed as a consequence of the redundant exploration of compact explorers , and reflects the fact that the limiting step to find a target is the time taken to reach its vicinity . We stress that the differences of MFPT can be very significant . For instance , the time needed to find a 10 nm target located at a distance of 250 nm is 68 times longer for c-Myc compared to P-TEFb ( 506 . 1 s for c-Myc and 7 . 4 s for P-TEFb ) . If the target is located at 5 μm of the TF , the difference in the search time is reduced to a factor of 8 ( 525 . 3 s for c-Myc and 64 . 6 for P-TEFb ) . Here , we considered that c-Myc has a full access to the nuclear volume . It is interesting to note that if , as suggested by the temporal variance of the angular distribution , c-Myc is confined to a smaller domain , the MFPT would scale linearly with this volume . We also considered the case of a factor susceptible to bind to two different targets T1 and T2 ( Figure 6B ) . To do so , we computed the splitting probability P , that is the probability to reach T1 before T2 as a function of the initial distance to T1 . For c-Myc , the probability was equal to 0 . 5 as soon as the initial distance was larger than a few tens of nanometers , in stark contrast with the case of P-TEFb , for which P varied almost linearly with the distance . Overall , our analysis of SM experiments of c-Myc and P-TEFb reveals two characteristics of TFs diffusion relevant to the understanding of transcription regulation kinetics . First , the exploration geometry of the nucleus by TFs is determined by the function and interactions of the nuclear factor . Rather than being subjected to a universal sampling geometry imposed by the nuclear architecture , c-Myc and P-TEFb adopt different modes of exploration leading to normal and anomalous diffusion , respectively . Second , despite apparently similar diffusion coefficients , the different exploration strategies of c-Myc and P-TEFb ( non-compact and compact , respectively ) can lead to opposite dependence of the search kinetics on the distance to the target and on the target size . The distance-dependence of the MFPT has direct implications on the probability of interaction of c-Myc and the P-TEFb with their respective partners , which in turn may affect transcriptional kinetics and regulation . With the PALM imaging assay adapted for SM detection of intracellular proteins in eukaryotic cells , we probed the spatial dynamics of different proteins in the nucleus of live human cells: ‘free’ Dendra2 , histone H2B , the proto-oncogene c-Myc , and the elongation factor P-TEFb . The analysis of individual trajectories , supported by numerical simulations of diffusive tracers on free , confined , and fractal structures , and switching between different regimes , shows that these nuclear proteins fundamentally differ in their exploration of the nucleoplasm . Our results on ‘free’ Dendra2 are along the lines of those obtained with microinjected fluorescent streptavidin , which explores all nuclear compartments with three subpopulations having different diffusion characteristics ( 0 . 15 , 0 . 8 , and 5 μm2/s ) ( Grünwald et al . , 2008 ) , possibly reflecting differences in viscosity and/or crowding in the nucleus . In contrast , FCS experiments using ‘free’ GFP-repeats or SPT tracking of QD aggregates suggested anomalous diffusion ( Bancaud et al . , 2009 ) . We determined that ‘free’ Dendra2 and the proto-oncogene c-Myc undergo normal diffusion in 3D , whereas the displacement of P-TEFb was accounted for by a subdiffusive movement . This finding was further supported by measurements of the distribution of angles between consecutive translocations . Importantly , this distribution was initially isotropic for Dendra2 and c-Myc and an asymmetry towards the return angles increased over time , as expected for confined Brownian motion . Conversely , P-TEFb showed a pronounced and time-invariant anisotropy consistent with the motion on a fractal structure . Thus , the nuclear geometry , or equivalently , the architecture of the space sampled by diffusing factors , is not unique but constitutes a protein-specific parameter . Furthermore , taking into consideration the diffusion parameters derived from the analysis of the MSD , together with the geometrical aspects of the exploration of c-Myc and P-TEFb , we determined the mode of exploration of these factors to be non-compact and compact , respectively . We stress that the distinction between compact and non-compact exploration , rather than the one between anomalous and normal diffusion , is the proper criterion to analyze the search dynamics of transcription factors . The notion of compactness is intimately linked to the geometry and the dimensionality of the sampled space . In this regard , there is a specificity of random motions in a three-dimensional medium with respect to the one- and bi-dimensional cases , for which the exploration is always compact since the fractal dimension Df ( less or equal to 1 and 2 , respectively ) is necessarily smaller than Dw . Only in the case of 3D search , can both compact and non-compact behaviors be observed . Our data demonstrate the relevance of the notion of compactness for the description of nuclear factor dynamics . One microscopic mechanism leading to a compact exploration of the nucleus could be a compartmentalization of the nucleoplasm into interconnected domains forming a fractal labyrinth in which molecules diffuse . In our view , such a model assuming that molecules encounter physical barriers is poorly compatible with the dynamic nature of nuclear organization and with the lack of correlation between protein size and mobility in the nucleus ( Sprague et al . , 2004; Mueller et al . , 2008 ) . Another interpretation is that of a fractal structure restricting the mobility of proteins at its surface . Chromatin has been described as a fractal globule ( Grosberg et al . , 2007; Lieberman-Aiden et al . , 2009 ) and transient , non-specific interactions to a continuum of binding sites would account for the diffusing factors not escaping from their interaction with chromatin . In this scenario , the number of binding sites with which c-Myc interacts is not sufficient to restrict its motion to chromatin ( 36 , 000 E-boxes in a diploid genome , representing less than 50 sites per μm3 ) . P-TEFb interacts with the CTD of the catalytic subunit of RNA Polymerase II , which contains 52 repetitions of a hepta-peptide motif ( Taube et al . , 2002 ) . The RNA Polymerase CTD is not folded and can occupy the space very efficiently , potentially forming a mesh offering a nuclear continuum of binding sites for P-TEFb . Such CTD matrix could have an intrinsic existence or be linked to the chromatin globular organization . The existence of a nuclear protein scaffold or matrix has been speculated for more than half a century ( Pederson , 2000 ) and both our works offer an observation of a functional role for such a structure . Several other studies support this hypothesis , showing that nuclear proteins are in constant interaction with their environment and their motion is governed by specific and non-specific bindings ( Phair et al . , 2004; Sprague et al . , 2004; Hager et al . , 2009; Speil et al . , 2011 ) , therefore opening the door for mechanisms where factors are guided on networks of binding sites ( Bénichou et al . , 2011 ) . From a general standpoint , the distance-dependence of the search kinetics could have strong implications for gene regulation . For example , it has been recently shown that , in Escherichia coli , the spatial distribution of TFs is determined by the local state of DNA ( Kuhlman and Cox , 2012 ) . Let us consider the case of TFs co-regulating multiple loci; the relative localization of these loci is an important parameter that will play different roles depending on the compact or non-compact exploration of the TFs . Non-compact TFs have a very similar probability to bind to all loci . In other words , all loci will have the same probability to be occupied , regardless of their spatial position . In contrast , compact factors will be preferentially shared between proximal loci , and therefore the probability of a locus to be occupied by a compact explorer is a function of the occupation history of its neighboring sites: it is distance and time dependent . Importantly , this indicates that two loci , such as two regulatory sites located a few tens of kbp away from each other , can transfer information and influence one another without direct physical contact . This spatial relation could underlie the process of sequestration of factors away from their targets ( Yao et al . , 2011 ) , which would occur only with compact explorers . Such geometrically controlled long-distance interactions are not detectable using conventional chromatin capture assays , which predominantly rely on the chemical crosslinking between contacting sites . A remarkable feature of compact searchers is their propensity to visit their neighboring sites multiple times . As a result , they have a probability equal to one to return to a site that they previously occupied , a property designated as the recurrence of compact trajectories . From a biochemical viewpoint , this property might affect our understanding of the kinetic stability of molecular complexes . Certainly , molecular machines controlling the nuclear functions such as transcription , splicing , and replication are composed of large numbers of molecules . Some of these molecules are stable constituents while others can be rapidly exchanged in order to control the specificity and modulate the activity of a particular complex ( Fong et al . , 2012 ) . It is therefore important to understand how these molecular machines can assemble from their principal components . For instance , we cannot yet reconcile the need for strong and stable interactions , believed to be required for the viability of such complexes , and the requisite of weak and transient interactions required for molecules to compete for the same target regulating their composition . The observation of compact modes suggests that strong binding , associated to small dissociation rates , is not required to ensure high occupancy . Recently , the role and importance of transcriptional fluctuations within a single cell have been extensively studied ( Raj et al . , 2008; Zenklusen et al . , 2008; Larson et al . , 2009; English et al . , 2011; Itzkovitz and van Oudenaarden , 2011 ) . Using a simple model in which the activation of a gene is controlled by the binding of a single TF to a locus , Meyer et al . ( 2012 ) have modeled how the search dynamics of these TFs affects the transcriptional response . In this model , for compact TFs , the recurrence of the trajectories and the facilitated re-association to the locus would result in transcriptional bursting . In contrast , for the non-compact case , the gene activation rate is determined by the total TF concentration in the nucleus , and the transcriptional activity is uncorrelated in time . This further illustrates how the translocation properties of nuclear factors might underlie the kinetics of functional cellular events . In this study , we have experimentally demonstrated that different nuclear proteins with different functions sample the nucleoplasm with different search strategies: the exploration geometry of the nucleus is protein-dependent . We have also determined that two different universality classes of search modes , namely compact and non-compact explorations , coexist in the nucleoplasm . Our current view of the nucleoplasm and chromatin is that of a structure whose condensation influences its accessibility to transacting factors . Here , we have established that the same target in the nucleoplasm can be visited with different probability and kinetics by different factors depending on how they sample space . In addition to chromatin condensation , the compactness of the exploration itself needs to be taken into account to understand how gene regulation operates . While the space-sampling mode of a random exploration is either compact or non-compact , the question to be answered in the future is whether one molecule manifests both types of search , exhibiting transitions between them , and whether different dynamics may still arise within each search mode . If that is the case , it will be of paramount importance to understand the level of regulation of such transitions , as well as the implications for the kinetics of the transcription process . The inverse first passage time is a measure of the reaction rate constant . Therefore , the different interaction kinetics that results from compact or non-compact explorations has profound implications in the understanding of the dynamic interactions and reactivity rates between TFs and corresponding regulated genes . For a non-compact explorer like c-Myc , the interaction rate is that of a homogenous solution , and thus it will bind with equal probability to any target in the nucleoplasmic volume . On the other hand , for a compact exploration such as the one of P-TEFb , the recurrent search of the local environment and the distance dependence of the search time translate into spatial and temporal correlations between binding events . Such spatial correlation can be seen as a mechanism that adds a level of control to the rapid assembly of molecular complexes , reconciling weak and transient interactions with functional stability . This last notion suggests the idea of a regulated level of compactness of TFs both in time and space . U2OS ( Human Osteosarcoma ) cells were grown in DMEM ( Life Technologies , Carlsbad , CA ) with 1 g/l glucose and glutamax supplemented with 10% FBS ( Fetal Bovine Serum , Life Technologies ) and 1% Penicillin/Streptomycin ( Life Technologies ) at 37°C with 5% CO2 . 48 hr prior to the imaging , cells were seeded at 30–40% confluence on a plasma-cleaned ( 2 min with air with Femto model , Diener Electronic , Ebhausen , Germany ) and collagen-coated ( Collagen I from Rat tail , Life Technologies ) coverslips ( N°1 25 mm , Marienfeld , Lauda-Königshofen , Germany ) . The C terminal of c-Myc and H2B were fused to Dendra2 and expressed under the control of the CMV promoter . Prior to experiments , U2OS cells were transfected 24 hr before imaging with the plasmid of interest ( 100 ng/25 mm coverslip ) using Fugene 6 ( Roche Applied Science , Penzberg , Germany ) according to manufacturer's instructions . Clones with very low over-expression of exogenous protein , as judged by low fluorescence intensity of pre-converted Dendra2 , were used . Experiments with P-TEFb ( Cyclin T1 fused to Dendra2 on N terminal ) and Dendra2 ( alone ) were performed on U2OS cell line stably transfected and selected with geneticin ( Life Technologies ) . Clones with very low expression of fluorescent protein ( CyclinT1-Dendra2 or Dendra2 ) , as judged by low fluorescence intensity of pre-converted Dendra2 , were used . Transient transfections of Cyclin T1 Dendra2 were also performed and gave the same results . Single-molecule imaging was performed on an inverted microscope Nikon Ti Eclipse ( Nikon Instruments , Tokyo , Japan ) , with a high numerical aperture objective ( 1 . 49 NA ) and 100X magnification; extra magnification of 1 . 5X was used in the tube lens of the microscope , resulting in a total magnification of 150X . We also used perfect focus system ( Nikon ) designed to avoid drift on the Z-axis ( focus ) of the objective , relative to the coverslip . The excitation ( 561 nm ) and activation ( 405 nm ) laser beams were injected into a fiber and focused in the back focal plane of the objective , using an appropriate dichroic ( Di01-R561-25x36 ) ( Figure 1—figure supplement 1A ) . A motorized mirror allowed us to choose between wide-field or inclined excitation configurations; a small angle , between 0 and 30° , was typically used to avoid stray-light reflections and reduce background from cell auto-fluorescence . Experiments were acquired under continuous excitation ( 561 nm laser , 5 kW/cm2 on the sample ) and pulsed activation ( 405 nm laser , 1 pulse of 10 ms per second , 0 . 01 kW/cm2 during the pulse on the sample ) . Fluorescence emission from individual Dendra2 molecules was filtered with a single band emission filter centered at 617 nm and a bandpass of 73 nm and recorded on an EMCCD camera ( iXon 897 Andor Technology , Belfast , Ireland ) . The pixel size of the EMCCD was 16 μm , and we imaged a small region of interest ( ROI ) of about 100 pixels × 100 pixels . This ROI was sufficient for imaging a large cross-section within the nucleus of single cells , and allowed acquisition rates as fast as 100 Hz ( 10 ms per frame ) . Images of the pre-converted ( green ) form of the ensemble fluorescence of Dendra2 were taken using a mercury lamp for illumination ( excitation: 485 nm , emission FF01-525/30 ) . Cells were imaged in Leibovitz's L15 medium ( Life Technologies ) containing 10% FBS ( Fetal Bovine Serum , Life Technologies ) . The sample was placed on the microscope , on a stage heated at 37°C on the microscope . Once an ROI was selected from the pre-converted ( Dendra2 green-form ) fluorescence imaging of the live cells , activation pulses were fired every 100 frames , and videos of several thousands of frames were acquired under continuous 561 nm illumination ( typically 2000 to 10 , 000 frames per cell ) . Each coverslip was used for a maximum of 45 min after placing them on the scope . The same conditions that were used for SM imaging were used to obtain the images of the pre-converted ensemble fluorescence of Dendra2 , but exchanging the light source for a mercury Lamp ( Intensilight , Nikon ) and appropriate excitation and emission filters ( 485/20 nm and 525/30 nm , respectively ) . In order to compensate the very weak expression levels , images were reconstituted averaging 100 images of a temporal sequence therefore minimizing the noise . Based on RNA Pol II chIP-SEQ data available in the laboratory , we selected genes that are expressed in U2OS cells . Those genes were: SPG21 , LMF1 , BEX1 , IGF2R , GAPDH , HMGB2 , SOD1 , RPL30 , ORC3 , CUL1 , TRAF5 , STX11 . Using RT-qPCR , we compared the mRNA expression of these genes in two conditions: wild type untransfected U2OS and c-Myc-Dendra2 transfected U2OS . In order to precisely compare the amount of RNA , we counted and fluorescence-activated cell sorted ( FACS ) the same number of untransfected and c-MYC-Dendra2-expressing cells . We then performed quantitative PCR experiments and compared the expression levels of the analyzed RNA in the two different conditions . RNA was purified using TRIzol Reagent ( 15596-018; Invitrogen , Life Technologies ) according to the manufacturer's instructions . Total RNA was quantified on a NanoDrop 2000c Spectrophotometer ( Thermo Fisher Scientific , Waltham , MA ) and their quality was evaluated on RNA Nano Chips ( 5067-1511; Agilent 2100 bioanalyzer , Agilent , Santa Clara , CA ) . Reverse transcription from total RNA to cDNA was done with oligo-dT ( 18418-020; Invitrogen , Life technologies ) using SuperScript III RT ( 18080-085; Invitrogen , Life technologies ) and RNAse OUT ( 10777-019; Invitrogen , Life technologies ) . Quantitative real-time PCR ( qPCR ) was done using 5 μl of 1:20 diluted cDNA on a LightCycler480 system ( Roche , Basel , Switzerland ) using Maxima SYBR Green qPCR Master Mix ( K0252; Fermentas , Thermo Fisher Scientific ) . A final concentration of 500 nM of primer pairs ( Eurofins , MWG , Huntsville , Al , designed according to Dugast-Darzacq and Grange , 2009 ) was used for each qPCR reaction . The cycling conditions were as follows: 95°C for 10 min , 45 cycles ( 95°C , 15 s; 58°C , 30 s; 72°C , 20 s ) and melting curve analysis . LightCycler 480 SW 1 . 5 was used to evaluate and to analyze the data .	Transcription factors are proteins that control the expression of genes in the nucleus , and they do this by binding to other proteins or DNA . First , however , these regulatory proteins need to overcome the challenge of finding their targets in the nucleus , which is crowded with other proteins and DNA . Much research to date has focused on measuring how fast proteins can diffuse and spread out throughout the nucleus . However these measurements only make sense if these proteins have access to the same space within the nucleus . Now , Izeddin , Récamier et al . have developed a new technique to track single protein molecules in the nucleus of mammalian cells . A transcription factor called c-Myc and another protein called P-TEFb were tracked and while they diffused at similar rates , they ‘explored’ the space inside the nucleus in very different ways . Izeddin , Récamier et al . found that c-Myc explores the nucleus in a so-called ‘non-compact’ manner: this means that it can move almost everywhere inside the nucleus , and has an equal chance of reaching any target regardless of its position in this space . P-TEFb , on the other hand , searches the nucleus in a ‘compact’ way . This means that it is constrained to follow a specific path through the nucleus and is therefore guided to its potential targets . Izeddin , Récamier et al . explain that the different ‘search strategies’ used by these two proteins influence how long it takes them to find their targets and how far they can travel in a given time . These findings , together with information about where and when different proteins interact in the nucleus , will be essential to understand how the organization of the genome within the nucleus can control the expression of genes . The next challenge will now be to uncover what determines a protein's search strategy in the nucleus , as well as the potential ways that this strategy might be regulated .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2014	Single-molecule tracking in live cells reveals distinct target-search strategies of transcription factors in the nucleus
All animals use olfactory information to perform tasks essential to their survival . Odors typically activate multiple olfactory receptor neuron ( ORN ) classes and are therefore represented by the patterns of active ORNs . How the patterns of active ORN classes are decoded to drive behavior is under intense investigation . In this study , using Drosophila as a model system , we investigate the logic by which odors modulate locomotion . We designed a novel behavioral arena in which we could examine a fly’s locomotion under precisely controlled stimulus condition . In this arena , in response to similarly attractive odors , flies modulate their locomotion differently implying that odors have a more diverse effect on locomotion than was anticipated . Three features underlie odor-guided locomotion: First , in response to odors , flies modulate a surprisingly large number of motor parameters . Second , similarly attractive odors elicit changes in different motor programs . Third , different ORN classes modulate different subset of motor parameters . Humans rely chiefly on vision for their description of the world around them . But for many organisms , the world is dominated by their sense of smell , in which every day activities like finding food depends on the ability to modulate behavior based on olfactory cues . To meet the challenges of detecting and discriminating between different olfactory cues , animals are endowed with large families of odorant receptors ( ORs ) ( Spehr and Munger , 2009; Touhara and Vosshall , 2009 ) that are expressed in olfactory receptor neurons ( ORNs ) and bind to odors . In phylogenetically diverse species , individual ORNs express only one or few ORs ( Buck and Axel , 1991; Su et al . , 2009 ) which largely determine their response specificity; thus ORNs can be classified into a discrete number of classes ( Vosshall et al . , 1999; Vosshall et al . , 2000; Clyne et al . , 1999; Clyne et al . , 1999 ) according to the OR gene they express . Each odor activates multiple ORN classes and is represented by an ensemble of active ORN classes ( Hallem and Carlson , 2006; Hallem et al . , 2004; Malnic et al . , 1999 ) . How activities from different ORN classes are combined to modulate behavior is under intense investigation . The clearest insights into odor modulation of behavior have come from the study of chemical communication . Initial work ( Karlson and Lüscher , 1959 ) suggested that there are 'specialist' ORNs that bind to odors of particular ecological importance and play a major role in inter- and intra-species chemical communication . Recent work has shown that only a minority of chemical communication occurs through the detection of rare , highly-specialized chemicals by a single ORN class ( Kaissling , 1996; Dorries et al . , 1995 ) . Most chemical communication involves integration of signals from multiple ORN classes ( Christensen and Sorensen , 1996 ) . Activation of different combinations of these ORNs can signal predator ( Mullersc , 1971; Endres and Fendt , 2009 ) , encourage approach ( Dorries et al . , 1995; Dorries et al . , 1997; Lin et al . , 2005; ) or aggression ( Ropartz , 1968 ) , cause aggregation ( Bartelt , 1986; Bartelt , 1985 ) and exert their effect by modulating a wide-range of specific motor programs implying that odors have diverse effect on an animal’s behavior . Most ORN classes are not involved in chemical communication but respond to a broad spectrum of chemicals and are 'generalists' ( Hildebrand , 1997 ) . According to current models of the function of generalist ORN classes , activity in the generalist ORNs are decoded by higher-order neurons to create an olfactory percept . Olfactory perception is probed by examining either an animal’s ability to discriminate between different odors or the hedonic valence ( Knaden and Hansson , 2014; Knaden et al . , 2012 ) it assigns to different odors . These approaches to the study of olfactory perception are rooted in the psychophysical literature where an animal is made to choose between a few discrete behaviors ( Green , 1966 ) . A limitation of these approaches is that it has led to a 'portmanteau teleological' ( Kennedy , 1978 ) description of generalist odors as attractants or repellents . As a result , even in a sophisticated analysis of an animal’s navigation to an odor source , the experiments invariably test a single odorant ( Albrecht and Bargmann , 2011; Budick , 2006; Johnsen and Teeter , 1985; Weissburg , 1994; Baker and Kuenen , 1982; Porter et al . , 2007; Gao et al . , 2013 ) . The possibility that , like the ORNs involved in chemical communication , the generalist ORNs also signal diverse behavioral goals and exert their effect by modulating specific motor programs has not been explored . In this study , using Drosophila as a model system , we directly assess whether generalist odors are classified into attractants or repellents or evoke a more diverse set of behaviors . We created a novel behavioral assay in which both the fly’s level of attraction to an odor and the change in locomotion in the presence of that odor could be measured . We investigated how different odors which activate different ORNs modulate locomotion , and how mutating different ORN classes affects a fly’s behavior in response to a natural odor . The null hypothesis was that based on the pattern of ORN activation flies would decide how attractive an odor is and modulate their locomotion according to the level of attraction . Our data is inconsistent with this simple model; and instead supports a different view of odor-guided locomotion that has three salient features . One salient feature is that odors independently modulate a surprising number of locomotor parameters . A second salient feature is that two similarly attractive odors can produce changes in completely different aspects of locomotion . A third salient feature is that a single ORN class can strongly affect some motor parameters ( like run duration ) without affecting other parameters ( like stop duration or angular speed ) . These data support a modular organization in which each ORN class affects a subset of motor parameters , and each motor parameter is affected by a subset of ORN classes . We designed a circular arena ( Figure 1A , B; details in Figure 1—figure supplement 1 and Materials and methods ) in which the flies are constrained to walk between two plexiglass plates . A push-pull arrangement whereby air is pushed into the arena via an inlet tube and pulled through the arena by vacuum creates a sharp interface between a central zone of constant odor concentration ( i . e . the odor-zone ) and a surrounding no-odor zone . To demonstrate that the odor is limited to the odor-zone , we performed smoke visualization , and found that smoke introduced through the inlet tube was confined to the odor-zone , implying that odor from the inlet tube should also be limited to the odor-zone ( Figure 1—figure supplement 2 ) . To directly assess the spread of odors in the arena , we performed field-potential recordings from flies’ antennae ( i . e . electroantennogram or EAG ) at different locations in the arena . The EAG responses were uniformly large inside the odor-zone and rapidly decreased with distance outside ( Figure 1C ) . We estimate that the odor concentration decreases to less than 10% of its peak 3 mm away from the boundary of the nominal odor-zone ( Figure 1D ) . Thus , in our walking arena an odor-zone is separated from a no-odor-zone by a sharp interface . 10 . 7554/eLife . 11092 . 003Figure 1 . A novel behavioral paradigm for measuring odor-evoked change in fly’s locomotion . ( A ) Schematic of the behavioral arena . ( B ) Top view of the chamber . ( C ) Electroantennogram ( EAG ) recording at different locations ( indicated by a dot ) shows a large EAG response when the measurement point lies within the odor-zone ( denoted by circle ) . Response decreases when just the head of the fly is outside the odor-zone and is completely abolished 3 mm away from the odor-zone . ( D ) EAG response plotted as a function of distance from the nominal interface ( n = 5 ) . Red dots correspond to the data points shown in C . ( E ) Sample tracks of a fly in 3 min periods before , during and after presentation of ACV ( left ) . ( F ) Boxplots ( n = 29 ) showing that flies spend more time inside the odor-zone . ( G ) Track of another fly during ACV presentation . Connected dots represent the fractional time spent inside by a single fly before and during odor presentation ( right ) . ( H ) . Sample track of a fly during 2-butanone presentation shows that it is qualitatively different from the tracks in presence of ACV ( left ) . Flies spend more time inside the odor-zone in the presence of 2-butanone ( right , n = 31 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 00310 . 7554/eLife . 11092 . 004Figure 1—figure supplement 1 . A detailed schematic showing the parts of the behavioral chamber . The final chamber consisted of top and bottom assemblies . The bottom assembly was always fixed in place . The top assembly was replaced after a fly was introduced in the arena . During the experiment , the top and bottom assemblies were held together by magnets . Top assembly: Air tube ( 1 ) ( ID 2 . 2 cm , OD 2 . 4 cm ) was glued to a plexiglass plate . The top plexiglass plate ( 2 ) is glued to a Mylar film ( 3 ) which was machined such that the center was lined with holes 0 . 5 mm diameter with 1 mm spacing . The film served as a thin , almost featureless mesh through which odor/air could pass from the air tube to the arena but flies could not escape . The space between the two plexiglass plates is 3 mm . A polypropylene ring ( 4 ) ( 6 . 4 cm in diameter and 3 mm in height ) was machined to have 1 mm holes separated by 1 . 5 mm to allow radial air flow . Bottom assembly: A plastic mesh ( 5 ) was attached to the bottom plate to create a surface on which a fly could walk . The polypropylene ring , plastic mesh and bottom plexiglass plate ( 6 ) were glued together . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 00410 . 7554/eLife . 11092 . 005Figure 1—figure supplement 2 . Flow visualization shows a precise interface between odor zone and no odor zone . ( A ) The left image shows a side view of the behavioral arena . The field of view is enlarged to visualize the region where the inlet air tube attached to the arena . On the right , a background subtracted image of the chamber taken while smoke was introduced through the air tube . The photo shows that smoke is limited to a disc about the same size as the diameter of the inlet tube , implying that flow through the inlet tube is spatially limited . The sharp interface between the odor zone and no-odor zone is marked with arrowheads . ( B ) . A plot of luminance ( due to the smoke ) along the line connecting the two arrowheads shows that the boundary between smoke-filled region and no odor zone is sharp . The nominal odor zone is shaded grey . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 00510 . 7554/eLife . 11092 . 006Figure 1—figure supplement 3 . Control experiments . ( A ) We assessed the effect of air and vacuum on the distribution of the flies in our arena . Each data point represents the fractional time spent inside the odor-zone by a single fly in a 3 min period when the air and vacuum is off ( No air , no vacuum ) , and when it is on ( air on , vacuum on ) . The dotted line represents the fractional time a fly would spend inside the odor zone if they were randomly distributed . There is no statistical difference in the fractional time a fly spends inside the odor-zone before and after air and vacuum were turned on . These data represent the effect of air right after it is turned on when the effect of air should be the largest . Because the air/vacuum are on throughout the course of the experiment , any effects of air during actual experiment should be even less . ( B ) Both our solvent controls - water and paraffin oil cause little change in the distribution of the fly as measured by the attraction index ( see Materials and methods ) . The dotten line represents no change from the before period . The median attraction index ( marked with crosses ) is not significantly different from 0 . ( C ) The probability that a fly is inside the odor zone ( n = 29 ) remains high throughout the 3 minutes when ACV0 is on . Thus , the flies are not desensitized in the 3 min that the odor is on . D . PSTH’s showing spike responses of three ORNs to ACV shows that there is little desensitization of the ORN response during the time frame of the behavioral experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 006 We measured the effect of odors on a fly’s behavior by comparing its locomotion during a 3 min period immediately before the odor was turned on ( i . e . the before period or control period ) and a 3 min period in the presence of the odor ( i . e . during period ) . Before odor onset , the fly spends most of its time at the outer border with occasional forays to the center of the arena ( Figure 1E ) . The fly encounters the odor only when it enters the odor-zone for the first time after the odor is turned on . In the presence of apple cider vinegar ( ACV ) , a strong attractant , flies spend 2 . 5-fold more time inside the odor-zone ( Figure 1F ) than during the control period . The fractional time inside is a measure of attraction . In later experiments , we will use a metric , attraction index ( see Materials and methods ) , based on the change in time spent inside the odor-zone to measure attraction . This is comparable to the measures of attraction used in previous research ( Semmelhack and Wang , 2009 ) . The attraction to the odor remains high throughout the 3 min period when the odor is on ( Figure 1—figure supplement 3 ) and there is no noticeable difference in the response of ORNs to ACV ( Figure 1—figure supplement 3 ) . The solvent used to dilute the odor , itself , did not result in a change in behavior ( Figure 1—figure supplement 3 ) . Do different attractive odors modulate a fly’s locomotion in similar or distinct ways ? In an initial screen for attractive odors , using fractional time spent inside as the metric for quantifying attraction , we found that 2-butanone was strongly attractive to the fly ( Figure 1H , statistical test for attraction shown in Figure 4 ) . We compared how undiluted ACV ( ACV0 ) and 2-butanone at 10-3 dilution ( BUN3 ) , both of which are strongly attractive to the fly , modulate locomotion . The trajectory of a fly in response to the two odors is noticeably different ( compare Figure 1G–H , also see Video 1 , 2 ) . In the presence of ACV0 , flies preferentially walk along the interface between the odor-zone and no-odor zone and make sharp turns outside the odor-zone to return to the odor-zone ( Figure 1E , G ) . In contrast , in the presence of BUN3 , the flies distribute more uniformly inside the odor-zone and return to the odor-zone largely via smooth turns ( Figure 1H ) . Thus , visual observation of the differences in a fly’s trajectory suggests that its response to BUN3 is not simply a scaled-down version of its response to ACV0 . Rather , the differences in the fly’s trajectories in response to the two odors suggest that different odors modulate a fly’s locomotion in qualitatively distinct ways . One difference is the rapid change in speed as the fly enters the odor-zone . Flies decrease their speed when they enter the odor-zone in the presence of ACV0 but not in the presence of BUN3 ( Figure 2A ; also see Video 1 ) . Conversely , as the flies exit the odor-zone there is an increase in speed in the presence of ACV0 but not in the presence of BUN3 ( Figure 2B ) . 10 . 7554/eLife . 11092 . 007Video 1 . This video shows the behavior of the fly to apple cider vinegar ( Video 1 ) . The tracks over the preceding 2 s are marked with dotted white line . The centroid is marked with green . The video also marks stops , sharp turns ( S-turns ) and curved walk . It shows that the fly slows down when it encounters apple cider vinegar ( Video 1 ) . Then , it explores the edge . This particular fly stays just outside the border for some time before entering . Other flies explore the border from the inside . Overall , the fly stays inside the odor-zone and explores the odor-zone with short frequent stops and increased turning . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 00710 . 7554/eLife . 11092 . 008Video 2 . This video shows the behavior of the fly to 2-butanone . The tracks over the preceding 2 s are marked with dotted white line . The centroid is marked with green . The video also marks stops , sharp turns ( S-turns ) and curved walk . Unlike in apple cider vinegar , although the fly keeps returning to the odor-zone , it often walks straight through it . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 00810 . 7554/eLife . 11092 . 009Figure 2 . Different attractive odors modulate different motor parameters . ( A ) Flies decrease their speed when they enter the odor-zone in the presence of ACV0 but not BUN3 . Left: Mean changes in speed between before and during periods . ( n = 29 for apple cider vinegar ( ACV0 ) , n = 31 for 2-butanone [BUN3] ) . Right: Box plots showing the distribution of speed differences upon entering the odor zone . Normalization process in Table 1 . ( B ) Same as in A except that change in speed as the fly leaves the odor-zone is plotted . ( C ) Sample trace showing speed as a function of time for a fly in response to ACV0 and another fly in response to BUN3 . Stop duration decreases for both ACV0 and BUN3 but run duration and run speed only decreases for ACV0 . Shaded regions represent the time during which odor is present . ( D , E , F ) Group statistics for stops and runs . ( Black: before period ( 726 runs and 725 stops from 60 flies ) , magenta: ACV0 ( n = 649 ) , Green: BUN3 [n = 296] ) . Distribution of stop durations is significantly different for both ACV0 and BUN3 ( p <10-9 for both ACV and 2-butanone , KS test ) . Run duration and run speed is only different for ACV0 ( p <10-9 for ACV , p=0 . 08 for BUN3 ) . ( G ) A fly preferentially executes sharp turns ( marked by open black circles , see Materials and methods ) at the interface between the odor-zone and no odor zone in the presence of ACV0 but not in the presence of BUN3 . Traces are tracks of a single fly . ( H ) Left: Group data showing that flies preferentially execute turns right at the odor interface in ACV0 but not BUN3 ( p <0 . 001 for both significantly different points ) . Right: Total turn frequency increases in response to both odors . Black: before period ( n = 60 ) , magenta: ACV ( n = 29 ) , Green: 2-butanone ( n = 31 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 009 Other differences in behavior are not immediately obvious and require further analysis ( Figure 2C–H ) . One example is the odor-dependent modulation of how a fly partitions its time into bouts of walking and stopping ( see Materials and methods ( Martin , 2004; Robie et al . , 2010 ) . In the absence of odor , median stop duration is 0 . 67 s interspersed with occasional long ( >10 s ) stops . In the presence of either odor , the long stops are largely abolished , and the median stop is only 0 . 3 s long ( Figure 2D ) . In the absence of odor , median run duration is 4 . 81 s but runs frequently last >10 s ( Figure 2E ) . However , the run duration ( Figure 2E , median 1 . 0 s ) and speed during runs ( Figure 2F , median speed decreased from 0 . 57 cm/s to 0 . 18 cm/s ) are only modulated in the presence of ACV0 . Modulation of turns by odor also differs between these two odors . In the presence of ACV0 , flies make sharp turns ( see Materials and methods for our definition ) at the border between the odor and no-odor zone ( Figure 2G , H ) . In the presence of BUN3 , there is an overall increase in the frequency of turns in the presence of odor but the increase does not strongly peak at the odor interface . One possible reason for the differences in behavior between ACV0 and BUN3 is that flies perceive ACV0 as more attractive than 2-butanone ( median time spent inside the odor-zone in ACV0 is 1 . 5 times that in the presence of 2-butanone ) and hence modulate more motor parameters in the presence of ACV0 . If this rationale were true , we would expect that at a concentration at which the attractiveness of ACV0 and BUN3 are matched , the motor programs should be matched too . To test this idea , we performed behavioral experiments at 5 concentrations of apple cider vinegar and found that apple cider vinegar at a concentration of 10-2 ( ACV2 ) had a similar level of attractiveness as BUN3 ( Figure 3A ) . Yet , the motor parameters modulated by ACV2 and BUN3 are different . Unlike their response to BUN3 , flies decreased their speed upon entering the odor-zone in the presence of ACV2 ( Figure 3B ) . Moreover , the increased rate of sharp turns at the odor border observed with ACV0 was still observed at the lower concentration of ACV ( Figure 3C ) . Thus , we concluded that equally attractive odors modulate distinct motor parameters . 10 . 7554/eLife . 11092 . 010Figure 3 . Two similarly attractive odors modulate different sets of motor parameters . ( A ) Attraction index showing that the fly is similarly attracted to two odors—apple cider vinegar ( 10-2 , ACV2 , n = 34 ) and BUN3 ( n = 31 ) . Dotted line marks expected value when there is no odor modulation . ( B ) Decrease in speed upon entering the odor-zone is observed with ACV2 but not with BUN3 ( left ) . Box plots showing the distribution of speed differences upon entering the odor zone ( right ) . Normalization described in Table 1 . ( C ) Left: 2 examples of walking trajectory in ACV2 trials during the presence of ACV2 with open black circles marking sharp turns . Right: Increased rate of sharp turns at the odor border is still observed with ACV2 . 2-butanone data is replicated from Figure 2 for comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 01010 . 7554/eLife . 11092 . 011Figure 4 . Odors independently modulate multiple behavioral parameters in an odor dependent manner . ( A ) 17 parameters which are all significantly modulated by ACV0 at p <0 . 05 ( each parameter is described in detail in Materials and methods ) . Bars on top indicate the variables that are significantly different from the solvent control in a rank sum test after Bonferroni correction for multiple comparisons ( p <0 . 003 ) . Dashed line marks the expected value when there is no odor modulation . ( B ) Principal component analysis on the 17-dimensional odor space shows that ACV0 and BUN3 both activate multiple independent motor programs . ( C ) Left: The first canonical variates for the response of individual flies to ACV0 ( magenta ) and BUN3 ( green ) . The responses due to BUN3 and ACV0 are clearly segregated along the first canonical variate . Each circle is a single fly . Right: The distribution of distances between medians in 50 , 000 trials in which odor labels were randomized . Less than 0 . 003 ( 0 . 3% ) of the trials had medians greater than the original distribution ( gray shaded area ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 01110 . 7554/eLife . 11092 . 012Figure 4—figure supplement 1 . Evidence for independence of different motor parameters . If attraction is a singular motor program , then we would expect other motor programs to scale with the level of attraction . We examine the linear correlations between these parameters for a fly’s response to ACV0 . ( A ) On a fly-by-fly basis , time to return is uncorrelated with time spent inside implying that these two programs are independent . Line represent the best linear fit to data . ( B ) Same data as in A but normalized to show fractional change in the two parameter . Line represent the best linear fit to data . ( C ) Cross-covariance between the parameters in our 17-parameter description of behavior show that most parameters are not significantly correlated ( left ) . Only 15 out 136 pairwise comparisons are significantly correlated ( right ) . 1 . attraction index , 2 . time inside/transit , 3 . time to return , 4 . radial density , 5 . speed in , 6 . speed out , 7 . speed crossing in , 8 . speed crossing out , 9 . run duration , 10 . stop duration , 11 . run probability in , 12 . run probability out , 13 . angular speed in , 14 . angular speed out , 15 . smooth turns in , 16 . smooth turns out , 17 . sharp turn at odor bounary . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 012 Another line of evidence supports the idea that the level of attractiveness does not define the motor parameters modulated by a fly . Level of attraction , measured by the fraction of time a fly spends inside the odor-zone , depends on two mutually exclusive programs: Level of attraction would increase if visits to the odor-zone last longer . This can be quantified by measuring time inside/transit ( see Materials and methods ) . Level of attraction will also increase if the time between successive visits becomes shorter ( time to return , see Materials and methods ) . If the level of attractiveness defines a fly’s behavior , we would expect that flies that are more attracted to ACV0 should strongly modulate both time inside/transit ( ti ) and time to return ( tr ) and hence how strongly these two parameters are modulated in a given fly should be correlated . Instead , we found virtually no correlation between ti and tr ( Figure 4—figure supplement 1 ) , implying that there is considerable flexibility in the motor parameters modulated by a given odor . All together , the differences in motor parameters modulated by ACV2 and BUN3 and the lack of correlation between modulation of time inside/transit and time to return shows that level of attraction is a poor descriptor of the change in behavior in response to a given odor . In the rest of this study , we create a quantitative framework to analyze the behavioral response of the fly . Using this framework , we analyze the role of different ORN classes in mediating the behavioral response to ACV and the overall logic by which olfactory cues modulate locomotion . Olfactory behaviors have been characterized quantitatively in simpler organisms such as bacteria ( Berg and Brown , 1972; Pierce-Shimomura , 1999 ) ; C . elegans ( Pierce-Shimomura , 1999 ) and Drosophila larvae ( Gomez-Marin et al . , 2011 ) . In all of these organisms , locomotion can be described in terms of straight runs punctuated by discrete stops during which the organism changes its orientation . Odors exert their effect on locomotion by modulating the duration of runs and by biasing orientation . This simple description fails to describe the behavior of an adult fly because a fly’s locomotion cannot be decomposed into a series of straight runs and stops . For instance , flies can change their orientation by turning smoothly during a run . Our attempts to fit a statistical model such as Hidden Markov Model , which has been employed to describe the behavior of simpler organisms ( Gallagher et al . , 2013 ) , were unsuccessful ( data not shown ) . Therefore , we employed an ad-hoc approach to parameterize the fly’s odor response and investigated how odors affect a large number of locomotor parameters Out of the parameters we investigated , we found 17 parameters ( Table 1 ) that are all significantly ( p <0 . 05 in a rank sum test ) modulated by ACV0 . These parameters describe how a fly’s position ( 4 parameters ) , speed ( 4 parameters ) , run and stop statistics ( 4 parameters ) and turns ( 5 parameters ) are modulated ( details in Materials and methods ) . We expected these parameters to be dependent on each other; and measured pairwise correlation to identify dependent parameters ( Figure 4—figure supplement 1 ) . Surprisingly , we found that the linear correlation between these 17 parameters is small , suggesting that they are largely independent of each other . The number of parameters required to describe a fly’s behavior is large because a fly can modulate its locomotion inside the odor-zone independently from its locomotion outside it and also because the same parameter ( like speed ) is modulated independently across different time windows . 10 . 7554/eLife . 11092 . 013Table 1 . 17 behavioral parameter . Parameters are defined in more details in the methods . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 013Motor parameterWhat it representsHow it is calculated ( see Materials and methods for details ) Attraction indexOverall time spent inside the odor zone ( Time spent inside in the during period ) − ( Time spent inside in the before period ) ( Time spent inside in the before period ) Time spent/transitMedian time spent inside the odor zone per visit ( median time inside per transit 'during' ) − ( median time inside per transit 'before' ) ( median time inside per transit 'before' ) Time to returnMedian time spent outside the odor zone between successive entries into the odor-zone ( median time outside per transit 'during' ) − ( median time outside per transit 'before' ) ( median time outside per transit 'before' ) Radial densityMean distance from the center of the arenaFly’s location as the distance from the center of the arena . Data were binned to 12 bins and normalized by area of each bin . Speed insideMean speed inside the odor zone over the entire period ( average speed inside 'during' ) − ( average speed inside 'before' ) ( average speed inside 'before' ) Speed outsideMean speed outside the odor zone over the entire period ( average speed outside 'during' ) − ( average speed outside 'before' ) ( average speed outside 'before' ) Speed crossing insideAcute change in speed in the first 3 s after entering the odor-zonespeed crossing inside 'during'−speed crossing inside 'before'speed crossing inside 'before'Speed crossing outsideAcute change in speed in the first 3 s after leaving the odor-zonespeed crossing outside 'during'−speed crossing outside 'before'speed crossing outside 'before'Run durationAverage duration of runsrun duration 'during'run duration 'before'Stop durationAverage duration of stopsstop duration 'during'stop duration 'before'Run probability inFraction of time a fly spends running while inside the odor zoneprobability that a fly is moving inside 'during'probability that a fly is moving inside 'before'Run probability outFraction of time a fly spends running when outside the odor zoneprobability that a fly is moving ourside 'during' probability that a fly is moving outside 'before' Angular speed insideAngular speed change inside the odor zone ( average angular speed inside 'during' ) − ( average angular speed inside 'before' ) ( average angular speed inside 'before' ) Angular speed outsideAngular speed change outside the odor zone ( average angular speed outside 'during' ) − ( average angular speed outside 'before' ) ( average angular speed outside 'before' ) Smooth turns inFraction of time a fly is performing a smooth turn inside the odor zoneNumber of frames of smooth turns inside the odor zoneNumber of frames of run inside the odor zoneSmooth turns outFraction of time a fly is performing smooth turns outside the odor zoneNumber of frames of smooth turns outside the odor zoneNumber of frames of run outside the odor zoneSharp turns at boundaryFraction of sharp turns near the odor boundaryThe fraction of sharp turns which took place at the odor border , i . e . in a ring 2 mm around the border To estimate how these 17 parameters are modulated by ACV0 and BUN3 , we performed a rank sum test with Bonferroni correction for multiple comparisons . The parameters which are significantly different at p <0 . 003 ( or 0 . 05/17 - —corresponding to testing whether a given parameter is different after correcting for multiple comparison ) were considered to be significantly modulated by a given odor ( marked by a bar in individual panels in Figure 4A ) . Overlapping but distinct sets of parameters are modulated by ACV0 and BUN3 ( Figure 4A ) . A fly’s response to BUN3 is characterized by a decrease in stop duration and a large increase in its propensity to return to the odor-zone along gently curved trajectories ( reflected as a change in smooth turns outside ) which results in a large drop in tr . The attraction of flies to BUN3 is primarily a result of flies returning to the odor-zone more often . Flies decrease their return time even in the presence of ACV0 , but they do so primarily by making sharp turns outside the odor-zone – an activity which leads them back to the odor . Additionally , the flies spend more time inside the odor-zone every time they enter inside by decreasing their speed inside the odor-zone and making sharp turns to stay inside it . To test how many independent parameters are sufficient to describe a fly’s behavior to odors , we performed principal component analysis ( PCA ) on the 17-parameter behavioral representation of an odor . PCA is a mathematical algorithm that reduces the dimensionality of data while retaining most of the variation in a particular data set . In an extreme case where the attractiveness of an odor determines how strongly every other parameter is modulated , we would expect that a single parameter ( i . e . attraction index ) would capture most of the variability in the behavior . Instead , we found that the first principal component only captured about 25% of the variance in data and the first seven principal components together contribute 90% to the variance in the data ( Figure 4B ) . These results suggest that multiple independent parameters are necessary to describe a fly’s behavior to an odor . Using the representation of the behavior of a fly to an odor in the 17-dimensional behavioral space , we can demonstrate that the behavioral response to BUN3 is different from the behavioral response to ACV0 . We performed the canonical variate analysis ( CVA ) , which finds the axis in the 17-dimensional space which maximizes the ratio of between-group and within-group variances , i . e . , finds the single dimension along which the behaviors due to the two odors is most different ( see Materials and methods ) . Along the first canonical variate , the fly’s response to ACV0 and BUN3 is clearly separable ( Figure 4C ) . We performed a permutation test by randomly assigning responses to the ACV0 or BUN3 group and then performed CVA . The median distance between the randomly assigned ACV0 and BUN3 groups was larger than the median distance between the original groups in only 0 . 3% of the trials ( Figure 4D ) . We performed a similar analysis using the fly’s response to ACV2 and BUN3 . Consistent with the data presented in Figure 3 , different parameters are modulated by ACV2 and BUN3 ( Figure 5 ) . Behavioral responses of a fly due to BUN3 and ACV2 are also separable along the first canonical variate , and support the hypothesis that two similarly attractive odors elicit very different motor responses ( Figure 5 ) . It is possible that the differences between ACV2 and BUN3 could be due to few flies which are strongly attracted to ACV . To control for this possibility we performed the same comparison , but with a subset of ACV2 flies whose distribution of attraction index closely matched that of BUN3 . The differences in behavior between ACV2 and BUN3 still persist ( Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 11092 . 014Figure 5 . Two similarly attractive odors modulate different sets of motor parameters . ( A ) 17 motor parameters show that the parameters modulated by two similarly attractive odor—ACV2 and BUN3 are different . Bars on top indicate the variables that are significantly different from the solvent control in a rank sum test after Bonferroni correction for multiple comparisons ( p <0 . 003 ) . Dashed line marks the expected value when there is no odor modulation . ( B ) Canonical variate analysis shows that the behavioral response to ACV2 and BUN3 are distinct along the first canonical variate . ( C ) Permutation tests show that less than 4% of trials in which the odor labels are randomized had median distances greater than that of the original grouping . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 01410 . 7554/eLife . 11092 . 015Figure 5—figure supplement 1 . Differences in behavior due to BUN3 and ACV2 is not due to a few flies which are very attractive to ACV or due to different temporal evolution of behavior in the two odors . ( A , B ) . Subsampling of ACV2 flies does not affect behavioral differences . ( A ) Shows the subsampling . We removed the flies which contributed to the long tail in the distribution of attraction index . This can be observed in the boxplot but is more clear in the cumulative probability distribution . ( B ) There is some change in the distribution . Even with the subsampling , all 6 parameters which are modulated more strongly by ACV2 than by BUN3 were still modulated significantly strongly in the subsampled population . Apart from being an excellent control experiment , this experiment adds to other evidence which suggests that attraction is not predictive of other behavioral parameters measured here . ( C ) Behavior due to ACV2 and BUN3 are even more distinct in the first 20 s after odor exposure . We performed our 17-parameter analysis but with only the first 20 s after odor exposure rather than the entire 3 min period . Canonical variate analysis on this 17-parameter space shows that the median distance between behaviors is larger in the first 20 s compared to the entire 3 min odor period . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 01510 . 7554/eLife . 11092 . 016Figure 5—figure supplement 2 . Differences in the motor parameters modulated are not simply due to differences between simple and complex odors . ( A ) Behavioral responses due to two complex food odors - ACV and banana are distinct along the first canonical variate . Two monomolecular odors , 2-butanone ( BUN3 ) and ethyl acetate ( at a concentration of 10-4 , ETA4 ) also exhibit different behaviors along the first canonical variate . ( B ) Left: Distribution of stop duration pooled from all the stops in multiple flies in no odor ( black , n = 60 ) , 2-butanone ( green , n = 31 ) and ethyl acetate ( red , n = 23 ) . Right: Box plots showing the distribution of median stop durations across multiple flies . Stop duration decreased in 2-butanone but not in ETA4 . ( C ) Left: Distribution of run duration pooled from all the runs in multiple flies . Right: Box plots showing the distribution of median run durations across multiple files . Run duration increases in the presence of ETA4 but not in BUN3 . Thus , a given parameter such as run duration could increase , decrease or be constant in the presence of different attractive odors . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 016 Taken together , the obvious visual differences in the trajectories of the fly in the presence of the two odors ( Figure 1F , G ) , differences in modulation of individual parameters in different odors ( Figures 2 , 3 ) and statistical analyses presented in Figures 4 , 5 and Figure 4—figure supplement 1 strongly support the idea that odors evoke changes in multiple motor parameters in an odor-dependent manner . It is possible that the differences in behavior arise from different temporal evolution of behavior . That is , the initial behavioral response is the same for all odors; but , behavioral response evolves differently for different odors resulting in behavior which is different for different odors . To control for this possibility , we evaluated the differences in behavior for BUN3 and ACV2 pair in the first 20 s after odor was turned on . We found that the differences in behavior between BUN3 and ACV2 are even greater in the first 20 s of exposure to odor ( Figure 5—figure supplement 1 ) . It is possible that the differences we observe between the two odors reflect the fact that ACV is a complex food odor and BUN is a monomolecular odor . To test for this possibility we compared the responses of the fly to ACV and banana . These responses are also distinct ( Figure 5—figure supplement 2 ) . Similarly , the response to BUN3 is different from the response to another monomolecular odor , ethyl acetate at 10-4 dilution ( ETA4 , Figure 5—figure supplement 2 ) . Like most other odors , ACV and BUN3 activate multiple ORN classes ( Semmelhack and Wang , 2009; Olsen et al . , 2010 ) . To understand how activities from multiple ORN classes are integrated to drive behavioral response to natural odors , we carried out a series of experiments to dissect the contribution of different ORN classes to the behavioral response of a fly to ACV . We first assessed which ORN classes are activated by ACV . ORNs are housed within three morphological classes of sensilla ( Shanbhag and Muller , 1999 ) – basiconic , coeloconic and trichoid . Because most of the generalist ORN classes reside in the basiconic and coeloconic sensilla ( Su et al . , 2009 ) , we measured odor-evoked responses from all basiconic and coeloconic sensilla using single sensillum recording . We matched the stimulus in the behavioral chamber to that on the electrophysiological rig by matching the amplitude of EAG response measured in the behavioral chamber to the EAG response in an electrophysiological rig . In all , we recorded responses from 10 basiconic sensilla which house 24 ORN classes , and 4 coeloconic sensilla which contain 11 ORN classes and from the antennal sacculus which houses Ir64a-ORNs . Out of these 36 ORN classes , we found that ACV0 activates 7 ORN classes ( Figure 6A , see Materials and methods for details ) . Because we calculate most motor parameters as an average over the entire 3 min period , and because we have not observed any effect of different levels of activity of ORN on behavior , we are only reporting which ORNs are activated by ACV . In the rest of this study , we will use mutants , specific odors and lower concentration of ACV to probe the contribution of different ORN classes activated by ACV to behavior . 10 . 7554/eLife . 11092 . 017Figure 6 . Activation of single ORN class leads to a change in a subset of motor parameters . ( A ) Schematic representation of ORNs activated by ACV0 . ( B ) Ethyl acetate at low concentrations activate only Or42b-ORNs . ( C ) Behavioral modulation by activation of Or42b-ORNs alone using low concentration ( 10–8 ) ethyl acetate ( ETA8 ) . ACV0 data is shown for comparison . Flies show increased angular speed inside the odor-zone and have a shorter run duration in response to ETA8 compared to the solvent control . Significantly modulated parameters ( in a ranksum test after Bonferroni correction ) are enclosed in a shaded box . ( D ) 2-butanone at 10–5 ( BUN5 ) activates only Or42a-ORNs . Activating Or42a-ORNs results in modulation of three parameters . No other parameters were modulated . Only the significantly modulated parameters of the 17 are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 01710 . 7554/eLife . 11092 . 018Figure 6—figure supplement 1 . Ethyl acetate at low concentrations nearly saturates Or42b-ORNs without activating any other receptor . To establish that low concentration of ethyl acetate activates Or42b-ORNs and no other ORNs , we first show that Or42b-ORNs are activated by low ethyl acetate concentration . Next we show that there is a large change in the field potential recordings from the antennae in the Or42b-mutants implying that Or42b . ( A ) Raster plot showing spikes of Or42b- and Or92b-ORNs , both of which are present on the ab1 sensilla . It was unnecessary to sort between these spikes . Ethyl acetate at 10-7 ( ETA7 ) results in a large increase in the firing rate which was completely abolished in the Or42b mutant implying that Or42b-ORNs are activated by ETA7 . Spontaneous activity is dramatically reduced in Or42b mutant because the Or42b-ORNs are not spontaneously active . ( B ) Average response of Or42b-ORN at low concentration of ethyl acetate . ( C ) Field potential recordings from the antennae that is a measure of responses from all the neurons in the antennae shows a robust response to low concentrations of ethyl acetate which is abolished in the mutant . ( D ) Field potential recording from the palp shows that no neuron in the palp responds to concentrations of ethyl acetate below 10-–6 dilution . ( E ) Recordings from ab2B sensilla which expresses Or85a receptor shows that ethyl acetate at 10–6 does not activate ab2B ORNs projecting to DM5 glomerulus . Raster plot on the left and PSTH on the right . This is significant because it rules out the possibility that the lack of attraction observed in response to ETA at low concentration is due to the activation of DM5 glomerulus resulting in repulsion and canceling attraction to activation of Or42b-ORNs . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 01810 . 7554/eLife . 11092 . 019Figure 6—figure supplement 2 . Behavioral response to ETA8 is abolished in Or42b mutants . In figure 6 , we show that activating Or42b-ORNs alone using ETA8 results in a change in run duration and angular speed inside . If Or42b-ORNs are indeed responsible for the behavioral effects of ETA8 , then the modulation of these parameters should be abolished in the Or42b mutant . In this supplement , we show that the behavioral response to ETA8 is abolished in the Or42b mutant . These data further support the idea that activation of Or42b-ORNs affect run duration and angular speed inside . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 019 We first investigated how activation of a single ORN class modulates behavior . A previous study has shown that activating Or42b-ORNs alone produces robust attraction ( Semmelhack and Wang , 2009 ) . Moreover , we found that of the 7 ORNs activated by ACV , Or42b-ORNs are the most sensitive ( see below ) . Therefore , we started by analyzing the role of Or42b-ORNs in behavior ( Figure 6B ) . We activated the Or42b-ORNs alone by using low concentrations of ethyl acetate ( Olsen et al . , 2010 ) . We have previously shown that ethyl acetate at a concentration <10-6 only activates Or42b-ORNs33 , 34 and confirmed that the same is true under the exact stimulus conditions used in this study ( Figure 6—figure supplement 1 ) . In behavioral experiments , we found that ethyl acetate at 10-8 ( ETA8 ) does not elicit attraction ( Figure 6C , first panel ) . The lack of attraction is not due to the activation of Or85a-ORNs ( Figure 6—figure supplement 1 ) which are important for odor-mediated repulsion ( Semmelhack and Wang , 2009 ) . However , the lack of attraction does not imply a lack of behavioral response; we measured how ETA8 affects each of the 17 parameters and found that ETA8 elicits increase in angular speed and decrease in run duration compared to the solvent control ( Figure 6C ) . Thus , activation of Or42b-ORNs alone affects only a small subset of motor parameters affected by ACV0 , and modulation of these parameters does not lead to attraction . Moreover , both the increase in angular speed and decrease in run duration by ETA8 are abolished in the Or42b null mutant ( Figure 6—figure supplement 2 ) . It is possible that the differences in behavioral response to ETA8 and ACV0 arise from the different firing rate these odors elicit in the Or42b-ORNs . This possibility is ruled out by the fact that ETA7 which elicits a similar firing rate as ACV2 ( see Figure 7 ) elicits the same behavioral response as ETA8 ( data not shown ) . In our behavioral paradigm , we did not observe any behavioral effect of increased firing rate in Or42b-ORNs on behavior . 10 . 7554/eLife . 11092 . 020Figure 7 . Higher concentration of ACV activates more ORNs and recruits more motor parameters . ( A ) PSTHs showing the response of the Or42b-ORN to ACV and ethyl acetate ( mean ± SEM , n = 5-7 ) . ( B ) At all spike rates , ACV is more attractive than ETA implying that attraction due to ACV is not due to activation of Or42b-ORNs alone . ( C–E ) Motor parameters modulated by increasing concentration of ACV . Only parameters that are modulated are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 020 We performed a similar experiment with another ORN class , Or42a-ORNs , which are also activated by ACV . To activate Or42a-ORNs alone , we used 10-5 dilution of 2-butanone ( BUN5 ) , a concentration at which it only strongly activates Or42a-ORNs ( Olsen et al . , 2010 ) . We found that BUN5 also results in the modulation of a small number of motor parameters ( Figure 6D ) . Importantly , BUN5 affects a distinct , but overlapping set of motor parameters compared to ETA8 . Like ETA8 , BUN5 affects angular speed inside the odor-zone . But unlike ETA8 , it does not affect run duration; rather it causes a decrease in stop duration . No other parameters ( of the 17 we investigated ) other than the three shown in Figure 6D were affected when Or42a-ORNs were activated . These experiments suggest that each ORN class modulates a distinct but overlapping set of motor parameters . The modular organization in which different ORN classes activate different motor parameters makes two predictions: 1 . the behavior of a fly at ACV concentrations which activates a single ORN class should be similar to the behavior when the same ORN class is activated by another odor , 2 . Since higher odor concentrations activate more ORN classes we anticipate that as the odor concentration is decreased only a subset of motor parameters should be affected . To test these predictions , we compared the behavioral response elicited by different ACV concentrations to the behavioral response elicited by activating the ORN class most sensitive to ACV . First , using the same strategy we employed to show that Or42b-ORNs are the only ones activated by low concentrations of ethyl acetate , we established that Or42b-ORNs are the most sensitive to ACV ( data not shown ) . ACV elicits a robust response in Or42b-ORNs even at a 10-4 dilution ( ACV4 ) . EAG response to ACV4 is abolished in the Or42b-mutant implying that the Or42b-ORNs are the only ORNs strongly activated by ACV4 . Next , in order to perform a direct comparison between the behavioral responses elicited by activating Or42b-ORNs alone to the behavioral response elicited by the entire complement of ORNs activated by ACV , we matched the spike rate elicited by ethyl acetate to that elicited by ACV ( Figure 7A ) . ACV4 activates Or42b-ORNs alone and expectedly does not elicit attraction ( Figure 7B ) . At concentrations below 10-6 ( ETA7 and ETA8 ) , ETA also activates Or42b-ORNs alone and fails to elicit attraction ( Figure 7B ) . At higher concentrations ACV is attractive to the fly ( Figure 7B ) because it activates more ORN classes ( see Figure 7C–E ) . ETA , too , at concentrations 10-6 or above activates multiple ORN classes . Consistent with the activation of more ORN classes , flies spend more time inside the odor-zone in the presence of ETA6 . As in the case of ETA8 ( Figure 6C ) lack of attraction does not imply a lack of behavioral response; ACV4 affects the same two parameters as ETA8 - run duration and angular speed inside ( Figure 7C ) . ACV3 activates Or59b-ORNs in addition to Or42b-ORNs . Behaviorally , ACV3 modulates more motor parameters ( Figure 7D ) ; this is likely due to the activation of more ORN classes . Importantly , run duration and angular speed inside continue to be modulated because Or42b-ORNs are still activated . In addition , ACV3 also elicits a decrease in the time to return , a likely consequence of increased turn-rate outside the odor-zone . The number of parameters modulated increases further when the concentration is increased to ACV2 . At this dilution , 10 parameters are affected ( Figure 7E ) and the behavior appears qualitatively similar to the fly’s behavior to ACV0 . These data further support the hypothesis that different ORN classes modulate different motor parameters . At low odor concentration when only Or42b-ORNs are activated by the odor , only run duration and angular speed in the odor-zone is affected . At higher odor concentrations , consistent with the activation of more ORN classes more motor parameters are modulated . Another prediction of our model is that if we perturb the activity in a subset of ORN classes , we should observe a change in a subset of motor parameters . To investigate this issue , we studied how the fly’s ACV-evoked behavior changes in Or42bEY14886 flies , a previously characterized null mutation in the Or42b receptor ( Bhandawat et al . , 2007 ) . Or42b-mutants were as attracted to ACV as control flies ( Figure 8B , E -attraction index panel , p >0 . 1 on a ranksum test ) ; but some of the individual motor parameters were affected in the mutant . Two examples are shown in Figure 8C , D . The mutant flies had a significantly sharper distribution at the odor border ( Figure 8B , C ) . The change in run duration was significantly diminished in the mutants compared to the control flies ( Figure 8D , E ) . Based on the fact that activating Or42b-ORNs resulted in an increase in angular speed ( Figure 6C , 7C ) , we expected a decrease in angular speed inside in the Or42b mutants . Although the median angular speed inside does decrease in the mutant , this decrease in angular speed is not significant . The lack of significant effect on angular speed implies that multiple ORNs activated by ACV can modulate angular speed . Overall , only 3 parameters out of 17 we investigated were affected in the mutants ( Figure 8E ) . To confirm that these differences are odor-dependent and do not simply represent differences between the two genotypes , we compared the responses of control and mutant flies to the solvent control and found no differences ( Figure 8—figure supplement 1 ) . Thus , consistent with experiments in which we activated Or42b-ORNs ( Figure 6 ) , a null mutation in the Or42b gene does not abolish the attraction of a fly to ACV . Instead , it selectively affects a small fraction of motor parameters . 10 . 7554/eLife . 11092 . 021Figure 8 . Mutating a single ORN class does not affect attraction to ACV0 but changes certain motor parameters . ( A ) In the Or42b mutant , a single ORN class is non-functional . ( B ) Sample tracks showing that both wild type and Or42b-/- flies are attracted to ACV . Tracks also show that the mutant fly is closer to the odor interface than the wild-type flies . ( C ) Mean radial density . Although both wild type and Or42b-/- flies are attracted to ACV , mutant flies show a sharper distribution at the interface ( KS test p <0 . 001 ) . ( n = 29 for control , n = 25 for Or42b mutant ) ( D ) Run duration is not modulated in the Or42b mutants . ( E ) 17 motor parameters of wild type and Or42b-/- flies in response to ACV . Run duration and speed while crossing out are modulated more strongly in the wild-type compared to the mutant . Shaded box show parameters which are significantly modulated in the wild type but not Or42b mutant . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 02110 . 7554/eLife . 11092 . 022Figure 8—figure supplement 1 . Responses of wild-type and Or42b mutant flies to the solvent control are not significantly different for any parameter even if not corrected for multiple comparisons . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 022 Since different ORN classes affect distinct motor parameters , inactivating multiple ORN classes should affect the modulation of more parameters than a single ORN class . Out of the 7 ORN classes activated by ACV0 , 3 ORNs require a co-receptor , Ir8a . In the Ir8a mutant ( Abuin et al . , 2011 ) , these 3 ORN classes are non-functional and only 4 ORN classes activated by ACV0 are functional ( Figure 9A ) . The Ir8a mutant flies are still attracted to ACV0 ( Figure 9D , p <0 . 0001 for both control and mutant in a ranksum test ) ; there is a small but statistically insignificant decrease in attraction to ACV0 in the mutants compared to the control flies . However , there is a large change in the modulation of 5 motor parameters out of the 17 under study ( Figure 9D ) . The most striking difference between the wild-type and mutant is that the immediate modulation of speed upon entering the odor-zone is completely abolished in the mutant ( Figure 9B , p >0 . 1 in a ranksum test for the mutants ) . Similarly , modulation of speed as the fly exits the odor-zone is also abolished ( Figure 9C ) . Most of the parameters affected in the mutants are those that , in control flies , are affected as the fly crosses into the odor-zone or are modulated inside the odor-zone . Thus , it is likely that the Ir8a-mutant strongly affects changes in motor program which characterizes a fly’s behavior in the presence of high concentration of food odor . Consistent with this idea , the time spent/transit decreases in the mutant without a significant change in time to return ( Figure 9D ) . The effect of Ir8a-ORNs shown above is not just on the fly’s response to ACV but is also observed in response to other odors ( Figure 9—figure supplement 1 ) , and supports the notion that different ORN classes affect different motor parameters . 10 . 7554/eLife . 11092 . 023Figure 9 . Modulation of motor programs by ACV inside the odor-zone is strongly affected in the Ir8a-mutant . ( A ) Three ORN classes activated by ACV are non-functional in Ir8a mutant . ( B ) The reduction in speed when the wild type flies enter the odor-zone in the presence of ACV is abolished in the Ir8a mutant . ( n = 29 for wild type and n = 33 for Ir8a mutant ) . ( C ) Increase in speed when the flies exit the odor-zone is also abolished . ( D ) 17-motor parameters in the wild type and the mutant . Many aspects of a fly’s locomotion are affected in the Ir8a mutant . Bars on top reflect whether the parameters are significantly different in the two genotypes compared to the solvent control . Shaded parameters are the ones affected in the Ir8a mutant . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 02310 . 7554/eLife . 11092 . 024Figure 9—figure supplement 1 . Ir8a-ORNs directly modulate speed . ( A ) Change in speed as a function of field potential response at the ORN layer in flies in which Ir8a-ORNs are the only ones active . Each data point represents a different odor . Δspeedis the difference in average speed in the 2 s after entering the odor-zone when an odor is present vs . control period . Ir8a response = Peak field potential response . Odors ( in order of the response they elicit from Ir8a ) = ethyl acetate ( 10-–5 ) , ethyl acetate ( 10–4 ) , 2-butanone ( 10–3 ) , ethyl butyrate ( 10–3 ) , ACV ( 10–2 ) , banana , ACV0 , acetic acid ( 10–2 ) , acetic acid ( 10–2 ) + propionic acid ( 10–2 ) . ( B ) Decrease in speed elicited by propionic acid is abolished in the Ir8a mutant . C . Decrease in speed elicited by banana is strongly attenuated in the Ir8a mutant . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 024 That the motor parameters inside the odor-zone are specifically affected in the Ir8a mutants suggests that two independent motor programs underlie a fly’s response to odors: One motor program corresponds to the fly’s search for the odor source; the other corresponds to the change in fly’s behavior once the odor source is near . It is possible that Ir8a-ORNs preferentially modulates the motor program which underlie changes in a fly’s search strategy from a global search in the absence of odor to a local search in the presence of high concentration of food odor . Consistent with this idea , Ir8a mutants spend less time in the odor-zone in the presence of ACV compared to control flies ( Figure 9D ) . To further probe this issue , we plotted the time spent inside as a function of time after odor onset ( Figure 10A ) . In the absence of the odor , both the mutant and the control flies spent a fifth of their time inside the odor-zone . Upon odor onset , there is a rapid increase in the time spent inside the odor-zone followed by a small decline . The initial increase in the time spent inside is similar for both the mutant and the control . However , the fractional time inside peaks faster in mutants suggesting that both mutant and control flies find the odor efficiently but the control flies stay inside the odor-zone for longer . We also plotted both the time inside/transit and time to return as a function of time after odor onset ( Figure 10B ) . Time inside per transit was longer for the wild-type flies throughout the entire 3 min that the odor was on . Time to return is similar for the two genotypes . These data suggest that the change in the Ir8a mutants’ behavior inside the odor-zone is manifested as decrease in time spent inside the odor-zone . 10 . 7554/eLife . 11092 . 025Figure 10 . Ir8a mutants approach ACV at the same rate as wild-type but spend less time in proximity to it . ( A ) Ir8a mutant find the odor as well as the wild-type flies but because they spend less time inside the odor-zone on each visit their attraction to odor decreases with time at a faster rate than wild-type flies . ( B ) . The time a fly spends inside the odor-zone on a single visit decreases at a much slower rate for wild-type than for the Ir8a mutant . In contrast , the time a fly takes to return to the odor-zone is the same for both genotypes . ( C ) A photo of the arena with the outer edge marked with a white line . Fly can be seen as a tiny white object ( marked with an arrowhead ) . The hole in the center is used to deliver odors . ( D ) Tracks of a control fly shows that it is strongly attracted to apple cider vinegar . ( E ) Radial density averaged over multiple flies show that Ir8a mutants find the odor as quickly as the wild-type but leave the odor much faster . ( F ) Speed ( between 0-–2 min after odor on ) near the odor source decreases in wild type but not in the Ir8a mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 025 To investigate whether the findings above extend to conditions in which a fly usually encounters odors , we measured how wild-type and Ir8a mutant flies redistributed in response to a point source of ACV placed in the center of a circular arena ( Figure 10C ) . Before the odor was presented , flies spent most of their time at the periphery ( Figure 10D , left panel ) . In the presence of ACV the wild-type flies navigate to the odor source within seconds ( Figure 10D , right panel ) . The Ir8a-mutant flies also navigated rapidly to the odor . But , the mutant flies dispersed from the odor source at a much faster rate than did the controls ( Figure 10E ) . We measured the mean speeds in 5 radial bins and found that in the control flies , but not in Ir8a mutant flies , speed decreased in bins closer to the odor source ( Figure 10F ) . No speed modulation was observed in the Ir8a mutants ( Figure 10F ) . These results are consistent with a modular organization of behavior and with the fact that Ir8a-ORNs strongly affects a fly’s behavior near the odor source but not how a fly finds that source . Two lines of evidence suggest that odors independently modulate an unexpectedly large number of motor parameters . One line of evidence is the low correlation among the 17 parameters modulated by ACV0: Of the 136 pair-wise correlations between these parameters , only 15 are significantly correlated ( Figure 4—figure supplement 1 ) . Consistent with this lack of correlation between motor parameters , PCA performed on the 17-dimensional behavioral description shows that 7 principal components are necessary to explain 90% of the variance in the data . Another line of evidence is that different odors modulate different parameters . BUN3 decreases stop duration but not run duration ( Figure 2 ) ; and ethyl acetate at 10-4 dilution increases run duration but has no effect on stop duration ( Figure 5—figure supplement 2 ) . These data imply that run duration and stop duration can be modulated independently . In addition , different odors have different effects on sharp and smooth turns ( Figure 4 ) . Experiments with Ir8a mutants suggest that the acute decrease in speed observed within the first second of encountering an odor might be yet another independently controlled parameter . Thus , it appears that at least 5 parameters: stop duration , run duration , sharp and smooth turns and acute changes in speed are all modulated independently . These results together support the idea that at least 5 independently controlled parameters underlie a fly’s response to odors . Despite the fact that we are characterizing the behavior using 17-parameters , this description is unlikely to be a complete description of the behavior . A limitation of our analysis is that it does not describe the dynamics of a fly’s locomotion . Simple generative models like Hidden Markov Models which has been successfully used to model locomotion of simpler animals ( Gallagher et al . , 2013 ) , are poor fits to a fly’s tracks ( data not shown ) because of the complexity of a fly’s locomotion . Ultimately , a hierarchical statistical model ( such as a Hierarchical Hidden Markov Model ) is necessary to model both how a fly searches its environment and how odors modulate this search . Even without a hierarchical model , it is clear that processes which govern a fly’s behavior outside the odor-zone are distinct from the processes which govern its behavior inside it . Consistent with this idea , most of the correlated parameters ( Figure 4—figure supplement 1 ) correspond to the fly’s behavior inside the odor-zone . Thus , a sharp decrease in speed as a fly enters the odor-zone is predictive of the overall decrease in speed , decrease in run duration and increased turning inside the odor-zone . It is likely that these 4 parameters form part of a coordinated response which constitutes a local search near high concentrations of food odor . Most studies of odor-guided locomotion in animals are designed with the inherent assumption that the problem an animal is solving in nature is that of navigating to or away from a single salient odor . The problem of finding a distant odor source is certainly one important class of problems that the olfactory system solves . Studies across multiple species have shown that a conserved mechanism is in place to detect and track a particularly salient odor source over long distances ( Johnsen and Teeter , 1985; Porter et al . , 2007; van Breugel and Dickinson , 2014; Kennedy , 1983; Cardé and Willis , 2008 ) . But it is unlikely that tracking an odor source over long distances is the only problem an olfactory system is trying to solve . A more general class of problems faced by the olfactory system is to navigate a complex odor environment ( Riffell et al . , 2014 ) consisting of multiple odor sources of varying salience . Navigating such a complex odor environment requires fine control over locomotion . It is likely that an animal’s overall search strategy is under multimodal control and is flexibly linked to the overall visual , wind and odor environment . Our experiments suggest , surprisingly , that an important feature of this control is a precise modulation of motor parameters by the exact pattern of ORNs activated . This component of an animal’s odor-guided locomotion has been overlooked so much so that in most experiments an odor is treated as a 'scalar' which either turns on a motor program or changes the gain of an ongoing behavior . There is general consensus that a combinatorial activation of multiple ORN classes drives behavior; however , there is little consensus regarding the role of a single ORN class in behavior and the principles by which activities from multiple ORN classes are combined to yield behavior . For instance , in one study it was shown that the Or42b-ORNs plays an important role in attraction to a natural odor ( Semmelhack and Wang , 2009 ) . In contrast , another study finds that Or42b-ORNs ( Knaden et al . , 2012 ) does not contribute to attraction at all and proposes a more modest contribution of a single ORN class to behavior . Consistent with the study in 36 we show that activation of Or42b-ORNs is not enough to attract flies to an odor . Similarly , Or42b mutants are still strongly attracted to ACV suggesting a modest role for Or42b-ORNs in mediating attraction to odor . These differences in behavioral response to Or42b-ORNs in different labs could result from differences in behavioral paradigm or the state of the fly . Apart from Or42b-ORNs , we have also performed experiments in which we specifically activate Or42a-ORNs , Or59b-ORNs and Or7a-ORNs but failed to observe strong attraction . Thus , it is unlikely that activating a single ORN class will result in the strong attraction observed in the presence of a natural attractant . On the other hand , most odors which activate multiple ORN classes are somewhat attractive to a hungry fly suggesting that a mild attraction to odor can be elicited readily and does not require a unique or rarely activated combination of ORN classes . Similarly , attraction to natural odors is robust: even the Ir8a mutant is attracted to ACV . These findings together suggest that there are many redundant paths to odor-mediated attraction . More importantly , our work suggests a novel framework for the study of relation between ORN activity and behavior . Although , activation of Or42b-ORNs by themselves does not elicit attraction , they have a significant effect on some motor parameters but not others . Another ORN class , Or42a-ORNs , affects another subset of motor parameters . These data suggest that individual ORN classes only affect a subset of motor parameters . Our experiments with different ACV concentrations show that as the ACV concentration is increased , more ORN classes are recruited which in turn result in the modulation of more motor parameters . These data led us to propose a fundamentally different model of odor-guided locomotion in which individual ORN classes affect a small number of motor parameters ( Figure 11B ) . The above model finds strong support in experiments aimed at dissecting the contribution of different ORN classes to the fly’s response to ACV . We used three complementary approaches to activate subsets of ORNs: activation of known ORN classes using low concentrations of odors , different ACV concentrations and different OR mutants . We find that a given pattern of ORN activates the same motor parameters irrespective of how the ORNs are activated: activation of Or42b-ORNs by ACV4 or ETA8 decreases run duration and increases turning . Similarly , ACV does not elicit a decrease in speed until the concentration is high enough to activate Ir8a dependent ORNs; this observation is consistent with the strong effect of Ir8a-ORNs on speed . Overall , activation of different subsets of ORN which are all activated by ACV results in the modulation of different subsets of motor parameters . Multiple lines of evidence suggest that Or42b-ORNs affect run duration . Similarly , Ir8a-ORNs strongly affects the modulation of speed . It is tempting to conclude that different motor parameters are simply modulated by instantaneous summed activity from a subset of active ORN classes . This model is unlikely to be correct for all motor parameters . In particular , the principles underlying modulation of both sharp and smooth turns appear to be impervious to this simple analysis . This is reflected in the fact that although Or42b-ORN activation can affect angular speed , modulation of angular speed is not significantly affected in the Or42b mutants . Moreover , the facts that BUN3 modulates sharp turns but with little spatial specificity; and , ACV modulates it with great spatial specificity which is increased in the Or42b mutants indicate that a more complex rule than simple summation of ORN activities is at work . In sum , our work indicates that each motor parameter is modulated by a subset of ORNs; but , the relation between ORN classes and motor parameter could be simple for some parameters and complex for others . In the same vein , in this study in deriving the relation between ORN activation and behavior we have focused on which ORN classes are active . Future experiments will determine whether the level of activation and temporal characteristics of ORN spike response is relevant to behavioral output . To delineate the relationship between the spatio-temporal pattern of ORN activity and motor parameters , we need both a hierarchical model of a fly’s locomotion and experiments aimed at understanding how inputs from different ORNs are integrated by higher-order neurons in the fly’s brain . Our experiments with the Ir8a mutant strengthen our conclusion regarding modular organization of olfactory control on locomotion , and also suggest one organizing principle . Removal of the three Ir8a dependent ORN classes reveals strong effect on some motor parameters while leaving other motor parameters completely unaffected . The motor parameters affected by Ir8a mutants strongly affect a fly’s behavior inside the odor-zone without strongly affecting its behavior outside the odor-zone . These results further support the idea that the modulation of locomotion which leads to attraction to the odor source and modulation of locomotion which results in a fly’s exploration of regions of high odor concentration are controlled independently . Our data suggests that activation of Ir8a-ORNs is important for initiating a local search characterized by short slow runs and frequent but short duration stops . This local search represents a part of an overall change in strategy that promotes a thorough search of a local area . Recent work has demonstrated that locomotion and proboscis extension to initiate feeding are antagonistically related ( Mann , et al . , 2013 ) . Thus , food odors and tastants could potentially work in concert to put brakes on locomotion and initiate feeding . An animal’s behavioral response depends not only on the immediate sensory stimulus but also on context and the animal’s goal . There are two competing models which account for flexible sensorimotor transformation . In the first and the dominant model , sensory processing creates an internal representation of the external world . Using this internal model , decision-making circuits can take a behavioral decision which is then executed by motor circuits . A student of neuroscience would be strongly inclined to this model after picking up any neuroscience textbook ( Kandel , 2000; Purves , 2008 ) . The strength and the allure of this model lie in the access to an internal representation . In this model , the versatility in animal behavior arises by endowing the nervous system with cognition - the ability to plan actions based on an internal model of the external world ( Cruse , 2003; Turvey and Fonseca , 2009 ) . This model has a centralized 'executive system or circuit' which takes into account sensory input and the state of the animal to plan action ( Cruse , 2003 ) . But , efforts to create robots with such a subsystem have failed ( Brooks , 1991 ) . Updating internal representations with every small act is just too slow . The alternative model proposes a modular organization containing parallel sensorimotor loops , each of which represents a solution to one aspect of an ecological problem ( Wehner , 1987; Wessnitzer and Webb , 2006 ) . In this model , the role of sensory systems is not to create an internal representation of the world but to extract behaviorally relevant features in the environment . A presumed limitation of the modular organization is that it would result in simple , stereotypical behaviors and hence this framework is considered relevant only for simple behaviors performed by simple organisms . Recent behavioral work has revealed that even animals with simple nervous systems are capable of complex behaviors . Fruit flies , for instance , can make use of both idiothetic ( Pick and Strauss , 2005 ) and allothetic cues to navigate and also exhibit spatial memory ( Ofstad et al . , 2011 ) . At the other extreme , recent work in the mammalian retina has revealed that many retinal ganglion cells are feature detectors and only affect specific behaviors ( Zhang et al . , 2012 ) . These studies have led to the realization that a modular organization is not necessarily inconsistent with complex behaviors or complex nervous systems . We propose a model for flexible sensorimotor transformation in the olfactory system which was proposed in the context of work in primate vision . This model suggests that perception ( or internal models ) of objects occur via circuits that are independent from action directed at the same object ( Goodale and Milner , 1992 ) . As proposed in vision , there are two separate computations – one for action and one for perception for every sensory system . The system for action , simple and complex , allows rapid online control . The system for perception is relatively slow and exerts control over action over a longer time course and modulates the system for action based on the state of the animal and its goal . Superficially , a modular system like the one we propose appears hardwired . But in reality , by modulating the relation between ORN activation and change in motor parameter one can flexibly couple different motor output to the same sensory input . The implication of our model is that at any instant , based on its assessment of its state and goal , an animal can rapidly transform sensory input into behavioral output . This study describes the sensorimotor transformation for an average hungry fly over a 3 min exposure to odor . An important feature of this transformation is that different ORN classes affect different motor parameters . But , the effect of activating a given ORN class on a motor parameter is likely to change depending on other variables such as the state of that individual and sensory context . Figure 10 shows an example of how the behavioral response changes with continuous exposure to odor . Future studies will determine whether this modulation implies a change in weighting between ORN spikes and a given motor parameter or a complete reorganization between ORN activity and locomotor output . Control flies were either w1118 , laboratory cultures of 200 wild caught Drosophila melanogaster used previously by other labs ( Ofstad et al . , 2011; Bhandawat , 2010 ) or Or42b+/- heterozygotes . Or42bmutant flies ( Or42bEY14886 ) were obtained from Bloomington stock center and backcrossed to w1118 for 10 generations . Flies used for the behavioral assay were raised in ‘sparse culture’ condition as described previously ( Bhandawat , 2010 ) . Briefly , 100~∼150 eggs were collected in a standard cornmeal-agar media bottle and left in the 25o°C incubator with 12 hr:12 hr light:dark cycle until the adult flies eclosed . Newly eclosed flies were transferred to fresh food vials . One day before behavioral experiments , 10∼20 flies were transferred to an empty vial containing a wet paper for starvation . The custom-built behavioral arena ( details in Figure 1—figure supplement 1 ) was connected to an odor-delivery system that was similar to the one described previously ( Bhandawat , 2010 ) . The construction of our behavioral arena is diagrammed in Figure 1—figure supplement 1 . The significant features of the arena are described below . A sharp interface between the odor-zone and no odor-zone is an important feature of the arena . The sharpness of the interface largely resulted from the vacuum being run at a much higher rate than the air flow thereby sucking air radially inwards into the arena . Vacuum pulled 6 liters/min while the flow through the air tube was 1 . 2 liters/min . To allow for a radial inward flow of air , we machined holes into the outer rim . Air flow rates were kept low to minimize anemotactic responses . The airflow at the interface of air tube and the arena is 0 . 07 m/s and the highest speed anywhere in the arena is 0 . 11 m/s . These speeds aretwofold lower than the lowest speeds used to induce anemotaxis in flight ( Budick et al . , 2007 ) . Consistent with these low flow rates , we saw little effect of turning on the air on the fly’s behavior ( Figure 1—figure supplement 3 ) . In constructing the junction between the air tube and the upper plexiglass plate , there were two important considerations . First , the joint had to be as clear as possible to minimize occlusion of the fly’s image on the camera . Second , the edge of the tube had to be flush with the upper plate so that the fly did not feel a significant mechanosensory edge . We also machined a thin Mylar sheet with 0 . 5 mm holes separated by 1 mm . The air/odor entered the arena through this Mylar sheet . The flies did not show a strong preference for exploring the junction between the air tube and the plate or the Mylar sheet . To enhance the visual contrast between a fly and the background , the bottom plate and mesh were sprayed black and the vacuum bottle was covered with black tape . Infrared LED lights were placed around the arena to achieve uniform illumination . As much as practicable , we covered or painted any reflective surface to prevent spurious reflections . The olfactometer we used was same as the one previously described ( Bhandawat , 2010 ) . The background flow was set at a rate of 1 liter/min . The flow through a secondary air stream at 200 ml/min was controlled by a valve which regulated whether the air passed through the odor vial or not ( Figure 1A ) . Ethyl acetate ( CAS # 141-78-6 ) , 2-butanone ( CAS # 78-93-3 ) and ethyl butyrate ( CAS # 105-54-5 ) ( Sigma-Aldrich ) were serially diluted in paraffin oil ( Baker ) . Organic apple cider ( Spectrum Naturals , Filtered ) vinegar was diluted in distilled water . We used EAG to quantify odor concentration in our behavioral arena . To measure EAG , we inserted a glass electrode into the fly’s eye . After inserting the electrode , we used a UV-cured glue ( KOA-300 from Kemxert Corporation ) to glue the electrode to the eye . This electrode served both as a tether and as a ground electrode for sensillum recording . This electrode was mounted on a micromanipulator to position the fly at different places in the arena . Next , another electrode ( a sharp electrode which is used to perform sensillum recording ) was inserted into the fly’s antennae . This second electrode was also mounted on a micromanipulator . This whole ensemble was moved around the arena . For each measurement , the electrode was held stationary at a given location and the odor was turned on . The ensemble must be horizontal and should also travel horizontally . For a given experiment , the electrode needs to be stable for all measurements . This is because the value of EAGs strongly depends on the place and depth of insertion . Each measurement was repeated 4 times a given location . Because absolute EAG values vary widely from experiment-to-experiment , the values were normalized by dividing by the peak response obtained inside the odor-zone . The normalized response in Figure 1D is obtained by integrating over a 15 s window after odor on . This window was chosen because the response reaches a plateau within 15 s . ORNs were recorded using single-sensillum recording as previously described ( Bhandawat et al . , 2007 ) . Briefly , flies were immobilized in the trimmed end of a plastic pipette tip under a 50X air objective mounted on an Olympus BX51WI microscope . A reference electrode filled with saline was inserted into the eye , and a sharp saline-filled glass capillary was inserted into a sensillum . Recordings were obtained with an A-M Systems Model 2400 amplifier , low-pass filtered at 2 kHz and digitized at 10 kHz . ORN spikes were detected using routines in IgorPro ( Wavemetrics ) . Recorded ORNs were matched with ORN class based on: ( 1 ) sensillum morphology and size , ( 2 ) sensillum position on the antenna , ( 3 ) spike amplitude , ( 4 ) spontaneous spike frequency , and ( 5 ) odor tuning of cells in a sensillum . Because all these properties are a stereotyped function of cell lineage , together they form an unambiguous signature of ORN identity ( Bhandawat et al . , 2007; de Bruyne , 1999; de Bruyne et al . , 2001 ) . Spike times were extracted from raw ORN recordings using routines in Igor Pro . Each odor was presented 4-–6 times at intervals of 40–60 s ( a 'block” of trials ) . The response to the first presentation was not included in our analysis . Each of the remaining trials was converted into a peri-stimulus-time histogram ( PSTH ) by counting the number of spikes in 50 ms bins that overlapped by 25 ms . These single-trial PSTHs were averaged together to generate a PSTH describing the response to an odor in a given experiment . Multiple cells corresponding to each ORN class and each cell were tested with a given odor in multiple experiments , each with a different fly . Average PSTHs represent the mean ± s . e . m computed across experiments . To obtain the pattern of ORNs activated by different concentrations of ACV , we performed single-sensillum recording from all basiconic and coeloconic sensilla . Single sensillum recording was performed as previously described ( de Bruyne , 1999; Bhandawat et al . , 2007; Olsen and Bhandawat , 2007 ) . Sensillum type were identified first by morphology and a panel of diagnostic odors . We measured the response of each sensilla to the highest concentration of ACV . ORNs which did not respond to the highest concentration in three different flies were deemed unresponsive to ACV . We repeated this process at each lower concentration of ACV until at ACV4 , only Or42b-ORNs respond to ACV . This was confirmed by showing that the EAG response to ACV4 is abolished in the Or42b mutant . Flies were starved for 18~∼24 hr before the behavioral assay . A single female fly was cold-anesthetized and placed in the behavioral arena . We usually waited for 30 min to give flies time to acclimatize to new environment . Although experiments were performed during the peak circadian activity periods , there were still some inactive flies and we discarded ones that did not walk for at least half of the acclimation period . Individual flies walked either on the floor or the ceiling of the arena without many transitions ( <1 on average ) between the two . As a population , flies had equal probability of being on the floor or the ceiling . Each trial consisted of 3 min of no odor ( before ) , 3 min of odor ( during ) and 3 min of no odor periods ( after ) . Behavior videos were acquired at a rate of 30 frames per second and processed offline to obtain the position of a fly using a custom-written MATLAB script ( available at https://github . com/bhandawat/fly-walking-behavior ) . During the behavioral assay , we waited for at least 3 min between different odor trials to avoid any effects caused by residual odor from the previous trial . We tested multiple odors for a single fly and used odors such as apple cider vinegar to which a fly has a strong behavioral response at the end of the experiment . The air tube made it impossible to image the arena with a single camera because images behind the air tube were severely distorted ( see Figure 1B ) . Instead , we imaged the arena with two cameras . The un-occluded sides of the image from the two cameras were registered and stitched to obtain an un-occluded view of the arena ( see videos ) . The resulting video was used to extract a fly’s position . We subtracted the average frame from each frame to obtain a background-subtracted video in which the fly is the only bright object in an otherwise dark background . The position of the fly was obtained as the centroid of the largest bright object ( usually the only bright object ) in a frame . When the fly was directly under the edge of the air tube , background-subtracted video showed two separate objects instead of one , which caused errors in tracking . To correct these errors , the centroid was calculated as the center of both objects when the fly was near the edge of the air tube . We checked auto tracking by measuring the fly’s instantaneous velocity . Frames on which the instantaneous velocity exceeded 3 mm/s represented frames where the fly’s position had been incorrectly assigned . These errors in tracking were less than 1% and they were manually corrected . Lastly we transformed the track such that it represents the view from a single camera placed directly above the arena . All image processing was performed in MATLAB using the Image Processing Toolbox . Flies enter the odor-zone at variable times after the odor is turned on . Thus , there are multiple reasonable ways to delineate 'before' , 'during' and 'after' period . We chose the time between the start of the trial to the nominal time when the odor was turned on as the ‘before’ period . Therefore , the ‘before odor’ period always had a fixed length of 3 min . The definition of ‘during odor’ period depended on whether the fly was inside the odor-zone when the odor was turned on or outside . If the fly was inside the odor-zone , the ‘during’ period started at 3 min and 5 s because based on the EAG measurements it took 5 s for the odor to reach the arena . If the fly was outside the odor-zone , the time at which it first entered the odor-zone marked the start of the ‘during’ period . Thus the length of ‘during’ period was variable and <3 min long . The ‘after’ period started 6 min after the start of the video and lasted till the end of video acquisition . From the EAG measurement ( Figure 1C , D ) and smoke test ( Figure 1—figure supplement 2 ) , we determined that there was a sharp odor boundary at the physical inner rim ( 1 . 2 cm away from the center of the arena ) and we estimated that there was no odor response at 1 . 5 cm from the center of the arena . Moreover because we used the fly’s centroid ( typically on the thorax or abdomen of a fly ) instead of actual antenna location , the fly could be at 1 . 65 cm ( adding 0 . 15 cm for the half-length of the fly ) facing radially inwards and still be experiencing odor . In the presence of ACV , flies often stopped just before entering the odor-zone . For ACV at 10-2 and 10-1 , these stop events occur within 1 . 65 cm of the arena center . In case of pure ACV , many flies stopped outside 1 . 65 cm , so we assumed that centroid locations at this point was a better measurement of the actual odor boundary . We found this to be at 1 . 9 cm away from the center of the arena and used it as a conservative estimate of the odor boundary . The raw coordinates of the fly were smoothed using a sliding window which was 10 frames long ( 0 . 33 s ) . The window size was chosen empirically by comparing the raw and smoothed tracks . To convert camera pixels into real world distances , we measured the diameter of the arena ( 6 . 4 cm ) in pixel units to obtain a conversion ratio of 1 pixel to 0 . 1 mm . The typical size of a fly in a video was about 30 pixels which corresponded correctly to the actual size of flies we used ( ~∼3 mm ) . Speed was measured from the smoothed coordinates by simply dividing the displacement in consecutive frames by the time between them ( 0 . 033 s ) . Runs and stops were assigned using the Schmidt trigger algorithm already introduced in earlier studies ( Martin , 2004; Robie et al . , 2010 ) . If the instantaneous walking speed of a fly was lower than 0 . 5 mm/sec , then the fly was stationary or at a stop and if the speed was faster than 1 mm/sec , then the fly was moving or on a run . If the speed was in between those two thresholds , it was classified as a continuation of the previous run or stop . Comparison of automated run-stop analysis with manually scored run-stop results showed a 100% agreement except for two systematic biases . First , very short stops lasting 1-–3 frames or less than 100 msec went undetected by our algorithm . Second , the algorithm is 1–3 frames late in recording the start and ends of runs . Angular speed was measured by calculating the curvature of the smoothed tracks during runs . To minimize errors occurring during fly’s slow movement , we disregarded angular speed changes when flies were moving slower than 2 mm/sec . In our experiment , flies have at least two qualitatively different modes of turning: they can either turn smoothly over hundreds of millisecond or sharply over much shorter time . To parse the fly’s turns into these two modes , we divided the turns into 'smooth turns' or 'sharp turns' . A fly was designated as performing a turn when the sum of it angular velocity over 5 frames was >0 . 3 radian . If the sum was greater than 1 . 3 radians , it constituted a sharp turn ( Figure 12 ) . 10 . 7554/eLife . 11092 . 027Figure 12 . Determination of curved walk and sharp turns . ( A ) A part of a fly’s walking trajectory . The curved walk is indicated with orange line and sharp turns are marked with black circles . The right panel shows the magnified walking trajectory from the left ( marked by yellow highlight ) . B ) The angular speed ( grey line ) and sum of 5 frame-long angular speed ( blue line ) for the frames corresponding to the highlighted track above . The orange lines mark the threshold for the curved walk and the grey lines mark the threshold for sharp turns . Frames detected as curved walks ( orange ) and sharp turns ( black circle ) are marked . DOI: http://dx . doi . org/10 . 7554/eLife . 11092 . 027 Behavioral studies with a point source of odor were performed in an arena that was a 10 cm in diameter and 4 mm in height . A hole with 1 . 5 mm diameter was drilled in the middle of the arena to create an odor delivery port . Odors were delivered by attaching an odor vial below the hole . Odors entered the arena passively , i . e . there was no airstream carrying the odor . The chamber was stabilized with magnets which held the chamber firmly in place against a metallic ring stand . Experiments were performed in dark under infrared illumination . A camera was placed 25 cm above the chamber . The chamber was painted black; a fly was the only bright object . Tracking was performed as in the ring assay . Preparation of the fly cultures , starvation and other details are as described for the ring assay in section 5 above . The behavior was performed one fly at a time to prevent distortions due to flies bumping against each other . Flies were tracked for 5 min before odor onset and 5 min during the odor period . 5 min was sufficient time for most flies to venture into the center even before odor onset . This was critical for assessment of statistics . To calculate the radial density , we simply averaged the instantaneous radial position and averaged over all the flies to obtain the radial density as a function of time . To calculate speed as a function of radial distance , we divided the arena into 10 concentric bins and measured the mean speed in each bin . The resulting distribution was normalized for individual flies by dividing by the maximum speed . The mean and standard deviation over flies is reported in Figure 10F . In conventional olfactory behavioral assays such as a T-maze test , results are typically represented as a performance index , which measures the proportion of time that flies spend inside the specifically defined odor zone ( reviewed by Davis 2005 ) . We used a parameter similar to the performance index , ‘attraction index’ , as a measure of the relative change in time a fly spends inside the odor zone between before and during the odor stimulation periods . We used the fly’s response to apple cider vinegar to design an appropriate parameterization of the fly’s odor-evoked behavior . Underlying the fly’s attraction to apple cider vinegar , both its time to return to the odor zone and time spent inside changed . This implies that the change in fly’s behavior is not limited to the time that a fly is actually in the presence of odor ( i . e . in the odor zone ) . We also observed that a fly’s speed increases outside the odor zone in the presence of odor and decreases inside . Thus , the same parameter can change in different ways outside and inside the odor zone . Therefore , we investigated the change in a given parameter both inside and outside the odor zone . In addition , we observed that flies sometimes showed acute behavioral changes on crossing the odor boundary , so we added certain parameters to depict the acute changes ( e . g . speed crossing in ) . Below , we describe each parameter in the 17-dimensional behavioral space . Two considerations influenced our statistical approach . First , the modulation of many parameters depended on the value of that parameter before odor onset . For instance , if one fly repeatedly went inside the odor-zone before odor onset , it would likely do so even when the odor is turned on . To account for these intrinsic differences in a fly’s tendency , we are reporting normalized parameters . Second , modulation of a given motor parameter by an odor can result from either the odor itself or the solvent control . Therefore , to establish whether a given parameter is modulated by a given odor , we will perform statistical tests between the distributions of a given parameter in the solvent control versus its distribution in the presence of the odor . Most parameters did not distribute normally . Wherever a measure of central tendency was necessary we have used medians . Because of non-normality , boxplots are used to represent the data . On each box , the central mark is the median , the edges of the box are the 25th and 75th percentiles , the whiskers extend to the most extreme data points not considered outliers , and outliers are plotted individually . In choosing the 17-dimensional space , we used p <0 . 05 as a criteria for considering a given parameter . To ascertain whether a given parameter is significantly modulated by odor , we corrected for multiple comparisons by applying the Bonferroni correction and used a p < ( 0 . 05/17 ) =0 . 003 . CVA is similar to PCA but explicitly designed to find linear combinations of variables which maximize the difference between groups . It does so by looking for the eigenvalues and vectors of the between-groups covariance matrix 'in the metric of' the within-groups covariance matrix . If B is the within-group covariance matrix and W is the within-groups covariance matrix , then the canonical variates are obtained by solving the equation ( B-1W ) a = 0 or an eigen decomposition of W-1B . Unlike PCA , the number of dimensions does not equal the number of variables . Rather the number of dimensions is one less than the number of groups . For comparison between two groups , all the 'variance' is captured in a single dimension , leading to simple one dimensional representation we have used in Figures 4 , 5 and Supplementary Figure 6 . To show that the separation between flies’ behavior to two odors along the first canonical variate is not just an artifact of over-fitting , we randomized the odor label 50000 times and calculated the median distance between odor representations in each iteration . To assess significance , we measured the fraction of randomized trial with median > original median .	Humans rely chiefly on vision to understand and navigate the world around them . But for many organisms , the world is dominated by their sense of smell . For these animals , everyday activities , like finding food , depend on being able to change behavior based on odor-based cues . To meet the challenges of detecting and discriminating between different odors , animals have many odorant receptors that bind to the odors , which are found on olfactory receptor neurons ( ORNs ) . Each odor activates multiple ORNs , and different odors activate different combinations of ORNs . But it is not clear how activities from different classes of ORN are combined to create the perception of an odor or to guide behavior . Now , Jung et al . have investigated the logic by which odors can alter a fruit fly’s movements . The olfactory system of the fruit fly is organized along similar lines to that of a mammal , but is much simpler . Moreover , many genetic tools are available in fruit flies to allow neuroscientists to activate and inactivate specific neurons and assess the effect this has on behavior . The results suggest that odor-guided movement in fruit flies has two noteworthy features . Firstly , in the presence of odors , flies alter their walking in unexpectedly large number of ways . Therefore , one needs to consider many different factors , or “motor parameters” , to describe how odors affect a fly’s movement . For instance , instead of just walking faster or slower , a fly can change how long it stops ( stop duration ) , how long it runs ( run duration ) and how fast it runs ( run speed ) – all of which will affect overall speed . Secondly , a single class of ORN can strongly affect some parameters ( like run duration ) without affecting others ( like stop duration ) . These data indicate that the neural circuits involved have a modular organization in which each ORN class affects a subset of motor parameters , and each motor parameter is affected by a subset of ORN classes . These findings were largely unexpected . Jung et al . ’s study focused on attractive odors . Future work will study repulsive odors to investigate if similar results are seen when studying repulsion versus attraction .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Odor-identity dependent motor programs underlie behavioral responses to odors
Spermatogenesis , the complex process of male germ cell proliferation , differentiation , and maturation , is the basis of male fertility . In the seminiferous tubules of the testes , spermatozoa are constantly generated from spermatogonial stem cells through a stereotyped sequence of mitotic and meiotic divisions . The basic physiological principles , however , that control both maturation and luminal transport of the still immotile spermatozoa within the seminiferous tubules remain poorly , if at all , defined . Here , we show that coordinated contractions of smooth muscle-like testicular peritubular cells provide the propulsive force for luminal sperm transport toward the rete testis . Using a mouse model for in vivo imaging , we describe and quantify spontaneous tubular contractions and show a causal relationship between peritubular Ca2+ waves and peristaltic transport . Moreover , we identify P2 receptor-dependent purinergic signaling pathways as physiological triggers of tubular contractions both in vitro and in vivo . When challenged with extracellular ATP , transport of luminal content inside the seminiferous tubules displays stage-dependent directionality . We thus suggest that paracrine purinergic signaling coordinates peristaltic recurrent contractions of the mouse seminiferous tubules to propel immotile spermatozoa to the rete testis . Spermatogenesis ranks among the most complex , yet least understood developmental processes in postnatal life . Inside the seminiferous tubules , which represent the functional units of the testis , this intricate course of mass cell proliferation and transformation events generates haploid spermatozoa from diploid spermatogonial stem cells . The seminiferous epithelium is composed of Sertoli cells , each intimately associated with ≥30 germ cells at different developmental stages ( Mruk and Cheng , 2004 ) . Sertoli cells provide the microenvironment critical for spermatogenesis by establishing the blood-testis barrier ( Cheng and Mruk , 2010 ) , forming the spermatogonial stem cell niche ( Oatley and Brinster , 2012 ) , and controlling epithelial cyclicity via auto- , para- , and endocrine feedback ( Heindel and Treinen , 1989 ) . Different types of spermatogonia ( type A , intermediate , type B ) are localized along the seminiferous tubule basement membrane ( Chiarini-Garcia and Russell , 2001 ) . Upon detachment , type B spermatogonia enter meiosis as preleptotene spermatocytes . During meiotic divisions and subsequent maturation steps , germ cells progress from primary to secondary spermatocytes and round to elongated spermatids . Accumulating evidence implicates purinergic signaling in testicular paracrine communication . While the general picture is still incomplete , cell- and stage-specific testicular expression of different purinoceptor isoforms has been reported in Leydig cells ( Antonio et al . , 2009; Foresta et al . , 1996 ) , Sertoli cells ( Veitinger et al . , 2011 ) , both pre- and postmeiotic germ cells ( Fleck et al . , 2016; Glass et al . , 2001 ) , testicular peritubular cells ( TPCs ) ( Walenta et al . , 2018 ) , as well as mature spermatozoa ( Navarro et al . , 2011 ) , albeit with contradictory results . Functionally , several studies have suggested purinergic paracrine control of gonadotropin effects on Leydig and Sertoli cells ( Filippini et al . , 1994; Gelain et al . , 2005; Gelain et al . , 2003; Lalevée et al . , 1999; Meroni et al . , 1998; Poletto Chaves et al . , 2006 ) , including steroidogenesis and testosterone/17β-estradiol secretion ( Foresta et al . , 1996; Rossato et al . , 2001 ) . Members of the P2 purinoceptor family are activated by extracellular ATP ( Burnstock , 1990 ) . P2 receptors subdivide into ionotropic P2X ( Bean , 1992; Bean and Friel , 1990 ) and metabotropic P2Y ( Barnard et al . , 1994 ) receptors , comprising seven ( P2X ) and eight ( P2Y ) isoforms , respectively ( Müller et al . , 2020 ) . All P2X channels display substantial Ca2+ permeability and show distinct pharmacological profiles , ligand affinities , and desensitization kinetics ( Khakh and North , 2012 ) . G-protein-coupled P2Y receptors are sensitive to both ATP and UTP and they form two subgroups that either activate phospholipase C via Gαq/Gα11 ( P2Y1 , 2 , 4 , 6 , and 11 ) or couple to Gαi/o ( P2Y12 , 13 , and 14 ) ( Müller et al . , 2020 ) . Notably , several P2 receptor isoforms affect smooth muscle cell physiology , with P2X1 , P2X2 , P2X4 , P2X7 , P2Y1 , and P2Y2 acting as the principle subunits ( Burnstock , 2014 ) . So far , the most prominent role for a specific subunit in reproductive physiology has been attributed to P2X1 , which is critical for vas deferens smooth muscle contraction and male fertility ( Mulryan et al . , 2000 ) . In mice , stimulation-dependent ATP secretion from both Sertoli and germ cells was reported ( Gelain et al . , 2005; Gelain et al . , 2003 ) and may itself be under endocrine control ( Gelain et al . , 2005; Lalevée et al . , 1999 ) . The mechanism ( s ) of cellular ATP release , however , remain subject to debate . ATP secretion via exocytotic release ( Bodin and Burnstock , 2001; Zhang et al . , 2007 ) has been proposed . Alternative ATP release pathways include connexin/pannexin hemichannels ( Bao et al . , 2004; Cotrina et al . , 1998 ) , transporters ( Lohman et al . , 2012 ) , voltage-gated ( Taruno et al . , 2013 ) or large-conductance anion ( Bell et al . , 2003 ) channels , or even P2X7 receptors ( Pellegatti et al . , 2005; Suadicani et al . , 2006 ) . Along the seminiferous epithelium , spermatogenesis has been conceptualized by attribution of sequential cellular ‘stages’ ( Figure 1A ) , which progress through coordinated cycles ( Hess and De Franca , 2008; Russell , 1990 ) . First initiated in mice about 7 days postpartum ( Kolasa et al . , 2012 ) , each spermatogenic cycle comprises 12 stages and lasts 8 . 7 days ( Hermo et al . , 2010 ) . After approximately 39 days ( 4 . 5 cycle repetitions ) , spermatogenesis completes with the release of immotile spermatozoa from the seminiferous epithelium into the lumen of the tubule ( spermiation ) . Once detached from the Sertoli cells , sperm must be transported to the rete testis and epididymis for final maturation . Precisely regulated tubular transport mechanisms are , thus , imperative for reproduction . While bulk movement of luminal content has been anecdotally reported ( Cross , 1958; Setchell et al . , 1978; Worley et al . , 1985 ) , no quantitative data on sperm transport within the seminiferous tubules is available . Early in vitro observations of apparent minute undulating motions of seminiferous tubule segments ( Roosen-Runge , 1951; Suvanto and Kormano , 1970 ) suggested that smooth muscle-like TPCs ( Clermont , 1958; Ross , 1967 ) could mediate contractile tubule movements . This concept has gained widespread support from several , mostly indirect , in vitro studies ( Ailenberg et al . , 1990; Filippini et al . , 1993; Miyake et al . , 1986; Tripiciano et al . , 1996 ) . However , quantitative direct ( i . e . live cell ) measurements of seminiferous tubule contractions are rare and controversial ( Ellis et al . , 1978; Harris and Nicholson , 1998; Losinno et al . , 2012; Worley and Leendertz , 1988 ) . Moreover , mechanistic in vivo evidence is lacking . Here , we demonstrate that , by acting on ionotropic and metabotropic P2 receptors , extracellular ATP activates TPC contractions that trigger directional sperm movement within the mouse seminiferous tubules both in vitro and in vivo . Accumulating data suggests that purinergic signaling constitutes a critical component of testicular paracrine communication ( Fleck et al . , 2016; Foresta et al . , 1995; Gelain et al . , 2003; Poletto Chaves et al . , 2006; Veitinger et al . , 2011; Walenta et al . , 2018 ) , with Sertoli cells acting as a primary source of ATP secretion ( Gelain et al . , 2005 ) . Therefore , we asked if mouse TPCs are sensitive to extracellular ATP . Primary TPC cultures retain high purity for ≥14 days in vitro ( Figure 1B , and Figure 1—figure supplement 1A&B ) and cells express transcripts for several ionotropic ( P2X2 , P2X4 , P2X7 ) and metabotropic ( P2Y2 , P2Y6 ) purinoceptors ( Figure 1C ) . The specific biophysical and pharmacological profile of ATP-dependent transmembrane currents ( Figure 1D–H ) strongly suggests functional expression of P2X2 and/or P2X4 , but not P2X7 receptors . As reported for both P2X2 and P2X4 ( North , 2002 ) , TPC currents are saturated at ≤100 µM ATP ( Figure 1D&E ) , whereas P2X7 receptors display strongly reduced ATP sensitivity ( Donnelly-Roberts et al . , 2009 ) . Moreover , currents recorded from TPCs showed modest but persistent desensitization ( Figure 1D&F ) , which is similarly observed for recombinant P2X2 and P2X4 , but not P2X7 receptors ( Coddou et al . , 2011 ) . TPCs also displayed reduced BzATP sensitivity ( data not shown ) , which is a potent activator of P2X7 receptors ( Donnelly-Roberts et al . , 2009 ) . Ivermectin ( Figure 1G&H ) , an agent selectively potentiating P2X4 receptor currents ( Khakh et al . , 1999; Silberberg et al . , 2007 ) , increased ATP-induced currents in a subpopulation of TPCs ( n = 7/12 ) , whereas suramin ( Figure 1G&H ) , a drug inhibiting P2X2 , but not P2X4 receptors ( Evans et al . , 1995 ) , inhibited a TPC subset ( n = 10/18 ) . Notably , live-cell ratiometric Ca2+ imaging in cultured TPCs revealed robust and repetitive cytosolic Ca2+ transients upon ATP exposure ( Figure 1I&J ) . We next reduced the extracellular Ca2+ concentration ( [Ca2+]e ) to 100 nM , a concentration approximately equimolar to cytosolic levels , by adding an appropriate chelator/ion concentration ratio ( 1 mM EGTA/0 . 5 mM CaCl2 ) . This treatment , which drastically diminishes the driving force for Ca2+ influx , did substantially reduce , but not abolish ATP response amplitudes ( Figure 1K&L ) . The selective P2Y receptor agonist UTP ( Alexander et al . , 2019b ) also triggered Ca2+ signals ( data not shown ) , indicating a role for G protein-dependent Ca2+ release from internal storage organelles ( Müller et al . , 2020 ) . Notably , ~46% of all ATP-sensitive TPCs additionally displayed a delayed , but long-lasting inward current that gradually developed over tens of seconds after ATP stimulation ended ( Figure 1M&N ) . We hypothesized that this slower current could result from P2Y receptor-/G protein-dependent Ca2+ release , likely mediated by the P2Y2 isoform since P2Y6 receptors lack substantial ATP sensitivity ( Alexander et al . , 2019a; Jacobson et al . , 2015 ) . Indeed , occurrence of the delayed current depends on presence of intracellular GTP ( Figure 1O ) . Moreover , selective recruitment of G protein-coupled P2Y receptors with UTP ( Figure 1P ) exclusively triggered such slowly developing currents . Largely carried by Cl– ( Figure 1—figure supplement 2 ) , this current likely results from P2Y receptor-mediated phosphoinositide turnover , Ca2+ release , and activation of Ca2+-gated Cl– channels . Together , these data suggest that mouse TPCs functionally express both ionotropic and metabotropic purinoceptors . Next , we asked if TPCs also exhibit ATP sensitivity in their physiological setting . Therefore , we examined purinergic Ca2+ signals from mouse TPCs in acute seminiferous tubule sections ( Fleck et al . , 2016 ) . In parallel approaches , we employed two different fluorescent Ca2+ reporters , a synthetic ratiometric Ca2+ sensor ( fura-2 ) as well as a genetically encoded Ca2+ indicator ( GCaMP6f ) . The dual excitation ratiometric indicator fura-2 allows semi-quantitative Ca2+ measurements ( Bootman et al . , 2013 ) , but lacks cell type specificity as tubules are bulk-loaded with a membrane-permeable acetoxymethyl ester conjugate . By contrast , conditional gene targeting via the Cre/Lox system ( Smith , 2011 ) allows TPC-specific expression of the single-wavelength indicator GCaMP6f . First , we confirmed inducible TPC-targeted testicular expression of fluorescent reporter proteins in SMMHC-CreERT2 x Ai14D mice ( Figure 2A–C , Video 1 ) . Tamoxifen-induced transgenic expression of CreERT2 under control of the mouse smooth muscle myosin , heavy polypeptide 11 ( a . k . a . SMMHC ) promoter drives Cre-mediated recombination of loxP-flanked reporters ( tdTomato ( Ai14D mice ) or GCaMP6f ( Ai95D ) ) in smooth muscle cells and TPCs ( Wirth et al . , 2008 ) . Second , TPC-specific GCaMP6f expression in SMMHC-CreERT2 x Ai95D mice revealed robust Ca2+ transients in cells of the tubular wall upon ATP exposure ( Figure 2D&E ) . Third , fura-2/AM loading of acute seminiferous tubule sections preferentially labeled the outermost cell layer ( Figure 2F ) , allowing semi-quantitative in situ imaging of ATP-dependent Ca2+ signals in mouse TPCs ( Figure 2G&H ) . So far , our results thus demonstrate that challenging TPCs with extracellular ATP triggers robust Ca2+ signals both in vitro and in situ . We hypothesized that ATP-induced Ca2+ signals in TPCs could mediate contractile motion of the seminiferous tubule . To address this , we established a fast , quasi-simultaneous image acquisition method that enables parallel recording of both peritubular Ca2+ responses and seminiferous tubule movement ( methods ) . Brief ATP exposure resulted in a peripheral band of Ca2+ activity at the edge of the tubule . Such signals usually coincided with a pronounced contractile motion of the seminiferous tubule ( Figure 3A , Video 2 ) . When movement is quantified as the time-lapse image flow field strength ( methods ) tubular contraction follows the Ca2+ signal onset with minimal delay , outlasts the Ca2+ signal peak , and recovers slowly ( Figure 3B ) . Both Ca2+ responses and tubular movement are dose-dependent and share an ATP threshold concentration of approximately 1 µM ( Figure 3C–E , Video 3 ) . Contractile smooth muscle plasticity ( Tuna et al . , 2012 ) likely underlies the apparent difference in signal saturation ( Figure 3D ) . Notably , in some tubules , we observed spontaneous low-amplitude ‘vibratory’ movements and local indentations ( Figure 4A ) , reminiscent of the relatively high frequency rippling previously described ( Ellis et al . , 1981; Worley et al . , 1985 ) . We next investigated the Ca2+ signaling mechanism ( s ) underlying ATP-dependent TPC contractions . First , we asked whether influx of external Ca2+ is involved in TPC force generation . Similar to in vitro observations ( Figure 1K&L ) , diminishing or even reversing the driving force for transmembrane Ca2+ flux by reducing [Ca2+]e to 100 nM or 12 nM , respectively , for variable durations , significantly decreased both TPC Ca2+ signals and tubular contractions ( Figure 4B–H ) . While , upon [Ca2+]e reduction , ATP-dependent responses ( both Ca2+ signals and contractions ) were still detected in the vast majority of cells/experiments ( Figure 4E ) , response strength was strongly diminished ( Figure 4F ) . These effects were independent of both the extent ( 12 nM or 100 nM ) and the duration ( 1–10 min ) of [Ca2+]e reduction and were fully reversible ( Figure 4H ) . Second , we examined a potential role of ATP-induced Ca2+ release from internal storage organelles . Ca2+ depletion of the sarcoplasmic reticulum via pharmacological inhibition of the sarco/endoplasmic reticulum Ca2+-ATPase by cyclopiazonic-acid ( CPA ) essentially abolished both ATP-dependent Ca2+ signals and contractions ( Figure 4D ) , with very few cells/tubules retaining some residual ATP sensitivity during CPA treatment ( Figure 4G ) . Importantly , all results from ratiometric fura-2 imaging were qualitatively indistinguishable from those obtained with genetically targeted GCaMP6f ( Figure 4E–G ) , showing that both approaches to TPC Ca2+ measurement provide comparable results . Third , given the pronounced effect of pharmacological store depletion , we aimed to quantify the specific contribution of metabotropic purinoceptors to the overall ATP-mediated effect . The P2Y receptor-selective agonist UTP ( Coddou et al . , 2011 ) evoked both TPC Ca2+ signals and tubular contractions ( Figure 4E&F ) . However , under control [Ca2+]e conditions , UTP-evoked responses were substantially reduced compared to control ATP stimulations ( Figure 4F ) . Notably , these UTP responses were statistically indistinguishable from the diminished ATP-dependent signals we observed under low [Ca2+]e conditions ( Figure 4F ) . Together , these data strongly suggest that ( i ) extracellular ATP acts as a potent TPC stimulus that triggers seminiferous tubule contractions in situ , that ( ii ) P2X and P2Y receptors act in concert to mediate TPC responses to ATP exposure , that ( iii ) , while P2X receptor-dependent external Ca2+ influx apparently boosts responses to ATP , P2Y receptor-mediated Ca2+ mobilization from the sarcoplasmic reticulum is necessary to evoke TPC responses , and consequently – since store depletion essentially abolishes ATP-dependent signals – that ( iv ) influx of external Ca2+ via ionotropic P2X receptors is not sufficient to drive TPC signals and evoke contractions . Notably , our general finding of ATP-induced mouse TPC contractions is likely transferable to human peritubular cells . When primary human TPC cultures ( Walenta et al . , 2018 ) were exposed to extracellular ATP , morphological changes were observed within seconds-to-minutes ( Figure 4—figure supplement 1A&B ) . Moreover , embedding cells in collagen gel lattices revealed considerable contractile force in response to ATP ( Figure 4—figure supplement 1C&D ) . We hypothesized that ATP-induced tubular contractions could impact the transport of luminal fluid and spermatozoa . To test this , we custom-built a whole-mount macroscopic imaging platform , designed to allow both widefield and fluorescence time-lapse imaging of intact seminiferous tubules ( Figure 5A ) . In addition , this setup enables visual categorization of the spermatogenic cycle into three distinct stage groups following published protocols ( Hess and De Franca , 2008 ) and allows precisely timed focal perfusion ( methods ) . First , we asked if brief focal purinergic stimulation triggers seminiferous tubule contractions and , consequently , luminal content movement . Flow field change analysis reveals some basal luminal motion independent of mechanical stimulation ( Figure 5B ) . However , ATP exposure triggered a strong increase in luminal flow that outlasted the presence of ATP for several tens of seconds ( Figure 5B , Video 4 ) . Second , we analyzed if luminal movement depends on the tubule’s cycle stage and , consequently , luminal sperm count . When we analyzed ATP-induced movement in directly stimulated luminal regions ( each designated as region-of-interest ( ROI ) 0 ) and compared stage groups with a high ( group II ) vs . a relatively low ( groups I and III ) amount of luminal sperm , we observed no difference in stimulation-dependent motion ( Figure 5CI ) . Thus , direct ATP exposure triggers tubular contractions independent of cycle stage and luminal sperm count . Third , we investigated if luminal movement is restricted to the area of stimulation or , by contrast , if fluid flow propagates beyond the directly stimulated tubule section . When we analyzed luminal motion in equidistant tubule sections adjacent to the directly stimulated area ROI 0 ( Figure 5A ) , we found a significant , though relatively small bidirectional wave of propagating movement in stage groups I and III , which exhibit a low luminal sperm count ( Figure 5CII ) . Strikingly , we observed strong unidirectional luminal movement upon ATP stimulation of stage group II tubule sections which show high luminal sperm density associated with spermiation ( Figure 5CII ) . In this stage group , luminal content is predominantly propelled toward areas of ascending spermatogenic cycle stages . These findings demonstrate directionality of sperm transport upon focal purinergic TPC stimulation in isolated seminiferous tubules . As expected , ATP-induced tubule contractions also manifest as Ca2+ signals in TPCs ( Figure 6A&B , Video 5 ) . However , these Ca2+ elevations appear to be limited to those areas directly exposed to ATP ( ROI 0 ) . We observed no such signals in adjacent tubule sections independent of the stimulated stage group or an ascending or descending stage direction ( Figure 6C&D ) . This finding indicates that , in the isolated seminiferous tubule , ATP acts as a local messenger that , by itself , is not sufficient to trigger a signal that propagates in a regenerative wave-like fashion along a tubule’s longitudinal axis . However , local contractions generate sufficient force to move luminal content beyond the directly stimulated area and , in turn , directionality of flow along short-to-medium distances ( ≤600 µm; Figure 5CII ) is not critically dependent on peristaltic contractility . To ultimately attribute a physiological role to ATP-dependent Ca2+ signals in TPCs , tubular contractions , and corresponding transport of luminal content , these phenomena must ( i ) occur spontaneously in living animals , and must ( ii ) be triggered experimentally by ATP exposure in vivo . Thus , to investigate any in vivo relevance of our findings , we designed a custom-built 3D printed in vivo imaging stage ( Figure 7—figure supplement 1 ) that allows both widefield epi-fluorescence and multiphoton microscopy of the mouse testis . Initially , we monitored spontaneous seminiferous tubule activity in SMMHC-CreERT2 x Ai95D mice . Multiphoton time-lapse imaging revealed spontaneous TPC Ca2+ signals that typically accompanied strong tubule contractions ( Figure 7A&B , Video 6 ) . Several characteristics emerged from quantitative analysis of these observations . First , during sufficiently long recording periods ( ≤30 min ) , contractions occur in essentially all seminiferous tubules ( Figure 7—figure supplement 2A ) . Second , contractions of individual tubules within the 2D confocal plane are not synchronized ( Figure 7B ) . Third , periods of enhanced activity ( ≥2 contractions within 90 s ) are interrupted by long episodes of quiescence ( Figure 7B , Figure 7—figure supplement 2B ) . Fourth , the durations of TPC Ca2+ signals and corresponding contractions are positively correlated ( Figure 7—figure supplement 2C ) , confirming a causal relationship . Next , we asked whether spontaneous in vivo contractions are coordinated along the longitudinal tubular axis . Low magnification incident light microscopy enabled simultaneous observation of several superficial seminiferous tubule segments ( Figure 7C ) . Movement analysis along the length of digitally straightened tubules demonstrates wave-like unidirectional motions that propagate with high velocities ( Figure 7C&D ) . These movements coincide with ‘macroscopic’ Ca2+ waves that travel at comparable speed and direction ( Figure 7—figure supplement 2D ) . Notably , the observed coordinated contractile movements provide sufficient force to ensure luminal sperm transport ( Video 7 ) . Finally , we examined if brief focal ATP stimulation also triggers peritubular Ca2+ signals and seminiferous tubule contractions in vivo . Therefore , we filled low resistance patch pipettes with fluorescently labeled ATP solution , penetrated the tunica albuginea , and targeted the interstitial space close to neighboring tubules ( Figure 7E , Video 8 ) . Nanoliter puffs of ATP-containing test solution induced both Ca2+ transients in genetically labeled TPCs and strong tubule contractions in the majority of experiments ( Figure 7F&G ) . By contrast , puffs of extracellular saline rarely stimulated any such response ( Figure 7—figure supplement 2E ) . Taken together , in vivo recordings demonstrate that robust recurrent seminiferous tubule contractions ( i ) occur spontaneously , ( ii ) are driven by cytosolic Ca2+ elevations in TPCs that propagate in a wave-like fashion , and ( iii ) can be triggered experimentally by ATP exposure . Consequently , paracrine purinergic signaling in the mouse testis is a mediator of luminal sperm transport within the seminiferous tubule network . The molecular and cellular mechanisms that control paracrine testicular communication have to a large extent remained controversial , if not elusive ( Schlatt and Ehmcke , 2014 ) . For TPCs in particular , a contractile function under paracrine control and , consequently , a critical role in male infertility have long been proposed ( Albrecht et al . , 2006; Romano et al . , 2005 ) , but direct experimental evidence has been lacking ( Mayerhofer , 2013 ) . While several signaling molecules , including vasopressin ( Pickering et al . , 1989 ) , oxytocin ( Worley et al . , 1985 ) , prostaglandins ( Hargrove et al . , 1975 ) , endothelin ( Filippini et al . , 1993 ) , and others ( Albrecht et al . , 2006; Mayerhofer , 2013 ) , have been proposed to act on TPCs , a role of ATP in seminiferous tubule contractility has been explicitly ruled out early on ( Hovatta , 1972 ) . By contrast , our data reveal ATP is a strong stimulus that activates TPCs via both P2X and P2Y receptors , mediating coordinated tubule contractions and luminal sperm transport in situ and in vivo . Both spontaneous and ATP-dependent contractions trigger fast , stage-dependent , and directional transport of luminal content . It is thus tempting to speculate that seminiferous tubule contractility in general , and purinergic TPC signaling in particular , are promising targets for male infertility treatment and/or contraceptive development . The site ( s ) /cellular origin of testicular ATP release as well as the mechanism ( s ) that trigger ATP secretion in vivo currently remain elusive . The apparent absence of efferent nerve endings in the seminiferous tubules and interstitial tissue ( Tripiciano et al . , 1996 ) suggests that tubule contractility is under endo-/paracrine control . By contrast , autonomic innervation of the testicular capsule mediates smooth muscle cell contraction of the tunica albuginea , using ATP as a ( co ) transmitter ( Banks et al . , 2006 ) . Regulated ATP release has been reported for both Sertoli and germ cells ( Gelain et al . , 2005; Gelain et al . , 2003 ) . Moreover , TPCs express P2Y6 receptors ( Figure 1C ) , which were reported to mediate ATP release upon activation ( Carneiro et al . , 2014 ) . Thus , TPCs could themselves participate in regenerative nucleotide release . During spermatogenesis , apoptosis is a vital process ( Print and Loveland , 2000 ) . In fact , up to 75% of germ cells undergo apoptosis under physiological conditions ( Huckins , 1978 ) . This substantial germ cell loss is called ‘density-dependent regulation’ ( Hess and De Franca , 2008 ) . Since ATP release from apoptotic cells is well documented ( Elliott et al . , 2009 ) it is likely that cell density-dependent waves of apoptosis could regularly generate local ATP surges . We have previously shown that one result of seminiferous extracellular ATP elevation is signal amplification by increased ATP release ( Fleck et al . , 2016 ) , although the mechanistic basis of this positive feedback pathway is yet unknown . Given ( i ) the robust cytosolic Ca2+ transients observed in response to ATP exposure in various testicular cell types ( Fleck et al . , 2016; Gelain et al . , 2005; Liévano et al . , 1996; Veitinger et al . , 2011; Walenta et al . , 2018 ) and ( ii ) the usually millimolar ATP content in secretory vesicles ( Bodin and Burnstock , 2001; Zhang et al . , 2007 ) , the most parsimonious explanation for ATP-induced ATP release would be an inevitable ATP ‘co-secretion’ upon any Ca2+-dependent exocytosis event . In addition , ATP release has been observed in several cell types as a result of mechanical deformation , shear stress , stretch , or osmotic swelling ( Button et al . , 2013 ) adding another putative mechanism of regenerative signaling in purinergic contraction control . Notably , extracellular ATP is rapidly degraded by ecto-nucleotidases ( Zimmermann et al . , 2012 ) , rendering its interstitial half-life relatively short and , thus , narrowing its paracrine radius to a few hundred micrometers ( Fitz , 2007 ) . Combined with its fast diffusion – approximately 1 μm in less than 10 ms ( Khakh , 2001 ) – extracellular ATP bears all characteristics of a fast paracrine agent in testicular communication ( Praetorius and Leipziger , 2009 ) . Excitation–contraction coupling in TPCs is poorly described . Our results strongly suggest that a combination of P2X ( isoforms 2 and/or 4 ) receptor-dependent Ca2+ influx and P2Y ( likely isoform 2 ) receptor-activated phospholipase Cβ-dependent Ca2+ release from the sarcoplasmic reticulum – the latter being critical and resembling the recently reported mechanism of vascular smooth muscle cell contraction in small pulmonary veins ( Henriquez et al . , 2018 ) – provides the [Ca2+]c elevation required for force generation ( Berridge , 2008 ) . P2X and P2Y receptors act on different time scales and display different ligand sensitivity , with EC50 values in the nanomolar ( P2Y ) vs . micromolar ( P2X ) range ( North , 2002 ) . It is possible that the co-activation of an ionotropic ( P2X ) and a metabotropic ( P2Y ) signaling pathway serves functions analogous to the concomitant exposure to both ATP and noradrenaline in mesenteric artery smooth muscle . Here , activation of P2X1 receptors generates a small initial contraction that is followed by larger noradrenaline-induced contraction ( Lamont et al . , 2006; Lamont et al . , 2003 ) . Regarding TPC [Ca2+]c elevation , our data suggest that P2X receptor activation also targets Ca2+ release from internal stores , as their depletion inhibits excitation–contraction coupling entirely . Therefore , it is likely that P2X receptors act as signal boosters that mediate Ca2+-induced Ca2+ release , possibly via activation of ryanodine receptors ( Berridge , 2008 ) . This way , the combined action of P2X and P2Y receptors might equip TPCs with a broader ‘two-step’ stimulus integration range . Whole-mount imaging of isolated seminiferous tubules reveals propagation of luminal content that extends beyond the confines of the stimulated/contracted area and displays stage-dependent directionality . While peristaltic contractions are driven by propagating wave-like Ca2+ signals in vivo , focal ATP stimulation appears insufficient to trigger a regenerative Ca2+ wave in isolated tubules . We , thus , conclude that the observed directionality results from other , likely structural characteristics , for example anatomical features of stage group II and III tubules that favor a specific flow direction ( increased tubule diameter and reduced luminal resistance along the stage II-to-III transition zone ) . We cannot , however , rule out that the use of large field-of-view/low numerical aperture objectives for ‘macroscopic’ imaging simply prevents the detection of low-amplitude Ca2+ signal spread . Translation of our findings from the mouse model to humans awaits further in-depth investigation . We have recently reported that ATP activates Ca2+ signals in human TPCs in vitro ( Walenta et al . , 2018 ) . Moreover , our present findings reveal ATP-induced contractions in cultured human TPCs . There are , however , notable differences between the human and the mouse tubular walls . While a single layer of TPCs surrounds the mouse seminiferous tubules , the human tubular wall architecture is more complex , containing several TPC layers , substantial amounts of extracellular matrix proteins , and immune cells ( Mayerhofer , 2013 ) . Impaired spermatogenesis in sub-/infertile men typically coincides with tubular wall remodeling and a partial loss of TPC contractility proteins has been reported in infertile men ( Welter et al . , 2013 ) . Accordingly , interference with TPC contractility had been proposed as a promising strategy for human male contraception ( Romano et al . , 2005 ) . However , a causal relationship between contractility ( or the lack thereof ) and male ( in ) fertility has never been established . In fact , seminiferous tubule contractions had , so far , never been observed in vivo and most in vitro reports were based on indirect and non-quantitative evidence , for example from post-hoc fluorescence or scanning electron microscopy ( Barone et al . , 2002; Fernández et al . , 2008; Losinno et al . , 2016; Losinno et al . , 2012; Tripiciano et al . , 1999; Tripiciano et al . , 1997; Tripiciano et al . , 1996 ) , morphometry of single cells in culture ( Rossi et al . , 2002; Santiemma et al . , 2001; Santiemma et al . , 1996; Tripiciano et al . , 1996 ) , or intraluminal pressure analysis ( Miyake et al . , 1986; Yamamoto et al . , 1989 ) . The fact that expression of TPC contractility proteins initiates with puberty under androgen control and that selective androgen receptor knock-out in TPCs renders mice infertile ( Welsh et al . , 2009 ) underscores a potential role of TPC contractions in male fertility . Accordingly , pharmacological targeting of purinergic signaling pathways to ( re ) gain control of TPC contractility represents an attractive approach for male infertility treatment or contraceptive development . Among several remaining questions , future experimental efforts will have to address ( i ) whether TPCs are coupled by gap junctions to display coordinated activity; ( ii ) whether and , if so , how the final ATP metabolite adenosine affects seminiferous tubule physiology; ( iii ) whether Rho/Rho kinase signaling pathways modulate TPC contractility as frequently observed in other smooth muscle cells ( Somlyo and Somlyo , 2003 ) ; ( iv ) what , if any , role is played by P2X receptor-dependent changes in membrane potential; ( v ) which function is served by the sustained Ca2+-gated Cl– current ( Figure 1—figure supplement 2 ) ; ( vi ) why periods of enhanced contractile activity are interrupted by longer quiescent episodes ( Figure 7—figure supplement 2B ) ; ( vii ) which additional or complementary roles in TPC physiology are played by previously proposed activators , including vasopressin , oxytocin , prostaglandins , and endothelin ( Albrecht et al . , 2006; Mayerhofer , 2013 ) ; ( viii ) whether an additional cytosolic and/or membrane Ca2+ oscillator ( Berridge , 2008 ) provides an endogenous pacemaker mechanism that acts independent of purinergic stimulation; and ( ix ) whether , similar to vascular smooth muscle cells , some specific tone is maintained between contractions by spatial averaging of asynchronous oscillations ( Berridge , 2008 ) , a mechanism that could explain the occurrence of spontaneous low-amplitude ‘vibratory’ movements and local indentations that we ( Figure 4A ) and others ( Ellis et al . , 1981; Worley et al . , 1985 ) have observed . All animal procedures were approved by local authorities and in compliance with both European Union legislation ( Directive 2010/63/EU ) and recommendations by the Federation of European Laboratory Animal Science Associations ( FELASA ) . When possible , mice were housed in littermate groups of both sexes ( room temperature ( RT ) ; 12:12 hr light-dark cycle; food and water available ad libitum ) . If not stated otherwise , experiments used adult ( >12 weeks ) males . Mice were killed by CO2 asphyxiation and decapitation using sharp surgical scissors . We used C57BL/6J mice ( Charles River Laboratories , Sulzfeld , Germany ) as well as offspring from crossing either SMMHC-CreERT2 ( JAX #019079 ) ( Wirth et al . , 2008 ) or 129S . FVB-Tg ( Amh-cre ) 8815Reb/J ( JAX #007915 ) ( Holdcraft and Braun , 2004 ) mice with either Ai95D ( JAX #028865 ) ( Madisen et al . , 2015 ) or Ai14D ( JAX #007914 ) ( Madisen et al . , 2010 ) mice , respectively . The following solutions were used: ( S1 ) 4- ( 2-Hydroxyethyl ) piperazine-1-ethanesulfonic acid ( HEPES ) buffered extracellular solution containing ( in mM ) 145 NaCl , 5 KCl , 1 CaCl2 , 0 . 5 MgCl2 , 10 HEPES; pH = 7 . 3 ( adjusted with NaOH ) ; osmolarity = 300 mOsm ( adjusted with glucose ) . ( S2 ) Oxygenated ( 95% O2 , 5% CO2 ) extracellular solution containing ( in mM ) 120 NaCl , 25 NaHCO3 , 5 KCl , 1 CaCl2 , 0 . 5 MgCl2 , 5 N , N-bis ( 2-hydroxyethyl ) −2-aminoethanesulfonic acid ( BES ) ; pH = 7 . 3; 300 mOsm ( glucose ) . ( S3 ) Extracellular low Ca2+ solution containing ( in mM ) 145 NaCl , 5 KCl , 0 . 5 MgCl2 , 10 HEPES; pH = 7 . 3 ( NaOH ) ; osmolarity = 300 mOsm ( glucose ) ; [Ca2+]free = ~110 nM ( 1 mM EGTA , 0 . 5 mM CaCl2 ) or ~12 nM ( 1 mM EGTA , 0 . 1 mM CaCl2 ) . ( S4 ) Oxygenated ( 95% O2 , 5% CO2 ) extracellular solution containing ( in mM ) 120 NaCl , 25 NaHCO3 , 5 KCl , 0 . 5 MgCl2 , 5 BES; pH = 7 . 3; 300 mOsm ( glucose ) ; [Ca2+]free = ~110 nM ( 1 mM EGTA , 0 . 5 mM CaCl2 ) or ~12 nM ( 1 mM EGTA , 0 . 1 mM CaCl2 ) . ( S5 ) Gluconate-based extracellular solution containing ( in mM ) 122 . 4 Na gluconate , 22 . 6 NaCl , 5 KCl , 1 CaCl2 , 0 . 5 MgCl2 , 10 HEPES; pH = 7 . 3 ( adjusted with NaOH ) ; osmolarity = 300 mOsm ( glucose ) . ( S6 ) Standard pipette solution containing ( in mM ) 143 KCl , 2 KOH , 1 EGTA , 0 . 3 CaCl2 , 10 HEPES ( [Ca2+]free = ~110 nM ) ; pH = 7 . 1 ( adjusted with KOH ) ; osmolarity = 290 mOsm ( glucose ) . ( S7 ) Gluconate-based pipette solution containing ( in mM ) 110 Cs gluconate , 30 CsCl , 2 CsOH , 1 EGTA , 0 . 3 CaCl2 , 10 HEPES ( [Ca2+]free = ~110 nM ) ; pH = 7 . 1 ( adjusted with CsOH ) ; osmolarity = 290 mOsm ( glucose ) . In some experiments Na-GTP ( 0 . 5 mM ) was added to the pipette solution . Free Ca2+ concentrations were calculated using WEBMAXCLITE v1 . 15 ( RRID:SCR_000459 ) . If not stated otherwise , chemicals were purchased from Sigma ( Schnelldorf , Germany ) . Cyclopiazonic-acid ( CPA ) and 2' ( 3' ) -O- ( 4-Benzoylbenzoyl ) adenosine-5'-triphosphate ( BzATP ) triethylammonium salt was purchased from Tocris Bioscience ( Bristol , UK ) . Fura-2/AM was purchased from Thermo Fisher Scientific ( Waltham , MA ) . Final solvent concentrations were ≤0 . 1% . When high ATP concentrations ( ≥1 mM ) were used , pH was readjusted . For focal stimulation , solutions and agents were applied from air pressure-driven reservoirs via an 8-in-1 multi-barrel ‘perfusion pencil’ ( AutoMate Scientific; Berkeley , CA ) . Changes in focal superfusion ( Veitinger et al . , 2011 ) were software-controlled and , if required , synchronized with data acquisition by TTL input to 12V DC solenoid valves using a TIB 14S digital output trigger interface ( HEKA Elektronik , Lambrecht/Pfalz , Germany ) . For focal stimulation during in vivo recordings , ATP was puffed from pulled glass pipettes using a microinjection dispense system ( Picospritzer III; Parker Hannifin , Hollis , NH ) . Low [Ca2+]e solutions ( S3 and S4 ) were applied via both the bath and perfusion pencil . To ensure depletion of Ca2+ stores by CPA we monitored intracellular Ca2+ levels during drug treatment ( 0 . 05 Hz frame rate ) . Transient CPA-dependent Ca2+ elevations lasted 10–40 min . After baseline Ca2+ levels were restored , cells/slices were again challenged with ATP . Control recordings , omitting CPA , were performed under the same conditions . Acute seminiferous tubule slices were prepared as previously described ( Fleck et al . , 2016 ) with minor modifications . Briefly , seminiferous tubules from young adults were isolated after tunica albuginea removal , embedded in 4% low-gelling temperature agarose ( VWR , Erlangen , Germany ) , and 250 µm slices were cut with a VT1000S vibratome ( RRID:SCR_016495; Leica Biosystems , Nussloch , Germany ) . Acute slices were stored in a submerged , oxygenated storage container ( S2; RT ) . When using testicular tissue from Ai95D mice , slices were protected from light during storage to avoid GCaMP6f bleaching . After mouse testis isolation and removal of the tunica albuginea , the seminiferous tubules were placed in Dulbecco's Modified Eagle Medium/Nutrient Mixture F-12 ( DMEM/F-12; Invitrogen ) containing 1 mg ml−1 collagenase A and 6 µg ml−1 DNase ( 10 min; 34°C; shaking water bath ( 60 cycles min−1 ) ) . Three times , the samples were washed ( DMEM/F-12; 5 ml ) , allowed to settle for 5 min , and the supernatant was discarded . Next , tubules were incubated DMEM/F-12 containing 1 mg ml−1 trypsin and 20 µg ml−1 DNase ( 20 min; 34°C; shaking water bath ( 60 cycles min−1 ) ) . Digestion was stopped by addition of 100 µg ml−1 soybean trypsin inhibitor ( SBTI ) and 20 µg ml−1 DNase in phosphate-buffered saline ( D-PBS ) . Then , samples were allowed to settle for 5 min and the supernatant was collected . After two more cycles of washing ( DMEM/F-12 ) , settling ( 5 min ) , and supernatant collection , the collected cell suspension was centrifuged ( 10 min; 400 g ) and the supernatant discarded . The pellet was resuspended in DMEM containing FBS ( 10% ) and penicillin G/streptomycin ( 1% ) , filtered ( cell strainer ( 100 µm ) ) , and cells were plated in 75 cm2 cell culture flask ( T75; Invitrogen ) and placed in a humidified incubator ( 37°C; 5% CO2 ) . Approximately 1/3 of medium volume was replaced every 3 days . Cells usually reached 100% confluence after 7 days in vitro ( DIV ) . Then , cells were washed twice ( DPBS-/-; 5 min; 37°C ) and incubated in 0 . 05% trypsin/EDTA ( 5 min; 37°C ) . Detachment of cells was checked visually and , if necessary , facilitated mechanically . The cell suspension was centrifuged ( 3 min; 800 g ) and the supernatant discarded . The pellet was resuspended in DMEM at cell densities of ~105 cells ml−1 and plated again either in culture flasks or on glass coverslips in 35 mm dishes for experimental use . Again , 1/3 of medium volume was replaced every 3 days . Experiments were performed for ≤5 days after passage . Human TPCs were isolated from small testicular tissue fragments derived from consenting donors with obstructive azoospermia and normal spermatogenesis as described ( Albrecht et al . , 2006; Walenta et al . , 2018 ) . The study was approved by the local ethical committee ( Ethikkommission , School of Medicine , TU Munich , project 169/18S ) . Total RNA was isolated and purified from cultured mouse TPCs ( passage 1 ) with Trizol followed by complementary DNA synthesis with RevertAid H Minus kit ( #K1632 Thermo Fisher ) according to the manufacturer’s instructions . Controls in which the reverse transcriptase was omitted were routinely performed . PCR amplification was performed during 30 thermal cycles ( 95°C , 20 s; 58°C , 20 s; 72°C , 20 s ) . The following specific primer pairs were used for PCR amplification: TargetForward primer 5´−3´Reverse primer 5´−3´P2X1CCGAAGCCTTGCTGAGAAGGTTTGCAGTGCCGTACATP2X2GACCTCCATCGGGGTGGGCTTGGGGTCCGTGGATGTGGAGTP2X3CTGCCTAACCTCACCGACAAGAATACCCAGAACGCCACCCP2X4CCCTTTGCCTGCCCAGATATCCGTACGCCTTGGTGAGTGTP2X5GCTGCCTCCCACTGCAACCC AAGCCCCAGCACCCATGAGCP2X6CCCAGAGCATCCTTCTGTTCCGGCACCAGCTCCAGATCTCAP2X7CCCAGATGGACTTCTCCGACGGACTTAGGGGCCACCTCTTP2Y1CGACAGGGTTTATGCCACTTTCGTGTCTCCATTCTGCTTGP2Y2CGTGCTCTACTTCGTCACCAGACCTCCTGTGGTCCCATAAP2Y4ACTGGCTTCTGCAAGTTCGTAGGCAGCCAGCTACTACCAAP2Y6CATTAGCTTCCAGCGCTACCGCTCAGGTCGTAGCACACAGP2Y12CATTGCTGTACACCGTCCTGAACTTGGCACACCAAGGTTCGAPDHCAAGGTCATCCATGACAACTTTGGTCCACCACCCTGTTGCTGTAG For immunochemistry of testicular cryosections , testes were fixed with 4% ( w/v ) paraformaldehyde ( PFA ) in PBS-/- ( 10 mM , pH 7 . 4; ≥12 hr; 4°C ) and subsequently cryoprotected in PBS-/- containing 30% sucrose ( ≥24 hr; 4°C ) . Samples were then embedded in Tissue Freezing Medium ( Leica Biosystems ) , sectioned at 20 µm on a Leica CM1950 cryostat ( RRID:SCR_018061; Leica Biosystems ) , and mounted on Superfrost Plus slides ( Menzel , Braunschweig , Germany ) . For immunostaining of cultured mouse TPCs , cells were washed ( 3x; PBS-/- ) , fixed with ice-cold 4% PFA in PBS-/- ( 20 min; RT ) , and washed again ( 3x; PBS-/- ) . For blocking , sections/cells were incubated in PBS-/- containing Tween-20 ( 0 . 1% ) /BSA ( 3% ) solution ( 1 hr; RT ) . After washing ( PBS-/-; 2 × 5 min ) , sections/cells were incubated FITC-conjugated monoclonal anti-actin , α-smooth muscle ( α-SMA-FITC , cat # F3777 , MilliporeSigma ) antibody ( 1:500 in 3% BSA; 1 hr; RT ) . Excess antibodies were removed by washing ( 2 × 5 min PBS-/- ) . For nuclear counterstaining , sections/cells were then incubated in PBS-/- containing either DAPI ( 5 µg ml−1; 10 min; RT; Thermo Fisher Scientific ) or DRAQ5 ( 1:500; 5 min; RT; Thermo Fisher Scientific ) . Fluorescent images were taken using either an inverted microscope ( Leica DMI4000B , Leica Microsystems ) or an upright fixed stage scanning confocal microscope ( TCS SP5 DM6000 CFS; Leica Microsystems ) equipped with a 20 × 1 . 0 NA water immersion objective ( HCX APO L; Leica Microsystems ) . To control for non-specific staining , experiments in which the primary antibody was omitted were performed in parallel with each procedure . Digital images were uniformly adjusted for brightness and contrast using Adobe Photoshop CS6 ( Adobe Systems , San Jose , CA , USA ) . For testicular tissue clearing we adopted the CLARITY method ( Chung et al . , 2013 ) with minor modifications ( Gretenkord et al . , 2019 ) . Briefly , testes from adult mice were fixed overnight at 4°C in hydrogel fixation solution containing 4% acrylamide , 0 . 05% bis-acrylamide , 0 . 25% VA-044 Initiator , 4% PFA in PBS-/- to maintain structural integrity . After hydrogel polymerization , lipids were removed by incubation in 4% sodium dodecyl phosphate ( SDS ) solution with 200 mM boric acid ( pH 8 . 5 ) over periods of two months . Solutions were changed bi-weekly . During the final incubation period , the nuclear marker DRAQ5 ( 1:1000 ) was added . After washing ( 2 d ) with PBST ( 0 . 1% TritonX ) , samples were incubated for 24 hr in RIMS80 containing 80 g Nycodenz , 20 mM PS , 0 . 1% Tween 20 , and 0 . 01% sodium acid . Cleared samples were imaged using a Leica TCS SP8 DLS confocal microscope , equipped with a digital light-sheet module , 552 nm and 633 nm diode lasers , a HC PL FLUOTAR 5x/0 . 15 IMM DLS objective ( observation ) , a L 1 . 6x/0 . 05 DLS objective ( illumination ) , a DLS TwinFlect 7 . 8 mm Gly mirror cap , and a DFC9000 sCMOS camera . Rendering and three-dimensional reconstruction of fluorescence images was performed using Imaris 8 microscopy image analysis software ( Bitplane , Zurich , Switzerland ) . Whole-cell patch-clamp recordings were performed as described ( Fleck et al . , 2016; Veitinger et al . , 2011 ) . Briefly , mouse TPCs were transferred to the stage of an inverse microscope ( DMI 4000B , Leica Microsystems ) , equipped with phase contrast objectives and a cooled CCD camera ( DFC365FX , Leica Microsystems ) . Cells were continuously superfused with solution S1 ( ∼3 ml min−1; gravity flow;~23°C ) . Patch pipettes ( ∼5 MΩ ) were pulled from borosilicate glass capillaries with filament ( 1 . 50 mm OD/0 . 86 mm ID; Science Products ) on a PC-10 vertical two-step micropipette puller ( Narishige Instruments , Tokyo , Japan ) , fire-polished ( MF-830 Microforge; Narishige Instruments ) and filled with S6 . An agar bridge ( 150 mM KCl ) connected reference electrode and bath solution . An EPC-10 amplifier ( RRID:SCR_018399 ) controlled by Patchmaster 2 . 9 software ( RRID:SCR_000034; HEKA Elektronik ) was used for data acquisition . We monitored and compensated pipette and membrane capacitance ( Cmem ) as well as series resistance ( Rseries ) . Cmem values served as a proxy for the cell surface area and , thus , for normalization of current amplitudes ( i . e . current density ) . Cells displaying unstable Rseries values were not considered for further analysis . Liquid junction potentials were calculated using JPCalcW software ( Barry , 1994 ) and corrected online . Signals were low-pass filtered [analog 3- and 4-pole Bessel filters ( –3 dB ) ; adjusted to 1/3 - 1/5 of the sampling rate ( 10 kHz ) ] . If not stated otherwise , holding potential ( Vhold ) was –60 mV . Cultured mouse TPCs were imaged as described ( Veitinger et al . , 2011 ) . Briefly , cells were loaded with fura-2/AM in the dark ( 5 μM; 30 min; RT; S1 ) and imaged with an upright microscope ( Leica DMI6000FS , Leica Microsystems ) equipped for ratiometric live-cell imaging with a 150 W xenon arc lamp , a motorized fast-change filter wheel illumination system for multi-wavelength excitation , a CCD camera ( DFC365 FX , Leica ) , and Leica LAS X imaging software . Ten to thirty cells in randomly selected fields of view were viewed at 20x magnification and illuminated sequentially at 340 nm and 380 nm ( cycle time 2 s ) . The average pixel intensity at 510 nm emission within user-selected ROIs was digitized and calculated as the f340/f380 intensity ratio . For parallel recordings of intracellular Ca2+ signals and tubular contractions , acute seminiferous tubule slices were bulk-loaded with fura-2/AM in the dark ( 30 µM; 30 min; RT ) . After washing ( 3x; S1 ) , slices were transferred to a recording chamber and imaged with an upright microscope ( Leica DMI6000FS , see above ) . We installed a custom-built reflective shield beneath the recording chamber for parallel monitoring of fluorescence and reflected light . At 1 Hz imaging cycles , we thus recorded two 510 nm fluorescence images ( 340/380 nm excitation ) and a ‘pseudo-brightfield’ reflected light image that allowed quasi simultaneous analysis of intracellular Ca2+ and tubular movement . To ensure effective store depletion by CPA treatment , we recorded intracellular Ca2+ levels during CPA incubation at low frequency to monitor Ca2+ release from the ER , but also prevent phototoxicity . Experiments were only conducted if ( i ) we detected a substantial gradual rise in intracellular Ca2+ upon CPA treatment , and if ( ii ) functional Ca2+ extrusion mechanisms ensured that Ca2+-dependent fluorescence signals returned to base level . The time-course of this Ca2+ release – Ca2+ extrusion process varied between samples and ranged between 5 . 3 and 44 . 0 min ( 18 . 8 ± 9 . 3 min; mean ± SD ) . Isolated tubules ( >1 cm length ) were placed onto a membrane within a custom-built 3D printed two-compartment recording chamber that was constantly superfused with S1 . Small membrane holes under the tubules and around a defined stimulation area allowed for ( i ) gentle fixation of the tubules and ( ii ) focal ATP perfusion of selected tubular regions by vacuum-generated negative pressure ( 80–180 mmHg ) in the submembraneous chamber compartment and continuous suction of S1 from the top compartment . After visual determination of tubular stages ( I – III ) ( Parvinen , 1982 ) , the perfusion pencil was positioned to selectively stimulate an area of known and homogeneous stage . Focal stimulation in the desired area was routinely confirmed by transient dye perfusion ( Fast Green ) prior to ATP exposure . ATP stimulations ( 100 µM; 10 s ) and corresponding negative controls were compared to determine ATP-dependent Ca2+ signals ( offspring from crossing SMMHC-CreERT2 and Ai95D mice ) or tubular contractions and sperm transport . For low-magnification brightfield or fluorescence imaging , we used a MacroFluo Z16 APO A system ( Leica Microsystems ) equipped with either a DFC450C camera and a PLANAPO 1 . 0x/WD 97 mm objective ( brightfield ) or with a monochrome DFC365FX camera and a 5 . 0x/0 . 50 LWD PLANAPO objective ( fluorescence ) . Images were acquired at 1 Hz . We administered tamoxifen ( 75 mg tamoxifen kg−1 body weight ) to double-positive adult male offspring ( SMMHC-CreERT2 x Ai95D ) via daily intraperitoneal injections for five consecutive days . Mice were closely monitored for any adverse reactions to the treatment . Experiments were performed 2–5 weeks after the first injection . For surgery , mice were anesthetized with ketamine-xylazine-buprenorphine ( 100 , 10 , 0 . 05–0 . 1 mg kg−1 , respectively; Reckitt Benckiser Healthcare , UK ) . First , we made an incision next to the linea alba in the hypogastric region , followed by a 5 mm incision into the peritoneum . One testis was gently lifted from the abdominal cavity . Its gubernaculum was cut and the testis – with the spermatic cord , its blood vessels and vas deferens still intact – was transferred to a temperature-controlled imaging chamber filled with extracellular solution ( S1; 35°C ) , mounted on a custom-designed 3D printed in vivo stage ( Figure 7—figure supplement 1 ) . Throughout each experiment , vital signs ( heartbeat , blood oxygen level , breathing rhythm ) were constantly monitored and recorded ( breathing ) . Moreover , we routinely checked unobstructed blood flow within testicular vessels during experiments . To avoid movement artifacts , the tunica was glued to two holding strings using Histoacryl tissue adhesive . After surgery , anesthesia was maintained by constant isoflurane inhalation ( 1–1 . 5% in air ) . Time-lapse intravital imaging was performed using a Leica TCS SP8 MP microscope . For incident light illumination/reflected light widefield recordings ( 5–10 Hz ) , we used N PLAN 5x/0 . 12 or HC APO L10x/0 . 30 W DLS objectives with large fields of view . Multiphoton time-lapse images were acquired at ~2 Hz frame rates using external hybrid detectors and the HCX IRAPO L25x/0 . 95 W objective at 930 nm excitation wavelength . Individual recording duration varied between 13 and 30 min ( mean = 25 min ) . For in vivo stimulation experiments , we used a Picospritzer III ( Parker Hannifin , Pine Brook , NJ ) to puff nanoliter volumes of control saline ( S1; containing Alexa Fluor 555 ( 4 µM ) ) or stimulus solution ( S1; containing Alexa Fluor 555 ( 4 µM ) and ATP ( 1 mM ) ) , respectively , from beveled glass micropipettes onto the surface of seminiferous tubules . All data were obtained from independent experiments performed on at least three days . Individual numbers of cells/tubules/experiments ( n ) are denoted in the respective figures and/or legends . If not stated otherwise , results are presented as means ± SEM . Statistical analyses were performed using paired or unpaired t-tests , one-way ANOVA with Tukey’s HSD post hoc test or the Fisher Exact test ( as dictated by data distribution and experimental design ) . Tests and corresponding p-values that report statistical significance ( ≤0 . 05 ) are individually specified in the legends . Data were analyzed offline using FitMaster 2 . 9 ( HEKA Elektronik ) , IGOR Pro 8 ( RRID:SCR_000325; WaveMetrics ) , Excel 2016 ( Microsoft , Seattle , WA ) , and Leica LAS X ( RRID:SCR_013673; Leica Microsystems ) software . Dose-response curves were fitted by the Hill-equation . Time-lapse live-cell imaging data displaying both Ca2+ signals and tubular contractions were analyzed using custom-written code in MATLAB ( RRID:SCR_001622; The MathWorks , Natick , MA ) . For quantitative image analysis , images from both reflected light and fluorescence time-lapse recordings were registered to their respective first image frame at time point t0 , using the registration algorithm from Liu et al . , 2015 ( implementation in Evangelidis , 2013 ) , resulting in stabilized recordings without movement . For fura-2 fluorescence recordings , we first performed a single registration on the combined image ( f340 + f380 ) and then applied the displacement vector field , computed by the registration algorithm , to both images ( f340 and f380 ) separately . ROIs were defined manually at t0 and superimposed onto all subsequent images of the stabilized recording . At each time point ti , the fluorescence signal F was computed as the mean f340/f380 ratio of all pixels within a given ROI . When measuring Ca2+-dependent changes in GCaMP6f intensity , the fluorescence signal F was normalized with respect to a baseline before stimulation , computing the intensity change for the ith time point as Fi − FbaselineFi . For clarity , linear baseline shifts were corrected in some example traces . Seminiferous tubule contractions and transport of luminal content were visualized by reflected light microscopy of acute slices or whole-mount macroscopic tubule imaging , respectively . Data from both types of time-lapse recordings were analyzed and quantified as either flow strength or flow change ( see below ) . For each frame at a given time point ti , the registration algorithm computed a flow or displacement vector field Vi= ( v1 , 1⋯v1 , n⋮⋱⋮vm , 1⋯vm , n ) , where v1 , 1= ( x , y ) is a vector indicating strength and direction of the displacement of pixel ( 1 , 1 ) between time points t0 and ti . The average norm \| Vi \| = 1mn∑p , q‖vp , q ‖ is a measure for the effort that is necessary to register the image at t0 to the image at ti . The flow field strength quantified by this measure is interpreted as the amount of visible changes that , dependent on the experiment , result from tubule contraction and / or luminal content movement . For analysis of contractions in acute seminiferous tubule slices ( Figures 3 , 4 , 7 ) , we quantified the flow strength si within an ROI as the average norm \|Vi\| computed only for the vp , q corresponding to pixels within the ROI defined at t0 . For whole-mount macroscopic imaging of luminal content movement in intact tubule segments ( Figure 5 ) , we quantified the flow change ci= si− si−1 as the change of flow strength between two consecutive time points / frames . Here , si values were preprocessed by smoothing with a moving average filter . Results are reported as the AUC , that is , the area under the ci curve . For analysis of in vivo data , we employed a custom set of ImageJ macros utilizing build-in functions of Fiji-ImageJ ( RRID:SCR_002285 ) ( Rueden et al . , 2017; Schindelin et al . , 2012 ) . Widefield imaging data was first corrected for brightness fluctuation caused by a 50 Hz AC power supply . Here , we used the bleach correction plugin in histogram matching mode ( Miura et al . , 2014 ) . Next , we applied Gaussian filter functions ( GausBlur ( five px radius ) and Gaussian Blur3D ( x = 0 , y = 0 , z = 5 ) ) . We calculated flow change via the Gaussian Window MSE function ( sigma = 1; max distance = 3 ) . Tubule selection used the polyline tool ( line width adjusted to tubule diameter ) . Selected tubules ranged from 200 µm to 3 . 4 mm length . Next , flow fields of individual tubules were straightened . Average movement intensity was calculated from transversal line profiles ( perpendicular to the straightened longitudinal axis of each tubule ) and plotted as kymographs ( space-time plots ) to measure movement progression speed from linear regressions . Multiphoton time-lapse imaging data was recorded in dual-channel mode , with ( i ) a target channel recording GCaMP6f fluorescence and some background signal ( 525\50 nm ) , and ( ii ) a background channel mainly recording autofluorescence ( 585\40 nm ) , allowing for background correction of the GCaMP6f signal using a dye separation routine . Slow constant movement in both channels was registered and removed to correct for steady drift . After Gaussian filtering ( GausBlur ( five px radius ) ; Gaussian Blur3D ( x = 0 , y = 0 , z = 5 ) ) , flow fields were calculated from the background signal . Again , flow change was calculated via the Gaussian Window MSE function ( sigma = 1; max distance = 3 ) . Time-lapse epifluorescence in vivo recordings were processed to isolate transient fluorescence signals from static background noise using custom ImageJ code with Fiji’s build-in functions ( see Data and materials availability ) .	As sperm develop in the testis , the immature cells must make their way through a maze of small tubes known as seminiferous tubules . However , at this stage , the cells do not yet move the long tails that normally allow them to ‘swim’; it is therefore unclear how they are able to move through the tubules . Now , Fleck , Kenzler et al . have showed that , in mice , muscle-like cells within the walls of seminiferous tubules can create waves of contractions that push sperm along . Further experiments were then conducted on cells grown in the laboratory . This revealed that a signaling molecule called ATP orchestrates the moving process by activating a cascade of molecular events that result in contractions . Fleck , Kenzler et al . then harnessed an advanced microscopy technique to demonstrate that this mechanism occurs in living mice . Together , these results provide a better understanding of how sperm mature , which could potentially be relevant for both male infertility and birth control .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2021	ATP activation of peritubular cells drives testicular sperm transport
Kinetochores , multi-subunit complexes that assemble at the interface with centromeres , bind spindle microtubules to ensure faithful delivery of chromosomes during cell division . The configuration and function of the kinetochore–centromere interface is poorly understood . We report that a protein at this interface , CENP-M , is structurally and evolutionarily related to small GTPases but is incapable of GTP-binding and conformational switching . We show that CENP-M is crucially required for the assembly and stability of a tetramer also comprising CENP-I , CENP-H , and CENP-K , the HIKM complex , which we extensively characterize through a combination of structural , biochemical , and cell biological approaches . A point mutant affecting the CENP-M/CENP-I interaction hampers kinetochore assembly and chromosome alignment and prevents kinetochore recruitment of the CENP-T/W complex , questioning a role of CENP-T/W as founder of an independent axis of kinetochore assembly . Our studies identify a single pathway having CENP-C as founder , and CENP-H/I/K/M and CENP-T/W as CENP-C-dependent followers . Mitosis creates two genetically identical daughter cells through the equal segregation of sister chromatids . Meiosis , on the other hand , aims to halve the genetic content of a mother cell . In either case , chromosome segregation is executed by molecular machinery that is largely conserved in the evolution of eukaryotes . Kinetochores are essential elements of such machinery ( Santaguida and Musacchio , 2009; Westermann and Schleiffer , 2013 ) . They are biochemically complex structures , which contain multiple copies of approximately 30 core subunits , in turn contributing to recruit many additional regulatory proteins ( Figure 1A ) . Schematically , kinetochores can be viewed as layered structures , whose inner and outer layers directly contact centromeric chromatin and spindle microtubules , respectively . 10 . 7554/eLife . 02978 . 003Figure 1 . Ablation of CENP-M perturbs kinetochore function . ( A ) Core kinetochore components . CENP-C and CENP-T/W may create independent connections between centromeres and outer kinetochores . Green lines indicate direct connections with centromeric DNA or chromatin . Black lines indicate recruitment dependencies . CENP-C binds directly to CENP-A and Mis12 complex ( Przewloka et al . , 2011; Screpanti et al . , 2011; Kato et al . , 2013 ) . CENP-T , together with CENP-W , S and X , may form a nucleosome-like structure interacting directly with DNA and the Ndc80 complex ( Hori et al . , 2008a; Gascoigne et al . , 2011; Schleiffer et al . , 2012; Nishino et al . , 2013 , 2012 ) . Sub-complexes of CCAN subunits were inferred from reconstitution or from similarity of depletion phenotypes ( see main text ) . ( B ) Representative immunofluorescence ( IF ) images showing endogenous CENP-M localization to kinetochores of HeLa cells in both interphase and mitosis . Kinetochores were visualized with CREST sera and DNA stained with DAPI . Insets show a higher magnification of regions outlined by the white boxes . Scale bar = 2 µm . ( C ) Whole cell protein extracts from HeLa cells treated with specific siRNAs ( showed in D ) were run on SDS-PAGE and immunoblotted for the indicated kinetochore proteins . Vinculin was the loading control . MWM , molecular weight marker . ( D ) HeLa cells depleted for CENP-M display significant chromosome congression defects . Following fixation , cells treated with CENP-M siRNA were imaged for endogenous CENP-M , CREST and DNA ( DAPI ) . Scale bars = 2 µm . ( E ) CENP-M kinetochore levels from the experiment in D . Quantifications are expressed as normalized CENP-M/CREST fluorescence intensity ratios . Graphs and bars indicate mean ± SEM . ( F ) Quantification of chromosome congression defects in D . As a positive control , cells treated with 500 nM Hesperadin were scored for alignment defects . ( G ) Quantification of the percentage of mitotic cells in the experiment in D . DOI: http://dx . doi . org/10 . 7554/eLife . 02978 . 00310 . 7554/eLife . 02978 . 004Figure 1—figure supplement 1 . Additional localization data . ( A ) Human CENP-M and CENP-I are constitutively associated with kinetochores during the cell cycle . Asynchronously growing HeLa cells were immunostained for endogenous CENP-M , endogenous CENP-I , and the centromeric marker CREST . DNA was stained with DAPI . The displayed interphase and metaphase cells are the same shown in Figure 1B but with the addition of the CENP-I staining ( left panels ) . Insets show a higher magnification of the regions outlined by the white boxes . Scale bar = 2 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 02978 . 004 The KMN network is the main constituent of the outer kinetochore ( Figure 1A ) . It is a 10-subunit assembly originating from the interaction of three sub-complexes , named the Knl1-complex , the Mis12-complex , and the Ndc80-complex ( Cheeseman and Desai , 2008 ) . It provides at least two sites for microtubule attachment , positioned in the Ndc80 and Knl1 subunits , and it regulates cell cycle progression through the recruitment , activation , and subsequent silencing of the spindle assembly checkpoint ( SAC ) ( Foley and Kapoor , 2013; Figure 1A ) . The inner kinetochore controls outer kinetochore assembly ( Liu et al . , 2006; McClelland et al . , 2007; Cheeseman et al . , 2008 ) , influences microtubule binding ( McClelland et al . , 2007; Amaro et al . , 2010 ) , and contributes to epigenetic specification of centromeres ( Takahashi et al . , 2000; Okada et al . , 2006 , 2009; Hori et al . , 2013 ) . The inner kinetochore is built on centromeres , genetic loci whose hallmark is the presence of CENP-A , a variant of histone H3 . CENP-A interacts with histone H2A , H2B and H4 in a nucleosome-like structure whose precise organization is under discussion ( Black and Cleveland , 2011; Figure 1A ) . In vertebrates , a group of at least 16 inner kinetochore proteins , collectively identified as constitutive centromere-associated network ( CCAN ) , neighbors the CENP-A nucleosome ( Obuse et al . , 2004; Liu et al . , 2005; Foltz et al . , 2006; Izuta et al . , 2006; Okada et al . , 2006 ) . The CCAN ( whose subunits are indicated as ‘CENP’—for centromeric protein—followed by a letter , see Figure 1A ) interacts with the outer kinetochore and contributes to maintain centromere identity by participating in CENP-A loading at every new cell division cycle ( Perpelescu and Fukagawa , 2011; Westhorpe and Straight , 2013 ) . Understanding the physical organization of the CCAN is instrumental to shed light on these vital functions of CCAN . Initial characterization of CCAN subunits after enrichment from cell lysates suggested that they engage in a complex network of interactions ( Cheeseman et al . , 2002; De Wulf , 2003; Obuse et al . , 2004; Liu et al . , 2005; Foltz et al . , 2006; Izuta et al . , 2006; Okada et al . , 2006 ) . Subsequent work led to biochemical reconstitution of at least three CCAN sub-complexes ( Figure 1A ) . First , the 4-subunit CENP-O/P/Q/U complex ( also known as COMA complex in S . cerevisiae ) has been implicated in microtubule binding and spindle checkpoint control ( Perpelescu and Fukagawa , 2011 ) , and is believed to interact with CENP-R , a protein of unknown function ( Hori et al . , 2008b ) . Second , the 2-subunit CENP-N/L complex binds directly to the CATD box , a specific segment of CENP-A that has been implicated in the epigenetic specification of centromeres ( Carroll et al . , 2009; Black and Cleveland , 2011; Hinshaw and Harrison , 2013 ) . Third , the 4-subunit CENP-T/W/S/X complex contains proteins with histone-fold domains that bind DNA and have been proposed to form a nucleosome-like structure ( Hori et al . , 2008a; Nishino et al . , 2012 ) . Together with CENP-C ( Earnshaw and Rothfield , 1985 ) , the CENP-T/W sub-complex has been shown to contribute to outer kinetochore assembly . CENP-C and CENP-T/W interact directly with the Mis12 and Ndc80 complexes , respectively ( Gascoigne et al . , 2011; Przewloka et al . , 2011; Screpanti et al . , 2011 ) . CENP-C provides additional important scaffolding functions , as it binds directly to the CENP-A nucleosome and to the CENP-N/L complex ( Hinshaw and Harrison , 2013; Kato et al . , 2013 ) . Several of these interactions among CCAN subunits have been recognized in S . cerevisiae and in S . pombe , suggesting a conserved plan of kinetochore assembly ( Cheeseman et al . , 2002; Measday et al . , 2002; De Wulf , 2003; Pidoux et al . , 2003; Pot et al . , 2003; Westermann et al . , 2003; Liu et al . , 2005; Tanaka et al . , 2009; Westermann and Schleiffer , 2013 ) . In this study , we concentrate on four additional CCAN subunits , CENP-H , CENP-I , CENP-K , and CENP-M . These proteins have been shown to be proximal to the subunits of the CENP-T/W/X/S , CENP-O/P/Q/U , and CENP-N/L complexes in cell lysates ( Obuse et al . , 2004; Foltz et al . , 2006; Izuta et al . , 2006; Okada et al . , 2006 ) , but their organization and pattern of interactions remain unclear ( Westhorpe and Straight , 2013 ) . Because similar chromosome alignment and kinetochore assembly defects were observed after their RNAi-based depletions , CENP-H , CENP-I and CENP-K were proposed to form a complex ( Okada et al . , 2006; McClelland et al . , 2007; Cheeseman et al . , 2008 ) . Conversely , CENP-M was classified in a distinct phenotypic class , and tentatively assigned as a subunit of the CENP-L/N complex ( Okada et al . , 2006; Westhorpe and Straight , 2013 ) . In this study , we elucidate the molecular basis of CENP-M's essential function in kinetochore assembly . Antibodies against CENP-M or CENP-I stained interphase and mitotic kinetochores ( Figure 1B , Figure 1—figure supplement 1 ) , confirming that CENP-M and CENP-I reside constitutively at kinetochores ( Foltz et al . , 2006; Okada et al . , 2006 ) . To gain insight into CENP-M function , we depleted it by RNA interference ( RNAi ) in HeLa cells ( Figure 1C–E ) . No change in the levels of several other kinetochore proteins was observed ( Figure 1C ) . In mitotic HeLa cells , depletion of CENP-M prevented chromosome alignment at the metaphase plate to a comparable degree to that caused by Hesperadin , a small-molecule competitive inhibitor of Aurora B kinase , whose activity is crucially required for chromosome alignment ( Hauf et al . , 2003; Figure 1F ) . CENP-M depletion also caused robust mitotic arrest , likely a consequence of spindle checkpoint activation caused by chromosome alignment defects ( Figure 1G ) . Thus , CENP-M is required for chromosome alignment and successful mitotic progression . To gain molecular insight on CENP-M , we purified recombinant CENP-M ( Figure 2A ) and asked if it interacted with recombinant versions of known kinetochore and centromere components in size-exclusion chromatography ( SEC ) co-elution experiments ( in which proteins or protein complexes are separated on the basis of size and shape and co-elute if interacting ) . No interaction of CENP-M with a collection of recombinant proteins or protein complexes covering most known inner and outer kinetochore proteins or protein complexes was observed ( summarized in Table 1 ) . 10 . 7554/eLife . 02978 . 005Figure 2 . Characterization of the HIKM complex . ( A ) SEC elution profile of CENP-M with associated SDS-PAGE separations of peak fractions indicated by the horizontal bar under the profile . CENP-M ( ∼20 kDa ) elutes as expected for a monomeric species . ( B ) Schematic representation of the primary sequence of CENP-H , CENP-I , CENP-K , and CENP-M . ( C ) SEC elution profile and SDS-PAGE separation of the CENP-H/K complex . CENP-H/K forms a 1:1 dimer ( ∼61 kDa ) ( Figure 2—figure supplement 1 , panel B ) but elutes near the 158 kDa marker , indicative of an elongated complex . ( D ) SEC elution profile and SDS-PAGE separation of the CENP-HI57–CKM complex . CENP-HI57–CKM ( ∼159 kDa ) elutes near the 158 kDa marker , suggesting that the complex contains a single copy of each subunit . ( E ) Summary of cross-links . Intra-molecular cross-links are shown in blue and outside the ideal perimeter designed by the four subunits of the complex . Inter-molecular cross-links are shown as black lines . ( F ) CENP-H/His-CENP-K complex and His-CENP-I57–281 , both at 10 µM , form a stoichiometric complex as shown by co-elution from SEC runs and corresponding SDS-PAGE separations . ( G ) Lack of co-elution from SEC runs and SDS-PAGE analysis indicate that CENP-H/K/I57–281 complex and CENP-M do not bind . DOI: http://dx . doi . org/10 . 7554/eLife . 02978 . 00510 . 7554/eLife . 02978 . 006Figure 2—source data 1 . List of intra- and inter-molecular crosslinks of the CENP-HIKM complex . DOI: http://dx . doi . org/10 . 7554/eLife . 02978 . 00610 . 7554/eLife . 02978 . 007Figure 2—figure supplement 1 . Additional biochemical characterization . ( A ) SEC elution profile of the CENP-H/His-CENP-K complex . ( B ) Static light scattering analysis of the CENP-H/K complex indicates that the complex has a 1:1 stoichiometry and a molecular weight of ∼61 kDa . ( C–E ) Summary of expression or co-expression tests with the indicated CENP-I constructs . Only the co-expression of CENP-I57–C with both CENP-M and CENP-H-K ( shown in panel E ) was able to promote full solubilisation and stabilization of CENP-I57–C . Co-expression with only CENP-H/K or CENP-M was insufficient . ( F ) SEC analysis shows that CENP-M does not interact with CENP-H/K complex . ( G ) CENP-M does not interact with His-CENP-I57–281 . DOI: http://dx . doi . org/10 . 7554/eLife . 02978 . 00710 . 7554/eLife . 02978 . 008Table 1 . List of potential binding interactions of CENP-M tested with purified proteins and complexesDOI: http://dx . doi . org/10 . 7554/eLife . 02978 . 008CENP-M incubation with:Binding:H3-containing mononucleosomes ( DNA 601–167 bp ) NOCENP-A-containing mononucleosomes ( DNA 601–167 bp ) NOCENP-C constructs ( 1–544 , 509–760 , 631–C-terminus ) NOCENP-L/CENP–N complexNOCENP-H/CENP–K complex , CENP-I57–281 , CENP-H/CENP-K/CENP-I57–281 complexNOCENP-O/CENP-P/CENP-Q/CENP–U complex , CENP-O/CENP-P and CENP-Q/CENP-U sub-complexesNOCENP-RNOCENP-T/CENP-W/CENP-S/CENP-X complex , CENP-T/CENP-W sub-complex ( phosphorylated by Cdk1 or not ) , CENP-S/CENP-X sub-complexNOMis12 complexNONdc80 complexNOKnl12000–2311NOZwintNOKMN networkNOMicrotubulesNO CENP-I ( known as Mis6 and Ctf3 in S . pombe and S . cerevisiae , respectively [Saitoh et al . , 1997; Measday et al . , 2002; Nishihashi et al . , 2002] ) had been initially excluded from these analyses because its expression in bacteria or insect cells had not resulted in a soluble product in cell lysates . Previous observations suggested that CENP-I might interact with CENP-H and CENP-K ( Pot et al . , 2003; Okada et al . , 2006 ) , but formal proof through reconstitution with recombinant proteins had been missing . CENP-H ( known as Fta3 and Mcm16 in S . pombe and S . cerevisiae , respectively [Sanyal et al . , 1998; Sugata et al . , 1999; Liu et al . , 2005] ) and CENP-K ( known as Sim4 and Mcm22 in S . pombe and S . cerevisiae , respectively [Poddar et al . , 1999; Pidoux et al . , 2003; Foltz et al . , 2006; Okada et al . , 2006] ) ( Figure 2B ) form a tight dimer when co-expressed in , and purified from , insect cells ( Figure 2C , Figure 2—figure supplement 1 , panel A ) . Co-expression of full-length CENP-I or CENP-I57–C ( a CENP-I fragment lacking the first 56 residues of CENP-I , predicted to be disordered ) with CENP-H and CENP-K led to a partial solubilization of CENP-I , but the resulting soluble material was unstable and could not be purified homogenously ( Figure 2—figure supplement 1 , panel D ) . We therefore asked if further co-expression of CENP-M favored solubilization of CENP-I57–C . Indeed , co-expression of CENP-M with CENP-I57–C , CENP-H and CENP-K resulted in a soluble and stable 4-subunit complex that could be purified to homogeneity ( Figure 2D , Figure 2—figure supplement 1 , panels C–E ) . Thus , CENP-M stabilizes a quaternary complex containing CENP-H , CENP-I57–C , CENP-K , and CENP-M , to which we refer as ‘HIKM complex’ . The theoretical molecular masses of the HIKM complex and of recombinant HI57–CKM complex used in our studies are 166 and 159 kDa , respectively . In SEC experiments , the HI57–CKM complex eluted close to the 158-kDa protein marker ( Figure 2D ) , suggesting that the recombinant complex contains a single copy of each subunit . To shed light on the organization of the HIKM complex , we subjected it to chemical cross-linking with the bifunctional reagent BS2G ( bis[sulfosuccinimidyl]glutarate ) , which cross-links the primary amines of lysine side chains within a distance compatible with the length of the cross-linker ( 7 . 7 Å ) ( Maiolica et al . , 2007; Herzog et al . , 2012 ) . Subsequent mass spectrometry analysis identified numerous cross-links between CENP-H and CENP-K , the majority of which mapped to the central domains of these proteins , which are likely to adopt an α-helical arrangement . The specific co-linear distribution of cross-links in this region suggests that CENP-H and CENP-K interact through an extended interface in a parallel arrangement ( Figure 2E; the list of crosslinks is available in Figure 2—source data 1 ) . In addition to interacting with each other , the central domains of CENP-H and CENP-K displayed an extensive network of cross-links with the N-terminal region of CENP-I57–C . Accordingly , we were able to reconstitute an interaction between a recombinant segment encompassing the N-terminal region of CENP-I ( residues 57–281 , which was expressed in a soluble form in the absence of other stabilizing components ) and the CENP-H/K complex ( Figure 2F ) , thus confirming that CENP-H/K binds the N-terminal region of CENP-I . We did not observe cross-links between CENP-M and the CENP-H/K complex ( Figure 2E ) , and indeed there was no detectable interaction between these proteins ( Figure 2—figure supplement 1 , panel F ) . Two cross-links , however , were observed between CENP-M and CENP-I ( Figure 2E ) . In SEC experiments , CENP-M did not interact with CENP-I57–281 ( Figure 2—figure supplement 1 , panel G ) or with the CENP-HI57–281K complex ( Figure 2G ) , collectively suggesting that the incorporation of CENP-M in the CENP-HIKM complex requires CENP-I , and that the interaction of CENP-M with CENP-I requires residues located C-terminally to residue 281 . However , co-expression of CENP-M with CENP-I was insufficient to solubilize CENP-I ( Figure 2—figure supplement 1 , left part of panel E ) . In summary , these observations suggest that CENP-I bridges CENP-H/K and CENP-M , and that both CENP-H/K and CENP-M contribute to CENP-I solubility and stabilization . We generated crystals of residues 1–171 of CENP-M ( a fragment identified by limited proteolysis ) ( Figure 3—figure supplement 1 , panel A ) and determined their crystal structure by the SAD ( single-wavelength anomalous diffraction ) method to a resolution of 1 . 5 Å using a SeMet derivative ( Table 2 ) . The structure of CENP-M1–171 is globular and consists of a five-stranded parallel β-sheet surrounded by six α-helices ( Figure 3A , Figure 3—figure supplement 1 , panel B ) . A search of the Protein Data Bank ( PDB ) using the Dali server ( Holm and Rosenström , 2010 ) indicated that CENP-M is structurally related to small GTPases ( for comparison , the structure and topology of the small GTPase Rab1 are shown in Figure 3B and Figure 3—figure supplement 1 , panel C , respectively ) , as previously predicted in a bioinformatics analysis ( Santaguida and Musacchio , 2009; Westermann and Schleiffer , 2013 ) . 10 . 7554/eLife . 02978 . 009Table 2 . Data collection , phasing and refinement statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 02978 . 009NativeDerivativeData collection BeamlineESRF ID14-4SLS X06DA ( PXIII ) SpacegroupP3P3 Unit cell parameters ( Å , ° ) a = b = 104 . 50 , c = 33 . 59a = b = 104 . 03 , c = 33 . 56 α = β = 90 , γ = 120α = β = 90 , γ = 120 Wavelength ( Å ) 0 . 919700 . 97942 Resolution limits ( Å ) 52 . 25–1 . 49 ( 1 . 54–1 . 49 ) 31 . 45–2 . 00 ( 2 . 06–2 . 00 ) Reflections observed/unique607786/64606419456/27271 Completeness ( % ) 98 . 3 ( 96 . 3 ) 99 . 9 ( 98 . 9 ) Rsym† ( % ) 5 . 6 ( 36 . 1 ) 9 . 4 ( 75 . 2 ) <I>/<σI>26 . 3 ( 6 . 4 ) 24 . 1 ( 3 . 9 ) Redundancy9 . 4 ( 8 . 8 ) 15 . 4 ( 13 . 9 ) SAD phasing BAYES-CC49 . 1 ± 18 . 5 Se sites found/expected5/6 FOM before solvent flattening and density modification0 . 35 FOM after solvent flattening and density modification0 . 69Refinement Resolution limits ( Å ) 52 . 25–1 . 49 ( 1 . 52–1 . 49 ) * Reflections for Rcryst/for Rfree59806/4800 Rcryst‡ ( % ) 12 . 4 ( 20 . 6 ) * Rfree‡ ( % ) 16 . 4 ( 23 . 0 ) * No . of protein atoms/water atoms2307/297 Average B factor protein atoms/water atoms ( Å2 ) 21 . 67/35 . 34 RMSD bond lengths ( Å ) 0 . 005 RMSD bond angles ( ° ) 0 . 854 Twin fraction ( operator −h , −k , l ) 0 . 49Ramachandran Plot Statistics§ Favoured region ( % ) 99 . 3 Outliers ( % ) 0 . 0BAYES-CC: Bayesian estimate of the correlation coefficient ( CC ) between the experimental map and an ideal map , reported as CC * 100 ± 2 standard deviations . FOM: figure of merit . RMSD: root mean square deviation . Values in parentheses refer to the highest resolution shell . †Rsym = ΣhΣi\|Ih , i − <Ih>\|/ΣhΣi Ih , i . ‡Rcryst and Rfree = Σ\|Fobs − Fcalc\|/Σ Fobs; Rfree calculated for a 7 . 4% subset of reflections not used in the refinement . §Calculated using MOLPROBITY within the PHENIX suite . 10 . 7554/eLife . 02978 . 010Figure 3 . A small G-protein fold in CENP-M . ( A ) Cartoon model of CENP-M1–171 in two orientations . ( B ) Cartoon model of Rab1A/GDP ( PDB ID 4FMC ) . ( C ) Sequence alignment based on the structural superposition of CENP-M1–171 with Rab1A . Conserved elements of small G proteins are in yellow . A conserved residue involved in catalysis and targeted by activating mutations in Ras is in light blue . Secondary structure elements of CENP-M1–171 not present in Rab1A/GDP are in red , while those present in Rab1A/GDP but not in CENP-M1–171 are in green . ( D ) Experiments with N-methylanthraniloyl ( MANT ) derivatives of GTP and ATP . Binding of MANT-GTP or MANT-ATP to the small GTPase Arl2 ( ‘Arl2’ ) or CENP-M ( ‘M’ ) was monitored at an emission wavelength of 440 nm ( See Figure 3—figure supplement 1 , panel D ) . The histogram shows the time-averaged fluorescence value after addition of the indicated proteins to a solution of the indicated MANT nucleotides , normalized against the time-averaged value prior to protein addition . Only the addition of Arl2 to MANT-GTP ( or MANT-GDP , see Figure 3—figure supplement 1 , panel D ) gave a clear increase in signal indicative of a physical interaction of the MANT nucleotide with Arl2 . ( E ) Unrooted maximum likelihood tree of 157 sequences . Shown are members of classical small GTPase families in many species , covering a wide range of evolutionary points . Gray names are reclassified families ( Rojas et al . , 2012 ) . Bold names are human sequences . CENP-M sequences are pink . Uppercase indicates Uniprot code for proteins . 3-code labels are: Nve ( Nematostella vectensis ) , Bfl ( Branchiostoma floridae ) , Xtr ( Xenopus tropicalis ) , and Cin ( Ciona intestinalis ) . Numbers on the left of 3-code labels are accession numbers corresponding to the DOE Joint Genome Institute ( JGI ) database . Red and underlined names are Genbank gi identifiers found in iterative hmmer searches against non-redundant database . Acanth . castellanii indicates Acantanthamoeba castellanii , where indicates that this entry is annotated as a RAS protein . Dicty . purpureum is Dictyostelium purpureum , Aae is Aedes aegypti , Aqu is Amphimedon queenslandica ( sponge ) , and Clu is Clavispora lusitaniae ( fungi ) . Red numbers within the tree indicate number of trees corresponding to more than 80% of statistical support for a given group , whereas black indicates values below 80% . Only representative numbers have been shown for clarity . DOI: http://dx . doi . org/10 . 7554/eLife . 02978 . 01010 . 7554/eLife . 02978 . 011Figure 3—figure supplement 1 . Additional analyses of CENP-M ( related to Figure 3 ) . ( A ) SEC elution profile of CENP-M1–171 . The protein ( ∼19 kDa ) elutes as expected for a monomer . ( B ) Topology diagram of CENP-M . ( C ) Topology diagram of the small GTPase Rab1A . ( D ) Experiments with N-methylanthraniloyl ( MANT ) derivatives of GTP and ATP ( Hiratsuka , 1983 ) demonstrate that CENP-M does not bind to GTP , GDP , ATP or ADP . The small GTPase Arl2 was used both as positive and as negative control . ( E ) Alignment of the sequences of 12 distant CENP-M orthologs . DOI: http://dx . doi . org/10 . 7554/eLife . 02978 . 01110 . 7554/eLife . 02978 . 012Figure 3—figure supplement 2 . Comparison of CENP-M with Rab-like GTPases . ( A ) The coordinates of CENP-M ( red ) and Rab1A ( green; PDB code 4FMC ) were superposed using the PDBeFold server ( http://www . ebi . ac . uk/msd-srv/ssm/cgi-bin/ssmserver; Krissinel and Henrick , 2004 ) and visualized using Pymol . ( B ) Table showing the root mean square deviation in the coordinates of Cα atoms after superposition of CENP-M with the indicated models of Rab-family GTPases ( PDB ID codes are reported ) . The superpositions were carried out using PDBeFOLD . ( C ) Cartoon models of CENP-M and the indicated Rab-family GTPases viewed with the same orientation . DOI: http://dx . doi . org/10 . 7554/eLife . 02978 . 012 Small GTPases are guanine nucleotide binding proteins that control a variety of essential cellular functions ( Vetter and Wittinghofer , 2001; Cherfils and Zeghouf , 2013 ) . They can act as molecular switches by engaging in different interactions with effectors that depend on the bound nucleotide ( GTP or GDP ) . Small GTPases describe a superfamily of proteins , which are generally classified in several families ( Rojas et al . , 2012 ) . Structural superposition with representative members of different families identified Rab proteins as the closest structural homologs of CENP-M ( Figure 3—figure supplement 2 ) . A sequence alignment derived from the structural superposition of HsCENP-M with HsRab1A ( PDB code 4FMC , [Dong et al . , 2012] ) shows that CENP-M has lost essential sequence motifs required for GTP binding and hydrolysis by small G-proteins ( Vetter and Wittinghofer , 2001; Figure 3C ) . For instance , the glycine-rich P-loop ( GX4GKS/T , where X is any aminoacid ) , which is involved in GTP binding by GTPases , lacks two essential Gly residues in CENP-M . The so-called switch I and switch II regions of small GTPases , which are sensitive to nucleotide hydrolysis and create a dependency on nucleotide status for the interaction of GTPases with their effectors ( Vetter and Wittinghofer , 2001; Cherfils and Zeghouf , 2013 ) , are also profoundly modified in CENP-M . A deletion of the entire α1-β2 loop and the β2 strand effectively removed the switch I region ( Figure 3C ) , while the sequence corresponding to switch II ( DxxG motif ) is highly divergent . Thus , HsCENP-M has the fold of a small GTPase , but may be unable to bind ( and therefore hydrolyze ) GTP . Experiments with MANT nucleotides ( Hiratsuka , 1983 ) confirmed this prediction directly ( Figure 3D , Figure 3—figure supplement 1 , panel D ) . Despite the divergence of CENP-M from bona fide GTPases , several elements indicate that CENP-M evolved from active GTPases . For instance , distant CENP-M family members retain a canonical P-loop ( Figure 3—figure supplement 1 , panel E , ‘Discussion’ ) . More importantly , phylogenetic analyses using a CENP-M/Rab1 structure-based sequence alignment ( Figure 3E , see ‘Materials and methods’ for details ) replicated the distribution of Ras sub-families previously described using a different template alignment ( Rojas et al . , 2012 ) . CENP-M family members cluster at the base of the RAB/RAS sub-families , suggesting that CENP-M might have originated at a similar time . The relative location of the ARF/SRPRB sub-families seems to exclude CENP-M as a basal member of the superfamily . Early divergent eukaryotes ( such as amoebas ) and early metazoans ( molluscs ) express both CENP-M and bona fide GTPases . CENP-M , however , has been lost from most fungi , where additional protein products of specific lineage expansions or existing Rab proteins may fulfill its role ( ‘Discussion’ ) . In conclusion , we identify CENP-M as a ‘pseudo G-protein’ , in analogy to inactive kinase domains , indicated with the term ‘pseudokinase’ . Because the structure of CENP-I is unknown , we submitted its sequence to the structure prediction servers I-TASSER , Phyre2 and Rosetta ( Das and Baker , 2008; Zhang , 2008; Kelley and Sternberg , 2009 ) . The α-solenoid fold of β-karyopherins such as Importin-β was consistently identified as a high-confidence template for CENP-I structural modeling ( Figure 4A ) . Similar results were obtained when structural predictions were carried out with the sequences of CENP-I homologs , including Mis6 ( S . pombe ) and Ctf3 ( S . cerevisiae ) ( Figure 4—figure supplement 1A–B ) . The structure of Importin-β consists of a tandem series of HEAT ( Huntingtin , elongation factor 3 , PR65/A subunit of protein phosphatase 2A and kinase TOR ) repeats , helical hairpins that stack against each other to create a twisted super-helical arrangement . We could not unequivocally demonstrate the existence of HEAT repeats in CENP-I . However , program RADAR ( Heger and Holm , 2000 ) predicted the presence of repeats within the presumed α-helical part of the protein ( Figure 4—figure supplement 1C ) . 10 . 7554/eLife . 02978 . 013Figure 4 . Structural organization of the HIKM complex . ( A ) Cartoon representation of the CENP-I model generated by program I-TASSER ( left ) , of the Importin-β/Ran complex ( middle ) , and of a hypothetical structure between CENP-I and CENP-M modeled on the Importin-β/Ran complex ( right ) . A scoring function ( C-score ) associated with I-TASSER models estimates accuracy of structure predictions . C-score is typically in a range from −5 to 2 , where a higher score reflects a model of better quality . Both false positive and false negative rates are estimated to be below 0 . 1 when a C-score >−1 . 5 is displayed ( Zhang , 2008 ) . The CENP-I model is associated with a C-score of −1 . ( B ) Representative class averages of the negatively stained HIKM complex . Figure 4—figure supplement 3 shows the complete set of class averages . Scale bar = 10 nm . ( C ) A 3D reconstruction of HIKM complex from negatively stained particles at ∼22 Å resolution . Scale bar = 10 nm . ( D ) Summary of interactions in the CENP-HIKM complex . The central regions of CENP-H and CENP-K may form an extended parallel interaction , possibly through an α-helical arrangement , which interacts more or less co-linearly with the N-terminal region of CENP-I ( IN ) . Additional globular domains may be present at the N- and C-termini of CENP-H and CENP-K . The entire sequence of CENP-I may fold as a helical solenoid . CENP-M does not interact with CENP-H/K and may bind near the concave surface of the predicted CENP-I solenoid , becoming largely buried . ( E ) siRNA depletion of endogenous CENP-M abrogates CENP-I kinetochore localization in HeLa cells . Representative cells displayed here are the same shown in Figure 1D , but with addition of CENP-I staining ( left panels ) . Insets display a higher magnification of regions outlined by white boxes . Scale bars = 2 µm . ( F ) CENP-M and CENP-I kinetochore levels from the experiment illustrated in E . Quantification for CENP-M kinetochore levels are the same shown in Figure 1D and were performed as previously described . Graphs and bars indicate mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 02978 . 01310 . 7554/eLife . 02978 . 014Figure 4—figure supplement 1 . Structural predictions on CENP-I orthologs . ( A ) The sequences of human CENP-I plus three orthologues were submitted to the I-TASSER server ( http://zhanglab . ccmb . med . umich . edu/I-TASSER/ ) . The best models from each search were superposed on the human model and displayed . The C-score is shown for each model . ( B ) Summary of the superposition for each homology model . Importantly , while similar to one another , the models are also sufficiently different , as a result of different templates being used for homology modeling . ( C ) Analysis of α-helical repeat elements . Repeat elements within the human CENP-I sequence were identified by the RADAR server ( http://www . ebi . ac . uk/Tools/pfa/radar/ ) ( Heger and Holm , 2000 ) . RADAR also detected repeat elements in the equivalent region for chicken CENP-I , Mis6 , and Ctf3 . From a multiple sequence alignment of 25 CENP-I orthologs , the three repeat regions were all aligned against each other . The multiple sequence alignment was then submitted to the WebLogo server ( http://weblogo . berkeley . edu/logo . cgi ) ( Crooks et al . , 2004 ) to generate a consensus sequence . The secondary structure for the repeats , based on the homology models , is shown above . Consistent α-helical elements are shown in pink; variable elements are shown in light pink; unstructured regions are represented as a black line . The repeat is rather divergent , and indeed is sometimes comprised of three rather than two α-helices . DOI: http://dx . doi . org/10 . 7554/eLife . 02978 . 01410 . 7554/eLife . 02978 . 015Figure 4—figure supplement 2 . Conservation mapped on the CENP-I model . ( A ) Surface representation of the CENP-M/I model ( as shown in Figure 4 ) . The C-terminal part of the CENP-I model has been removed to show the concave surface with clarity . Residues are colored from white to magenta according to conservation within 23 different CENP-I orthologues . ( B ) A representative sequence alignment of eight CENP-I sequences from different organisms is shown , with a focus on the region that , according to the model , may be involved in CENP-M binding . Surface residues predicted to be in contact with CENP-M are highlighted with green circles . A comparison of four importin-β ( ImpB ) orthologues is shown to demonstrate that while many residues are conserved between CENP-I and ImpB ( consistent with a similar fold ) those residues that might contact CENP-M are divergent ( consistent with binding a different ligand ) . Species abbreviations are as follows H . s . , Homo sapiens; O . s . , Ornithorhynchus anatinus; G . g . , Gallus gallus; X . t . , Xenopus tropicalis; D . r . , Danio rerio; N . v . ; Nematostella vectensis; S . c . , Saccharomyces cerevisiae S288C; S . p . , Schizosaccharomyces pombe . DOI: http://dx . doi . org/10 . 7554/eLife . 02978 . 01510 . 7554/eLife . 02978 . 016Figure 4—figure supplement 3 . EM analysis . ( A ) Representative electron micrograph area of the negatively stained CENP-HIKM complex . Scale bar = 100 nm . ( B ) Collection of class averages of the CENP-HIKM complex derived from a data set of 5958 single particles . Selected classes are shown in Figure 4B . Scale bar = 10 nm . ( C ) Fourier shell correlation ( FSC ) curves of the negative stain reconstruction of the CENP-HIKM complex . The resolution was estimated by the FSC 0 . 5 criterion to be 22 Å . ( D ) Reprojections of the 3D reconstruction paired with their corresponding class averages . Scale bar = 10 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 02978 . 01610 . 7554/eLife . 02978 . 017Figure 4—figure supplement 4 . Fitting the CENP-I/M model in the EM density . ( A ) A 3D reconstruction of HIKM complex from negatively stained particles at ∼22 Å resolution ( already shown in Figure 4C ) . Scale bar = 10 nm . ( B ) Four orientations of the CENP-I/M model built by I-TASSER ( see Figure 4A , right ) . ( C ) Tentative manual fitting of the CENP-I/M model into the EM density . The CENP-I/M model fits snugly in the ‘base’ density , leaving empty space in the ‘head’ and ‘nose’ domains , which is therefore predicted to host CENP-H/K . DOI: http://dx . doi . org/10 . 7554/eLife . 02978 . 017 Lysines clustering near the N- and C-termini of CENP-I are involved in numerous intra-CENP-I cross-links ( Figure 2E ) , suggesting that the N- and C-terminal regions of CENP-I are in relatively close proximity , possibly reflecting a super-helical arrangement . Furthermore , the only two cross-links between CENP-M and CENP-I mapped near the N- and the C-termini of CENP-I ( Figure 2E ) . Thus , we speculate that CENP-M may bind , becoming largely buried , near the concave surface of the predicted CENP-I solenoid , in a manner that is reminiscent of the interaction of β-importin with the small GTPase Ran ( Vetter et al . , 1999; Figure 4A ) . In agreement with this hypothesis , there is strong sequence conservation near the surface of the I-TASSER model of CENP-I predicted to interact with CENP-M ( Figure 4—figure supplement 2A , B ) . We determined the three-dimensional ( 3D ) structure of the HIKM complex by negative stain electron microscopy and single particle analysis ( Figure 4B , C ) . The resulting map ( Figure 4C ) extended to a maximal resolution of 22 Å ( Figure 4—figure supplement 3 ) . The HIKM complex has a long axis of ∼14 nm and short axes of ∼5–7 nm , and a rather irregular shape , with a broader ‘base’ and a ‘head’ domain from which a prominent ‘nose’ domain emerges . The model of the CENP-I/M complex fits snugly in the base domain of this density ( Figure 4—figure supplement 4 ) , suggesting that CENP-H/K occupy the head and nose domains . Our future studies will address the validity of this fitting . Figure 4D summarizes the hypothetical topological organization of the HIKM complex emerging from the battery of structural and biochemical analyses reported above . Collectively , our observations identify a crucial role of CENP-M in the stabilization of CENP-I . We therefore asked if CENP-M is required for kinetochore localization of CENP-I . In line with this expectation , RNAi-based depletion of CENP-M resulted in the complete disappearance of CENP-I from kinetochores ( Figure 4E , F ) . In principle , the loss of CENP-I from kinetochores upon CENP-M depletion might reflect the loss of the interaction between CENP-M and CENP-I , but also the loss of additional interactions required for kinetochore stability . To overcome this objection , we sought to identify point mutations in CENP-M affecting its interaction with CENP-I . To identify such mutants , we focused on conserved residues exposed at the surface of CENP-M ( indicated with an asterisk in Figure 5A and displayed in Figure 5B; see also the alignment in Figure 3—figure supplement 1 , panel E ) , reasoning that at least a subset of such conserved surface-exposed residues might be involved in CENP-I binding . We then co-expressed GST-CENP-M or its point mutants with CENP-H , CENP-I , and CENP-K in insect cells , and monitored the amount of CENP-I co-purifying with GST-CENP-M on a glutathione–sepharose resin ( data not shown ) . A double point mutant of GST-CENP-M , GST-CENP-ML94A–L163E , was unable to precipitate CENP-H , CENP-I , and CENP-K when co-expressed in insect cells , contrarily to GST-CENP-Mwt ( Figure 5C ) . Mutation of L94 and L163 did not affect the solubility of CENP-M , nor its behavior during the purification procedure ( Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 02978 . 018Figure 5 . CENP-M residues required for kinetochore targeting of CENP-I . ( A ) The CENP-M alignment identifies highly conserved residues ( based on alignment in Figure 3—figure supplement 1 , panel D ) , a subset of which ( asterisks ) is exposed at the surface of CENP-M . ( B ) Position of conserved residues shown in A on two opposite faces of the CENP-M surface . ( C ) After insect cell co-expression of indicated proteins , affinity purification with GST-CENP-M ( if present ) led to isolation of associated proteins shown , after SDS-PAGE separation , in the ‘beads’ fraction . Material eluted from beads was collected and shown in lanes labeled ‘elution’ . Co-expression of all four CENP-HI57–CKM subunits ( positive control ) is necessary for identification of the CENP-HI57–CKM complex on beads and in elution fraction . CENP-ML94A–L163E fails to assemble CENP-HI57–CKM despite the expression of CENP-H , CENP-I57–C and CENP-K . ( D ) GFP-CENP-Mwt , but not GFP-CENP-ML94A+L163E , co-immunoprecipitates CENP-I , CENP-T and Mis12 from mitotic cells . Panels represent the α-GFP co-immunoprecipitation analysis of protein extracts obtained from mitotic HeLa Flp-In T-REx cells stably expressing GFP , GFP-CENP-Mwt or GFP-CENP-ML94A+L163E from an inducible promoter . Total protein extracts ( Input ) and immunoprecipitates ( α-GFP IP ) were run on SDS-PAGE and subjected to WB with indicated antibodies . Vinculin was used as a loading control . ( E ) Representative images of HeLa Flp-In T-REx cells treated with siRNA for endogenous CENP-M and expressing the indicated siRNA-resistant GFP-CENP-M fusions . Expression of GFP-CENP-Mwt , but not of GFP-CENP-ML94A+L163E , rescues chromosome alignment defects and loss of CENP-I kinetochore localization observed upon depletion of endogenous CENP-M . Following fixation , cells were immunostained and imaged for GFP , CENP-I , CREST and DNA ( DAPI ) . Insets show a higher magnification of regions outlined by white boxes . Scale bars = 2 µm . ( F ) Quantification , for experiment in E , of the CENP-I kinetochore levels normalized to CREST kinetochore signal . Graphs and bars indicate mean ± SEM . See ‘Materials and methods’ section for details on quantification . DOI: http://dx . doi . org/10 . 7554/eLife . 02978 . 01810 . 7554/eLife . 02978 . 019Figure 5—figure supplement 1 . Stability of CENP-M mutant . SEC runs and corresponding SDS-PAGE of CENP-Mwt or CENP-ML94A+L163E purified to homogeneity after expression in E . coli indicate essentially identical elution profiles , suggesting that the point mutations do not affect the stability of the CENP-M mutant . DOI: http://dx . doi . org/10 . 7554/eLife . 02978 . 01910 . 7554/eLife . 02978 . 020Figure 5—figure supplement 2 . Inducible expression and localization of CENP-M . ( A ) HeLa Flp-In T-REx cells expressing siRNA resistant GFP-CENP-M fusions and treated with siRNA specific for endogenous CENP-M show significantly reduced levels of the target protein . The effect of CENP-M siRNA treatment on the cellular levels of CENP-M , GFP , and GFP-CENP-M fusions was monitored by Western blotting . Protein extracts from the indicated conditions were run on SDS-PAGE and immunoblotted for the indicated proteins . Vinculin was used as a loading control . ( B–D ) Representative images of the cellular localization of CENP-M fusions to GFP . HeLa Flp-In T-REx cells expressing GFP fused to CENP-Mwt ( either N-terminally [B] or C-terminally [C] ) or CENP-ML94A+L163E ( D ) were analysed for the localization of CENP-I and the GFP fusions . Cultures were enriched for G2 cells by inhibiting Cdk1 with RO-3306 ( Vassilev et al . , 2006 ) . G2 cells were then released into mitosis with a washout of the inhibitor in pre-warmed media , fixed after 1 hr and imaged for GFP , endogenous CENP-I , CREST and DNA ( DAPI ) . Insets show a higher magnification of the regions outlined by the white boxes . Scale bar represents 2 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 02978 . 020 To assess the effect of these mutations in cells , GFP-CENP-Mwt or GFP-CENP-ML94A–L163E were expressed from an inducible promoter after stable integration in HeLa cells . Precipitates of GFP-CENP-Mwt contained CENP-I , the CCAN subunits CENP-T/W , and the KMN protein Mis12 . Conversely , GFP-CENP-ML94A–L163E was unable to establish any of these interactions ( Figure 5D ) . Thus , the CENP-I binding surface of CENP-M is required to establish interactions with other kinetochore proteins . RNAi-resistant GFP-CENP-Mwt localized to kinetochores ( Figure 5—figure supplement 2 , panels A , B ) and rescued the severe chromosome alignment defect caused by the depletion of endogenous CENP-M by RNAi , as well as kinetochore localization of CENP-I , demonstrating the specificity of the RNAi-based depletion of CENP-M . GFP-CENP-ML94A–L163E , on the other hand , did not rescue chromosome alignment or CENP-I localization ( Figure 5E , F ) . Furthermore , GFP-CENP-ML94A–L163E did not localize to kinetochores ( Figure 5—figure supplement 2 , panel D ) , indicating that CENP-M and CENP-I are mutually required for kinetochore localization . Collectively , our analysis highlights the importance of CENP-M as a stabilization factor for CENP-I in vitro and in vivo . Next , we wished to study the CENP-M/I interaction in the context of inner and outer kinetochore assembly ( Figure 1A ) , focusing in particular on CENP-T and CENP-C . Both these proteins likely span a large fraction of the physical distance that separates chromatin from microtubules ( Hori et al . , 2008a , 2013; Wan et al . , 2009; Gascoigne et al . , 2011; Przewloka et al . , 2011; Screpanti et al . , 2011; Suzuki et al . , 2011; Schleiffer et al . , 2012; Nishino et al . , 2013 ) . CENP-C interacts directly with the CENP-A nucleosome ( Carroll et al . , 2010; Kato et al . , 2013 ) . CENP-T , on the other hand , has been proposed to form a nucleosome-like structure that might flank the CENP-A nucleosome and other canonical H3 nucleosomes in centromeres ( Hori et al . , 2008a; Nishino et al . , 2012; Takeuchi et al . , 2014 ) . Mutual independence in their kinetochore recruitment has led to suggest that CENP-C and CENP-T contribute two major independent axes for outer kinetochore assembly ( Hori et al . , 2008a , 2013; Gascoigne et al . , 2011 ) . Indeed , both CENP-T and CENP-C interact directly with outer kinetochore components , and the CENP-T/W complex has been proposed to contribute to a pathway of Ndc80 recruitment that is independent of the Mis12 complex ( Gascoigne et al . , 2011; Schleiffer et al . , 2012; Hori et al . , 2013 ) . Depletion of CENP-M did not perturb the kinetochore levels of CENP-C ( Figure 6A; see Figure 6E for quantification of fluorescence intensity data for panels A–D ) . CENP-C has been implicated in a direct interaction with the Mis12 complex ( Przewloka et al . , 2011; Screpanti et al . , 2011; Figure 1A ) . Consistently with the retention of CENP-C in cells depleted of CENP-M , the kinetochore levels of Nsl1 , a subunit of the Mis12 complex , remained largely normal after CENP-M depletion ( Figure 6B ) . Depletion of CENP-M , on the other hand , resulted in co-depletion of CENP-T/W from kinetochores . Importantly , this defect was rescued by GFP-CENP-Mwt but not CENP-ML94A–L163E ( Figure 6C ) . In line with the idea that CENP-T/W contributes to recruit the Ndc80 complex to kinetochores , loss of CENP-T/W in CENP-M depleted cells correlated with a severe reduction of the kinetochore levels of Ndc80 ( also known as Hec1 ) , a subunit of the Ndc80 complex ( Figure 6D ) . Collectively , these results suggest that reduced kinetochore levels of CENP-T/W in cells depleted of CENP-M reduce Ndc80 localization and generate chromosome alignment defects . 10 . 7554/eLife . 02978 . 021Figure 6 . Significance of the CENP-M/CENP-I interaction for kinetochore assembly . ( A–D ) Representative images of the localization of kinetochore proteins in HeLa Flp-In T-REx cells treated with siRNA for endogenous CENP-M and expressing the indicated siRNA-resistant GFP-CENP-M fusions . Scale bars = 2 µm . ( E ) Quantification , for experiments A–D , of the kinetochore levels of the indicated proteins normalized to CREST . Graphs and bars indicate mean ± SEM . ( F ) Depletion of CENP-C abrogates kinetochore accumulation of CENP-T/W . Representative images of HeLa cells treated with siRNA for CENP-C or CENP-T and arrested in G2 with the Cdk1 inhibitor RO-3306 ( ‘Materials and methods’ ) . Following fixation , cells were immunostained for CENP-C , CENP-T/W and CREST . DNA was stained with DAPI . Scale bars = 10 µm . ( G ) Quantification , for experiment F , of the kinetochore levels of the indicated proteins normalized to CREST kinetochore signal . Graphs and bars indicate mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 02978 . 021 Previously , kinetochore localization of at least a subset of HIKM subunits has been shown to depend on CENP-C ( Milks et al . , 2009; Carroll et al . , 2010; Gascoigne et al . , 2011 ) . Because kinetochore localization of CENP-T/W depends on the HIKM complex ( Figure 6C ) , we therefore hypothesized that kinetochore accumulation of CENP-T/W might also rely on CENP-C . Indeed , depletion of CENP-C prevented kinetochore localization of CENP-T/W both in interphase ( Figure 6F; quantified in Figure 6G ) and in mitotis ( not shown ) , whereas depletion of CENP-T/W did not alter CENP-C localization ( Figure 6F , G ) . These results are in agreement with a previous report ( Carroll et al . , 2010 ) and highlight the essential role of CENP-C as the basis of the pathway of kinetochore assembly of the HIKM and CENP-T/W complexes . Finally , we asked if the loss of CENP-T/W from kinetochores upon depletion of CENP-M could reflect a direct interaction between these proteins . We immobilized GST ( as control ) or a GST-tagged version of the HIKM complex ( containing GST-CENP-M ) on solid phase as baits , and exposed them to different prays . Untagged CENP-T/W/X/S complex ( Nishino et al . , 2012 ) bound directly to the HIKM complex and was retained on GST-HIKM beads ( Figure 7A . Note that the band corresponding to CENP-X is rather diffuse and poorly visible ) . A version of the CENP-T/W/S/X complex containing CENP-T458–C , which only contains the histone-fold domain of CENP-T , rather than full-length CENP-T , bound to GST-HIKM equally effectively ( Figure 7B ) . Indeed , the CENP-T458–C/W complex was sufficient for an interaction with GST-HIKM ( Figure 7C ) , whereas no binding to GST-HIKM was observed with CENP-S/X in the absence of CENP-T/W ( Figure 7D ) . Collectively , these results clarify that the CENP-T/W complex interacts directly with the HIKM complex , whose subunits are required for the recruitment of the CENP-T/W complex to kinetochores . 10 . 7554/eLife . 02978 . 022Figure 7 . Direct interaction of HIKM complex with the CENP-T/W complex . ( A–D ) GST or GST-HIKM baits were immobilized on beads and incubated with ( A ) CENP-T/W/S/X complex , ( B ) CENP-T452–C/W/S/X complex , ( C ) CENP-T452–C/W , or CENP-S/X . For each sample , both the input and the solid phase bound material ( indicated as ‘pull-down’ ) are shown after separation by SDS-PAGE and staining with Coomassie brilliant blue . Note that full-length CENP-T and CENP-I57–C migrated indistinguishably . ( E ) Model of kinetochore assembly supported by our analysis . CENP-C and possibly CENP-N/L interact directly with the CENP-A nucleosome . The presence of CENP-C at the centromere is essential for the recruitment of CENP-T/W and CENP-HIKM complex . CENP-T/W and CENP-HIKM complex are co-dependent and interact physically with each other . DOI: http://dx . doi . org/10 . 7554/eLife . 02978 . 022 CENP-M was initially identified for its high expression in proliferating cells , and named accordingly PANE1 , for proliferation associated nuclear element 1 ( Bierie et al . , 2004 ) . Subsequently , CENP-M was shown to be closely associated with CENP-A , as well as with CENP-H , CENP-K , CENP-I , CENP-L , CENP-N , and CENP-T ( Obuse et al . , 2004; Foltz et al . , 2006; Izuta et al . , 2006; Okada et al . , 2006 ) . RNAi-based depletion of CENP-M caused mis-localization of other CCAN subunits , thus pointing to an important role of CENP-M in inner kinetochore stability ( Foltz et al . , 2006; Okada et al . , 2006 ) . However , the molecular mechanisms subtending to the function of CENP-M and the physical interactions in which CENP-M engages at the kinetochore had remained unknown . Kinetochores , like many other cellular structures , are patchworks of different protein–protein interaction motifs and domains . For instance , at least six kinetochore proteins , including the CCAN subunit CENP-O and CENP-P ( respectively orthologs of Mcm21 and Ctf19 of S . cerevisiae ) , contain RWD domains ( Schmitzberger and Harrison , 2012; Petrovic et al . , 2014 ) . More recently , the structure of the Chl4/Iml3 complex of S . cerevisiae , respectively orthologs of CENP-N and CENP-L , revealed structural similarity with the bacterial recombination-association protein RdgC and with TATA-binding protein TBP ( Guo et al . , 2013; Hinshaw and Harrison , 2013 ) . In this study , we have extended the collection of kinetochore folds by demonstrating that CENP-M folds like a small GTPase , as recently postulated based on prediction methods ( Westermann and Schleiffer , 2013 ) . CENP-M is devoid of all essential sequence signatures associated with nucleotide binding , hydrolysis and conformational switching of small GTPases ( Vetter and Wittinghofer , 2001; Rojas et al . , 2012; Cherfils and Zeghouf , 2013 ) . An invariant feature of CENP-M is the presence of leucine at position 52 ( of HsCENP-M ) , equivalent to Gln61 ( Q61 ) at the end of the Switch II region of Ras ( Figure 3C ) . Mutation of Gln61 to Leu impairs GTP hydrolysis ( Scheidig et al . , 1999 ) and unleashes the transforming potential of Ras . Thus , the first step leading CENP-M to diverge from genuine GTPases might have been an impairment of GTP hydrolysis , followed by the loss of other features associated with GTP handling . Consistent with this hypothesis , a substantially conserved P-loop is present in distant CENP-M family members ( Figure 3—figure supplement 1 , panel E ) . Our phylogenetic analysis identified CENP-M as a bona fide member of the small GTPase tree and suggests that CENP-M might have evolved from an ancestor shared with Rab-family GTPases . Interestingly , CENP-M can only be identified in metazoans but not in fungal genomes . Conversely , the other components of the complex , CENP-H , CENP-I , and CENP-K , are nearly ubiquitously conserved and are clearly identified in fungi , where they appear to interact ( Measday et al . , 2002 ) . Given the importance of CENP-M in kinetochore assembly and stability , and its interaction with evolutionarily conserved proteins , its absence in several representatives of Opisthokonta is puzzling . We consider it unlikely that the fission yeast protein Mis17 is a CENP-M ortholog , as recently proposed ( Shiroiwa et al . , 2011 ) , because all computational prediction tools we tested failed to identify a domain related to GTPases in Mis17 ( data not shown ) . It is possible , however , that Mis17 acts as a functional analogue of CENP-M . It is also legitimate to speculate that a functional GTPase , possibly a Rab-family GTPase given its evolutionary proximity to CENP-M , might take up the function of CENP-M in those organisms in which CENP-M cannot be identified . Modeling of CENP-I suggests that it adopts an α-solenoid fold analogous to that observed in β-importin ( Cingolani et al . , 1999; Vetter et al . , 1999 ) . An N-terminal domain of CENP-I ( residues 57–281 ) is sufficient to bind the CENP-H/K sub-complex , while the C-terminal half ( residues 282-C ) is necessary to bind CENP-M . Contiguity between CENP-H , CENP-I , and CENP-K had been previously hypothesized based on proteomic analysis of precipitates from cellular lysates , from the similarity of phenotypes caused by depletion of individual subunits , and from 2-hybrid interaction data ( Measday et al . , 2002; Okada et al . , 2006 ) . However , that the interactions among these subunits were direct , and that CENP-M was also part of the complex , had remained unclear . Thus , our reconstitution of a stable quaternary complex , the HIKM complex , significantly extends these previous analyses and identifies a new stable sub-complex of crucial importance for kinetochore stability . The HIKM complex flanks other previously recognized stable kinetochore sub-complexes , including the CENP-O/P/Q/U complex , the Mis12 complex , the Ndc80 complex , the CENP-L/N complex and the CENP-T/W/S/X complex ( Meraldi et al . , 2006; Hemmerich et al . , 2008; Perpelescu and Fukagawa , 2011; Westermann and Schleiffer , 2013; Westhorpe and Straight , 2013 ) . In this study , we have identified a direct interaction with the CENP-T/W/S/X complex . Our future studies will address whether HIKM forms direct interactions with other kinetochore sub-complexes or subunits . Also of interest is the role of the HIKM complex in recruiting the chromatin remodeling FACT complex , contributing to CENP-A deposition ( Obuse et al . , 2004; Foltz et al . , 2006; Okada et al . , 2009 ) , as well as in the recruitment of spindle checkpoint components ( Liu et al . , 2003 , 2006; Matson et al . , 2012 ) . CENP-T depletion hampers kinetochore localization of the subunits of the HIKM complex ( Foltz et al . , 2006; Hori et al . , 2008a; Gascoigne et al . , 2011 ) . However , we now show that the HIKM complex is mutually required for stable association of CENP-T/W with kinetochores . The identification of a direct interaction between the HIKM complex and the CENP-T/W complex may configure a mechanistic basis for this phenomenon . Thus , the sole interaction with centromeric chromatin ( ‘Introduction’ ) is not sufficient to recruit or retain CENP-T/W at kinetochores in the absence of the subunits of the HIKM complex . Collectively , our observations , together with ( a ) recent structural data on DNA-bound CENP-T/W/S/X ( Takeuchi et al . , 2014 ) , ( b ) previous observations that CENP-T and CENP-W undergo relatively rapid turnover times at centromeres ( Prendergast et al . , 2011 ) , and ( c ) previous observations that CENP-S/X occupies a rather peripheral position within kinetochores ( Amano et al . , 2009 ) suggest that the hypothesis that CENP-T/W/S/X are embedded in a nucleosome-like structure ( Nishino et al . , 2012 ) might require further scrutiny . Because CENP-C is required for recruitment of the HIKM complex ( Milks et al . , 2009; Carroll et al . , 2010; Gascoigne et al . , 2011 ) , and HIKM is in turn required for CENP-T recruitment ( this study ) , we tested and confirmed the prediction that CENP-T recruitment is also dependent on CENP-C . Conversely , disruption of the CENP-M/CENP-I interaction , or depletion of CENP-T , did not have major effects on the kinetochore localization of CENP-C , as previously suggested ( Goshima et al . , 2003; Liu et al . , 2006 ) . These findings unequivocally position CENP-T/W downstream of CENP-C , as previously proposed ( Carroll et al . , 2010 ) , and imply that the interaction of CENP-C with CENP-A , possibly together with the interaction of CENP-L/N with CENP-A , represents the apex of the CCAN recruitment pathway ( Figure 7E ) . Our future studies will aim to identify the molecular basis for this plan of kinetochore assembly . A cDNA segment encoding human CENP-M isoform 1 was subcloned in pGEX-6P-2rbs , a modified pGEX-6P vector ( GE Healthcare , Piscataway , NJ ) , as a 3′ fusion to the sequence encoding GST . The construct CENP-M1–171 was created by insertion of a stop codon with the QuikChange kit ( Agilent Technologies , Inc . , Santa Clara , CA ) . Constructs were sequence verified . The expression and purification procedure was the same for both CENP-M constructs . Escherichia coli C41 ( DE3 ) cells harbouring vectors expressing CENP-M or CENP-M1–171 were grown in Terrific Broth at 37°C to an OD600 of 0 . 6–0 . 8 , when 0 . 2 mM IPTG was added and the culture was grown at 18°C for ∼15 hr . Cell pellets were resuspended in lysis buffer ( 50 mM Tris/HCl pH 7 . 4 , 300 mM NaCl , 5% glycerol , 1 mM DTT ) supplemented with protease inhibitor cocktail ( Serva , Heidelberg , Germany ) , lysed by sonication and cleared by centrifugation at 48 , 000×g at 4°C for 1 hr . The cleared lysate was applied to Glutathione Sepharose 4 Fast Flow beads ( GE Healthcare ) pre-equilibrated in lysis buffer , was incubated at 4°C for 2 hr , washed with 70 vol of lysis buffer and subjected to an overnight cleavage reaction with 3C protease to separate CENP-M from GST . Resource S cation exchange column ( GE Healthcare ) was pre-equilibrated in 20 mM MES pH 6 . 0 , 50 mM NaCl , 5% glycerol , 1 mM DTT . The eluate from Glutathione beads was adjusted to a final salt concentration of 50 mM , loaded onto the Resource S column and eluted with a linear gradient of 50–500 mM NaCl in 10 bed column volumes . Fractions containing CENP-M were concentrated and loaded onto a Superdex75 SEC column ( GE Healthcare ) pre-equilibrated in SEC buffer ( 10 mM MES pH 6 . 0 , 150 mM NaCl , 1 mM TCEP ) . Fractions containing CENP-M were concentrated , flash-frozen in liquid nitrogen and stored at −80°C . CENP-M1–171 ( 10 mg/ml ) was crystallized by sitting drop vapor diffusion using a Honeybee Cartesian robot and 96-well plates . Diffraction-quality crystals were obtained by optimizing the initial conditions in hanging drops . The optimal reservoir buffer contained 100 mM bicine pH 8 . 5 , 11% MPD and 8 mM spermidine . Crystals were transferred to a cryobuffer containing the reservoir liquor supplemented with 15% glycerol and flash-frozen in liquid nitrogen . Selenomethionine ( SeMet ) derivatives were crystallized under similar conditions . X-ray diffraction data were collected with synchrotron radiation at beamline ID14-4 at the European Synchrotron Radiation Facility ( ESRF , Grenoble , France ) for the native crystal , and beamline X06DA ( PXIII ) , Swiss Light Source ( Villigen , Switzerland ) for the SeMet crystal . X-ray diffraction data were processed with xia2 ( version 0 . 3 . 3 . 1 ) ( Winter et al . , 2013 ) . Analysis of data quality and crystal defects was performed using phenix . xtriage ( Adams et al . , 2010 ) . Although the CENP-M crystals suffer from merohedral twinning , with a twinning fraction close to 50% , SAD phases obtained using Phenix AutoSol yielded an interpretable 2 Å experimental map . Model building was carried out in Coot ( Emsley et al . , 2010 ) , with the help of fragments built automatically by Phenix AutoBuild , ARP/wARP ( Morris et al . , 2004 ) and Buccaneer ( Cowtan , 2006 ) . The model was then used for molecular replacement into the native dataset using Phenix AutoMR . Iterative model building with Coot and refinement with phenix . refine yielded a final model with two molecules covering the full asymmetric unit . The Collaborative Computational Project 4 ( CCP4 ) suite ( Collaborative Computational Project , Number 4 , 1994 ) was also used at several stages . The structure was illustrated with PyMOL ( www . pymol . org ) . A cDNA segment encoding human CENP-K was subcloned in a MultiBac pFL-derived vector ( Fitzgerald et al . , 2006 ) , with an N-terminal TEV cleavable 6xHis tag , under the control of the polh promoter . A cDNA segment encoding human CENP-H was subcloned in pUCDM vector , without any tag , under the control of the p10 promoter . Constructs were sequence verified . The two vectors were then fused via in vitro Cre-loxP recombination . Baculovirus was then produced as described previously ( Trowitzsch et al . , 2010 ) , and amplified with three rounds of amplification . Expression of CENP-H/K complex was carried out in Tnao38 cells , using a virus: culture ratio of 1: 50 at 27°C for 72 hr . Cell pellets were resuspended in lysis buffer ( 50 mM Tris/HCl pH 8 . 0 , 300 mM NaCl , 20 mM imidazole , 5% glycerol , 2 mM β-mercaptoethanol ) supplemented with protease inhibitor cocktail ( Serva ) , lysed by sonication and cleared by centrifugation at 48 , 000×g at 4°C for 1 hr . The cleared lysate was applied to Ni-NTA Agarose beads ( Qiagen , Venlo , The Netherlands ) pre-equilibrated in lysis buffer , was incubated at 4°C for 2 hr and washed with 70 vol of lysis buffer . Bound proteins were eluted with lysis buffer supplemented with 200 mM imidazole and then dialysed against 50 mM Tris/HCl pH 8 . 0 , 150 mM NaCl , 5% glycerol , 0 . 5 mM EDTA , 1 mM DTT at 4°C overnight . During this dialysis step , tag cleavage with TEV protease was also performed . Resource Q anion exchange chromatography column ( GE Healthcare ) was pre-equilibrated in 50 mM Tris/HCl pH 8 . 0 , 75 mM NaCl , 5% glycerol , 0 . 5 mM EDTA , 1 mM DTT . The dialysed sample was adjusted to a salt concentration of 75 mM , loaded onto the Resource Q column and eluted with a linear gradient of 75–500 mM NaCl in 10 bed column volumes . Fractions containing CENP-H/K complex were concentrated and loaded onto a Superdex200 SEC column ( GE Healthcare ) pre-equilibrated in SEC buffer ( 10 mM HEPES pH 7 . 5 , 150 mM NaCl , 1 mM TCEP ) . Fractions containing CENP-H/K complex were concentrated , flash-frozen in liquid nitrogen and stored at −80°C . Codon optimised human CENP-I 57-756 was subcloned in a MultiBac pFL-derived vector ( Fitzgerald et al . , 2006 ) , with an N-terminal TEV cleavable 6xHis tag , under the control of the polh promoter . A cDNA segment encoding human CENP-M isoform 1 was subcloned in the 2nd MCS of the same vector , under the control of the p10 promoter . Simultaneously , a second pFL-based vector was created with untagged CENP-H and CENP-K under the control of the polh and p10 promoters , respectively . The CENP-I/M vector was then linearized with BstZ171 , and the expression region corresponding to CENP-H/K was PCR amplified with primers designed for sequence and ligation independent cloning ( SLIC ) of the PCR fragment into the linearized CENP-I/M vector . The SLIC reaction was then carried out to produce a single pFL-based vector with four expression cassettes . Constructs were sequence verified . Baculovirus was then produced as described previously ( Trowitzsch et al . , 2010 ) , and amplified with three rounds of amplification . Expression of CENP-HI57–CKM complex was carried out in TnAo38 cells , using a virus:culture ratio of 1:40 . Infected cells were incubated for 72 hr at 27°C . Cell pellets were harvested , washed in 1xPBS , and finally resuspended in a buffer containing 50 mM HEPES 7 . 5 , 300 mM NaCl , 1 mM MgCl2 , 10% glycerol , 5 mM imidazole , 2 mM BME , 0 . 1 mM AEBSF , and 2 . 5 units/ml Benzonase ( Millipore , Billerica , MA ) . Cells were lysed by sonication , and cleared for 1 hr at 100 , 000g . Cleared cell lysate was then run over a 5-ml Talon superflow column ( Clontech , part of Takara Bio group , Shiga , Japan ) and then washed with 50 mM HEPES 7 . 5 , 1 M NaCl , 10% glycerol , 5 mM imidazole 2 mM BME . CENP-HIKM complex was eluted with a gradient of 5–300 mM imidazole , and the fractions containing HIKM pooled , and the His tag cleaved overnight at 4°C . HIKM in solution was then adjusted to a salt concentration of 100 mM and a pH of 6 . 5 , prior to loading on a 6-ml Resource S ion-exchange column ( GE Healthcare ) , equilibrated in 20 mM MES 6 . 5 , 100 mM NaCl , 2 mM BME . CENP-HIKM was then eluted with a gradient of 100–1000 mM NaCl over 20 column volumes , and peak fractions corresponding to CENP-HIKM were pooled and concentrated in a 50 kDa MW Amicon concentrator ( Millipore ) . CENP-HIKM was then loaded onto a Superdex 200 16/600 ( GE healthcare ) in 20 mM HEPES 7 . 5 , 150 mM NaCl , 2 . 5% glycerol , 2 mM TCEP . Sample was concentrated and flash frozen in liquid N2 prior to use . A cDNA segment encoding human CENP-M isoform 1 was subcloned in a MultiBac pFL-derived vector , with an N-terminal TEV cleavable GST tag , under the control of the polh promoter . Codon optimised human CENP-I 57-756 was subcloned in the 2nd MCS of the same vector , under the control of the p10 promoter . Mutant CENP-M constructs were created by site-directed mutagenesis using the QuikChange kit ( Stratagene , La Jolla , CA ) . Constructs were sequence verified . Baculovirus was then produced and amplified with three rounds of amplification . The baculovirus encoding CENP-H/His-CENP-K complex , which has been detailed in the previous paragraph , was also employed . For each GST-pull-down experiment , 25 ml of freshly diluted Tnao38 cells at a density of 1 × 106 cells/ml in serum-free medium ( Sf-900 II SFM; Life Technologies , Carlsbad , CA ) were co-infected with GST-CENP-M/CENP-I57–C and CENP-H/His-CENP-K viruses using a virus: culture ratio of 1: 10 for each virus at 27°C for 72 hr . Cell pellets were resuspended in lysis buffer ( 20 mM HEPES pH 7 . 5 , 300 mM NaCl , 1 mM TCEP ) supplemented with protease inhibitor cocktail ( Serva ) , lysed by sonication and cleared by centrifugation at 20 , 000×g at 4°C for 30 min . The cleared lysate was applied to Glutathione Sepharose 4 Fast Flow beads ( GE Healthcare ) pre-equilibrated in lysis buffer , was incubated at 4°C for 2 hr , washed with 60 vol of lysis buffer and eluted with lysis buffer supplemented with 30 mM reduced Glutathione . Samples of total lysate , supernatant , beads before elution and elution were analysed by SDS-PAGE and Coomassie blue staining and by western blotting . The following antibodies were used: anti-CENP-M ( in house made rabbit polyclonal antibody SI0868 , raised against the full length protein; 1:1000 ) , anti-CENP-I ( rabbit polyclonal , ab28844; 1:100; Abcam ) , anti-CENP-H ( goat polyclonal , sc-11297; 1:200; Santa Cruz , Dallas , Texas ) , anti penta-His ( mouse monoclonal; 1:2000; Qiagen ) . N-methylanthraniloyl ( MANT ) -labeled nucleotides ( ADP , ATP , GDP , GTP ) ( Hiratsuka , 1983 ) , were purchased from Pharma Waldhof ( Düsseldorf , Germany ) . The fluorescence quantum yield of MANT nucleotides increases in nonpolar solvents and upon binding to proteins . Fluorescence data were recorded with a Fluoromax-4 spectrophotometer ( Jobin Yvon , Horiba , Kyoto , Japan ) , with excitation and emission wavelengths of MANT-nucleotides at 366 and 450 nm , respectively . Arl2 , a member of the Ras superfamily of small GTPases , was used as control ( kind gift of Mandy Miertzschke , Max Planck Institute of Dortmund , Germany ) . 500 µl of 1 . 0 µM MANT-labeled nucleotides in CENP-M SEC buffer were used . After 7 min , when the fluorescence baseline signal was stabilized , recombinant purified CENP-M or Arl2 ( at 10 µM ) was added and the fluorescence signal was monitored for 1 hr . For each experiment , the fluorescence signal was normalized to the fluorescence signal at time zero . Analytical SEC experiments were performed on calibrated Superdex200 5/150 or Superose6 5/150 columns ( GE Healthcare ) . All samples were eluted under isocratic conditions at 4°C in SEC buffer at a flow rate of 0 . 2 ml/min for Superdex200 5/150 or 0 . 1 ml/min for Superose6 5/150 . Elution of proteins was monitored at 280 nm . 100 µl fractions were collected and analysed by SDS-PAGE and Coomassie blue staining . To detect the formation of a complex , proteins were mixed at the indicated concentrations in 50 µl , incubated for at least 2 hr on ice and then subjected to SEC . For binding assays with nucleosomes , a SEC buffer containing 10 mM HEPES pH 7 . 5 , 50 mM NaCl , 1 mM TCEP was used . For the other binding assays , a SEC buffer containing 150 mM NaCl was used when possible ( namely , with CENP-M , CENP-H/K , His-CENP-I57–281 , CENP-H/K/I57–281 , Mis12 complex , Ndc80 complex , Knl12000–2311 , Zwint ) . A SEC buffer containing 300 mM NaCl was instead employed with proteins that were not stable in lower NaCl concentrations ( specifically , CENP-C constructs , CENP-T/W/S/X , CENP-L/N , CENP-O/P/Q/U , CENP-R ) . HeLa cells were grown in Dulbecco's Modified Eagle's Medium ( DMEM; PAN Biotech ) at 37°C in the presence of 5% CO2 and supplemented with 10% Fetal Bovine Serum ( FBS; Clontech ) , penicillin and streptomycin ( GIBCO , Carlsbad , CA ) . Parental Flp-In T-REx HeLa cells used to generate stable doxycycline-inducible cell lines were a gift from Stephen Taylor ( University of Manchester , Manchester , England , UK ) . Flp-In T-REx HeLa cells expressing CENP-M fusions to GFP were generated as previously described ( Tighe , 2004 ) and maintained in DMEM with 10% tetracycline-free FBS supplemented with 250 µg/ml hygromycin and 4 µg/ml blastidicin ( Invitrogen , Carlsbad , CA ) . GFP-CENP-M fusions were expressed by addition of 1 ng/ml or 50 ng/ml doxycycline ( Sigma , St . Louis , MO ) for 24 or 48 hr . For CENP-M silencing , we used a combination of three siRNA duplexes ( target sequences: 5′-ACAAAAGGUCUGUGGCUAA-3′; 5′-UUAAGCAGCUGGCGUGUUA-3′; 5′-GUGCUGACUCCAUAAACAU-3′; Thermo Scientific , Carlsbad , CA ) targeting the 3′-UTR of endogenous CENP-M . CENP-M siRNA duplexes were used at 20 nM each . For CENP-T silencing , we used a combination of two siRNA duplexes ( target sequences: 5′-GUGGAGAAGUGCCUAGAUA-3′ from AMBION , and 5′-AAGUAGAGCCCUUACACGA-3′ from Thermo Scientific ) at a concentration of 5 nM each . For CENP-C silencing , we used a single siRNA ( target sequence: 5′-GGAUCAUCUCAGAAUAGAA-3′ from AMBION , Austin , TX ) at a concentration of 7 . 5 nM . All transfections were performed with HyPerFect ( Qiagen ) according to the manufacturer's instructions . Phenotypes were analysed 66 hr ( for CENP-T and CENP-C depletions ) , and 72 or 96 hr ( for CENP-M depletions ) after siRNA addition and protein depletion was monitored by western blotting or immunofluorescence . Where indicated , nocodazole ( Sigma-Aldrich ) was used at 0 . 3 µM for 16 hr , RO-3306 ( Calbiochem , part of EMD Biosciences , Darmstadt , Germany ) was used at 9 µM for 18 hr and MG-132 ( Calbiochem ) at 5 µM for 3 hr . A cDNA segment encoding human CENP-M isoform 1 was subcloned in pcDNA5/FRT/TO-EGFP-IRES vector , a modified version of pcDNA5/FRT/TO vector ( Invitrogen , Carlsbad , CA ) generated in house ( Petrovic et al . , 2010 ) , as a C-terminal fusion to EGFP . Mutant CENP-M constructs were created by site-directed mutagenesis using the QuikChange kit ( Stratagene ) . CENP-M cDNA was also subcloned in pcDNA5/FRT/TO vector ( Invitrogen ) as an N-terminal fusion to EGFP . Constructs were sequence verified . To enrich cultures for mitotic cells , nocodazole was added to the cell culture media . Mitotic cells were then harvested by shake off and lysed by incubation in lysis buffer ( 75 mM HEPES pH 7 . 5 , 150 mM KCl , 1 . 5 mM EGTA , 1 . 5 mM MgCl2 , 10% glycerol , 0 . 075% NP-40 , 90 U/ml benzonase [Sigma] , protease inhibitor cocktail [Serva] and PhosSTOP phosphatase inhibitors [Roche , Basel , Switzerland] ) at 4°C for 15 min followed by sonication and centrifugation . Extracts were pre-cleared with a mixture of protein A-Sepharose ( CL-4B; GE Healthcare ) and protein G-Sepharose ( rec-Protein G-Sepharose 4B; Invitrogen ) at 4°C for 1 hr . Subsequently , extracts were incubated with GFP-Traps ( ChromoTek , Martinsried , Germany ) at 4°C for 2–4 hr . Immunoprecipitates were washed with lysis buffer , resuspended in Laemmli sample buffer , boiled and analyzed by western blotting using 4–12% or 4–20% gradient gels ( Life technologies ) . The following antibodies were used: anti-GFP ( in house made rabbit polyclonal antibody; 1:4000 ) , anti-Hec1 ( mouse monoclonal , clone 9G3 . 23; 1:1000; Gene-Tex , Irvine , CA ) , anti-Mis12 ( clone QA21; 1:1000; [Petrovic et al . , 2014] ) , anti-Knl1 ( in house made rabbit polyclonal antibody SI0787 , raised against amino acids 1-22; 1:1000 ) , anti-Vinculin ( mouse monoclonal , clone hVIN-1; 1:15000; Sigma-Aldrich ) , anti-CENP-M ( in house made rabbit polyclonal antibody SI0868 , raised against the full length protein; 1:500 ) , anti-CENP-I ( in house made rabbit polyclonal antibody SI0887 , raised against amino acids 57–281; 1:500 ) , anti-CENP-T/W ( in house made rabbit polyclonal antibody SI0882 , raised against the full length protein complex; 1:800 ) , anti-CENP-C ( rabbit polyclonal antibody SI410; 1:1200; [Trazzi et al . , 2009] ) . Secondary antibodies were affinity purified anti-mouse ( Amersham , part of GE Healthcare ) , anti-rabbit ( Amersham ) , anti-goat ( Santa Cruz ) conjugated to horseradish peroxidase ( 1:10000 ) . After incubation with ECL western blotting system ( GE Healthcare ) , images were acquired with ChemiBIS 3 . 2 ( DNR Bio-Imaging Systems , Jerusalem , Israel ) . Levels were adjusted with ImageJ and Photoshop and images were cropped accordingly . HeLa cells and Flp-In T-REx HeLa cells were grown on coverslips pre-coated with 15 µg/ml poly-D-Lysine ( Millipore ) and 0 . 01% poly-L-Lysine ( Sigma ) , respectively . Cells were either fixed with methanol and rehydrated with PBS or fixed with PBS/PHEM-paraformaldehyde 4% followed by permeabilisation with PBS/PHEM-Triton 0 . 3% . The following antibodies were used for immunostaining: anti-Nsl1 [clone QM9-13; 1:1000] , anti-Hec1 ( mouse monoclonal , clone 9G3 . 23; 1:1000; Gene-Tex ) , anti-Knl1 ( in house made rabbit polyclonal antibody SI0787 , raised against amino acids 1-22; 1:1000 ) , anti-CENP-M cross-linked to Alexa568 ( in house made affinity purified rabbit polyclonal antibody SI0868 , raised against the full length protein; 1:200 ) , anti-CENP-I ( 1:700; a kind gift from Song-Tao Liu , University of Toledo , Ohio , USA ) , anti-CENP-T/W ( in house made rabbit polyclonal antibody SI0882 , raised against the full length protein complex; 1:800 ) , CREST/anti-centromere antibodies ( 1:100; Antibodies Inc . , Davis , CA ) , anti-Tubulin ( mouse , 1:8000; T9026; Sigma ) . For CENP-C , either polyclonal anti-CENP-C ( in house made rabbit antibody SI410; 1:1200 ) or monoclonal anti-CENP-C ( mouse monoclonal , 2159C5a; 1:200; Abcam ) were used . Cy3-conjugated , RhodamineRed-X-conjugated , Cy5-conjugated , and DyLigth649-conjugated secondary antibodies were purchased from Jackson ImmunoResearch Laboratories , West Grove , PA . Alexa 488-labeled and 568-labeled secondary antibodies were from Invitrogen . DNA was stained with 0 . 5 µg/ml DAPI ( Serva ) and coverslips mounted with Mowiol mounting media ( Calbiochem ) or ProLong Gold Antifade reagent ( Life Technologies ) . All experiments were imaged at room temperature and , with the exception of Figure 6F , using the spinning disk confocal microscopy of a 3i Marianas system ( Intelligent Imaging Innovations , Denver , CO ) equipped with an Axio Observer Z1 microscope ( Zeiss , Oberkochen , Germany ) , a CSU-X1 confocal scanner unit ( Yokogawa Electric Corporation , Tokyo , Japan ) , Plan-Apochromat 63x or 100x/1 . 4NA objectives ( Zeiss ) and Orca Flash 4 . 0 sCMOS Camera ( Hamamatsu , Hamamatsu City , Japan ) . Data for Figure 6F were acquired using a confocal microscope ( model TCS SP2; Leica ) equipped with a 63x NA 1 . 4 objective lens . Images were acquired as 0 . 27-µm Z-sections ( using Slidebook Software 5 . 5 from Intelligent Imaging Innovations or using LCS 3D software from Leica , Solms , Germany ) and converted into maximal intensity projections TIFF files for illustrative purposes . Quantification of kinetochore signals was performed on unmodified Z-series images using Imaris 7 . 3 . 4 software ( Bitplane , Zurich , Switzerland ) . After background subtraction , all signals were normalized to CREST and values obtained for control cells were set to 1 . Quantifications are based on two or three independent experiments where a minimum of 10 cells and 300 kinetochores per condition were analyzed . Prior to EM experiments , the CENP-HI57–CKM complex was separated on a Superdex 200 10/300 SEC column ( GE Healthcare ) ( pre-equilibrated in 20 mM Tris/HCl pH 8 . 0 , 150 mM NaCl , 1 mM TCEP ) and diluted . 4 µl of the sample were adsorbed at 25°C for 40 s onto glow-discharged carbon-coated grids . The grids were washed twice with SEC buffer and negatively stained with 0 . 07% uranyl formate ( SPI supplies/Structure probe , West Chester , PA ) for about 120 s as described previously ( Bröcker et al . , 2012 ) . Samples were imaged with a JEOL1400 microscope equipped with a LaB6 cathode operated at 120 kV . Images were recorded at low-dose conditions ( 19 electrons/Å2 ) at a corrected magnification of 82553x on a 4k × 4k CMOS camera F416 ( TVIPS , Oslo , Norway ) . Single particles were manually selected , aligned , and classified using reference-free alignment and k-means classification procedures as well as the iterative stable alignment and clustering approach ( ISAC ) ( Yang et al . , 2012 ) implemented in SPARX ( Hohn et al . , 2007 ) and EMAN2 ( Ludtke , 2010 ) . The data set used for the analysis of the HIKM complex contained 5859 particles . For the 3D reconstruction of the CENP-HIKM complex an initial 3D model was calculated from the best ISAC class averages using the SHC approach implemented in SPARX ( Hohn et al . , 2007 ) . Then , using the complete set of 5859 raw particle images , the reconstruction was further refined by the iterative projection matching approach implemented in SPARX until convergence was achieved . The resolution of the final reconstruction was estimated by the 0 . 5 FSC criterion to be 22 Å ( Figure 4—figure supplement 3 , panel C ) . Chimera ( Pettersen et al . , 2004 ) was used for visualization , analysis , and preparation of EM figures . To search for CENP-M homologous sequences , we first conducted PSI-blast searches using HsCENP-M . This identified a distant CENP-M ortholog from Crassostrea gigas ( K1QTP9_CRAGI ) , whose sequence was then used for further iterative searches of the NCBI non-redundant database using hidden Markov models ( HMM ) ( Eddy , 2011 ) . At the third iteration , four sequences gave a significant match above threshold: a predicted protein from the sponge Amphimedon queenslandica ( gi\|340373098 , e-value 0 . 00094 ) , the GTP binding protein SAS1 from the fungus Lodderomyces elongisporus ( gi\|146447565 , e-value 0 . 0036 ) , a hypothetical protein from the amoeba Dictyostelum purpureum ( gi\|330794657 , e-value 0 . 0017 ) , and a Ras-related protein from the arthropod Aedes aegypti ( gi\|157117562 , e-value 0 . 0092 ) . We removed the redundancy of this alignment ( 50% ) to keep the most representative sequences from the profile . Only common regions to the retrieved and CENP-M proteins present in the alignment were extracted ( ∼140 amino acids ) . We then removed the four retrieved sequences from the alignment , built a new profile consisting only of 18 CENP-M sequences and searched the non-redundant database using hmmsearch ( Eddy , 2011 ) . The discarded sequences were again identified above threshold . Reciprocally , we then removed the CENP-M sequences from the alignment keeping only the four distant orthologs and generated a new HMM profile that was used to search the non-redundant database . The CENP-M protein from Crassostrea gigas ( gi\|405975648 , or K1QTP9_CRAGI ) was recovered at e-value 0 . 00034 . Additional Blast ( delta ) searches with this protein provided CENP-M proteins . Given the large divergence in sequence similarity and the inability to obtain reliable sequence alignments for the N-terminal part of the CENP-M sequences with bona fide small GTPases , we created a sequence alignment from structural alignments . Rab1A ( PDB 4FMC_D ) was the best template found using CE algorithm ( Shindyalov and Bourne , 1998 ) ( 3 . 6 Å root mean square deviation ) . The region spanning residues 66–156 of HsCENP-M displays the highest structural conservation between GTPases and CENP-M and identified reciprocal sequence similarities in iterative HMM searches . We derived a sequence alignment from the CENP-M/Rab1A structural alignment and built a profile from it . CENP-M , Rab1A , the identified GTPases , additional hits , and a selected group of sequences from the five classical GTPase families were aligned to the profile using hmmalign ( Eddy , 2011 ) . The alignment was visually inspected to detect inaccuracies and was used as input for phylogenetic inference using Maximum Likelihood ( Guindon et al . , 2009 ) . PhyML was run using mpi implementation , with SPR and 5 starting trees . All parameters were optimized and 1002 bootstrap replicates were obtained from the tree ( commands: mpirun -n 6 phyml-mpi -i 175toPhyML . phylip -b 1002 -d aa -m LG -f m -s SPR --rand_start --n_rand_starts 5 -v e -a e -o tlr ) . The tree was visualized using iTOL ( Letunic and Bork , 2007 ) and is depicted in Figure 3D . CENP-HI57–CKM complex was cross-linked with isotope-labeled disuccinimidyl suberate and digested with Lys-C and trypsin after quenching with ammonium bicarbonate . Cross-linked peptides were enriched using size-exclusion chromatography , analyzed by liquid chromatography coupled to tandem mass spectrometry and identified by the search algorithm , xQuest . Cross-linking , MS analysis and database searching were performed as described ( Herzog et al . , 2012 ) . Visualization of the crosslinks was done by converting the raw data ( in form of Excel spreadsheets ) to the GEXF data format ( Graph Exchange XML Format ) using custom shell scripts . The data were then imported into the Gephi software ( http://gephi . org ) that was modified to allow simultaneous calculation and display of curved and straight connectors ( i . e . , intra- and intermolecular crosslinks ) . The Gephi graph was exported as an Adobe Illustrator file for final processing . The final model of CENP-M has been submitted to the Protein Data Bank under the accession number 4P0T .	When a human cell divides to make new cells , its 46 chromosomes must be replicated and then separated evenly between the two daughter cells . The process of separation is performed by the spindle—a network of fibres that form inside the cell , attach to the chromosomes and pull the copies to the opposite ends of the cell . The spindle fibres attach to a structure called a kinetochore , which forms at a region of the chromosomes called the centromere . The kinetochore has a layered structure with multiple copies of many proteins , and the inner layer is composed of at least 16 centromeric proteins . These proteins interact directly with the centromere and influence the formation of the rest of the kinetochore and the spindle fibres . While some of the interactions between centromeric proteins have been uncovered , the roles of several of them—including one called CENP-M—remain unknown . Now , Basilico , Maffini , Weir et al . reveal that CENP-M is essential for assembling and stabilizing the inner layer of the kinetochore . However , while it is structurally and evolutionarily related to enzymes called GTPases , CENP-M is not an enzyme . Instead , the CENP-M protein interacts with three other centromeric proteins to form a complex that becomes part of the inner layer of the kinetochore . Basilico , Maffini , Weir et al . also find that another centromeric protein , CENP-C , appears to start the assembly of the inner layer . This protein then recruits two complexes made of other centromeric proteins to the kinetochore , including the complex that contains CENP-M . The next challenge will be to reconstitute larger protein complexes that contain more proteins from the inner layer of the kinetochore , so that these assemblies can be studied in greater detail . It will also be important to investigate how CENP-C acts as a scaffold to organize the interface between the kinetochore and the centromere .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2014	The pseudo GTPase CENP-M drives human kinetochore assembly

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.